Merge tag 'drm-next-2018-06-06-1' of git://anongit.freedesktop.org/drm/drm

Pull drm updates from Dave Airlie:
 "This starts to support NVIDIA volta hardware with nouveau, and adds
  amdgpu support for the GPU in the Kabylake-G (the intel + radeon
  single package chip), along with some initial Intel icelake enabling.

  Summary:

  New Drivers:
   - v3d - driver for broadcom V3D V3.x+ hardware
   - xen-front - XEN PV display frontend

  core:
   - handle zpos normalization in the core
   - stop looking at legacy pointers in atomic paths
   - improved scheduler documentation
   - improved aspect ratio validation
   - aspect ratio support for 64:27 and 256:135
   - drop unused control node code.

  i915:
   - Icelake (ICL) enabling
   - GuC/HuC refactoring
   - PSR/PSR2 enabling and fixes
   - DPLL management refactoring
   - DP MST fixes
   - NV12 enabling
   - HDCP improvements
   - GEM/Execlist/reset improvements
   - GVT improvements
   - stolen memory first 4k fix

  amdgpu:
   - Vega 20 support
   - VEGAM support (Kabylake-G)
   - preOS scanout buffer reservation
   - power management gfxoff support for raven
   - SR-IOV fixes
   - Vega10 power profiles and clock voltage control
   - scatter/gather display support on CZ/ST

  amdkfd:
   - GFX9 dGPU support
   - userptr memory mapping

  nouveau:
   - major refactoring for Volta GV100 support

  tda998x:
   - HDMI i2c CEC support

  etnaviv:
   - removed unused logging code
   - license text cleanups
   - MMU handling improvements
   - timeout fence fix for 50 days uptime

  tegra:
   - IOMMU support in gr2d/gr3d drivers
   - zpos support

  vc4:
   - syncobj support
   - CTM, plane alpha and async cursor support

  analogix_dp:
   - HPD and aux chan fixes

  sun4i:
   - MIPI DSI support

  tilcdc:
   - clock divider fixes for OMAP-l138 LCDK board

  rcar-du:
   - R8A77965 support
   - dma-buf fences fixes
   - hardware indexed crtc/du group handling
   - generic zplane property support

  atmel-hclcdc:
   - generic zplane property support

  mediatek:
   - use generic video mode function

  exynos:
   - S5PV210 FIMD variant support
   - IPP v2 framework
   - more HW overlays support"

* tag 'drm-next-2018-06-06-1' of git://anongit.freedesktop.org/drm/drm: (1286 commits)
  drm/amdgpu: fix 32-bit build warning
  drm/exynos: fimc: signedness bug in fimc_setup_clocks()
  drm/exynos: scaler: fix static checker warning
  drm/amdgpu: Use dev_info() to report amdkfd is not supported for this ASIC
  drm/amd/display: Remove use of division operator for long longs
  drm/amdgpu: Update GFX info structure to match what vega20 used
  drm/amdgpu/pp: remove duplicate assignment
  drm/sched: add rcu_barrier after entity fini
  drm/amdgpu: move VM BOs on LRU again
  drm/amdgpu: consistenly use VM moved flag
  drm/amdgpu: kmap PDs/PTs in amdgpu_vm_update_directories
  drm/amdgpu: further optimize amdgpu_vm_handle_moved
  drm/amdgpu: cleanup amdgpu_vm_validate_pt_bos v2
  drm/amdgpu: rework VM state machine lock handling v2
  drm/amdgpu: Add runtime VCN PG support
  drm/amdgpu: Enable VCN static PG by default on RV
  drm/amdgpu: Add VCN static PG support on RV
  drm/amdgpu: Enable VCN CG by default on RV
  drm/amdgpu: Add static CG control for VCN on RV
  drm/exynos: Fix default value for zpos plane property
  ...

Documentation/devicetree/bindings/display/bridge/adi,adv7511.txt

@@ -14,7 +14,13 @@ Required properties:
 		"adi,adv7513"
 		"adi,adv7533"
 
-- reg: I2C slave address
+- reg: I2C slave addresses
+  The ADV7511 internal registers are split into four pages exposed through
+  different I2C addresses, creating four register maps. Each map has it own
+  I2C address and acts as a standard slave device on the I2C bus. The main
+  address is mandatory, others are optional and revert to defaults if not
+  specified.
+
 
 The ADV7511 supports a large number of input data formats that differ by their
 color depth, color format, clock mode, bit justification and random
@@ -70,6 +76,9 @@ Optional properties:
   rather than generate its own timings for HDMI output.
 - clocks: from common clock binding: reference to the CEC clock.
 - clock-names: from common clock binding: must be "cec".
+- reg-names : Names of maps with programmable addresses.
+	It can contain any map needing a non-default address.
+	Possible maps names are : "main", "edid", "cec", "packet"
 
 Required nodes:
 
@@ -88,7 +97,12 @@ Example
 
 	adv7511w: hdmi@39 {
 		compatible = "adi,adv7511w";
-		reg = <39>;
+		/*
+		 * The EDID page will be accessible on address 0x66 on the I2C
+		 * bus. All other maps continue to use their default addresses.
+		 */
+		reg = <0x39>, <0x66>;
+		reg-names = "main", "edid";
 		interrupt-parent = <&gpio3>;
 		interrupts = <29 IRQ_TYPE_EDGE_FALLING>;
 		clocks = <&cec_clock>;

Documentation/devicetree/bindings/display/bridge/cdns,dsi.txt

@@ -0,0 +1,133 @@
+Cadence DSI bridge
+==================
+
+The Cadence DSI bridge is a DPI to DSI bridge supporting up to 4 DSI lanes.
+
+Required properties:
+- compatible: should be set to "cdns,dsi".
+- reg: physical base address and length of the controller's registers.
+- interrupts: interrupt line connected to the DSI bridge.
+- clocks: DSI bridge clocks.
+- clock-names: must contain "dsi_p_clk" and "dsi_sys_clk".
+- phys: phandle link to the MIPI D-PHY controller.
+- phy-names: must contain "dphy".
+- #address-cells: must be set to 1.
+- #size-cells: must be set to 0.
+
+Optional properties:
+- resets: DSI reset lines.
+- reset-names: can contain "dsi_p_rst".
+
+Required subnodes:
+- ports: Ports as described in Documentation/devicetree/bindings/graph.txt.
+  2 ports are available:
+  * port 0: this port is only needed if some of your DSI devices are
+	    controlled through  an external bus like I2C or SPI. Can have at
+	    most 4 endpoints. The endpoint number is directly encoding the
+	    DSI virtual channel used by this device.
+  * port 1: represents the DPI input.
+  Other ports will be added later to support the new kind of inputs.
+
+- one subnode per DSI device connected on the DSI bus. Each DSI device should
+  contain a reg property encoding its virtual channel.
+
+Cadence DPHY
+============
+
+Cadence DPHY block.
+
+Required properties:
+- compatible: should be set to "cdns,dphy".
+- reg: physical base address and length of the DPHY registers.
+- clocks: DPHY reference clocks.
+- clock-names: must contain "psm" and "pll_ref".
+- #phy-cells: must be set to 0.
+
+
+Example:
+	dphy0: dphy@fd0e0000{
+		compatible = "cdns,dphy";
+		reg = <0x0 0xfd0e0000 0x0 0x1000>;
+		clocks = <&psm_clk>, <&pll_ref_clk>;
+		clock-names = "psm", "pll_ref";
+		#phy-cells = <0>;
+	};
+
+	dsi0: dsi@fd0c0000 {
+		compatible = "cdns,dsi";
+		reg = <0x0 0xfd0c0000 0x0 0x1000>;
+		clocks = <&pclk>, <&sysclk>;
+		clock-names = "dsi_p_clk", "dsi_sys_clk";
+		interrupts = <1>;
+		phys = <&dphy0>;
+		phy-names = "dphy";
+		#address-cells = <1>;
+		#size-cells = <0>;
+
+		ports {
+			#address-cells = <1>;
+			#size-cells = <0>;
+
+			port@1 {
+				reg = <1>;
+				dsi0_dpi_input: endpoint {
+					remote-endpoint = <&xxx_dpi_output>;
+				};
+			};
+		};
+
+		panel: dsi-dev@0 {
+			compatible = "<vendor,panel>";
+			reg = <0>;
+		};
+	};
+
+or
+
+	dsi0: dsi@fd0c0000 {
+		compatible = "cdns,dsi";
+		reg = <0x0 0xfd0c0000 0x0 0x1000>;
+		clocks = <&pclk>, <&sysclk>;
+		clock-names = "dsi_p_clk", "dsi_sys_clk";
+		interrupts = <1>;
+		phys = <&dphy1>;
+		phy-names = "dphy";
+		#address-cells = <1>;
+		#size-cells = <0>;
+
+		ports {
+			#address-cells = <1>;
+			#size-cells = <0>;
+
+			port@0 {
+				reg = <0>;
+				#address-cells = <1>;
+				#size-cells = <0>;
+
+				dsi0_output: endpoint@0 {
+					reg = <0>;
+					remote-endpoint = <&dsi_panel_input>;
+				};
+			};
+
+			port@1 {
+				reg = <1>;
+				dsi0_dpi_input: endpoint {
+					remote-endpoint = <&xxx_dpi_output>;
+				};
+			};
+		};
+	};
+
+	i2c@xxx {
+		panel: panel@59 {
+			compatible = "<vendor,panel>";
+			reg = <0x59>;
+
+			port {
+				dsi_panel_input: endpoint {
+					remote-endpoint = <&dsi0_output>;
+				};
+			};
+		};
+	};

Documentation/devicetree/bindings/display/bridge/renesas,dw-hdmi.txt

@@ -14,6 +14,7 @@ Required properties:
 - compatible : Shall contain one or more of
   - "renesas,r8a7795-hdmi" for R8A7795 (R-Car H3) compatible HDMI TX
   - "renesas,r8a7796-hdmi" for R8A7796 (R-Car M3-W) compatible HDMI TX
+  - "renesas,r8a77965-hdmi" for R8A77965 (R-Car M3-N) compatible HDMI TX
   - "renesas,rcar-gen3-hdmi" for the generic R-Car Gen3 compatible HDMI TX
 
     When compatible with generic versions, nodes must list the SoC-specific

Documentation/devicetree/bindings/display/bridge/tda998x.txt

@@ -27,6 +27,9 @@ Optional properties:
 	in question is used. The implementation allows one or two DAIs. If two
 	DAIs are defined, they must be of different type.
 
+  - nxp,calib-gpios: calibration GPIO, which must correspond with the
+	gpio used for the TDA998x interrupt pin.
+
 [1] Documentation/sound/alsa/soc/DAI.txt
 [2] include/dt-bindings/display/tda998x.h
 

Documentation/devicetree/bindings/display/bridge/thine,thc63lvd1024.txt

@@ -0,0 +1,60 @@
+Thine Electronics THC63LVD1024 LVDS decoder
+-------------------------------------------
+
+The THC63LVD1024 is a dual link LVDS receiver designed to convert LVDS streams
+to parallel data outputs. The chip supports single/dual input/output modes,
+handling up to two LVDS input streams and up to two digital CMOS/TTL outputs.
+
+Single or dual operation mode, output data mapping and DDR output modes are
+configured through input signals and the chip does not expose any control bus.
+
+Required properties:
+- compatible: Shall be "thine,thc63lvd1024"
+- vcc-supply: Power supply for TTL output, TTL CLOCKOUT signal, LVDS input,
+  PPL and digital circuitry
+
+Optional properties:
+- powerdown-gpios: Power down GPIO signal, pin name "/PDWN". Active low
+- oe-gpios: Output enable GPIO signal, pin name "OE". Active high
+
+The THC63LVD1024 video port connections are modeled according
+to OF graph bindings specified by Documentation/devicetree/bindings/graph.txt
+
+Required video port nodes:
+- port@0: First LVDS input port
+- port@2: First digital CMOS/TTL parallel output
+
+Optional video port nodes:
+- port@1: Second LVDS input port
+- port@3: Second digital CMOS/TTL parallel output
+
+Example:
+--------
+
+	thc63lvd1024: lvds-decoder {
+		compatible = "thine,thc63lvd1024";
+
+		vcc-supply = <&reg_lvds_vcc>;
+		powerdown-gpios = <&gpio4 15 GPIO_ACTIVE_LOW>;
+
+		ports {
+			#address-cells = <1>;
+			#size-cells = <0>;
+
+			port@0 {
+				reg = <0>;
+
+				lvds_dec_in_0: endpoint {
+					remote-endpoint = <&lvds_out>;
+				};
+			};
+
+			port@2{
+				reg = <2>;
+
+				lvds_dec_out_2: endpoint {
+					remote-endpoint = <&adv7511_in>;
+				};
+			};
+		};
+	};

Documentation/devicetree/bindings/display/exynos/exynos5433-decon.txt

@@ -19,7 +19,8 @@ Required properties:
 	  clock-names property.
 - clock-names: list of clock names sorted in the same order as the clocks
 	       property. Must contain "pclk", "aclk_decon", "aclk_smmu_decon0x",
-	       "aclk_xiu_decon0x", "pclk_smmu_decon0x", clk_decon_vclk",
+	       "aclk_xiu_decon0x", "pclk_smmu_decon0x", "aclk_smmu_decon1x",
+	       "aclk_xiu_decon1x", "pclk_smmu_decon1x", clk_decon_vclk",
 	       "sclk_decon_eclk"
 - ports: contains a port which is connected to mic node. address-cells and
 	 size-cells must 1 and 0, respectively.
@@ -34,10 +35,14 @@ decon: decon@13800000 {
 	clocks = <&cmu_disp CLK_ACLK_DECON>, <&cmu_disp CLK_ACLK_SMMU_DECON0X>,
 		<&cmu_disp CLK_ACLK_XIU_DECON0X>,
 		<&cmu_disp CLK_PCLK_SMMU_DECON0X>,
+		<&cmu_disp CLK_ACLK_SMMU_DECON1X>,
+		<&cmu_disp CLK_ACLK_XIU_DECON1X>,
+		<&cmu_disp CLK_PCLK_SMMU_DECON1X>,
 		<&cmu_disp CLK_SCLK_DECON_VCLK>,
 		<&cmu_disp CLK_SCLK_DECON_ECLK>;
 	clock-names = "aclk_decon", "aclk_smmu_decon0x", "aclk_xiu_decon0x",
-		"pclk_smmu_decon0x", "sclk_decon_vclk", "sclk_decon_eclk";
+		"pclk_smmu_decon0x", "aclk_smmu_decon1x", "aclk_xiu_decon1x",
+		"pclk_smmu_decon1x", "sclk_decon_vclk", "sclk_decon_eclk";
 	interrupt-names = "vsync", "lcd_sys";
 	interrupts = <0 202 0>, <0 203 0>;
 

Documentation/devicetree/bindings/display/renesas,du.txt

@@ -13,6 +13,7 @@ Required Properties:
     - "renesas,du-r8a7794" for R8A7794 (R-Car E2) compatible DU
     - "renesas,du-r8a7795" for R8A7795 (R-Car H3) compatible DU
     - "renesas,du-r8a7796" for R8A7796 (R-Car M3-W) compatible DU
+    - "renesas,du-r8a77965" for R8A77965 (R-Car M3-N) compatible DU
     - "renesas,du-r8a77970" for R8A77970 (R-Car V3M) compatible DU
     - "renesas,du-r8a77995" for R8A77995 (R-Car D3) compatible DU
 
@@ -47,20 +48,21 @@ bindings specified in Documentation/devicetree/bindings/graph.txt.
 The following table lists for each supported model the port number
 corresponding to each DU output.
 
-                      Port0          Port1          Port2          Port3
+                        Port0          Port1          Port2          Port3
 -----------------------------------------------------------------------------
- R8A7743 (RZ/G1M)     DPAD 0         LVDS 0         -              -
- R8A7745 (RZ/G1E)     DPAD 0         DPAD 1         -              -
- R8A7779 (R-Car H1)   DPAD 0         DPAD 1         -              -
- R8A7790 (R-Car H2)   DPAD 0         LVDS 0         LVDS 1         -
- R8A7791 (R-Car M2-W) DPAD 0         LVDS 0         -              -
- R8A7792 (R-Car V2H)  DPAD 0         DPAD 1         -              -
- R8A7793 (R-Car M2-N) DPAD 0         LVDS 0         -              -
- R8A7794 (R-Car E2)   DPAD 0         DPAD 1         -              -
- R8A7795 (R-Car H3)   DPAD 0         HDMI 0         HDMI 1         LVDS 0
- R8A7796 (R-Car M3-W) DPAD 0         HDMI 0         LVDS 0         -
- R8A77970 (R-Car V3M) DPAD 0         LVDS 0         -              -
- R8A77995 (R-Car D3)  DPAD 0         LVDS 0         LVDS 1         -
+ R8A7743 (RZ/G1M)       DPAD 0         LVDS 0         -              -
+ R8A7745 (RZ/G1E)       DPAD 0         DPAD 1         -              -
+ R8A7779 (R-Car H1)     DPAD 0         DPAD 1         -              -
+ R8A7790 (R-Car H2)     DPAD 0         LVDS 0         LVDS 1         -
+ R8A7791 (R-Car M2-W)   DPAD 0         LVDS 0         -              -
+ R8A7792 (R-Car V2H)    DPAD 0         DPAD 1         -              -
+ R8A7793 (R-Car M2-N)   DPAD 0         LVDS 0         -              -
+ R8A7794 (R-Car E2)     DPAD 0         DPAD 1         -              -
+ R8A7795 (R-Car H3)     DPAD 0         HDMI 0         HDMI 1         LVDS 0
+ R8A7796 (R-Car M3-W)   DPAD 0         HDMI 0         LVDS 0         -
+ R8A77965 (R-Car M3-N)  DPAD 0         HDMI 0         LVDS 0         -
+ R8A77970 (R-Car V3M)   DPAD 0         LVDS 0         -              -
+ R8A77995 (R-Car D3)    DPAD 0         LVDS 0         LVDS 1         -
 
 
 Example: R8A7795 (R-Car H3) ES2.0 DU

Documentation/devicetree/bindings/display/sunxi/sun6i-dsi.txt

@@ -0,0 +1,93 @@
+Allwinner A31 DSI Encoder
+=========================
+
+The DSI pipeline consists of two separate blocks: the DSI controller
+itself, and its associated D-PHY.
+
+DSI Encoder
+-----------
+
+The DSI Encoder generates the DSI signal from the TCON's.
+
+Required properties:
+  - compatible: value must be one of:
+    * allwinner,sun6i-a31-mipi-dsi
+  - reg: base address and size of memory-mapped region
+  - interrupts: interrupt associated to this IP
+  - clocks: phandles to the clocks feeding the DSI encoder
+    * bus: the DSI interface clock
+    * mod: the DSI module clock
+  - clock-names: the clock names mentioned above
+  - phys: phandle to the D-PHY
+  - phy-names: must be "dphy"
+  - resets: phandle to the reset controller driving the encoder
+
+  - ports: A ports node with endpoint definitions as defined in
+    Documentation/devicetree/bindings/media/video-interfaces.txt. The
+    first port should be the input endpoint, usually coming from the
+    associated TCON.
+
+Any MIPI-DSI device attached to this should be described according to
+the bindings defined in ../mipi-dsi-bus.txt
+
+D-PHY
+-----
+
+Required properties:
+  - compatible: value must be one of:
+    * allwinner,sun6i-a31-mipi-dphy
+  - reg: base address and size of memory-mapped region
+  - clocks: phandles to the clocks feeding the DSI encoder
+    * bus: the DSI interface clock
+    * mod: the DSI module clock
+  - clock-names: the clock names mentioned above
+  - resets: phandle to the reset controller driving the encoder
+
+Example:
+
+dsi0: dsi@1ca0000 {
+	compatible = "allwinner,sun6i-a31-mipi-dsi";
+	reg = <0x01ca0000 0x1000>;
+	interrupts = <GIC_SPI 89 IRQ_TYPE_LEVEL_HIGH>;
+	clocks = <&ccu CLK_BUS_MIPI_DSI>,
+		 <&ccu CLK_DSI_SCLK>;
+	clock-names = "bus", "mod";
+	resets = <&ccu RST_BUS_MIPI_DSI>;
+	phys = <&dphy0>;
+	phy-names = "dphy";
+	#address-cells = <1>;
+	#size-cells = <0>;
+
+	panel@0 {
+		compatible = "bananapi,lhr050h41", "ilitek,ili9881c";
+		reg = <0>;
+		power-gpios = <&pio 1 7 GPIO_ACTIVE_HIGH>; /* PB07 */
+		reset-gpios = <&r_pio 0 5 GPIO_ACTIVE_LOW>; /* PL05 */
+		backlight = <&pwm_bl>;
+	};
+
+	ports {
+		#address-cells = <1>;
+		#size-cells = <0>;
+
+		port@0 {
+			#address-cells = <1>;
+			#size-cells = <0>;
+			reg = <0>;
+
+			dsi0_in_tcon0: endpoint {
+				remote-endpoint = <&tcon0_out_dsi0>;
+			};
+		};
+	};
+};
+
+dphy0: d-phy@1ca1000 {
+	compatible = "allwinner,sun6i-a31-mipi-dphy";
+	reg = <0x01ca1000 0x1000>;
+	clocks = <&ccu CLK_BUS_MIPI_DSI>,
+		 <&ccu CLK_DSI_DPHY>;
+	clock-names = "bus", "mod";
+	resets = <&ccu RST_BUS_MIPI_DSI>;
+	#phy-cells = <0>;
+};

Documentation/devicetree/bindings/gpu/brcm,bcm-v3d.txt

@@ -0,0 +1,28 @@
+Broadcom V3D GPU
+
+Only the Broadcom V3D 3.x and newer GPUs are covered by this binding.
+For V3D 2.x, see brcm,bcm-vc4.txt.
+
+Required properties:
+- compatible:	Should be "brcm,7268-v3d" or "brcm,7278-v3d"
+- reg:		Physical base addresses and lengths of the register areas
+- reg-names:	Names for the register areas.  The "hub", "bridge", and "core0"
+		  register areas are always required.  The "gca" register area
+		  is required if the GCA cache controller is present.
+- interrupts:	The interrupt numbers.  The first interrupt is for the hub,
+		  while the following interrupts are for the cores.
+		  See bindings/interrupt-controller/interrupts.txt
+
+Optional properties:
+- clocks:	The core clock the unit runs on
+
+v3d {
+	compatible = "brcm,7268-v3d";
+	reg = <0xf1204000 0x100>,
+	      <0xf1200000 0x4000>,
+	      <0xf1208000 0x4000>,
+	      <0xf1204100 0x100>;
+	reg-names = "bridge", "hub", "core0", "gca";
+	interrupts = <0 78 4>,
+		     <0 77 4>;
+};

Documentation/devicetree/bindings/gpu/samsung-scaler.txt

@@ -0,0 +1,27 @@
+* Samsung Exynos Image Scaler
+
+Required properties:
+  - compatible : value should be one of the following:
+	(a) "samsung,exynos5420-scaler" for Scaler IP in Exynos5420
+	(b) "samsung,exynos5433-scaler" for Scaler IP in Exynos5433
+
+  - reg : Physical base address of the IP registers and length of memory
+	  mapped region.
+
+  - interrupts : Interrupt specifier for scaler interrupt, according to format
+		 specific to interrupt parent.
+
+  - clocks : Clock specifier for scaler clock, according to generic clock
+	     bindings. (See Documentation/devicetree/bindings/clock/exynos*.txt)
+
+  - clock-names : Names of clocks. For exynos scaler, it should be "mscl"
+		  on 5420 and "pclk", "aclk" and "aclk_xiu" on 5433.
+
+Example:
+	scaler@12800000 {
+		compatible = "samsung,exynos5420-scaler";
+		reg = <0x12800000 0x1294>;
+		interrupts = <0 220 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&clock CLK_MSCL0>;
+		clock-names = "mscl";
+	};

Documentation/gpu/drivers.rst

@@ -10,8 +10,10 @@ GPU Driver Documentation
    tegra
    tinydrm
    tve200
+   v3d
    vc4
    bridge/dw-hdmi
+   xen-front
 
 .. only::  subproject and html
 

Documentation/gpu/i915.rst

@@ -58,6 +58,12 @@ Intel GVT-g Host Support(vGPU device model)
 .. kernel-doc:: drivers/gpu/drm/i915/intel_gvt.c
    :internal:
 
+Workarounds
+-----------
+
+.. kernel-doc:: drivers/gpu/drm/i915/intel_workarounds.c
+   :doc: Hardware workarounds
+
 Display Hardware Handling
 =========================
 
@@ -249,6 +255,103 @@ Memory Management and Command Submission
 This sections covers all things related to the GEM implementation in the
 i915 driver.
 
+Intel GPU Basics
+----------------
+
+An Intel GPU has multiple engines. There are several engine types.
+
+- RCS engine is for rendering 3D and performing compute, this is named
+  `I915_EXEC_RENDER` in user space.
+- BCS is a blitting (copy) engine, this is named `I915_EXEC_BLT` in user
+  space.
+- VCS is a video encode and decode engine, this is named `I915_EXEC_BSD`
+  in user space
+- VECS is video enhancement engine, this is named `I915_EXEC_VEBOX` in user
+  space.
+- The enumeration `I915_EXEC_DEFAULT` does not refer to specific engine;
+  instead it is to be used by user space to specify a default rendering
+  engine (for 3D) that may or may not be the same as RCS.
+
+The Intel GPU family is a family of integrated GPU's using Unified
+Memory Access. For having the GPU "do work", user space will feed the
+GPU batch buffers via one of the ioctls `DRM_IOCTL_I915_GEM_EXECBUFFER2`
+or `DRM_IOCTL_I915_GEM_EXECBUFFER2_WR`. Most such batchbuffers will
+instruct the GPU to perform work (for example rendering) and that work
+needs memory from which to read and memory to which to write. All memory
+is encapsulated within GEM buffer objects (usually created with the ioctl
+`DRM_IOCTL_I915_GEM_CREATE`). An ioctl providing a batchbuffer for the GPU
+to create will also list all GEM buffer objects that the batchbuffer reads
+and/or writes. For implementation details of memory management see
+`GEM BO Management Implementation Details`_.
+
+The i915 driver allows user space to create a context via the ioctl
+`DRM_IOCTL_I915_GEM_CONTEXT_CREATE` which is identified by a 32-bit
+integer. Such a context should be viewed by user-space as -loosely-
+analogous to the idea of a CPU process of an operating system. The i915
+driver guarantees that commands issued to a fixed context are to be
+executed so that writes of a previously issued command are seen by
+reads of following commands. Actions issued between different contexts
+(even if from the same file descriptor) are NOT given that guarantee
+and the only way to synchronize across contexts (even from the same
+file descriptor) is through the use of fences. At least as far back as
+Gen4, also have that a context carries with it a GPU HW context;
+the HW context is essentially (most of atleast) the state of a GPU.
+In addition to the ordering guarantees, the kernel will restore GPU
+state via HW context when commands are issued to a context, this saves
+user space the need to restore (most of atleast) the GPU state at the
+start of each batchbuffer. The non-deprecated ioctls to submit batchbuffer
+work can pass that ID (in the lower bits of drm_i915_gem_execbuffer2::rsvd1)
+to identify what context to use with the command.
+
+The GPU has its own memory management and address space. The kernel
+driver maintains the memory translation table for the GPU. For older
+GPUs (i.e. those before Gen8), there is a single global such translation
+table, a global Graphics Translation Table (GTT). For newer generation
+GPUs each context has its own translation table, called Per-Process
+Graphics Translation Table (PPGTT). Of important note, is that although
+PPGTT is named per-process it is actually per context. When user space
+submits a batchbuffer, the kernel walks the list of GEM buffer objects
+used by the batchbuffer and guarantees that not only is the memory of
+each such GEM buffer object resident but it is also present in the
+(PP)GTT. If the GEM buffer object is not yet placed in the (PP)GTT,
+then it is given an address. Two consequences of this are: the kernel
+needs to edit the batchbuffer submitted to write the correct value of
+the GPU address when a GEM BO is assigned a GPU address and the kernel
+might evict a different GEM BO from the (PP)GTT to make address room
+for another GEM BO. Consequently, the ioctls submitting a batchbuffer
+for execution also include a list of all locations within buffers that
+refer to GPU-addresses so that the kernel can edit the buffer correctly.
+This process is dubbed relocation.
+
+GEM BO Management Implementation Details
+----------------------------------------
+
+.. kernel-doc:: drivers/gpu/drm/i915/i915_vma.h
+   :doc: Virtual Memory Address
+
+Buffer Object Eviction
+----------------------
+
+This section documents the interface functions for evicting buffer
+objects to make space available in the virtual gpu address spaces. Note
+that this is mostly orthogonal to shrinking buffer objects caches, which
+has the goal to make main memory (shared with the gpu through the
+unified memory architecture) available.
+
+.. kernel-doc:: drivers/gpu/drm/i915/i915_gem_evict.c
+   :internal:
+
+Buffer Object Memory Shrinking
+------------------------------
+
+This section documents the interface function for shrinking memory usage
+of buffer object caches. Shrinking is used to make main memory
+available. Note that this is mostly orthogonal to evicting buffer
+objects, which has the goal to make space in gpu virtual address spaces.
+
+.. kernel-doc:: drivers/gpu/drm/i915/i915_gem_shrinker.c
+   :internal:
+
 Batchbuffer Parsing
 -------------------
 
@@ -267,6 +370,12 @@ Batchbuffer Pools
 .. kernel-doc:: drivers/gpu/drm/i915/i915_gem_batch_pool.c
    :internal:
 
+User Batchbuffer Execution
+--------------------------
+
+.. kernel-doc:: drivers/gpu/drm/i915/i915_gem_execbuffer.c
+   :doc: User command execution
+
 Logical Rings, Logical Ring Contexts and Execlists
 --------------------------------------------------
 
@@ -312,28 +421,14 @@ Object Tiling IOCTLs
 .. kernel-doc:: drivers/gpu/drm/i915/i915_gem_tiling.c
    :doc: buffer object tiling
 
-Buffer Object Eviction
-----------------------
-
-This section documents the interface functions for evicting buffer
-objects to make space available in the virtual gpu address spaces. Note
-that this is mostly orthogonal to shrinking buffer objects caches, which
-has the goal to make main memory (shared with the gpu through the
-unified memory architecture) available.
-
-.. kernel-doc:: drivers/gpu/drm/i915/i915_gem_evict.c
-   :internal:
-
-Buffer Object Memory Shrinking
-------------------------------
+WOPCM
+=====
 
-This section documents the interface function for shrinking memory usage
-of buffer object caches. Shrinking is used to make main memory
-available. Note that this is mostly orthogonal to evicting buffer
-objects, which has the goal to make space in gpu virtual address spaces.
+WOPCM Layout
+------------
 
-.. kernel-doc:: drivers/gpu/drm/i915/i915_gem_shrinker.c
-   :internal:
+.. kernel-doc:: drivers/gpu/drm/i915/intel_wopcm.c
+   :doc: WOPCM Layout
 
 GuC
 ===
@@ -359,6 +454,12 @@ GuC Firmware Layout
 .. kernel-doc:: drivers/gpu/drm/i915/intel_guc_fwif.h
    :doc: GuC Firmware Layout
 
+GuC Address Space
+-----------------
+
+.. kernel-doc:: drivers/gpu/drm/i915/intel_guc.c
+   :doc: GuC Address Space
+
 Tracing
 =======
 

Documentation/gpu/kms-properties.csv

@@ -98,5 +98,4 @@ radeon,DVI-I,“coherent”,RANGE,"Min=0, Max=1",Connector,TBD
 ,,"""underscan vborder""",RANGE,"Min=0, Max=128",Connector,TBD
 ,Audio,“audio”,ENUM,"{ ""off"", ""on"", ""auto"" }",Connector,TBD
 ,FMT Dithering,“dither”,ENUM,"{ ""off"", ""on"" }",Connector,TBD
-rcar-du,Generic,"""alpha""",RANGE,"Min=0, Max=255",Plane,TBD
 ,,"""colorkey""",RANGE,"Min=0, Max=0x01ffffff",Plane,TBD

Documentation/gpu/todo.rst

@@ -212,6 +212,24 @@ probably use drm_fb_helper_fbdev_teardown().
 
 Contact: Maintainer of the driver you plan to convert
 
+Clean up mmap forwarding
+------------------------
+
+A lot of drivers forward gem mmap calls to dma-buf mmap for imported buffers.
+And also a lot of them forward dma-buf mmap to the gem mmap implementations.
+Would be great to refactor this all into a set of small common helpers.
+
+Contact: Daniel Vetter
+
+Put a reservation_object into drm_gem_object
+--------------------------------------------
+
+This would remove the need for the ->gem_prime_res_obj callback. It would also
+allow us to implement generic helpers for waiting for a bo, allowing for quite a
+bit of refactoring in the various wait ioctl implementations.
+
+Contact: Daniel Vetter
+
 idr_init_base()
 ---------------
 

Documentation/gpu/xen-front.rst

@@ -0,0 +1,31 @@
+====================================================
+ drm/xen-front Xen para-virtualized frontend driver
+====================================================
+
+This frontend driver implements Xen para-virtualized display
+according to the display protocol described at
+include/xen/interface/io/displif.h
+
+Driver modes of operation in terms of display buffers used
+==========================================================
+
+.. kernel-doc:: drivers/gpu/drm/xen/xen_drm_front.h
+   :doc: Driver modes of operation in terms of display buffers used
+
+Buffers allocated by the frontend driver
+----------------------------------------
+
+.. kernel-doc:: drivers/gpu/drm/xen/xen_drm_front.h
+   :doc: Buffers allocated by the frontend driver
+
+Buffers allocated by the backend
+--------------------------------
+
+.. kernel-doc:: drivers/gpu/drm/xen/xen_drm_front.h
+   :doc: Buffers allocated by the backend
+
+Driver limitations
+==================
+
+.. kernel-doc:: drivers/gpu/drm/xen/xen_drm_front.h
+   :doc: Driver limitations

MAINTAINERS

@@ -767,12 +767,14 @@ F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
 F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
 F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
 F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
 F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c
 F:	drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
 F:	drivers/gpu/drm/amd/amdkfd/
 F:	drivers/gpu/drm/amd/include/cik_structs.h
 F:	drivers/gpu/drm/amd/include/kgd_kfd_interface.h
 F:	drivers/gpu/drm/amd/include/vi_structs.h
+F:	drivers/gpu/drm/amd/include/v9_structs.h
 F:	include/uapi/linux/kfd_ioctl.h
 
 AMD SEATTLE DEVICE TREE SUPPORT
@@ -4685,7 +4687,7 @@ F:	Documentation/devicetree/bindings/display/exynos/
 
 DRM DRIVERS FOR FREESCALE DCU
 M:	Stefan Agner <stefan@agner.ch>
-M:	Alison Wang <alison.wang@freescale.com>
+M:	Alison Wang <alison.wang@nxp.com>
 L:	dri-devel@lists.freedesktop.org
 S:	Supported
 F:	drivers/gpu/drm/fsl-dcu/
@@ -4796,6 +4798,14 @@ S:	Maintained
 F:	drivers/gpu/drm/omapdrm/
 F:	Documentation/devicetree/bindings/display/ti/
 
+DRM DRIVERS FOR V3D
+M:	Eric Anholt <eric@anholt.net>
+S:	Supported
+F:	drivers/gpu/drm/v3d/
+F:	include/uapi/drm/v3d_drm.h
+F:	Documentation/devicetree/bindings/display/brcm,bcm-v3d.txt
+T:	git git://anongit.freedesktop.org/drm/drm-misc
+
 DRM DRIVERS FOR VC4
 M:	Eric Anholt <eric@anholt.net>
 T:	git git://github.com/anholt/linux
@@ -4842,6 +4852,15 @@ S:	Maintained
 F:	drivers/gpu/drm/tinydrm/
 F:	include/drm/tinydrm/
 
+DRM DRIVERS FOR XEN
+M:	Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
+T:	git git://anongit.freedesktop.org/drm/drm-misc
+L:	dri-devel@lists.freedesktop.org
+L:	xen-devel@lists.xen.org
+S:	Supported
+F:	drivers/gpu/drm/xen/
+F:	Documentation/gpu/xen-front.rst
+
 DRM TTM SUBSYSTEM
 M:	Christian Koenig <christian.koenig@amd.com>
 M:	Roger He <Hongbo.He@amd.com>

drivers/dma-buf/sync_debug.h

@@ -62,8 +62,6 @@ struct sync_pt {
 	struct rb_node node;
 };
 
-#ifdef CONFIG_SW_SYNC
-
 extern const struct file_operations sw_sync_debugfs_fops;
 
 void sync_timeline_debug_add(struct sync_timeline *obj);
@@ -72,12 +70,4 @@ void sync_file_debug_add(struct sync_file *fence);
 void sync_file_debug_remove(struct sync_file *fence);
 void sync_dump(void);
 
-#else
-# define sync_timeline_debug_add(obj)
-# define sync_timeline_debug_remove(obj)
-# define sync_file_debug_add(fence)
-# define sync_file_debug_remove(fence)
-# define sync_dump()
-#endif
-
 #endif /* _LINUX_SYNC_H */

drivers/gpu/drm/Kconfig

@@ -49,16 +49,17 @@ config DRM_DEBUG_MM
 
 	  If in doubt, say "N".
 
-config DRM_DEBUG_MM_SELFTEST
-	tristate "kselftests for DRM range manager (struct drm_mm)"
+config DRM_DEBUG_SELFTEST
+	tristate "kselftests for DRM"
 	depends on DRM
 	depends on DEBUG_KERNEL
 	select PRIME_NUMBERS
 	select DRM_LIB_RANDOM
+	select DRM_KMS_HELPER
 	default n
 	help
-	  This option provides a kernel module that can be used to test
-	  the DRM range manager (drm_mm) and its API. This option is not
+	  This option provides kernel modules that can be used to run
+	  various selftests on parts of the DRM api. This option is not
 	  useful for distributions or general kernels, but only for kernel
 	  developers working on DRM and associated drivers.
 
@@ -267,6 +268,8 @@ source "drivers/gpu/drm/amd/amdkfd/Kconfig"
 
 source "drivers/gpu/drm/imx/Kconfig"
 
+source "drivers/gpu/drm/v3d/Kconfig"
+
 source "drivers/gpu/drm/vc4/Kconfig"
 
 source "drivers/gpu/drm/etnaviv/Kconfig"
@@ -289,6 +292,8 @@ source "drivers/gpu/drm/pl111/Kconfig"
 
 source "drivers/gpu/drm/tve200/Kconfig"
 
+source "drivers/gpu/drm/xen/Kconfig"
+
 # Keep legacy drivers last
 
 menuconfig DRM_LEGACY

drivers/gpu/drm/Makefile

@@ -43,7 +43,7 @@ drm_kms_helper-$(CONFIG_DRM_KMS_CMA_HELPER) += drm_fb_cma_helper.o
 drm_kms_helper-$(CONFIG_DRM_DP_AUX_CHARDEV) += drm_dp_aux_dev.o
 
 obj-$(CONFIG_DRM_KMS_HELPER) += drm_kms_helper.o
-obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/
+obj-$(CONFIG_DRM_DEBUG_SELFTEST) += selftests/
 
 obj-$(CONFIG_DRM)	+= drm.o
 obj-$(CONFIG_DRM_MIPI_DSI) += drm_mipi_dsi.o
@@ -61,6 +61,7 @@ obj-$(CONFIG_DRM_MGA)	+= mga/
 obj-$(CONFIG_DRM_I810)	+= i810/
 obj-$(CONFIG_DRM_I915)	+= i915/
 obj-$(CONFIG_DRM_MGAG200) += mgag200/
+obj-$(CONFIG_DRM_V3D)  += v3d/
 obj-$(CONFIG_DRM_VC4)  += vc4/
 obj-$(CONFIG_DRM_CIRRUS_QEMU) += cirrus/
 obj-$(CONFIG_DRM_SIS)   += sis/
@@ -103,3 +104,4 @@ obj-$(CONFIG_DRM_MXSFB)	+= mxsfb/
 obj-$(CONFIG_DRM_TINYDRM) += tinydrm/
 obj-$(CONFIG_DRM_PL111) += pl111/
 obj-$(CONFIG_DRM_TVE200) += tve200/
+obj-$(CONFIG_DRM_XEN) += xen/

drivers/gpu/drm/amd/amdgpu/Makefile

@@ -56,13 +56,18 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 
 # add asic specific block
 amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
-	ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o \
-	amdgpu_amdkfd_gfx_v7.o
+	ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o
 
 amdgpu-$(CONFIG_DRM_AMDGPU_SI)+= si.o gmc_v6_0.o gfx_v6_0.o si_ih.o si_dma.o dce_v6_0.o si_dpm.o si_smc.o
 
 amdgpu-y += \
-	vi.o mxgpu_vi.o nbio_v6_1.o soc15.o emu_soc.o mxgpu_ai.o nbio_v7_0.o vega10_reg_init.o
+	vi.o mxgpu_vi.o nbio_v6_1.o soc15.o emu_soc.o mxgpu_ai.o nbio_v7_0.o vega10_reg_init.o \
+	vega20_reg_init.o
+
+# add DF block
+amdgpu-y += \
+	df_v1_7.o \
+	df_v3_6.o
 
 # add GMC block
 amdgpu-y += \
@@ -126,11 +131,20 @@ amdgpu-y += \
 	vcn_v1_0.o
 
 # add amdkfd interfaces
+amdgpu-y += amdgpu_amdkfd.o
+
+ifneq ($(CONFIG_HSA_AMD),)
 amdgpu-y += \
-	 amdgpu_amdkfd.o \
 	 amdgpu_amdkfd_fence.o \
 	 amdgpu_amdkfd_gpuvm.o \
-	 amdgpu_amdkfd_gfx_v8.o
+	 amdgpu_amdkfd_gfx_v8.o \
+	 amdgpu_amdkfd_gfx_v9.o
+
+ifneq ($(CONFIG_DRM_AMDGPU_CIK),)
+amdgpu-y += amdgpu_amdkfd_gfx_v7.o
+endif
+
+endif
 
 # add cgs
 amdgpu-y += amdgpu_cgs.o

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -129,6 +129,7 @@ extern int amdgpu_lbpw;
 extern int amdgpu_compute_multipipe;
 extern int amdgpu_gpu_recovery;
 extern int amdgpu_emu_mode;
+extern uint amdgpu_smu_memory_pool_size;
 
 #ifdef CONFIG_DRM_AMDGPU_SI
 extern int amdgpu_si_support;
@@ -137,6 +138,7 @@ extern int amdgpu_si_support;
 extern int amdgpu_cik_support;
 #endif
 
+#define AMDGPU_SG_THRESHOLD			(256*1024*1024)
 #define AMDGPU_DEFAULT_GTT_SIZE_MB		3072ULL /* 3GB by default */
 #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS	        3000
 #define AMDGPU_MAX_USEC_TIMEOUT			100000	/* 100 ms */
@@ -222,10 +224,10 @@ enum amdgpu_kiq_irq {
 	AMDGPU_CP_KIQ_IRQ_LAST
 };
 
-int amdgpu_device_ip_set_clockgating_state(struct amdgpu_device *adev,
+int amdgpu_device_ip_set_clockgating_state(void *dev,
 					   enum amd_ip_block_type block_type,
 					   enum amd_clockgating_state state);
-int amdgpu_device_ip_set_powergating_state(struct amdgpu_device *adev,
+int amdgpu_device_ip_set_powergating_state(void *dev,
 					   enum amd_ip_block_type block_type,
 					   enum amd_powergating_state state);
 void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
@@ -681,6 +683,8 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned ring_id);
 
 void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
+void amdgpu_ctx_mgr_entity_cleanup(struct amdgpu_ctx_mgr *mgr);
+void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr);
 void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
 
 
@@ -771,9 +775,18 @@ struct amdgpu_rlc {
 	u32 starting_offsets_start;
 	u32 reg_list_format_size_bytes;
 	u32 reg_list_size_bytes;
+	u32 reg_list_format_direct_reg_list_length;
+	u32 save_restore_list_cntl_size_bytes;
+	u32 save_restore_list_gpm_size_bytes;
+	u32 save_restore_list_srm_size_bytes;
 
 	u32 *register_list_format;
 	u32 *register_restore;
+	u8 *save_restore_list_cntl;
+	u8 *save_restore_list_gpm;
+	u8 *save_restore_list_srm;
+
+	bool is_rlc_v2_1;
 };
 
 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
@@ -867,6 +880,8 @@ struct amdgpu_gfx_config {
 
 	/* gfx configure feature */
 	uint32_t double_offchip_lds_buf;
+	/* cached value of DB_DEBUG2 */
+	uint32_t db_debug2;
 };
 
 struct amdgpu_cu_info {
@@ -938,6 +953,12 @@ struct amdgpu_gfx {
 	uint32_t			ce_feature_version;
 	uint32_t			pfp_feature_version;
 	uint32_t			rlc_feature_version;
+	uint32_t			rlc_srlc_fw_version;
+	uint32_t			rlc_srlc_feature_version;
+	uint32_t			rlc_srlg_fw_version;
+	uint32_t			rlc_srlg_feature_version;
+	uint32_t			rlc_srls_fw_version;
+	uint32_t			rlc_srls_feature_version;
 	uint32_t			mec_feature_version;
 	uint32_t			mec2_feature_version;
 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
@@ -1204,6 +1225,8 @@ struct amdgpu_asic_funcs {
 	/* invalidate hdp read cache */
 	void (*invalidate_hdp)(struct amdgpu_device *adev,
 			       struct amdgpu_ring *ring);
+	/* check if the asic needs a full reset of if soft reset will work */
+	bool (*need_full_reset)(struct amdgpu_device *adev);
 };
 
 /*
@@ -1368,7 +1391,19 @@ struct amdgpu_nbio_funcs {
 	void (*detect_hw_virt)(struct amdgpu_device *adev);
 };
 
-
+struct amdgpu_df_funcs {
+	void (*init)(struct amdgpu_device *adev);
+	void (*enable_broadcast_mode)(struct amdgpu_device *adev,
+				      bool enable);
+	u32 (*get_fb_channel_number)(struct amdgpu_device *adev);
+	u32 (*get_hbm_channel_number)(struct amdgpu_device *adev);
+	void (*update_medium_grain_clock_gating)(struct amdgpu_device *adev,
+						 bool enable);
+	void (*get_clockgating_state)(struct amdgpu_device *adev,
+				      u32 *flags);
+	void (*enable_ecc_force_par_wr_rmw)(struct amdgpu_device *adev,
+					    bool enable);
+};
 /* Define the HW IP blocks will be used in driver , add more if necessary */
 enum amd_hw_ip_block_type {
 	GC_HWIP = 1,
@@ -1398,6 +1433,7 @@ enum amd_hw_ip_block_type {
 struct amd_powerplay {
 	void *pp_handle;
 	const struct amd_pm_funcs *pp_funcs;
+	uint32_t pp_feature;
 };
 
 #define AMDGPU_RESET_MAGIC_NUM 64
@@ -1590,6 +1626,7 @@ struct amdgpu_device {
 	uint32_t 		*reg_offset[MAX_HWIP][HWIP_MAX_INSTANCE];
 
 	const struct amdgpu_nbio_funcs	*nbio_funcs;
+	const struct amdgpu_df_funcs	*df_funcs;
 
 	/* delayed work_func for deferring clockgating during resume */
 	struct delayed_work     late_init_work;
@@ -1764,6 +1801,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_asic_get_config_memsize(adev) (adev)->asic_funcs->get_config_memsize((adev))
 #define amdgpu_asic_flush_hdp(adev, r) (adev)->asic_funcs->flush_hdp((adev), (r))
 #define amdgpu_asic_invalidate_hdp(adev, r) (adev)->asic_funcs->invalidate_hdp((adev), (r))
+#define amdgpu_asic_need_full_reset(adev) (adev)->asic_funcs->need_full_reset((adev))
 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid))
 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
@@ -1790,6 +1828,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d))
 #define amdgpu_ring_emit_wreg(r, d, v) (r)->funcs->emit_wreg((r), (d), (v))
 #define amdgpu_ring_emit_reg_wait(r, d, v, m) (r)->funcs->emit_reg_wait((r), (d), (v), (m))
+#define amdgpu_ring_emit_reg_write_reg_wait(r, d0, d1, v, m) (r)->funcs->emit_reg_write_reg_wait((r), (d0), (d1), (v), (m))
 #define amdgpu_ring_emit_tmz(r, b) (r)->funcs->emit_tmz((r), (b))
 #define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib)))
 #define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r))

drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c

@@ -290,12 +290,11 @@ static int acp_hw_init(void *handle)
 	else if (r)
 		return r;
 
-	r = cgs_get_pci_resource(adev->acp.cgs_device, CGS_RESOURCE_TYPE_MMIO,
-			0x5289, 0, &acp_base);
-	if (r == -ENODEV)
-		return 0;
-	else if (r)
-		return r;
+	if (adev->rmmio_size == 0 || adev->rmmio_size < 0x5289)
+		return -EINVAL;
+
+	acp_base = adev->rmmio_base;
+
 	if (adev->asic_type != CHIP_STONEY) {
 		adev->acp.acp_genpd = kzalloc(sizeof(struct acp_pm_domain), GFP_KERNEL);
 		if (adev->acp.acp_genpd == NULL)

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c

@@ -50,15 +50,21 @@ int amdgpu_amdkfd_init(void)
 		kgd2kfd = NULL;
 	}
 
+
 #elif defined(CONFIG_HSA_AMD)
+
 	ret = kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd);
 	if (ret)
 		kgd2kfd = NULL;
 
 #else
+	kgd2kfd = NULL;
 	ret = -ENOENT;
 #endif
+
+#if defined(CONFIG_HSA_AMD_MODULE) || defined(CONFIG_HSA_AMD)
 	amdgpu_amdkfd_gpuvm_init_mem_limits();
+#endif
 
 	return ret;
 }
@@ -92,8 +98,12 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev)
 	case CHIP_POLARIS11:
 		kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
 		break;
+	case CHIP_VEGA10:
+	case CHIP_RAVEN:
+		kfd2kgd = amdgpu_amdkfd_gfx_9_0_get_functions();
+		break;
 	default:
-		dev_dbg(adev->dev, "kfd not supported on this ASIC\n");
+		dev_info(adev->dev, "kfd not supported on this ASIC\n");
 		return;
 	}
 
@@ -175,6 +185,28 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
 				&gpu_resources.doorbell_physical_address,
 				&gpu_resources.doorbell_aperture_size,
 				&gpu_resources.doorbell_start_offset);
+		if (adev->asic_type >= CHIP_VEGA10) {
+			/* On SOC15 the BIF is involved in routing
+			 * doorbells using the low 12 bits of the
+			 * address. Communicate the assignments to
+			 * KFD. KFD uses two doorbell pages per
+			 * process in case of 64-bit doorbells so we
+			 * can use each doorbell assignment twice.
+			 */
+			gpu_resources.sdma_doorbell[0][0] =
+				AMDGPU_DOORBELL64_sDMA_ENGINE0;
+			gpu_resources.sdma_doorbell[0][1] =
+				AMDGPU_DOORBELL64_sDMA_ENGINE0 + 0x200;
+			gpu_resources.sdma_doorbell[1][0] =
+				AMDGPU_DOORBELL64_sDMA_ENGINE1;
+			gpu_resources.sdma_doorbell[1][1] =
+				AMDGPU_DOORBELL64_sDMA_ENGINE1 + 0x200;
+			/* Doorbells 0x0f0-0ff and 0x2f0-2ff are reserved for
+			 * SDMA, IH and VCN. So don't use them for the CP.
+			 */
+			gpu_resources.reserved_doorbell_mask = 0x1f0;
+			gpu_resources.reserved_doorbell_val  = 0x0f0;
+		}
 
 		kgd2kfd->device_init(adev->kfd, &gpu_resources);
 	}
@@ -217,13 +249,19 @@ int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
 	struct amdgpu_bo *bo = NULL;
+	struct amdgpu_bo_param bp;
 	int r;
 	uint64_t gpu_addr_tmp = 0;
 	void *cpu_ptr_tmp = NULL;
 
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
-			     AMDGPU_GEM_CREATE_CPU_GTT_USWC, ttm_bo_type_kernel,
-			     NULL, &bo);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	r = amdgpu_bo_create(adev, &bp, &bo);
 	if (r) {
 		dev_err(adev->dev,
 			"failed to allocate BO for amdkfd (%d)\n", r);
@@ -432,3 +470,44 @@ bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
 
 	return false;
 }
+
+#if !defined(CONFIG_HSA_AMD_MODULE) && !defined(CONFIG_HSA_AMD)
+bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm)
+{
+	return false;
+}
+
+void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo)
+{
+}
+
+void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
+					struct amdgpu_vm *vm)
+{
+}
+
+struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f)
+{
+	return NULL;
+}
+
+int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem, struct mm_struct *mm)
+{
+	return 0;
+}
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void)
+{
+	return NULL;
+}
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void)
+{
+	return NULL;
+}
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_9_0_get_functions(void)
+{
+	return NULL;
+}
+#endif

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h

@@ -28,6 +28,7 @@
 #include <linux/types.h>
 #include <linux/mm.h>
 #include <linux/mmu_context.h>
+#include <linux/workqueue.h>
 #include <kgd_kfd_interface.h>
 #include <drm/ttm/ttm_execbuf_util.h>
 #include "amdgpu_sync.h"
@@ -59,7 +60,9 @@ struct kgd_mem {
 
 	uint32_t mapping_flags;
 
+	atomic_t invalid;
 	struct amdkfd_process_info *process_info;
+	struct page **user_pages;
 
 	struct amdgpu_sync sync;
 
@@ -84,6 +87,9 @@ struct amdkfd_process_info {
 	struct list_head vm_list_head;
 	/* List head for all KFD BOs that belong to a KFD process. */
 	struct list_head kfd_bo_list;
+	/* List of userptr BOs that are valid or invalid */
+	struct list_head userptr_valid_list;
+	struct list_head userptr_inval_list;
 	/* Lock to protect kfd_bo_list */
 	struct mutex lock;
 
@@ -91,6 +97,11 @@ struct amdkfd_process_info {
 	unsigned int n_vms;
 	/* Eviction Fence */
 	struct amdgpu_amdkfd_fence *eviction_fence;
+
+	/* MMU-notifier related fields */
+	atomic_t evicted_bos;
+	struct delayed_work restore_userptr_work;
+	struct pid *pid;
 };
 
 int amdgpu_amdkfd_init(void);
@@ -104,12 +115,14 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);
 void amdgpu_amdkfd_device_init(struct amdgpu_device *adev);
 void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev);
 
+int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem, struct mm_struct *mm);
 int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
 				uint32_t vmid, uint64_t gpu_addr,
 				uint32_t *ib_cmd, uint32_t ib_len);
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_9_0_get_functions(void);
 
 bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
 
@@ -143,14 +156,14 @@ uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd);
 
 /* GPUVM API */
 int amdgpu_amdkfd_gpuvm_create_process_vm(struct kgd_dev *kgd, void **vm,
-					  void **process_info,
-					  struct dma_fence **ef);
+					void **process_info,
+					struct dma_fence **ef);
 int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd,
-					   struct file *filp,
-					   void **vm, void **process_info,
-					   struct dma_fence **ef);
+					struct file *filp,
+					void **vm, void **process_info,
+					struct dma_fence **ef);
 void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
-				    struct amdgpu_vm *vm);
+				struct amdgpu_vm *vm);
 void amdgpu_amdkfd_gpuvm_destroy_process_vm(struct kgd_dev *kgd, void *vm);
 uint32_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm);
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c

@@ -98,8 +98,6 @@ static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
 static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 					unsigned int vmid);
 
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr);
 static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
 static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 			uint32_t queue_id, uint32_t __user *wptr,
@@ -183,7 +181,6 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.free_pasid = amdgpu_pasid_free,
 	.program_sh_mem_settings = kgd_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
-	.init_pipeline = kgd_init_pipeline,
 	.init_interrupts = kgd_init_interrupts,
 	.hqd_load = kgd_hqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
@@ -309,13 +306,6 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 	return 0;
 }
 
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr)
-{
-	/* amdgpu owns the per-pipe state */
-	return 0;
-}
-
 static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c

@@ -57,8 +57,6 @@ static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
 		uint32_t sh_mem_bases);
 static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 		unsigned int vmid);
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-		uint32_t hpd_size, uint64_t hpd_gpu_addr);
 static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
 static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 			uint32_t queue_id, uint32_t __user *wptr,
@@ -141,7 +139,6 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.free_pasid = amdgpu_pasid_free,
 	.program_sh_mem_settings = kgd_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
-	.init_pipeline = kgd_init_pipeline,
 	.init_interrupts = kgd_init_interrupts,
 	.hqd_load = kgd_hqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
@@ -270,13 +267,6 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 	return 0;
 }
 
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr)
-{
-	/* amdgpu owns the per-pipe state */
-	return 0;
-}
-
 static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c

@@ -0,0 +1,1043 @@
+/*
+ * Copyright 2014-2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#define pr_fmt(fmt) "kfd2kgd: " fmt
+
+#include <linux/module.h>
+#include <linux/fdtable.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_ucode.h"
+#include "soc15_hw_ip.h"
+#include "gc/gc_9_0_offset.h"
+#include "gc/gc_9_0_sh_mask.h"
+#include "vega10_enum.h"
+#include "sdma0/sdma0_4_0_offset.h"
+#include "sdma0/sdma0_4_0_sh_mask.h"
+#include "sdma1/sdma1_4_0_offset.h"
+#include "sdma1/sdma1_4_0_sh_mask.h"
+#include "athub/athub_1_0_offset.h"
+#include "athub/athub_1_0_sh_mask.h"
+#include "oss/osssys_4_0_offset.h"
+#include "oss/osssys_4_0_sh_mask.h"
+#include "soc15_common.h"
+#include "v9_structs.h"
+#include "soc15.h"
+#include "soc15d.h"
+
+/* HACK: MMHUB and GC both have VM-related register with the same
+ * names but different offsets. Define the MMHUB register we need here
+ * with a prefix. A proper solution would be to move the functions
+ * programming these registers into gfx_v9_0.c and mmhub_v1_0.c
+ * respectively.
+ */
+#define mmMMHUB_VM_INVALIDATE_ENG16_REQ				0x06f3
+#define mmMMHUB_VM_INVALIDATE_ENG16_REQ_BASE_IDX		0
+
+#define mmMMHUB_VM_INVALIDATE_ENG16_ACK				0x0705
+#define mmMMHUB_VM_INVALIDATE_ENG16_ACK_BASE_IDX		0
+
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32		0x072b
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32_BASE_IDX	0
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32		0x072c
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32_BASE_IDX	0
+
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32		0x074b
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32_BASE_IDX	0
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32		0x074c
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32_BASE_IDX	0
+
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32		0x076b
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32_BASE_IDX	0
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32		0x076c
+#define mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32_BASE_IDX	0
+
+#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32		0x0727
+#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32_BASE_IDX	0
+#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32		0x0728
+#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32_BASE_IDX	0
+
+#define V9_PIPE_PER_MEC		(4)
+#define V9_QUEUES_PER_PIPE_MEC	(8)
+
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES
+};
+
+/*
+ * Register access functions
+ */
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t sh_mem_config,
+		uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
+		uint32_t sh_mem_bases);
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+		unsigned int vmid);
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr,
+			uint32_t wptr_shift, uint32_t wptr_mask,
+			struct mm_struct *mm);
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm);
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs);
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+		uint32_t pipe_id, uint32_t queue_id);
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id);
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int utimeout);
+static int kgd_address_watch_disable(struct kgd_dev *kgd);
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo);
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd);
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset);
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+		uint8_t vmid);
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+		uint8_t vmid);
+static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t page_table_base);
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+static void set_scratch_backing_va(struct kgd_dev *kgd,
+					uint64_t va, uint32_t vmid);
+static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
+static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
+
+/* Because of REG_GET_FIELD() being used, we put this function in the
+ * asic specific file.
+ */
+static int amdgpu_amdkfd_get_tile_config(struct kgd_dev *kgd,
+		struct tile_config *config)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+
+	config->gb_addr_config = adev->gfx.config.gb_addr_config;
+
+	config->tile_config_ptr = adev->gfx.config.tile_mode_array;
+	config->num_tile_configs =
+			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
+	config->macro_tile_config_ptr =
+			adev->gfx.config.macrotile_mode_array;
+	config->num_macro_tile_configs =
+			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
+
+	return 0;
+}
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_gtt_mem_allocation = alloc_gtt_mem,
+	.free_gtt_mem = free_gtt_mem,
+	.get_local_mem_info = get_local_mem_info,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.alloc_pasid = amdgpu_pasid_alloc,
+	.free_pasid = amdgpu_pasid_free,
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.address_watch_disable = kgd_address_watch_disable,
+	.address_watch_execute = kgd_address_watch_execute,
+	.wave_control_execute = kgd_wave_control_execute,
+	.address_watch_get_offset = kgd_address_watch_get_offset,
+	.get_atc_vmid_pasid_mapping_pasid =
+			get_atc_vmid_pasid_mapping_pasid,
+	.get_atc_vmid_pasid_mapping_valid =
+			get_atc_vmid_pasid_mapping_valid,
+	.get_fw_version = get_fw_version,
+	.set_scratch_backing_va = set_scratch_backing_va,
+	.get_tile_config = amdgpu_amdkfd_get_tile_config,
+	.get_cu_info = get_cu_info,
+	.get_vram_usage = amdgpu_amdkfd_get_vram_usage,
+	.create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm,
+	.acquire_process_vm = amdgpu_amdkfd_gpuvm_acquire_process_vm,
+	.destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm,
+	.get_process_page_dir = amdgpu_amdkfd_gpuvm_get_process_page_dir,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base,
+	.alloc_memory_of_gpu = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu,
+	.free_memory_of_gpu = amdgpu_amdkfd_gpuvm_free_memory_of_gpu,
+	.map_memory_to_gpu = amdgpu_amdkfd_gpuvm_map_memory_to_gpu,
+	.unmap_memory_to_gpu = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu,
+	.sync_memory = amdgpu_amdkfd_gpuvm_sync_memory,
+	.map_gtt_bo_to_kernel = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel,
+	.restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos,
+	.invalidate_tlbs = invalidate_tlbs,
+	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
+	.submit_ib = amdgpu_amdkfd_submit_ib,
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_9_0_get_functions(void)
+{
+	return (struct kfd2kgd_calls *)&kfd2kgd;
+}
+
+static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
+{
+	return (struct amdgpu_device *)kgd;
+}
+
+static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	mutex_lock(&adev->srbm_mutex);
+	soc15_grbm_select(adev, mec, pipe, queue, vmid);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	soc15_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+static uint32_t get_queue_mask(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id)
+{
+	unsigned int bit = (pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			    queue_id) & 31;
+
+	return ((uint32_t)1) << bit;
+}
+
+static void release_queue(struct kgd_dev *kgd)
+{
+	unlock_srbm(kgd);
+}
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	lock_srbm(kgd, 0, 0, 0, vmid);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG), sh_mem_config);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_BASES), sh_mem_bases);
+	/* APE1 no longer exists on GFX9 */
+
+	unlock_srbm(kgd);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished
+	 * and the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	/*
+	 * need to do this twice, once for gfx and once for mmhub
+	 * for ATC add 16 to VMID for mmhub, for IH different registers.
+	 * ATC_VMID0..15 registers are separate from ATC_VMID16..31.
+	 */
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + vmid,
+	       pasid_mapping);
+
+	while (!(RREG32(SOC15_REG_OFFSET(
+				ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+		 (1U << vmid)))
+		cpu_relax();
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+	       1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT) + vmid,
+	       pasid_mapping);
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID16_PASID_MAPPING) + vmid,
+	       pasid_mapping);
+
+	while (!(RREG32(SOC15_REG_OFFSET(
+				ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+		 (1U << (vmid + 16))))
+		cpu_relax();
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+	       1U << (vmid + 16));
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT_MM) + vmid,
+	       pasid_mapping);
+	return 0;
+}
+
+/* TODO - RING0 form of field is obsolete, seems to date back to SI
+ * but still works
+ */
+
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCPC_INT_CNTL),
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static uint32_t get_sdma_base_addr(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t base[2] = {
+		SOC15_REG_OFFSET(SDMA0, 0,
+				 mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL,
+		SOC15_REG_OFFSET(SDMA1, 0,
+				 mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL
+	};
+	uint32_t retval;
+
+	retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL -
+					       mmSDMA0_RLC0_RB_CNTL);
+
+	pr_debug("sdma base address: 0x%x\n", retval);
+
+	return retval;
+}
+
+static inline struct v9_mqd *get_mqd(void *mqd)
+{
+	return (struct v9_mqd *)mqd;
+}
+
+static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v9_sdma_mqd *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr,
+			uint32_t wptr_shift, uint32_t wptr_mask,
+			struct mm_struct *mm)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct v9_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, data;
+
+	m = get_mqd(mqd);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	/* HIQ is set during driver init period with vmid set to 0*/
+	if (m->cp_hqd_vmid == 0) {
+		uint32_t value, mec, pipe;
+
+		mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+		pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+		pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+			mec, pipe, queue_id);
+		value = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS));
+		value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
+			((mec << 5) | (pipe << 3) | queue_id | 0x80));
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS), value);
+	}
+
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+
+	for (reg = hqd_base;
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		WREG32(reg, mqd_hqd[reg - hqd_base]);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL), data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_LO),
+		       lower_32_bits(guessed_wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI),
+		       upper_32_bits(guessed_wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR),
+		       lower_32_bits((uintptr_t)wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI),
+		       upper_32_bits((uintptr_t)wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1),
+		       get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_EOP_RPTR),
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr,
+			     CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE), data);
+
+	release_queue(kgd);
+
+	return 0;
+}
+
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	for (reg = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	release_queue(kgd);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_base_addr, sdmax_gfx_context_cntl;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+	sdmax_gfx_context_cntl = m->sdma_engine_id ?
+		SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_GFX_CONTEXT_CNTL) :
+		SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_GFX_CONTEXT_CNTL);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies))
+			return -ETIME;
+		usleep_range(500, 1000);
+	}
+	data = RREG32(sdmax_gfx_context_cntl);
+	data = REG_SET_FIELD(data, SDMA0_GFX_CONTEXT_CNTL,
+			     RESUME_CTX, 0);
+	WREG32(sdmax_gfx_context_cntl, data);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t sdma_base_addr = get_sdma_base_addr(adev, engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+10)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_base_addr + reg);
+	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_base_addr + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
+	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_base_addr + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
+	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_base_addr + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	act = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE));
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_BASE)) &&
+		   high == RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_BASE_HI)))
+			retval = true;
+	}
+	release_queue(kgd);
+	return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	enum hqd_dequeue_request_type type;
+	unsigned long end_jiffies;
+	uint32_t temp;
+	struct v9_mqd *m = get_mqd(mqd);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD15(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0);
+
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_DEQUEUE_REQUEST), type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
+	while (true) {
+		temp = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE));
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue preemption time out.\n");
+			release_queue(kgd);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	release_queue(kgd);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int utimeout)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_base_addr;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies))
+			return -ETIME;
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+							uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+								uint8_t vmid)
+{
+	uint32_t reg;
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	reg = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	uint32_t req = (1 << vmid) |
+		(0 << VM_INVALIDATE_ENG16_REQ__FLUSH_TYPE__SHIFT) | /* legacy */
+		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PTES_MASK |
+		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE0_MASK |
+		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE1_MASK |
+		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE2_MASK |
+		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L1_PTES_MASK;
+
+	mutex_lock(&adev->srbm_mutex);
+
+	/* Use legacy mode tlb invalidation.
+	 *
+	 * Currently on Raven the code below is broken for anything but
+	 * legacy mode due to a MMHUB power gating problem. A workaround
+	 * is for MMHUB to wait until the condition PER_VMID_INVALIDATE_REQ
+	 * == PER_VMID_INVALIDATE_ACK instead of simply waiting for the ack
+	 * bit.
+	 *
+	 * TODO 1: agree on the right set of invalidation registers for
+	 * KFD use. Use the last one for now. Invalidate both GC and
+	 * MMHUB.
+	 *
+	 * TODO 2: support range-based invalidation, requires kfg2kgd
+	 * interface change
+	 */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
+				0xffffffff);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
+				0x0000001f);
+
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
+				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
+				0xffffffff);
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
+				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
+				0x0000001f);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_REQ), req);
+
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_INVALIDATE_ENG16_REQ),
+				req);
+
+	while (!(RREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ACK)) &
+					(1 << vmid)))
+		cpu_relax();
+
+	while (!(RREG32(SOC15_REG_OFFSET(MMHUB, 0,
+					mmMMHUB_VM_INVALIDATE_ENG16_ACK)) &
+					(1 << vmid)))
+		cpu_relax();
+
+	mutex_unlock(&adev->srbm_mutex);
+
+}
+
+static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid)
+{
+	signed long r;
+	uint32_t seq;
+	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
+
+	spin_lock(&adev->gfx.kiq.ring_lock);
+	amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/
+	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
+	amdgpu_ring_write(ring,
+			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
+			PACKET3_INVALIDATE_TLBS_ALL_HUB(1) |
+			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
+			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(0)); /* legacy */
+	amdgpu_fence_emit_polling(ring, &seq);
+	amdgpu_ring_commit(ring);
+	spin_unlock(&adev->gfx.kiq.ring_lock);
+
+	r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
+	if (r < 1) {
+		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
+		return -ETIME;
+	}
+
+	return 0;
+}
+
+static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	int vmid;
+	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
+
+	if (ring->ready)
+		return invalidate_tlbs_with_kiq(adev, pasid);
+
+	for (vmid = 0; vmid < 16; vmid++) {
+		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
+			continue;
+		if (get_atc_vmid_pasid_mapping_valid(kgd, vmid)) {
+			if (get_atc_vmid_pasid_mapping_pasid(kgd, vmid)
+				== pasid) {
+				write_vmid_invalidate_request(kgd, vmid);
+				break;
+			}
+		}
+	}
+
+	return 0;
+}
+
+static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("non kfd vmid %d\n", vmid);
+		return 0;
+	}
+
+	write_vmid_invalidate_request(kgd, vmid);
+	return 0;
+}
+
+static int kgd_address_watch_disable(struct kgd_dev *kgd)
+{
+	return 0;
+}
+
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					uint32_t cntl_val,
+					uint32_t addr_hi,
+					uint32_t addr_lo)
+{
+	return 0;
+}
+
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX), gfx_index_val);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CMD), sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SH_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX), data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+					unsigned int watch_point_id,
+					unsigned int reg_offset)
+{
+	return 0;
+}
+
+static void set_scratch_backing_va(struct kgd_dev *kgd,
+					uint64_t va, uint32_t vmid)
+{
+	/* No longer needed on GFXv9. The scratch base address is
+	 * passed to the shader by the CP. It's the user mode driver's
+	 * responsibility.
+	 */
+}
+
+/* FIXME: Does this need to be ASIC-specific code? */
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	const union amdgpu_firmware_header *hdr;
+
+	switch (type) {
+	case KGD_ENGINE_PFP:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.pfp_fw->data;
+		break;
+
+	case KGD_ENGINE_ME:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.me_fw->data;
+		break;
+
+	case KGD_ENGINE_CE:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.ce_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC1:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.mec_fw->data;
+		break;
+
+	case KGD_ENGINE_MEC2:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.mec2_fw->data;
+		break;
+
+	case KGD_ENGINE_RLC:
+		hdr = (const union amdgpu_firmware_header *)adev->gfx.rlc_fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA1:
+		hdr = (const union amdgpu_firmware_header *)adev->sdma.instance[0].fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA2:
+		hdr = (const union amdgpu_firmware_header *)adev->sdma.instance[1].fw->data;
+		break;
+
+	default:
+		return 0;
+	}
+
+	if (hdr == NULL)
+		return 0;
+
+	/* Only 12 bit in use*/
+	return hdr->common.ucode_version;
+}
+
+static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t page_table_base)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint64_t base = (uint64_t)page_table_base << PAGE_SHIFT |
+		AMDGPU_PTE_VALID;
+
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID %u\n",
+		       vmid);
+		return;
+	}
+
+	/* TODO: take advantage of per-process address space size. For
+	 * now, all processes share the same address space size, like
+	 * on GFX8 and older.
+	 */
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32) + (vmid*2), 0);
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32) + (vmid*2), 0);
+
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32) + (vmid*2),
+			lower_32_bits(adev->vm_manager.max_pfn - 1));
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32) + (vmid*2),
+			upper_32_bits(adev->vm_manager.max_pfn - 1));
+
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32) + (vmid*2), lower_32_bits(base));
+	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32) + (vmid*2), upper_32_bits(base));
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32) + (vmid*2), 0);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32) + (vmid*2), 0);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32) + (vmid*2),
+			lower_32_bits(adev->vm_manager.max_pfn - 1));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32) + (vmid*2),
+			upper_32_bits(adev->vm_manager.max_pfn - 1));
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32) + (vmid*2), lower_32_bits(base));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32) + (vmid*2), upper_32_bits(base));
+}

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -23,6 +23,8 @@
 #define pr_fmt(fmt) "kfd2kgd: " fmt
 
 #include <linux/list.h>
+#include <linux/pagemap.h>
+#include <linux/sched/mm.h>
 #include <drm/drmP.h>
 #include "amdgpu_object.h"
 #include "amdgpu_vm.h"
@@ -33,10 +35,20 @@
  */
 #define VI_BO_SIZE_ALIGN (0x8000)
 
+/* BO flag to indicate a KFD userptr BO */
+#define AMDGPU_AMDKFD_USERPTR_BO (1ULL << 63)
+
+/* Userptr restore delay, just long enough to allow consecutive VM
+ * changes to accumulate
+ */
+#define AMDGPU_USERPTR_RESTORE_DELAY_MS 1
+
 /* Impose limit on how much memory KFD can use */
 static struct {
 	uint64_t max_system_mem_limit;
+	uint64_t max_userptr_mem_limit;
 	int64_t system_mem_used;
+	int64_t userptr_mem_used;
 	spinlock_t mem_limit_lock;
 } kfd_mem_limit;
 
@@ -57,6 +69,7 @@ static const char * const domain_bit_to_string[] = {
 
 #define domain_string(domain) domain_bit_to_string[ffs(domain)-1]
 
+static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work);
 
 
 static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
@@ -78,6 +91,7 @@ static bool check_if_add_bo_to_vm(struct amdgpu_vm *avm,
 
 /* Set memory usage limits. Current, limits are
  *  System (kernel) memory - 3/8th System RAM
+ *  Userptr memory - 3/4th System RAM
  */
 void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
 {
@@ -90,8 +104,10 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
 
 	spin_lock_init(&kfd_mem_limit.mem_limit_lock);
 	kfd_mem_limit.max_system_mem_limit = (mem >> 1) - (mem >> 3);
-	pr_debug("Kernel memory limit %lluM\n",
-		(kfd_mem_limit.max_system_mem_limit >> 20));
+	kfd_mem_limit.max_userptr_mem_limit = mem - (mem >> 2);
+	pr_debug("Kernel memory limit %lluM, userptr limit %lluM\n",
+		(kfd_mem_limit.max_system_mem_limit >> 20),
+		(kfd_mem_limit.max_userptr_mem_limit >> 20));
 }
 
 static int amdgpu_amdkfd_reserve_system_mem_limit(struct amdgpu_device *adev,
@@ -111,6 +127,16 @@ static int amdgpu_amdkfd_reserve_system_mem_limit(struct amdgpu_device *adev,
 			goto err_no_mem;
 		}
 		kfd_mem_limit.system_mem_used += (acc_size + size);
+	} else if (domain == AMDGPU_GEM_DOMAIN_CPU) {
+		if ((kfd_mem_limit.system_mem_used + acc_size >
+			kfd_mem_limit.max_system_mem_limit) ||
+			(kfd_mem_limit.userptr_mem_used + (size + acc_size) >
+			kfd_mem_limit.max_userptr_mem_limit)) {
+			ret = -ENOMEM;
+			goto err_no_mem;
+		}
+		kfd_mem_limit.system_mem_used += acc_size;
+		kfd_mem_limit.userptr_mem_used += size;
 	}
 err_no_mem:
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
@@ -126,10 +152,16 @@ static void unreserve_system_mem_limit(struct amdgpu_device *adev,
 				       sizeof(struct amdgpu_bo));
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
-	if (domain == AMDGPU_GEM_DOMAIN_GTT)
+	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
 		kfd_mem_limit.system_mem_used -= (acc_size + size);
+	} else if (domain == AMDGPU_GEM_DOMAIN_CPU) {
+		kfd_mem_limit.system_mem_used -= acc_size;
+		kfd_mem_limit.userptr_mem_used -= size;
+	}
 	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
 		  "kfd system memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0,
+		  "kfd userptr memory accounting unbalanced");
 
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 }
@@ -138,12 +170,17 @@ void amdgpu_amdkfd_unreserve_system_memory_limit(struct amdgpu_bo *bo)
 {
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 
-	if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT) {
+	if (bo->flags & AMDGPU_AMDKFD_USERPTR_BO) {
+		kfd_mem_limit.system_mem_used -= bo->tbo.acc_size;
+		kfd_mem_limit.userptr_mem_used -= amdgpu_bo_size(bo);
+	} else if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT) {
 		kfd_mem_limit.system_mem_used -=
 			(bo->tbo.acc_size + amdgpu_bo_size(bo));
 	}
 	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
 		  "kfd system memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.userptr_mem_used < 0,
+		  "kfd userptr memory accounting unbalanced");
 
 	spin_unlock(&kfd_mem_limit.mem_limit_lock);
 }
@@ -506,7 +543,8 @@ static void remove_bo_from_vm(struct amdgpu_device *adev,
 }
 
 static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem,
-				struct amdkfd_process_info *process_info)
+				struct amdkfd_process_info *process_info,
+				bool userptr)
 {
 	struct ttm_validate_buffer *entry = &mem->validate_list;
 	struct amdgpu_bo *bo = mem->bo;
@@ -515,10 +553,95 @@ static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem,
 	entry->shared = true;
 	entry->bo = &bo->tbo;
 	mutex_lock(&process_info->lock);
-	list_add_tail(&entry->head, &process_info->kfd_bo_list);
+	if (userptr)
+		list_add_tail(&entry->head, &process_info->userptr_valid_list);
+	else
+		list_add_tail(&entry->head, &process_info->kfd_bo_list);
 	mutex_unlock(&process_info->lock);
 }
 
+/* Initializes user pages. It registers the MMU notifier and validates
+ * the userptr BO in the GTT domain.
+ *
+ * The BO must already be on the userptr_valid_list. Otherwise an
+ * eviction and restore may happen that leaves the new BO unmapped
+ * with the user mode queues running.
+ *
+ * Takes the process_info->lock to protect against concurrent restore
+ * workers.
+ *
+ * Returns 0 for success, negative errno for errors.
+ */
+static int init_user_pages(struct kgd_mem *mem, struct mm_struct *mm,
+			   uint64_t user_addr)
+{
+	struct amdkfd_process_info *process_info = mem->process_info;
+	struct amdgpu_bo *bo = mem->bo;
+	struct ttm_operation_ctx ctx = { true, false };
+	int ret = 0;
+
+	mutex_lock(&process_info->lock);
+
+	ret = amdgpu_ttm_tt_set_userptr(bo->tbo.ttm, user_addr, 0);
+	if (ret) {
+		pr_err("%s: Failed to set userptr: %d\n", __func__, ret);
+		goto out;
+	}
+
+	ret = amdgpu_mn_register(bo, user_addr);
+	if (ret) {
+		pr_err("%s: Failed to register MMU notifier: %d\n",
+		       __func__, ret);
+		goto out;
+	}
+
+	/* If no restore worker is running concurrently, user_pages
+	 * should not be allocated
+	 */
+	WARN(mem->user_pages, "Leaking user_pages array");
+
+	mem->user_pages = kvmalloc_array(bo->tbo.ttm->num_pages,
+					   sizeof(struct page *),
+					   GFP_KERNEL | __GFP_ZERO);
+	if (!mem->user_pages) {
+		pr_err("%s: Failed to allocate pages array\n", __func__);
+		ret = -ENOMEM;
+		goto unregister_out;
+	}
+
+	ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, mem->user_pages);
+	if (ret) {
+		pr_err("%s: Failed to get user pages: %d\n", __func__, ret);
+		goto free_out;
+	}
+
+	amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, mem->user_pages);
+
+	ret = amdgpu_bo_reserve(bo, true);
+	if (ret) {
+		pr_err("%s: Failed to reserve BO\n", __func__);
+		goto release_out;
+	}
+	amdgpu_ttm_placement_from_domain(bo, mem->domain);
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (ret)
+		pr_err("%s: failed to validate BO\n", __func__);
+	amdgpu_bo_unreserve(bo);
+
+release_out:
+	if (ret)
+		release_pages(mem->user_pages, bo->tbo.ttm->num_pages);
+free_out:
+	kvfree(mem->user_pages);
+	mem->user_pages = NULL;
+unregister_out:
+	if (ret)
+		amdgpu_mn_unregister(bo);
+out:
+	mutex_unlock(&process_info->lock);
+	return ret;
+}
+
 /* Reserving a BO and its page table BOs must happen atomically to
  * avoid deadlocks. Some operations update multiple VMs at once. Track
  * all the reservation info in a context structure. Optionally a sync
@@ -748,7 +871,8 @@ static int update_gpuvm_pte(struct amdgpu_device *adev,
 }
 
 static int map_bo_to_gpuvm(struct amdgpu_device *adev,
-		struct kfd_bo_va_list *entry, struct amdgpu_sync *sync)
+		struct kfd_bo_va_list *entry, struct amdgpu_sync *sync,
+		bool no_update_pte)
 {
 	int ret;
 
@@ -762,6 +886,9 @@ static int map_bo_to_gpuvm(struct amdgpu_device *adev,
 		return ret;
 	}
 
+	if (no_update_pte)
+		return 0;
+
 	ret = update_gpuvm_pte(adev, entry, sync);
 	if (ret) {
 		pr_err("update_gpuvm_pte() failed\n");
@@ -820,6 +947,8 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 		mutex_init(&info->lock);
 		INIT_LIST_HEAD(&info->vm_list_head);
 		INIT_LIST_HEAD(&info->kfd_bo_list);
+		INIT_LIST_HEAD(&info->userptr_valid_list);
+		INIT_LIST_HEAD(&info->userptr_inval_list);
 
 		info->eviction_fence =
 			amdgpu_amdkfd_fence_create(dma_fence_context_alloc(1),
@@ -830,6 +959,11 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 			goto create_evict_fence_fail;
 		}
 
+		info->pid = get_task_pid(current->group_leader, PIDTYPE_PID);
+		atomic_set(&info->evicted_bos, 0);
+		INIT_DELAYED_WORK(&info->restore_userptr_work,
+				  amdgpu_amdkfd_restore_userptr_worker);
+
 		*process_info = info;
 		*ef = dma_fence_get(&info->eviction_fence->base);
 	}
@@ -872,6 +1006,7 @@ reserve_pd_fail:
 		dma_fence_put(*ef);
 		*ef = NULL;
 		*process_info = NULL;
+		put_pid(info->pid);
 create_evict_fence_fail:
 		mutex_destroy(&info->lock);
 		kfree(info);
@@ -967,8 +1102,12 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
 	/* Release per-process resources when last compute VM is destroyed */
 	if (!process_info->n_vms) {
 		WARN_ON(!list_empty(&process_info->kfd_bo_list));
+		WARN_ON(!list_empty(&process_info->userptr_valid_list));
+		WARN_ON(!list_empty(&process_info->userptr_inval_list));
 
 		dma_fence_put(&process_info->eviction_fence->base);
+		cancel_delayed_work_sync(&process_info->restore_userptr_work);
+		put_pid(process_info->pid);
 		mutex_destroy(&process_info->lock);
 		kfree(process_info);
 	}
@@ -1003,9 +1142,11 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	uint64_t user_addr = 0;
 	struct amdgpu_bo *bo;
+	struct amdgpu_bo_param bp;
 	int byte_align;
-	u32 alloc_domain;
+	u32 domain, alloc_domain;
 	u64 alloc_flags;
 	uint32_t mapping_flags;
 	int ret;
@@ -1014,14 +1155,21 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	 * Check on which domain to allocate BO
 	 */
 	if (flags & ALLOC_MEM_FLAGS_VRAM) {
-		alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
+		domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
 		alloc_flags = AMDGPU_GEM_CREATE_VRAM_CLEARED;
 		alloc_flags |= (flags & ALLOC_MEM_FLAGS_PUBLIC) ?
 			AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED :
 			AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
 	} else if (flags & ALLOC_MEM_FLAGS_GTT) {
-		alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
+		domain = alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
+		alloc_flags = 0;
+	} else if (flags & ALLOC_MEM_FLAGS_USERPTR) {
+		domain = AMDGPU_GEM_DOMAIN_GTT;
+		alloc_domain = AMDGPU_GEM_DOMAIN_CPU;
 		alloc_flags = 0;
+		if (!offset || !*offset)
+			return -EINVAL;
+		user_addr = *offset;
 	} else {
 		return -EINVAL;
 	}
@@ -1069,8 +1217,14 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s\n",
 			va, size, domain_string(alloc_domain));
 
-	ret = amdgpu_bo_create(adev, size, byte_align,
-				alloc_domain, alloc_flags, ttm_bo_type_device, NULL, &bo);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = byte_align;
+	bp.domain = alloc_domain;
+	bp.flags = alloc_flags;
+	bp.type = ttm_bo_type_device;
+	bp.resv = NULL;
+	ret = amdgpu_bo_create(adev, &bp, &bo);
 	if (ret) {
 		pr_debug("Failed to create BO on domain %s. ret %d\n",
 				domain_string(alloc_domain), ret);
@@ -1078,18 +1232,34 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 	}
 	bo->kfd_bo = *mem;
 	(*mem)->bo = bo;
+	if (user_addr)
+		bo->flags |= AMDGPU_AMDKFD_USERPTR_BO;
 
 	(*mem)->va = va;
-	(*mem)->domain = alloc_domain;
+	(*mem)->domain = domain;
 	(*mem)->mapped_to_gpu_memory = 0;
 	(*mem)->process_info = avm->process_info;
-	add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info);
+	add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, user_addr);
+
+	if (user_addr) {
+		ret = init_user_pages(*mem, current->mm, user_addr);
+		if (ret) {
+			mutex_lock(&avm->process_info->lock);
+			list_del(&(*mem)->validate_list.head);
+			mutex_unlock(&avm->process_info->lock);
+			goto allocate_init_user_pages_failed;
+		}
+	}
 
 	if (offset)
 		*offset = amdgpu_bo_mmap_offset(bo);
 
 	return 0;
 
+allocate_init_user_pages_failed:
+	amdgpu_bo_unref(&bo);
+	/* Don't unreserve system mem limit twice */
+	goto err_reserve_system_mem;
 err_bo_create:
 	unreserve_system_mem_limit(adev, size, alloc_domain);
 err_reserve_system_mem:
@@ -1122,12 +1292,24 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	 * be freed anyway
 	 */
 
+	/* No more MMU notifiers */
+	amdgpu_mn_unregister(mem->bo);
+
 	/* Make sure restore workers don't access the BO any more */
 	bo_list_entry = &mem->validate_list;
 	mutex_lock(&process_info->lock);
 	list_del(&bo_list_entry->head);
 	mutex_unlock(&process_info->lock);
 
+	/* Free user pages if necessary */
+	if (mem->user_pages) {
+		pr_debug("%s: Freeing user_pages array\n", __func__);
+		if (mem->user_pages[0])
+			release_pages(mem->user_pages,
+					mem->bo->tbo.ttm->num_pages);
+		kvfree(mem->user_pages);
+	}
+
 	ret = reserve_bo_and_cond_vms(mem, NULL, BO_VM_ALL, &ctx);
 	if (unlikely(ret))
 		return ret;
@@ -1173,21 +1355,32 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	struct kfd_bo_va_list *bo_va_entry = NULL;
 	struct kfd_bo_va_list *bo_va_entry_aql = NULL;
 	unsigned long bo_size;
-
-	/* Make sure restore is not running concurrently.
-	 */
-	mutex_lock(&mem->process_info->lock);
-
-	mutex_lock(&mem->lock);
+	bool is_invalid_userptr = false;
 
 	bo = mem->bo;
-
 	if (!bo) {
 		pr_err("Invalid BO when mapping memory to GPU\n");
-		ret = -EINVAL;
-		goto out;
+		return -EINVAL;
 	}
 
+	/* Make sure restore is not running concurrently. Since we
+	 * don't map invalid userptr BOs, we rely on the next restore
+	 * worker to do the mapping
+	 */
+	mutex_lock(&mem->process_info->lock);
+
+	/* Lock mmap-sem. If we find an invalid userptr BO, we can be
+	 * sure that the MMU notifier is no longer running
+	 * concurrently and the queues are actually stopped
+	 */
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
+		down_write(&current->mm->mmap_sem);
+		is_invalid_userptr = atomic_read(&mem->invalid);
+		up_write(&current->mm->mmap_sem);
+	}
+
+	mutex_lock(&mem->lock);
+
 	domain = mem->domain;
 	bo_size = bo->tbo.mem.size;
 
@@ -1200,6 +1393,14 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	if (unlikely(ret))
 		goto out;
 
+	/* Userptr can be marked as "not invalid", but not actually be
+	 * validated yet (still in the system domain). In that case
+	 * the queues are still stopped and we can leave mapping for
+	 * the next restore worker
+	 */
+	if (bo->tbo.mem.mem_type == TTM_PL_SYSTEM)
+		is_invalid_userptr = true;
+
 	if (check_if_add_bo_to_vm(avm, mem)) {
 		ret = add_bo_to_vm(adev, mem, avm, false,
 				&bo_va_entry);
@@ -1217,7 +1418,8 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 			goto add_bo_to_vm_failed;
 	}
 
-	if (mem->mapped_to_gpu_memory == 0) {
+	if (mem->mapped_to_gpu_memory == 0 &&
+	    !amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
 		/* Validate BO only once. The eviction fence gets added to BO
 		 * the first time it is mapped. Validate will wait for all
 		 * background evictions to complete.
@@ -1235,7 +1437,8 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 					entry->va, entry->va + bo_size,
 					entry);
 
-			ret = map_bo_to_gpuvm(adev, entry, ctx.sync);
+			ret = map_bo_to_gpuvm(adev, entry, ctx.sync,
+					      is_invalid_userptr);
 			if (ret) {
 				pr_err("Failed to map radeon bo to gpuvm\n");
 				goto map_bo_to_gpuvm_failed;
@@ -1418,6 +1621,337 @@ bo_reserve_failed:
 	return ret;
 }
 
+/* Evict a userptr BO by stopping the queues if necessary
+ *
+ * Runs in MMU notifier, may be in RECLAIM_FS context. This means it
+ * cannot do any memory allocations, and cannot take any locks that
+ * are held elsewhere while allocating memory. Therefore this is as
+ * simple as possible, using atomic counters.
+ *
+ * It doesn't do anything to the BO itself. The real work happens in
+ * restore, where we get updated page addresses. This function only
+ * ensures that GPU access to the BO is stopped.
+ */
+int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem,
+				struct mm_struct *mm)
+{
+	struct amdkfd_process_info *process_info = mem->process_info;
+	int invalid, evicted_bos;
+	int r = 0;
+
+	invalid = atomic_inc_return(&mem->invalid);
+	evicted_bos = atomic_inc_return(&process_info->evicted_bos);
+	if (evicted_bos == 1) {
+		/* First eviction, stop the queues */
+		r = kgd2kfd->quiesce_mm(mm);
+		if (r)
+			pr_err("Failed to quiesce KFD\n");
+		schedule_delayed_work(&process_info->restore_userptr_work,
+			msecs_to_jiffies(AMDGPU_USERPTR_RESTORE_DELAY_MS));
+	}
+
+	return r;
+}
+
+/* Update invalid userptr BOs
+ *
+ * Moves invalidated (evicted) userptr BOs from userptr_valid_list to
+ * userptr_inval_list and updates user pages for all BOs that have
+ * been invalidated since their last update.
+ */
+static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
+				     struct mm_struct *mm)
+{
+	struct kgd_mem *mem, *tmp_mem;
+	struct amdgpu_bo *bo;
+	struct ttm_operation_ctx ctx = { false, false };
+	int invalid, ret;
+
+	/* Move all invalidated BOs to the userptr_inval_list and
+	 * release their user pages by migration to the CPU domain
+	 */
+	list_for_each_entry_safe(mem, tmp_mem,
+				 &process_info->userptr_valid_list,
+				 validate_list.head) {
+		if (!atomic_read(&mem->invalid))
+			continue; /* BO is still valid */
+
+		bo = mem->bo;
+
+		if (amdgpu_bo_reserve(bo, true))
+			return -EAGAIN;
+		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+		ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+		amdgpu_bo_unreserve(bo);
+		if (ret) {
+			pr_err("%s: Failed to invalidate userptr BO\n",
+			       __func__);
+			return -EAGAIN;
+		}
+
+		list_move_tail(&mem->validate_list.head,
+			       &process_info->userptr_inval_list);
+	}
+
+	if (list_empty(&process_info->userptr_inval_list))
+		return 0; /* All evicted userptr BOs were freed */
+
+	/* Go through userptr_inval_list and update any invalid user_pages */
+	list_for_each_entry(mem, &process_info->userptr_inval_list,
+			    validate_list.head) {
+		invalid = atomic_read(&mem->invalid);
+		if (!invalid)
+			/* BO hasn't been invalidated since the last
+			 * revalidation attempt. Keep its BO list.
+			 */
+			continue;
+
+		bo = mem->bo;
+
+		if (!mem->user_pages) {
+			mem->user_pages =
+				kvmalloc_array(bo->tbo.ttm->num_pages,
+						 sizeof(struct page *),
+						 GFP_KERNEL | __GFP_ZERO);
+			if (!mem->user_pages) {
+				pr_err("%s: Failed to allocate pages array\n",
+				       __func__);
+				return -ENOMEM;
+			}
+		} else if (mem->user_pages[0]) {
+			release_pages(mem->user_pages, bo->tbo.ttm->num_pages);
+		}
+
+		/* Get updated user pages */
+		ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm,
+						   mem->user_pages);
+		if (ret) {
+			mem->user_pages[0] = NULL;
+			pr_info("%s: Failed to get user pages: %d\n",
+				__func__, ret);
+			/* Pretend it succeeded. It will fail later
+			 * with a VM fault if the GPU tries to access
+			 * it. Better than hanging indefinitely with
+			 * stalled user mode queues.
+			 */
+		}
+
+		/* Mark the BO as valid unless it was invalidated
+		 * again concurrently
+		 */
+		if (atomic_cmpxchg(&mem->invalid, invalid, 0) != invalid)
+			return -EAGAIN;
+	}
+
+	return 0;
+}
+
+/* Validate invalid userptr BOs
+ *
+ * Validates BOs on the userptr_inval_list, and moves them back to the
+ * userptr_valid_list. Also updates GPUVM page tables with new page
+ * addresses and waits for the page table updates to complete.
+ */
+static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
+{
+	struct amdgpu_bo_list_entry *pd_bo_list_entries;
+	struct list_head resv_list, duplicates;
+	struct ww_acquire_ctx ticket;
+	struct amdgpu_sync sync;
+
+	struct amdgpu_vm *peer_vm;
+	struct kgd_mem *mem, *tmp_mem;
+	struct amdgpu_bo *bo;
+	struct ttm_operation_ctx ctx = { false, false };
+	int i, ret;
+
+	pd_bo_list_entries = kcalloc(process_info->n_vms,
+				     sizeof(struct amdgpu_bo_list_entry),
+				     GFP_KERNEL);
+	if (!pd_bo_list_entries) {
+		pr_err("%s: Failed to allocate PD BO list entries\n", __func__);
+		return -ENOMEM;
+	}
+
+	INIT_LIST_HEAD(&resv_list);
+	INIT_LIST_HEAD(&duplicates);
+
+	/* Get all the page directory BOs that need to be reserved */
+	i = 0;
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node)
+		amdgpu_vm_get_pd_bo(peer_vm, &resv_list,
+				    &pd_bo_list_entries[i++]);
+	/* Add the userptr_inval_list entries to resv_list */
+	list_for_each_entry(mem, &process_info->userptr_inval_list,
+			    validate_list.head) {
+		list_add_tail(&mem->resv_list.head, &resv_list);
+		mem->resv_list.bo = mem->validate_list.bo;
+		mem->resv_list.shared = mem->validate_list.shared;
+	}
+
+	/* Reserve all BOs and page tables for validation */
+	ret = ttm_eu_reserve_buffers(&ticket, &resv_list, false, &duplicates);
+	WARN(!list_empty(&duplicates), "Duplicates should be empty");
+	if (ret)
+		goto out;
+
+	amdgpu_sync_create(&sync);
+
+	/* Avoid triggering eviction fences when unmapping invalid
+	 * userptr BOs (waits for all fences, doesn't use
+	 * FENCE_OWNER_VM)
+	 */
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node)
+		amdgpu_amdkfd_remove_eviction_fence(peer_vm->root.base.bo,
+						process_info->eviction_fence,
+						NULL, NULL);
+
+	ret = process_validate_vms(process_info);
+	if (ret)
+		goto unreserve_out;
+
+	/* Validate BOs and update GPUVM page tables */
+	list_for_each_entry_safe(mem, tmp_mem,
+				 &process_info->userptr_inval_list,
+				 validate_list.head) {
+		struct kfd_bo_va_list *bo_va_entry;
+
+		bo = mem->bo;
+
+		/* Copy pages array and validate the BO if we got user pages */
+		if (mem->user_pages[0]) {
+			amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm,
+						     mem->user_pages);
+			amdgpu_ttm_placement_from_domain(bo, mem->domain);
+			ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+			if (ret) {
+				pr_err("%s: failed to validate BO\n", __func__);
+				goto unreserve_out;
+			}
+		}
+
+		/* Validate succeeded, now the BO owns the pages, free
+		 * our copy of the pointer array. Put this BO back on
+		 * the userptr_valid_list. If we need to revalidate
+		 * it, we need to start from scratch.
+		 */
+		kvfree(mem->user_pages);
+		mem->user_pages = NULL;
+		list_move_tail(&mem->validate_list.head,
+			       &process_info->userptr_valid_list);
+
+		/* Update mapping. If the BO was not validated
+		 * (because we couldn't get user pages), this will
+		 * clear the page table entries, which will result in
+		 * VM faults if the GPU tries to access the invalid
+		 * memory.
+		 */
+		list_for_each_entry(bo_va_entry, &mem->bo_va_list, bo_list) {
+			if (!bo_va_entry->is_mapped)
+				continue;
+
+			ret = update_gpuvm_pte((struct amdgpu_device *)
+					       bo_va_entry->kgd_dev,
+					       bo_va_entry, &sync);
+			if (ret) {
+				pr_err("%s: update PTE failed\n", __func__);
+				/* make sure this gets validated again */
+				atomic_inc(&mem->invalid);
+				goto unreserve_out;
+			}
+		}
+	}
+
+	/* Update page directories */
+	ret = process_update_pds(process_info, &sync);
+
+unreserve_out:
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node)
+		amdgpu_bo_fence(peer_vm->root.base.bo,
+				&process_info->eviction_fence->base, true);
+	ttm_eu_backoff_reservation(&ticket, &resv_list);
+	amdgpu_sync_wait(&sync, false);
+	amdgpu_sync_free(&sync);
+out:
+	kfree(pd_bo_list_entries);
+
+	return ret;
+}
+
+/* Worker callback to restore evicted userptr BOs
+ *
+ * Tries to update and validate all userptr BOs. If successful and no
+ * concurrent evictions happened, the queues are restarted. Otherwise,
+ * reschedule for another attempt later.
+ */
+static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work)
+{
+	struct delayed_work *dwork = to_delayed_work(work);
+	struct amdkfd_process_info *process_info =
+		container_of(dwork, struct amdkfd_process_info,
+			     restore_userptr_work);
+	struct task_struct *usertask;
+	struct mm_struct *mm;
+	int evicted_bos;
+
+	evicted_bos = atomic_read(&process_info->evicted_bos);
+	if (!evicted_bos)
+		return;
+
+	/* Reference task and mm in case of concurrent process termination */
+	usertask = get_pid_task(process_info->pid, PIDTYPE_PID);
+	if (!usertask)
+		return;
+	mm = get_task_mm(usertask);
+	if (!mm) {
+		put_task_struct(usertask);
+		return;
+	}
+
+	mutex_lock(&process_info->lock);
+
+	if (update_invalid_user_pages(process_info, mm))
+		goto unlock_out;
+	/* userptr_inval_list can be empty if all evicted userptr BOs
+	 * have been freed. In that case there is nothing to validate
+	 * and we can just restart the queues.
+	 */
+	if (!list_empty(&process_info->userptr_inval_list)) {
+		if (atomic_read(&process_info->evicted_bos) != evicted_bos)
+			goto unlock_out; /* Concurrent eviction, try again */
+
+		if (validate_invalid_user_pages(process_info))
+			goto unlock_out;
+	}
+	/* Final check for concurrent evicton and atomic update. If
+	 * another eviction happens after successful update, it will
+	 * be a first eviction that calls quiesce_mm. The eviction
+	 * reference counting inside KFD will handle this case.
+	 */
+	if (atomic_cmpxchg(&process_info->evicted_bos, evicted_bos, 0) !=
+	    evicted_bos)
+		goto unlock_out;
+	evicted_bos = 0;
+	if (kgd2kfd->resume_mm(mm)) {
+		pr_err("%s: Failed to resume KFD\n", __func__);
+		/* No recovery from this failure. Probably the CP is
+		 * hanging. No point trying again.
+		 */
+	}
+unlock_out:
+	mutex_unlock(&process_info->lock);
+	mmput(mm);
+	put_task_struct(usertask);
+
+	/* If validation failed, reschedule another attempt */
+	if (evicted_bos)
+		schedule_delayed_work(&process_info->restore_userptr_work,
+			msecs_to_jiffies(AMDGPU_USERPTR_RESTORE_DELAY_MS));
+}
+
 /** amdgpu_amdkfd_gpuvm_restore_process_bos - Restore all BOs for the given
  *   KFD process identified by process_info
  *

drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c

@@ -322,3 +322,47 @@ int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev)
 
 	return ret;
 }
+
+union gfx_info {
+	struct  atom_gfx_info_v2_4 v24;
+};
+
+int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index;
+	uint8_t frev, crev;
+	uint16_t data_offset;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    gfx_info);
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		union gfx_info *gfx_info = (union gfx_info *)
+			(mode_info->atom_context->bios + data_offset);
+		switch (crev) {
+		case 4:
+			adev->gfx.config.max_shader_engines = gfx_info->v24.gc_num_se;
+			adev->gfx.config.max_cu_per_sh = gfx_info->v24.gc_num_cu_per_sh;
+			adev->gfx.config.max_sh_per_se = gfx_info->v24.gc_num_sh_per_se;
+			adev->gfx.config.max_backends_per_se = gfx_info->v24.gc_num_rb_per_se;
+			adev->gfx.config.max_texture_channel_caches = gfx_info->v24.gc_num_tccs;
+			adev->gfx.config.max_gprs = le16_to_cpu(gfx_info->v24.gc_num_gprs);
+			adev->gfx.config.max_gs_threads = gfx_info->v24.gc_num_max_gs_thds;
+			adev->gfx.config.gs_vgt_table_depth = gfx_info->v24.gc_gs_table_depth;
+			adev->gfx.config.gs_prim_buffer_depth =
+				le16_to_cpu(gfx_info->v24.gc_gsprim_buff_depth);
+			adev->gfx.config.double_offchip_lds_buf =
+				gfx_info->v24.gc_double_offchip_lds_buffer;
+			adev->gfx.cu_info.wave_front_size = le16_to_cpu(gfx_info->v24.gc_wave_size);
+			adev->gfx.cu_info.max_waves_per_simd = le16_to_cpu(gfx_info->v24.gc_max_waves_per_simd);
+			adev->gfx.cu_info.max_scratch_slots_per_cu = gfx_info->v24.gc_max_scratch_slots_per_cu;
+			adev->gfx.cu_info.lds_size = le16_to_cpu(gfx_info->v24.gc_lds_size);
+			return 0;
+		default:
+			return -EINVAL;
+		}
+
+	}
+	return -EINVAL;
+}

drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.h

@@ -30,5 +30,6 @@ int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_vram_width(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_vram_type(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev);
 
 #endif

drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c

@@ -550,7 +550,7 @@ static int amdgpu_atpx_init(void)
  * look up whether we are the integrated or discrete GPU (all asics).
  * Returns the client id.
  */
-static int amdgpu_atpx_get_client_id(struct pci_dev *pdev)
+static enum vga_switcheroo_client_id amdgpu_atpx_get_client_id(struct pci_dev *pdev)
 {
 	if (amdgpu_atpx_priv.dhandle == ACPI_HANDLE(&pdev->dev))
 		return VGA_SWITCHEROO_IGD;

drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c

@@ -75,13 +75,20 @@ static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
 {
 	struct amdgpu_bo *dobj = NULL;
 	struct amdgpu_bo *sobj = NULL;
+	struct amdgpu_bo_param bp;
 	uint64_t saddr, daddr;
 	int r, n;
 	int time;
 
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = sdomain;
+	bp.flags = 0;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
 	n = AMDGPU_BENCHMARK_ITERATIONS;
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE,sdomain, 0,
-			     ttm_bo_type_kernel, NULL, &sobj);
+	r = amdgpu_bo_create(adev, &bp, &sobj);
 	if (r) {
 		goto out_cleanup;
 	}
@@ -93,8 +100,8 @@ static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
 	if (r) {
 		goto out_cleanup;
 	}
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE, ddomain, 0,
-			     ttm_bo_type_kernel, NULL, &dobj);
+	bp.domain = ddomain;
+	r = amdgpu_bo_create(adev, &bp, &dobj);
 	if (r) {
 		goto out_cleanup;
 	}

drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c

@@ -23,7 +23,6 @@
  */
 #include <linux/list.h>
 #include <linux/slab.h>
-#include <linux/pci.h>
 #include <drm/drmP.h>
 #include <linux/firmware.h>
 #include <drm/amdgpu_drm.h>
@@ -109,121 +108,6 @@ static void amdgpu_cgs_write_ind_register(struct cgs_device *cgs_device,
 	WARN(1, "Invalid indirect register space");
 }
 
-static int amdgpu_cgs_get_pci_resource(struct cgs_device *cgs_device,
-				       enum cgs_resource_type resource_type,
-				       uint64_t size,
-				       uint64_t offset,
-				       uint64_t *resource_base)
-{
-	CGS_FUNC_ADEV;
-
-	if (resource_base == NULL)
-		return -EINVAL;
-
-	switch (resource_type) {
-	case CGS_RESOURCE_TYPE_MMIO:
-		if (adev->rmmio_size == 0)
-			return -ENOENT;
-		if ((offset + size) > adev->rmmio_size)
-			return -EINVAL;
-		*resource_base = adev->rmmio_base;
-		return 0;
-	case CGS_RESOURCE_TYPE_DOORBELL:
-		if (adev->doorbell.size == 0)
-			return -ENOENT;
-		if ((offset + size) > adev->doorbell.size)
-			return -EINVAL;
-		*resource_base = adev->doorbell.base;
-		return 0;
-	case CGS_RESOURCE_TYPE_FB:
-	case CGS_RESOURCE_TYPE_IO:
-	case CGS_RESOURCE_TYPE_ROM:
-	default:
-		return -EINVAL;
-	}
-}
-
-static const void *amdgpu_cgs_atom_get_data_table(struct cgs_device *cgs_device,
-						  unsigned table, uint16_t *size,
-						  uint8_t *frev, uint8_t *crev)
-{
-	CGS_FUNC_ADEV;
-	uint16_t data_start;
-
-	if (amdgpu_atom_parse_data_header(
-		    adev->mode_info.atom_context, table, size,
-		    frev, crev, &data_start))
-		return (uint8_t*)adev->mode_info.atom_context->bios +
-			data_start;
-
-	return NULL;
-}
-
-static int amdgpu_cgs_atom_get_cmd_table_revs(struct cgs_device *cgs_device, unsigned table,
-					      uint8_t *frev, uint8_t *crev)
-{
-	CGS_FUNC_ADEV;
-
-	if (amdgpu_atom_parse_cmd_header(
-		    adev->mode_info.atom_context, table,
-		    frev, crev))
-		return 0;
-
-	return -EINVAL;
-}
-
-static int amdgpu_cgs_atom_exec_cmd_table(struct cgs_device *cgs_device, unsigned table,
-					  void *args)
-{
-	CGS_FUNC_ADEV;
-
-	return amdgpu_atom_execute_table(
-		adev->mode_info.atom_context, table, args);
-}
-
-static int amdgpu_cgs_set_clockgating_state(struct cgs_device *cgs_device,
-				  enum amd_ip_block_type block_type,
-				  enum amd_clockgating_state state)
-{
-	CGS_FUNC_ADEV;
-	int i, r = -1;
-
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
-			continue;
-
-		if (adev->ip_blocks[i].version->type == block_type) {
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state(
-								(void *)adev,
-									state);
-			break;
-		}
-	}
-	return r;
-}
-
-static int amdgpu_cgs_set_powergating_state(struct cgs_device *cgs_device,
-				  enum amd_ip_block_type block_type,
-				  enum amd_powergating_state state)
-{
-	CGS_FUNC_ADEV;
-	int i, r = -1;
-
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
-			continue;
-
-		if (adev->ip_blocks[i].version->type == block_type) {
-			r = adev->ip_blocks[i].version->funcs->set_powergating_state(
-								(void *)adev,
-									state);
-			break;
-		}
-	}
-	return r;
-}
-
-
 static uint32_t fw_type_convert(struct cgs_device *cgs_device, uint32_t fw_type)
 {
 	CGS_FUNC_ADEV;
@@ -271,18 +155,6 @@ static uint32_t fw_type_convert(struct cgs_device *cgs_device, uint32_t fw_type)
 	return result;
 }
 
-static int amdgpu_cgs_rel_firmware(struct cgs_device *cgs_device, enum cgs_ucode_id type)
-{
-	CGS_FUNC_ADEV;
-	if ((CGS_UCODE_ID_SMU == type) || (CGS_UCODE_ID_SMU_SK == type)) {
-		release_firmware(adev->pm.fw);
-		adev->pm.fw = NULL;
-		return 0;
-	}
-	/* cannot release other firmware because they are not created by cgs */
-	return -EINVAL;
-}
-
 static uint16_t amdgpu_get_firmware_version(struct cgs_device *cgs_device,
 					enum cgs_ucode_id type)
 {
@@ -326,34 +198,6 @@ static uint16_t amdgpu_get_firmware_version(struct cgs_device *cgs_device,
 	return fw_version;
 }
 
-static int amdgpu_cgs_enter_safe_mode(struct cgs_device *cgs_device,
-					bool en)
-{
-	CGS_FUNC_ADEV;
-
-	if (adev->gfx.rlc.funcs->enter_safe_mode == NULL ||
-		adev->gfx.rlc.funcs->exit_safe_mode == NULL)
-		return 0;
-
-	if (en)
-		adev->gfx.rlc.funcs->enter_safe_mode(adev);
-	else
-		adev->gfx.rlc.funcs->exit_safe_mode(adev);
-
-	return 0;
-}
-
-static void amdgpu_cgs_lock_grbm_idx(struct cgs_device *cgs_device,
-					bool lock)
-{
-	CGS_FUNC_ADEV;
-
-	if (lock)
-		mutex_lock(&adev->grbm_idx_mutex);
-	else
-		mutex_unlock(&adev->grbm_idx_mutex);
-}
-
 static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 					enum cgs_ucode_id type,
 					struct cgs_firmware_info *info)
@@ -541,6 +385,9 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 			case CHIP_POLARIS12:
 				strcpy(fw_name, "amdgpu/polaris12_smc.bin");
 				break;
+			case CHIP_VEGAM:
+				strcpy(fw_name, "amdgpu/vegam_smc.bin");
+				break;
 			case CHIP_VEGA10:
 				if ((adev->pdev->device == 0x687f) &&
 					((adev->pdev->revision == 0xc0) ||
@@ -553,6 +400,9 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 			case CHIP_VEGA12:
 				strcpy(fw_name, "amdgpu/vega12_smc.bin");
 				break;
+			case CHIP_VEGA20:
+				strcpy(fw_name, "amdgpu/vega20_smc.bin");
+				break;
 			default:
 				DRM_ERROR("SMC firmware not supported\n");
 				return -EINVAL;
@@ -598,97 +448,12 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 	return 0;
 }
 
-static int amdgpu_cgs_is_virtualization_enabled(void *cgs_device)
-{
-	CGS_FUNC_ADEV;
-	return amdgpu_sriov_vf(adev);
-}
-
-static int amdgpu_cgs_get_active_displays_info(struct cgs_device *cgs_device,
-					  struct cgs_display_info *info)
-{
-	CGS_FUNC_ADEV;
-	struct cgs_mode_info *mode_info;
-
-	if (info == NULL)
-		return -EINVAL;
-
-	mode_info = info->mode_info;
-	if (mode_info)
-		/* if the displays are off, vblank time is max */
-		mode_info->vblank_time_us = 0xffffffff;
-
-	if (!amdgpu_device_has_dc_support(adev)) {
-		struct amdgpu_crtc *amdgpu_crtc;
-		struct drm_device *ddev = adev->ddev;
-		struct drm_crtc *crtc;
-		uint32_t line_time_us, vblank_lines;
-
-		if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-			list_for_each_entry(crtc,
-					&ddev->mode_config.crtc_list, head) {
-				amdgpu_crtc = to_amdgpu_crtc(crtc);
-				if (crtc->enabled) {
-					info->active_display_mask |= (1 << amdgpu_crtc->crtc_id);
-					info->display_count++;
-				}
-				if (mode_info != NULL &&
-					crtc->enabled && amdgpu_crtc->enabled &&
-					amdgpu_crtc->hw_mode.clock) {
-					line_time_us = (amdgpu_crtc->hw_mode.crtc_htotal * 1000) /
-								amdgpu_crtc->hw_mode.clock;
-					vblank_lines = amdgpu_crtc->hw_mode.crtc_vblank_end -
-								amdgpu_crtc->hw_mode.crtc_vdisplay +
-								(amdgpu_crtc->v_border * 2);
-					mode_info->vblank_time_us = vblank_lines * line_time_us;
-					mode_info->refresh_rate = drm_mode_vrefresh(&amdgpu_crtc->hw_mode);
-					/* we have issues with mclk switching with refresh rates
-					 * over 120 hz on the non-DC code.
-					 */
-					if (mode_info->refresh_rate > 120)
-						mode_info->vblank_time_us = 0;
-					mode_info = NULL;
-				}
-			}
-		}
-	} else {
-		info->display_count = adev->pm.pm_display_cfg.num_display;
-		if (mode_info != NULL) {
-			mode_info->vblank_time_us = adev->pm.pm_display_cfg.min_vblank_time;
-			mode_info->refresh_rate = adev->pm.pm_display_cfg.vrefresh;
-		}
-	}
-	return 0;
-}
-
-
-static int amdgpu_cgs_notify_dpm_enabled(struct cgs_device *cgs_device, bool enabled)
-{
-	CGS_FUNC_ADEV;
-
-	adev->pm.dpm_enabled = enabled;
-
-	return 0;
-}
-
 static const struct cgs_ops amdgpu_cgs_ops = {
 	.read_register = amdgpu_cgs_read_register,
 	.write_register = amdgpu_cgs_write_register,
 	.read_ind_register = amdgpu_cgs_read_ind_register,
 	.write_ind_register = amdgpu_cgs_write_ind_register,
-	.get_pci_resource = amdgpu_cgs_get_pci_resource,
-	.atom_get_data_table = amdgpu_cgs_atom_get_data_table,
-	.atom_get_cmd_table_revs = amdgpu_cgs_atom_get_cmd_table_revs,
-	.atom_exec_cmd_table = amdgpu_cgs_atom_exec_cmd_table,
 	.get_firmware_info = amdgpu_cgs_get_firmware_info,
-	.rel_firmware = amdgpu_cgs_rel_firmware,
-	.set_powergating_state = amdgpu_cgs_set_powergating_state,
-	.set_clockgating_state = amdgpu_cgs_set_clockgating_state,
-	.get_active_displays_info = amdgpu_cgs_get_active_displays_info,
-	.notify_dpm_enabled = amdgpu_cgs_notify_dpm_enabled,
-	.is_virtualization_enabled = amdgpu_cgs_is_virtualization_enabled,
-	.enter_safe_mode = amdgpu_cgs_enter_safe_mode,
-	.lock_grbm_idx = amdgpu_cgs_lock_grbm_idx,
 };
 
 struct cgs_device *amdgpu_cgs_create_device(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c

@@ -691,7 +691,7 @@ static int amdgpu_connector_lvds_get_modes(struct drm_connector *connector)
 	return ret;
 }
 
-static int amdgpu_connector_lvds_mode_valid(struct drm_connector *connector,
+static enum drm_mode_status amdgpu_connector_lvds_mode_valid(struct drm_connector *connector,
 					     struct drm_display_mode *mode)
 {
 	struct drm_encoder *encoder = amdgpu_connector_best_single_encoder(connector);
@@ -843,7 +843,7 @@ static int amdgpu_connector_vga_get_modes(struct drm_connector *connector)
 	return ret;
 }
 
-static int amdgpu_connector_vga_mode_valid(struct drm_connector *connector,
+static enum drm_mode_status amdgpu_connector_vga_mode_valid(struct drm_connector *connector,
 					    struct drm_display_mode *mode)
 {
 	struct drm_device *dev = connector->dev;
@@ -1172,7 +1172,7 @@ static void amdgpu_connector_dvi_force(struct drm_connector *connector)
 		amdgpu_connector->use_digital = true;
 }
 
-static int amdgpu_connector_dvi_mode_valid(struct drm_connector *connector,
+static enum drm_mode_status amdgpu_connector_dvi_mode_valid(struct drm_connector *connector,
 					    struct drm_display_mode *mode)
 {
 	struct drm_device *dev = connector->dev;
@@ -1448,7 +1448,7 @@ out:
 	return ret;
 }
 
-static int amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
+static enum drm_mode_status amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
 					   struct drm_display_mode *mode)
 {
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -382,8 +382,7 @@ retry:
 
 	p->bytes_moved += ctx.bytes_moved;
 	if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-	    bo->tbo.mem.mem_type == TTM_PL_VRAM &&
-	    bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT)
+	    amdgpu_bo_in_cpu_visible_vram(bo))
 		p->bytes_moved_vis += ctx.bytes_moved;
 
 	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
@@ -411,7 +410,6 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
 		struct amdgpu_bo_list_entry *candidate = p->evictable;
 		struct amdgpu_bo *bo = candidate->robj;
 		struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-		u64 initial_bytes_moved, bytes_moved;
 		bool update_bytes_moved_vis;
 		uint32_t other;
 
@@ -435,18 +433,14 @@ static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
 			continue;
 
 		/* Good we can try to move this BO somewhere else */
-		amdgpu_ttm_placement_from_domain(bo, other);
 		update_bytes_moved_vis =
 			adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-			bo->tbo.mem.mem_type == TTM_PL_VRAM &&
-			bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT;
-		initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
+			amdgpu_bo_in_cpu_visible_vram(bo);
+		amdgpu_ttm_placement_from_domain(bo, other);
 		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
-		bytes_moved = atomic64_read(&adev->num_bytes_moved) -
-			initial_bytes_moved;
-		p->bytes_moved += bytes_moved;
+		p->bytes_moved += ctx.bytes_moved;
 		if (update_bytes_moved_vis)
-			p->bytes_moved_vis += bytes_moved;
+			p->bytes_moved_vis += ctx.bytes_moved;
 
 		if (unlikely(r))
 			break;
@@ -536,7 +530,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 	if (p->bo_list) {
 		amdgpu_bo_list_get_list(p->bo_list, &p->validated);
 		if (p->bo_list->first_userptr != p->bo_list->num_entries)
-			p->mn = amdgpu_mn_get(p->adev);
+			p->mn = amdgpu_mn_get(p->adev, AMDGPU_MN_TYPE_GFX);
 	}
 
 	INIT_LIST_HEAD(&duplicates);

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c

@@ -91,7 +91,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			continue;
 
 		r = drm_sched_entity_init(&ring->sched, &ctx->rings[i].entity,
-					  rq, amdgpu_sched_jobs, &ctx->guilty);
+					  rq, &ctx->guilty);
 		if (r)
 			goto failed;
 	}
@@ -111,8 +111,9 @@ failed:
 	return r;
 }
 
-static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
+static void amdgpu_ctx_fini(struct kref *ref)
 {
+	struct amdgpu_ctx *ctx = container_of(ref, struct amdgpu_ctx, refcount);
 	struct amdgpu_device *adev = ctx->adev;
 	unsigned i, j;
 
@@ -125,13 +126,11 @@ static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
 	kfree(ctx->fences);
 	ctx->fences = NULL;
 
-	for (i = 0; i < adev->num_rings; i++)
-		drm_sched_entity_fini(&adev->rings[i]->sched,
-				      &ctx->rings[i].entity);
-
 	amdgpu_queue_mgr_fini(adev, &ctx->queue_mgr);
 
 	mutex_destroy(&ctx->lock);
+
+	kfree(ctx);
 }
 
 static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
@@ -170,12 +169,20 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 static void amdgpu_ctx_do_release(struct kref *ref)
 {
 	struct amdgpu_ctx *ctx;
+	u32 i;
 
 	ctx = container_of(ref, struct amdgpu_ctx, refcount);
 
-	amdgpu_ctx_fini(ctx);
+	for (i = 0; i < ctx->adev->num_rings; i++) {
 
-	kfree(ctx);
+		if (ctx->adev->rings[i] == &ctx->adev->gfx.kiq.ring)
+			continue;
+
+		drm_sched_entity_fini(&ctx->adev->rings[i]->sched,
+			&ctx->rings[i].entity);
+	}
+
+	amdgpu_ctx_fini(ref);
 }
 
 static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id)
@@ -437,16 +444,72 @@ void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
 	idr_init(&mgr->ctx_handles);
 }
 
+void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr)
+{
+	struct amdgpu_ctx *ctx;
+	struct idr *idp;
+	uint32_t id, i;
+
+	idp = &mgr->ctx_handles;
+
+	idr_for_each_entry(idp, ctx, id) {
+
+		if (!ctx->adev)
+			return;
+
+		for (i = 0; i < ctx->adev->num_rings; i++) {
+
+			if (ctx->adev->rings[i] == &ctx->adev->gfx.kiq.ring)
+				continue;
+
+			if (kref_read(&ctx->refcount) == 1)
+				drm_sched_entity_do_release(&ctx->adev->rings[i]->sched,
+						  &ctx->rings[i].entity);
+			else
+				DRM_ERROR("ctx %p is still alive\n", ctx);
+		}
+	}
+}
+
+void amdgpu_ctx_mgr_entity_cleanup(struct amdgpu_ctx_mgr *mgr)
+{
+	struct amdgpu_ctx *ctx;
+	struct idr *idp;
+	uint32_t id, i;
+
+	idp = &mgr->ctx_handles;
+
+	idr_for_each_entry(idp, ctx, id) {
+
+		if (!ctx->adev)
+			return;
+
+		for (i = 0; i < ctx->adev->num_rings; i++) {
+
+			if (ctx->adev->rings[i] == &ctx->adev->gfx.kiq.ring)
+				continue;
+
+			if (kref_read(&ctx->refcount) == 1)
+				drm_sched_entity_cleanup(&ctx->adev->rings[i]->sched,
+					&ctx->rings[i].entity);
+			else
+				DRM_ERROR("ctx %p is still alive\n", ctx);
+		}
+	}
+}
+
 void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
 {
 	struct amdgpu_ctx *ctx;
 	struct idr *idp;
 	uint32_t id;
 
+	amdgpu_ctx_mgr_entity_cleanup(mgr);
+
 	idp = &mgr->ctx_handles;
 
 	idr_for_each_entry(idp, ctx, id) {
-		if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
+		if (kref_put(&ctx->refcount, amdgpu_ctx_fini) != 1)
 			DRM_ERROR("ctx %p is still alive\n", ctx);
 	}
 

drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c

@@ -28,8 +28,13 @@
 #include <linux/debugfs.h>
 #include "amdgpu.h"
 
-/*
- * Debugfs
+/**
+ * amdgpu_debugfs_add_files - Add simple debugfs entries
+ *
+ * @adev:  Device to attach debugfs entries to
+ * @files:  Array of function callbacks that respond to reads
+ * @nfiles: Number of callbacks to register
+ *
  */
 int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
 			     const struct drm_info_list *files,
@@ -64,7 +69,33 @@ int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
 
 #if defined(CONFIG_DEBUG_FS)
 
-
+/**
+ * amdgpu_debugfs_process_reg_op - Handle MMIO register reads/writes
+ *
+ * @read: True if reading
+ * @f: open file handle
+ * @buf: User buffer to write/read to
+ * @size: Number of bytes to write/read
+ * @pos:  Offset to seek to
+ *
+ * This debugfs entry has special meaning on the offset being sought.
+ * Various bits have different meanings:
+ *
+ * Bit 62:  Indicates a GRBM bank switch is needed
+ * Bit 61:  Indicates a SRBM bank switch is needed (implies bit 62 is
+ * 			zero)
+ * Bits 24..33: The SE or ME selector if needed
+ * Bits 34..43: The SH (or SA) or PIPE selector if needed
+ * Bits 44..53: The INSTANCE (or CU/WGP) or QUEUE selector if needed
+ *
+ * Bit 23:  Indicates that the PM power gating lock should be held
+ * 			This is necessary to read registers that might be
+ * 			unreliable during a power gating transistion.
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static int  amdgpu_debugfs_process_reg_op(bool read, struct file *f,
 		char __user *buf, size_t size, loff_t *pos)
 {
@@ -164,19 +195,37 @@ end:
 	return result;
 }
 
-
+/**
+ * amdgpu_debugfs_regs_read - Callback for reading MMIO registers
+ */
 static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
 	return amdgpu_debugfs_process_reg_op(true, f, buf, size, pos);
 }
 
+/**
+ * amdgpu_debugfs_regs_write - Callback for writing MMIO registers
+ */
 static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf,
 					 size_t size, loff_t *pos)
 {
 	return amdgpu_debugfs_process_reg_op(false, f, (char __user *)buf, size, pos);
 }
 
+
+/**
+ * amdgpu_debugfs_regs_pcie_read - Read from a PCIE register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -204,6 +253,18 @@ static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_regs_pcie_write - Write to a PCIE register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user *buf,
 					 size_t size, loff_t *pos)
 {
@@ -232,6 +293,18 @@ static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_regs_didt_read - Read from a DIDT register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -259,6 +332,18 @@ static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_regs_didt_write - Write to a DIDT register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user *buf,
 					 size_t size, loff_t *pos)
 {
@@ -287,6 +372,18 @@ static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_regs_smc_read - Read from a SMC register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -314,6 +411,18 @@ static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_regs_smc_write - Write to a SMC register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
 static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *buf,
 					 size_t size, loff_t *pos)
 {
@@ -342,6 +451,20 @@ static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_gca_config_read - Read from gfx config data
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * This file is used to access configuration data in a somewhat
+ * stable fashion.  The format is a series of DWORDs with the first
+ * indicating which revision it is.  New content is appended to the
+ * end so that older software can still read the data.
+ */
+
 static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -418,6 +541,19 @@ static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_debugfs_sensor_read - Read from the powerplay sensors
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset is treated as the BYTE address of one of the sensors
+ * enumerated in amd/include/kgd_pp_interface.h under the
+ * 'amd_pp_sensors' enumeration.  For instance to read the UVD VCLK
+ * you would use the offset 3 * 4 = 12.
+ */
 static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -428,7 +564,7 @@ static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
 	if (size & 3 || *pos & 0x3)
 		return -EINVAL;
 
-	if (amdgpu_dpm == 0)
+	if (!adev->pm.dpm_enabled)
 		return -EINVAL;
 
 	/* convert offset to sensor number */
@@ -457,6 +593,27 @@ static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
 	return !r ? outsize : r;
 }
 
+/** amdgpu_debugfs_wave_read - Read WAVE STATUS data
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset being sought changes which wave that the status data
+ * will be returned for.  The bits are used as follows:
+ *
+ * Bits 0..6: 	Byte offset into data
+ * Bits 7..14:	SE selector
+ * Bits 15..22:	SH/SA selector
+ * Bits 23..30: CU/{WGP+SIMD} selector
+ * Bits 31..36: WAVE ID selector
+ * Bits 37..44: SIMD ID selector
+ *
+ * The returned data begins with one DWORD of version information
+ * Followed by WAVE STATUS registers relevant to the GFX IP version
+ * being used.  See gfx_v8_0_read_wave_data() for an example output.
+ */
 static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -507,6 +664,28 @@ static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/** amdgpu_debugfs_gpr_read - Read wave gprs
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset being sought changes which wave that the status data
+ * will be returned for.  The bits are used as follows:
+ *
+ * Bits 0..11:	Byte offset into data
+ * Bits 12..19:	SE selector
+ * Bits 20..27:	SH/SA selector
+ * Bits 28..35: CU/{WGP+SIMD} selector
+ * Bits 36..43: WAVE ID selector
+ * Bits 37..44: SIMD ID selector
+ * Bits 52..59: Thread selector
+ * Bits 60..61: Bank selector (VGPR=0,SGPR=1)
+ *
+ * The return data comes from the SGPR or VGPR register bank for
+ * the selected operational unit.
+ */
 static ssize_t amdgpu_debugfs_gpr_read(struct file *f, char __user *buf,
 					size_t size, loff_t *pos)
 {
@@ -637,6 +816,12 @@ static const char *debugfs_regs_names[] = {
 	"amdgpu_gpr",
 };
 
+/**
+ * amdgpu_debugfs_regs_init -	Initialize debugfs entries that provide
+ * 								register access.
+ *
+ * @adev: The device to attach the debugfs entries to
+ */
 int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
 {
 	struct drm_minor *minor = adev->ddev->primary;

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -83,8 +83,10 @@ static const char *amdgpu_asic_name[] = {
 	"POLARIS10",
 	"POLARIS11",
 	"POLARIS12",
+	"VEGAM",
 	"VEGA10",
 	"VEGA12",
+	"VEGA20",
 	"RAVEN",
 	"LAST",
 };
@@ -690,6 +692,8 @@ void amdgpu_device_gart_location(struct amdgpu_device *adev,
 {
 	u64 size_af, size_bf;
 
+	mc->gart_size += adev->pm.smu_prv_buffer_size;
+
 	size_af = adev->gmc.mc_mask - mc->vram_end;
 	size_bf = mc->vram_start;
 	if (size_bf > size_af) {
@@ -907,6 +911,46 @@ static void amdgpu_device_check_vm_size(struct amdgpu_device *adev)
 	}
 }
 
+static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
+{
+	struct sysinfo si;
+	bool is_os_64 = (sizeof(void *) == 8) ? true : false;
+	uint64_t total_memory;
+	uint64_t dram_size_seven_GB = 0x1B8000000;
+	uint64_t dram_size_three_GB = 0xB8000000;
+
+	if (amdgpu_smu_memory_pool_size == 0)
+		return;
+
+	if (!is_os_64) {
+		DRM_WARN("Not 64-bit OS, feature not supported\n");
+		goto def_value;
+	}
+	si_meminfo(&si);
+	total_memory = (uint64_t)si.totalram * si.mem_unit;
+
+	if ((amdgpu_smu_memory_pool_size == 1) ||
+		(amdgpu_smu_memory_pool_size == 2)) {
+		if (total_memory < dram_size_three_GB)
+			goto def_value1;
+	} else if ((amdgpu_smu_memory_pool_size == 4) ||
+		(amdgpu_smu_memory_pool_size == 8)) {
+		if (total_memory < dram_size_seven_GB)
+			goto def_value1;
+	} else {
+		DRM_WARN("Smu memory pool size not supported\n");
+		goto def_value;
+	}
+	adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28;
+
+	return;
+
+def_value1:
+	DRM_WARN("No enough system memory\n");
+def_value:
+	adev->pm.smu_prv_buffer_size = 0;
+}
+
 /**
  * amdgpu_device_check_arguments - validate module params
  *
@@ -948,6 +992,8 @@ static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
 		amdgpu_vm_fragment_size = -1;
 	}
 
+	amdgpu_device_check_smu_prv_buffer_size(adev);
+
 	amdgpu_device_check_vm_size(adev);
 
 	amdgpu_device_check_block_size(adev);
@@ -1039,10 +1085,11 @@ static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = {
  * the hardware IP specified.
  * Returns the error code from the last instance.
  */
-int amdgpu_device_ip_set_clockgating_state(struct amdgpu_device *adev,
+int amdgpu_device_ip_set_clockgating_state(void *dev,
 					   enum amd_ip_block_type block_type,
 					   enum amd_clockgating_state state)
 {
+	struct amdgpu_device *adev = dev;
 	int i, r = 0;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
@@ -1072,10 +1119,11 @@ int amdgpu_device_ip_set_clockgating_state(struct amdgpu_device *adev,
  * the hardware IP specified.
  * Returns the error code from the last instance.
  */
-int amdgpu_device_ip_set_powergating_state(struct amdgpu_device *adev,
+int amdgpu_device_ip_set_powergating_state(void *dev,
 					   enum amd_ip_block_type block_type,
 					   enum amd_powergating_state state)
 {
+	struct amdgpu_device *adev = dev;
 	int i, r = 0;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
@@ -1320,9 +1368,10 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
 	case CHIP_TOPAZ:
 	case CHIP_TONGA:
 	case CHIP_FIJI:
-	case CHIP_POLARIS11:
 	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 	case CHIP_CARRIZO:
 	case CHIP_STONEY:
 #ifdef CONFIG_DRM_AMDGPU_SI
@@ -1339,6 +1388,7 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
 	case CHIP_KABINI:
 	case CHIP_MULLINS:
 #endif
+	case CHIP_VEGA20:
 	default:
 		return 0;
 	case CHIP_VEGA10:
@@ -1428,9 +1478,10 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 	case CHIP_TOPAZ:
 	case CHIP_TONGA:
 	case CHIP_FIJI:
-	case CHIP_POLARIS11:
 	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 	case CHIP_CARRIZO:
 	case CHIP_STONEY:
 		if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY)
@@ -1472,6 +1523,7 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 #endif
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		if (adev->asic_type == CHIP_RAVEN)
 			adev->family = AMDGPU_FAMILY_RV;
@@ -1499,6 +1551,8 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 			return -EAGAIN;
 	}
 
+	adev->powerplay.pp_feature = amdgpu_pp_feature_mask;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
 			DRM_ERROR("disabled ip block: %d <%s>\n",
@@ -1654,12 +1708,17 @@ static int amdgpu_device_ip_late_set_cg_state(struct amdgpu_device *adev)
 	if (amdgpu_emu_mode == 1)
 		return 0;
 
+	r = amdgpu_ib_ring_tests(adev);
+	if (r)
+		DRM_ERROR("ib ring test failed (%d).\n", r);
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
 		/* skip CG for VCE/UVD, it's handled specially */
 		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
 		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
 		    adev->ip_blocks[i].version->funcs->set_clockgating_state) {
 			/* enable clockgating to save power */
 			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
@@ -1704,8 +1763,8 @@ static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
 		}
 	}
 
-	mod_delayed_work(system_wq, &adev->late_init_work,
-			msecs_to_jiffies(AMDGPU_RESUME_MS));
+	queue_delayed_work(system_wq, &adev->late_init_work,
+			   msecs_to_jiffies(AMDGPU_RESUME_MS));
 
 	amdgpu_device_fill_reset_magic(adev);
 
@@ -1759,6 +1818,7 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
 
 		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
 			adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
+			adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
 			adev->ip_blocks[i].version->funcs->set_clockgating_state) {
 			/* ungate blocks before hw fini so that we can shutdown the blocks safely */
 			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
@@ -1850,6 +1910,12 @@ int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
 	if (amdgpu_sriov_vf(adev))
 		amdgpu_virt_request_full_gpu(adev, false);
 
+	/* ungate SMC block powergating */
+	if (adev->powerplay.pp_feature & PP_GFXOFF_MASK)
+		amdgpu_device_ip_set_powergating_state(adev,
+						       AMD_IP_BLOCK_TYPE_SMC,
+						       AMD_CG_STATE_UNGATE);
+
 	/* ungate SMC block first */
 	r = amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_SMC,
 						   AMD_CG_STATE_UNGATE);
@@ -2086,16 +2152,15 @@ bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type)
 	case CHIP_MULLINS:
 	case CHIP_CARRIZO:
 	case CHIP_STONEY:
-	case CHIP_POLARIS11:
 	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 	case CHIP_TONGA:
 	case CHIP_FIJI:
-#if defined(CONFIG_DRM_AMD_DC_PRE_VEGA)
-		return amdgpu_dc != 0;
-#endif
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 #if defined(CONFIG_DRM_AMD_DC_DCN1_0)
 	case CHIP_RAVEN:
 #endif
@@ -2375,10 +2440,6 @@ fence_driver_init:
 		goto failed;
 	}
 
-	r = amdgpu_ib_ring_tests(adev);
-	if (r)
-		DRM_ERROR("ib ring test failed (%d).\n", r);
-
 	if (amdgpu_sriov_vf(adev))
 		amdgpu_virt_init_data_exchange(adev);
 
@@ -2539,7 +2600,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
 	/* unpin the front buffers and cursors */
 	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
 		struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-		struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb);
+		struct drm_framebuffer *fb = crtc->primary->fb;
 		struct amdgpu_bo *robj;
 
 		if (amdgpu_crtc->cursor_bo) {
@@ -2551,10 +2612,10 @@ int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
 			}
 		}
 
-		if (rfb == NULL || rfb->obj == NULL) {
+		if (fb == NULL || fb->obj[0] == NULL) {
 			continue;
 		}
-		robj = gem_to_amdgpu_bo(rfb->obj);
+		robj = gem_to_amdgpu_bo(fb->obj[0]);
 		/* don't unpin kernel fb objects */
 		if (!amdgpu_fbdev_robj_is_fb(adev, robj)) {
 			r = amdgpu_bo_reserve(robj, true);
@@ -2640,11 +2701,6 @@ int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
 	}
 	amdgpu_fence_driver_resume(adev);
 
-	if (resume) {
-		r = amdgpu_ib_ring_tests(adev);
-		if (r)
-			DRM_ERROR("ib ring test failed (%d).\n", r);
-	}
 
 	r = amdgpu_device_ip_late_init(adev);
 	if (r)
@@ -2736,6 +2792,9 @@ static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)
 	if (amdgpu_sriov_vf(adev))
 		return true;
 
+	if (amdgpu_asic_need_full_reset(adev))
+		return true;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
@@ -2792,6 +2851,9 @@ static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev)
 {
 	int i;
 
+	if (amdgpu_asic_need_full_reset(adev))
+		return true;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
@@ -3087,20 +3149,19 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
 
 	/* now we are okay to resume SMC/CP/SDMA */
 	r = amdgpu_device_ip_reinit_late_sriov(adev);
-	amdgpu_virt_release_full_gpu(adev, true);
 	if (r)
 		goto error;
 
 	amdgpu_irq_gpu_reset_resume_helper(adev);
 	r = amdgpu_ib_ring_tests(adev);
 
+error:
+	amdgpu_virt_release_full_gpu(adev, true);
 	if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) {
 		atomic_inc(&adev->vram_lost_counter);
 		r = amdgpu_device_handle_vram_lost(adev);
 	}
 
-error:
-
 	return r;
 }
 
@@ -3117,7 +3178,6 @@ error:
 int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 			      struct amdgpu_job *job, bool force)
 {
-	struct drm_atomic_state *state = NULL;
 	int i, r, resched;
 
 	if (!force && !amdgpu_device_ip_check_soft_reset(adev)) {
@@ -3140,10 +3200,6 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	/* block TTM */
 	resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
 
-	/* store modesetting */
-	if (amdgpu_device_has_dc_support(adev))
-		state = drm_atomic_helper_suspend(adev->ddev);
-
 	/* block all schedulers and reset given job's ring */
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
 		struct amdgpu_ring *ring = adev->rings[i];
@@ -3183,10 +3239,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 		kthread_unpark(ring->sched.thread);
 	}
 
-	if (amdgpu_device_has_dc_support(adev)) {
-		if (drm_atomic_helper_resume(adev->ddev, state))
-			dev_info(adev->dev, "drm resume failed:%d\n", r);
-	} else {
+	if (!amdgpu_device_has_dc_support(adev)) {
 		drm_helper_resume_force_mode(adev->ddev);
 	}
 

drivers/gpu/drm/amd/amdgpu/amdgpu_display.c

@@ -35,6 +35,7 @@
 #include <linux/pm_runtime.h>
 #include <drm/drm_crtc_helper.h>
 #include <drm/drm_edid.h>
+#include <drm/drm_gem_framebuffer_helper.h>
 #include <drm/drm_fb_helper.h>
 
 static void amdgpu_display_flip_callback(struct dma_fence *f,
@@ -151,8 +152,6 @@ int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc,
 	struct drm_device *dev = crtc->dev;
 	struct amdgpu_device *adev = dev->dev_private;
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-	struct amdgpu_framebuffer *old_amdgpu_fb;
-	struct amdgpu_framebuffer *new_amdgpu_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_flip_work *work;
 	struct amdgpu_bo *new_abo;
@@ -174,15 +173,13 @@ int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc,
 	work->async = (page_flip_flags & DRM_MODE_PAGE_FLIP_ASYNC) != 0;
 
 	/* schedule unpin of the old buffer */
-	old_amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-	obj = old_amdgpu_fb->obj;
+	obj = crtc->primary->fb->obj[0];
 
 	/* take a reference to the old object */
 	work->old_abo = gem_to_amdgpu_bo(obj);
 	amdgpu_bo_ref(work->old_abo);
 
-	new_amdgpu_fb = to_amdgpu_framebuffer(fb);
-	obj = new_amdgpu_fb->obj;
+	obj = fb->obj[0];
 	new_abo = gem_to_amdgpu_bo(obj);
 
 	/* pin the new buffer */
@@ -192,7 +189,7 @@ int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc,
 		goto cleanup;
 	}
 
-	r = amdgpu_bo_pin(new_abo, amdgpu_display_framebuffer_domains(adev), &base);
+	r = amdgpu_bo_pin(new_abo, amdgpu_display_supported_domains(adev), &base);
 	if (unlikely(r != 0)) {
 		DRM_ERROR("failed to pin new abo buffer before flip\n");
 		goto unreserve;
@@ -482,31 +479,12 @@ bool amdgpu_display_ddc_probe(struct amdgpu_connector *amdgpu_connector,
 	return true;
 }
 
-static void amdgpu_display_user_framebuffer_destroy(struct drm_framebuffer *fb)
-{
-	struct amdgpu_framebuffer *amdgpu_fb = to_amdgpu_framebuffer(fb);
-
-	drm_gem_object_put_unlocked(amdgpu_fb->obj);
-	drm_framebuffer_cleanup(fb);
-	kfree(amdgpu_fb);
-}
-
-static int amdgpu_display_user_framebuffer_create_handle(
-			struct drm_framebuffer *fb,
-			struct drm_file *file_priv,
-			unsigned int *handle)
-{
-	struct amdgpu_framebuffer *amdgpu_fb = to_amdgpu_framebuffer(fb);
-
-	return drm_gem_handle_create(file_priv, amdgpu_fb->obj, handle);
-}
-
 static const struct drm_framebuffer_funcs amdgpu_fb_funcs = {
-	.destroy = amdgpu_display_user_framebuffer_destroy,
-	.create_handle = amdgpu_display_user_framebuffer_create_handle,
+	.destroy = drm_gem_fb_destroy,
+	.create_handle = drm_gem_fb_create_handle,
 };
 
-uint32_t amdgpu_display_framebuffer_domains(struct amdgpu_device *adev)
+uint32_t amdgpu_display_supported_domains(struct amdgpu_device *adev)
 {
 	uint32_t domain = AMDGPU_GEM_DOMAIN_VRAM;
 
@@ -526,11 +504,11 @@ int amdgpu_display_framebuffer_init(struct drm_device *dev,
 				    struct drm_gem_object *obj)
 {
 	int ret;
-	rfb->obj = obj;
+	rfb->base.obj[0] = obj;
 	drm_helper_mode_fill_fb_struct(dev, &rfb->base, mode_cmd);
 	ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
 	if (ret) {
-		rfb->obj = NULL;
+		rfb->base.obj[0] = NULL;
 		return ret;
 	}
 	return 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_display.h

@@ -23,7 +23,7 @@
 #ifndef __AMDGPU_DISPLAY_H__
 #define __AMDGPU_DISPLAY_H__
 
-uint32_t amdgpu_display_framebuffer_domains(struct amdgpu_device *adev);
+uint32_t amdgpu_display_supported_domains(struct amdgpu_device *adev);
 struct drm_framebuffer *
 amdgpu_display_user_framebuffer_create(struct drm_device *dev,
 				       struct drm_file *file_priv,

drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c

@@ -115,6 +115,26 @@ void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev,
 	pr_cont("\n");
 }
 
+void amdgpu_dpm_get_active_displays(struct amdgpu_device *adev)
+{
+	struct drm_device *ddev = adev->ddev;
+	struct drm_crtc *crtc;
+	struct amdgpu_crtc *amdgpu_crtc;
+
+	adev->pm.dpm.new_active_crtcs = 0;
+	adev->pm.dpm.new_active_crtc_count = 0;
+	if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
+		list_for_each_entry(crtc,
+				    &ddev->mode_config.crtc_list, head) {
+			amdgpu_crtc = to_amdgpu_crtc(crtc);
+			if (amdgpu_crtc->enabled) {
+				adev->pm.dpm.new_active_crtcs |= (1 << amdgpu_crtc->crtc_id);
+				adev->pm.dpm.new_active_crtc_count++;
+			}
+		}
+	}
+}
+
 
 u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev)
 {

drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h

@@ -52,8 +52,6 @@ enum amdgpu_dpm_event_src {
 	AMDGPU_DPM_EVENT_SRC_DIGIAL_OR_EXTERNAL = 4
 };
 
-#define SCLK_DEEP_SLEEP_MASK 0x8
-
 struct amdgpu_ps {
 	u32 caps; /* vbios flags */
 	u32 class; /* vbios flags */
@@ -349,12 +347,6 @@ enum amdgpu_pcie_gen {
 		((adev)->powerplay.pp_funcs->set_clockgating_by_smu(\
 			(adev)->powerplay.pp_handle, msg_id))
 
-#define amdgpu_dpm_notify_smu_memory_info(adev, virtual_addr_low, \
-			virtual_addr_hi, mc_addr_low, mc_addr_hi, size) \
-		((adev)->powerplay.pp_funcs->notify_smu_memory_info)( \
-			(adev)->powerplay.pp_handle, virtual_addr_low, \
-			virtual_addr_hi, mc_addr_low, mc_addr_hi, size)
-
 #define amdgpu_dpm_get_power_profile_mode(adev, buf) \
 		((adev)->powerplay.pp_funcs->get_power_profile_mode(\
 			(adev)->powerplay.pp_handle, buf))
@@ -445,6 +437,8 @@ struct amdgpu_pm {
 	uint32_t                pcie_gen_mask;
 	uint32_t                pcie_mlw_mask;
 	struct amd_pp_display_configuration pm_display_cfg;/* set by dc */
+	uint32_t                smu_prv_buffer_size;
+	struct amdgpu_bo        *smu_prv_buffer;
 };
 
 #define R600_SSTU_DFLT                               0
@@ -482,6 +476,7 @@ void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev,
 				struct amdgpu_ps *rps);
 u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev);
 u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev);
+void amdgpu_dpm_get_active_displays(struct amdgpu_device *adev);
 bool amdgpu_is_uvd_state(u32 class, u32 class2);
 void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b,
 			      u32 *p, u32 *u);

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -75,9 +75,10 @@
  * - 3.23.0 - Add query for VRAM lost counter
  * - 3.24.0 - Add high priority compute support for gfx9
  * - 3.25.0 - Add support for sensor query info (stable pstate sclk/mclk).
+ * - 3.26.0 - GFX9: Process AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE.
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	25
+#define KMS_DRIVER_MINOR	26
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit = 0;
@@ -121,7 +122,7 @@ uint amdgpu_pg_mask = 0xffffffff;
 uint amdgpu_sdma_phase_quantum = 32;
 char *amdgpu_disable_cu = NULL;
 char *amdgpu_virtual_display = NULL;
-uint amdgpu_pp_feature_mask = 0xffffbfff;
+uint amdgpu_pp_feature_mask = 0xffff3fff; /* gfxoff (bit 15) disabled by default */
 int amdgpu_ngg = 0;
 int amdgpu_prim_buf_per_se = 0;
 int amdgpu_pos_buf_per_se = 0;
@@ -132,6 +133,7 @@ int amdgpu_lbpw = -1;
 int amdgpu_compute_multipipe = -1;
 int amdgpu_gpu_recovery = -1; /* auto */
 int amdgpu_emu_mode = 0;
+uint amdgpu_smu_memory_pool_size = 0;
 
 MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
 module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);
@@ -316,6 +318,11 @@ MODULE_PARM_DESC(cik_support, "CIK support (1 = enabled (default), 0 = disabled)
 module_param_named(cik_support, amdgpu_cik_support, int, 0444);
 #endif
 
+MODULE_PARM_DESC(smu_memory_pool_size,
+	"reserve gtt for smu debug usage, 0 = disable,"
+		"0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
+module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);
+
 static const struct pci_device_id pciidlist[] = {
 #ifdef  CONFIG_DRM_AMDGPU_SI
 	{0x1002, 0x6780, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TAHITI},
@@ -534,6 +541,9 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x6995, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x6997, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x699F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
+	/* VEGAM */
+	{0x1002, 0x694C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM},
+	{0x1002, 0x694E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM},
 	/* Vega 10 */
 	{0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
 	{0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
@@ -550,6 +560,13 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x69A2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
 	{0x1002, 0x69A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
 	{0x1002, 0x69AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	/* Vega 20 */
+	{0x1002, 0x66A0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x66A1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x66A2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x66A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x66A7, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x66AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20|AMD_EXP_HW_SUPPORT},
 	/* Raven */
 	{0x1002, 0x15dd, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RAVEN|AMD_IS_APU},
 

drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c

@@ -137,7 +137,7 @@ static int amdgpufb_create_pinned_object(struct amdgpu_fbdev *rfbdev,
 	/* need to align pitch with crtc limits */
 	mode_cmd->pitches[0] = amdgpu_align_pitch(adev, mode_cmd->width, cpp,
 						  fb_tiled);
-	domain = amdgpu_display_framebuffer_domains(adev);
+	domain = amdgpu_display_supported_domains(adev);
 
 	height = ALIGN(mode_cmd->height, 8);
 	size = mode_cmd->pitches[0] * height;
@@ -292,9 +292,9 @@ static int amdgpu_fbdev_destroy(struct drm_device *dev, struct amdgpu_fbdev *rfb
 
 	drm_fb_helper_unregister_fbi(&rfbdev->helper);
 
-	if (rfb->obj) {
-		amdgpufb_destroy_pinned_object(rfb->obj);
-		rfb->obj = NULL;
+	if (rfb->base.obj[0]) {
+		amdgpufb_destroy_pinned_object(rfb->base.obj[0]);
+		rfb->base.obj[0] = NULL;
 		drm_framebuffer_unregister_private(&rfb->base);
 		drm_framebuffer_cleanup(&rfb->base);
 	}
@@ -377,7 +377,7 @@ int amdgpu_fbdev_total_size(struct amdgpu_device *adev)
 	if (!adev->mode_info.rfbdev)
 		return 0;
 
-	robj = gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.obj);
+	robj = gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.base.obj[0]);
 	size += amdgpu_bo_size(robj);
 	return size;
 }
@@ -386,7 +386,7 @@ bool amdgpu_fbdev_robj_is_fb(struct amdgpu_device *adev, struct amdgpu_bo *robj)
 {
 	if (!adev->mode_info.rfbdev)
 		return false;
-	if (robj == gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.obj))
+	if (robj == gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.base.obj[0]))
 		return true;
 	return false;
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

@@ -131,7 +131,8 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
  * Emits a fence command on the requested ring (all asics).
  * Returns 0 on success, -ENOMEM on failure.
  */
-int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f)
+int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f,
+		      unsigned flags)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_fence *fence;
@@ -149,7 +150,7 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f)
 		       adev->fence_context + ring->idx,
 		       seq);
 	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
-			       seq, AMDGPU_FENCE_FLAG_INT);
+			       seq, flags | AMDGPU_FENCE_FLAG_INT);
 
 	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
 	/* This function can't be called concurrently anyway, otherwise
@@ -375,14 +376,14 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 	struct amdgpu_device *adev = ring->adev;
 	uint64_t index;
 
-	if (ring != &adev->uvd.ring) {
+	if (ring != &adev->uvd.inst[ring->me].ring) {
 		ring->fence_drv.cpu_addr = &adev->wb.wb[ring->fence_offs];
 		ring->fence_drv.gpu_addr = adev->wb.gpu_addr + (ring->fence_offs * 4);
 	} else {
 		/* put fence directly behind firmware */
 		index = ALIGN(adev->uvd.fw->size, 8);
-		ring->fence_drv.cpu_addr = adev->uvd.cpu_addr + index;
-		ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index;
+		ring->fence_drv.cpu_addr = adev->uvd.inst[ring->me].cpu_addr + index;
+		ring->fence_drv.gpu_addr = adev->uvd.inst[ring->me].gpu_addr + index;
 	}
 	amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq));
 	amdgpu_irq_get(adev, irq_src, irq_type);

drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c

@@ -113,12 +113,17 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 	int r;
 
 	if (adev->gart.robj == NULL) {
-		r = amdgpu_bo_create(adev, adev->gart.table_size, PAGE_SIZE,
-				     AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-				     ttm_bo_type_kernel, NULL,
-				     &adev->gart.robj);
+		struct amdgpu_bo_param bp;
+
+		memset(&bp, 0, sizeof(bp));
+		bp.size = adev->gart.table_size;
+		bp.byte_align = PAGE_SIZE;
+		bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+		bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+		bp.type = ttm_bo_type_kernel;
+		bp.resv = NULL;
+		r = amdgpu_bo_create(adev, &bp, &adev->gart.robj);
 		if (r) {
 			return r;
 		}

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

@@ -48,17 +48,25 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
 			     struct drm_gem_object **obj)
 {
 	struct amdgpu_bo *bo;
+	struct amdgpu_bo_param bp;
 	int r;
 
+	memset(&bp, 0, sizeof(bp));
 	*obj = NULL;
 	/* At least align on page size */
 	if (alignment < PAGE_SIZE) {
 		alignment = PAGE_SIZE;
 	}
 
+	bp.size = size;
+	bp.byte_align = alignment;
+	bp.type = type;
+	bp.resv = resv;
+	bp.preferred_domain = initial_domain;
 retry:
-	r = amdgpu_bo_create(adev, size, alignment, initial_domain,
-			     flags, type, resv, &bo);
+	bp.flags = flags;
+	bp.domain = initial_domain;
+	r = amdgpu_bo_create(adev, &bp, &bo);
 	if (r) {
 		if (r != -ERESTARTSYS) {
 			if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) {
@@ -221,12 +229,7 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 
 	/* reject invalid gem domains */
-	if (args->in.domains & ~(AMDGPU_GEM_DOMAIN_CPU |
-				 AMDGPU_GEM_DOMAIN_GTT |
-				 AMDGPU_GEM_DOMAIN_VRAM |
-				 AMDGPU_GEM_DOMAIN_GDS |
-				 AMDGPU_GEM_DOMAIN_GWS |
-				 AMDGPU_GEM_DOMAIN_OA))
+	if (args->in.domains & ~AMDGPU_GEM_DOMAIN_MASK)
 		return -EINVAL;
 
 	/* create a gem object to contain this object in */
@@ -771,16 +774,23 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv,
 }
 
 #if defined(CONFIG_DEBUG_FS)
+
+#define amdgpu_debugfs_gem_bo_print_flag(m, bo, flag)	\
+	if (bo->flags & (AMDGPU_GEM_CREATE_ ## flag)) {	\
+		seq_printf((m), " " #flag);		\
+	}
+
 static int amdgpu_debugfs_gem_bo_info(int id, void *ptr, void *data)
 {
 	struct drm_gem_object *gobj = ptr;
 	struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
 	struct seq_file *m = data;
 
+	struct dma_buf_attachment *attachment;
+	struct dma_buf *dma_buf;
 	unsigned domain;
 	const char *placement;
 	unsigned pin_count;
-	uint64_t offset;
 
 	domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
 	switch (domain) {
@@ -798,13 +808,27 @@ static int amdgpu_debugfs_gem_bo_info(int id, void *ptr, void *data)
 	seq_printf(m, "\t0x%08x: %12ld byte %s",
 		   id, amdgpu_bo_size(bo), placement);
 
-	offset = READ_ONCE(bo->tbo.mem.start);
-	if (offset != AMDGPU_BO_INVALID_OFFSET)
-		seq_printf(m, " @ 0x%010Lx", offset);
-
 	pin_count = READ_ONCE(bo->pin_count);
 	if (pin_count)
 		seq_printf(m, " pin count %d", pin_count);
+
+	dma_buf = READ_ONCE(bo->gem_base.dma_buf);
+	attachment = READ_ONCE(bo->gem_base.import_attach);
+
+	if (attachment)
+		seq_printf(m, " imported from %p", dma_buf);
+	else if (dma_buf)
+		seq_printf(m, " exported as %p", dma_buf);
+
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, CPU_ACCESS_REQUIRED);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, NO_CPU_ACCESS);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, CPU_GTT_USWC);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, VRAM_CLEARED);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, SHADOW);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, VRAM_CONTIGUOUS);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, VM_ALWAYS_VALID);
+	amdgpu_debugfs_gem_bo_print_flag(m, bo, EXPLICIT_SYNC);
+
 	seq_printf(m, "\n");
 
 	return 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c

@@ -127,6 +127,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	struct amdgpu_vm *vm;
 	uint64_t fence_ctx;
 	uint32_t status = 0, alloc_size;
+	unsigned fence_flags = 0;
 
 	unsigned i;
 	int r = 0;
@@ -227,7 +228,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 #endif
 		amdgpu_asic_invalidate_hdp(adev, ring);
 
-	r = amdgpu_fence_emit(ring, f);
+	if (ib->flags & AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE)
+		fence_flags |= AMDGPU_FENCE_FLAG_TC_WB_ONLY;
+
+	r = amdgpu_fence_emit(ring, f, fence_flags);
 	if (r) {
 		dev_err(adev->dev, "failed to emit fence (%d)\n", r);
 		if (job && job->vmid)
@@ -242,7 +246,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	/* wrap the last IB with fence */
 	if (job && job->uf_addr) {
 		amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
-				       AMDGPU_FENCE_FLAG_64BIT);
+				       fence_flags | AMDGPU_FENCE_FLAG_64BIT);
 	}
 
 	if (patch_offset != ~0 && ring->funcs->patch_cond_exec)

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c

@@ -31,6 +31,7 @@
 #include "amdgpu_sched.h"
 #include "amdgpu_uvd.h"
 #include "amdgpu_vce.h"
+#include "atom.h"
 
 #include <linux/vga_switcheroo.h>
 #include <linux/slab.h>
@@ -214,6 +215,18 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->ver = adev->gfx.rlc_fw_version;
 		fw_info->feature = adev->gfx.rlc_feature_version;
 		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL:
+		fw_info->ver = adev->gfx.rlc_srlc_fw_version;
+		fw_info->feature = adev->gfx.rlc_srlc_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM:
+		fw_info->ver = adev->gfx.rlc_srlg_fw_version;
+		fw_info->feature = adev->gfx.rlc_srlg_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM:
+		fw_info->ver = adev->gfx.rlc_srls_fw_version;
+		fw_info->feature = adev->gfx.rlc_srls_feature_version;
+		break;
 	case AMDGPU_INFO_FW_GFX_MEC:
 		if (query_fw->index == 0) {
 			fw_info->ver = adev->gfx.mec_fw_version;
@@ -273,12 +286,15 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	struct drm_crtc *crtc;
 	uint32_t ui32 = 0;
 	uint64_t ui64 = 0;
-	int i, found;
+	int i, j, found;
 	int ui32_size = sizeof(ui32);
 
 	if (!info->return_size || !info->return_pointer)
 		return -EINVAL;
 
+	/* Ensure IB tests are run on ring */
+	flush_delayed_work(&adev->late_init_work);
+
 	switch (info->query) {
 	case AMDGPU_INFO_ACCEL_WORKING:
 		ui32 = adev->accel_working;
@@ -332,7 +348,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			break;
 		case AMDGPU_HW_IP_UVD:
 			type = AMD_IP_BLOCK_TYPE_UVD;
-			ring_mask = adev->uvd.ring.ready ? 1 : 0;
+			for (i = 0; i < adev->uvd.num_uvd_inst; i++)
+				ring_mask |= ((adev->uvd.inst[i].ring.ready ? 1 : 0) << i);
 			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
 			ib_size_alignment = 16;
 			break;
@@ -345,8 +362,11 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			break;
 		case AMDGPU_HW_IP_UVD_ENC:
 			type = AMD_IP_BLOCK_TYPE_UVD;
-			for (i = 0; i < adev->uvd.num_enc_rings; i++)
-				ring_mask |= ((adev->uvd.ring_enc[i].ready ? 1 : 0) << i);
+			for (i = 0; i < adev->uvd.num_uvd_inst; i++)
+				for (j = 0; j < adev->uvd.num_enc_rings; j++)
+					ring_mask |=
+					((adev->uvd.inst[i].ring_enc[j].ready ? 1 : 0) <<
+					(j + i * adev->uvd.num_enc_rings));
 			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
 			ib_size_alignment = 1;
 			break;
@@ -701,10 +721,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		}
 	}
 	case AMDGPU_INFO_SENSOR: {
-		struct pp_gpu_power query = {0};
-		int query_size = sizeof(query);
-
-		if (amdgpu_dpm == 0)
+		if (!adev->pm.dpm_enabled)
 			return -ENOENT;
 
 		switch (info->sensor_info.type) {
@@ -746,10 +763,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			/* get average GPU power */
 			if (amdgpu_dpm_read_sensor(adev,
 						   AMDGPU_PP_SENSOR_GPU_POWER,
-						   (void *)&query, &query_size)) {
+						   (void *)&ui32, &ui32_size)) {
 				return -EINVAL;
 			}
-			ui32 = query.average_gpu_power >> 8;
+			ui32 >>= 8;
 			break;
 		case AMDGPU_INFO_SENSOR_VDDNB:
 			/* get VDDNB in millivolts */
@@ -913,8 +930,7 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 		return;
 
 	pm_runtime_get_sync(dev->dev);
-
-	amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
+	amdgpu_ctx_mgr_entity_fini(&fpriv->ctx_mgr);
 
 	if (adev->asic_type != CHIP_RAVEN) {
 		amdgpu_uvd_free_handles(adev, file_priv);
@@ -935,6 +951,8 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 	pd = amdgpu_bo_ref(fpriv->vm.root.base.bo);
 
 	amdgpu_vm_fini(adev, &fpriv->vm);
+	amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
+
 	if (pasid)
 		amdgpu_pasid_free_delayed(pd->tbo.resv, pasid);
 	amdgpu_bo_unref(&pd);
@@ -1088,6 +1106,7 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	struct amdgpu_device *adev = dev->dev_private;
 	struct drm_amdgpu_info_firmware fw_info;
 	struct drm_amdgpu_query_fw query_fw;
+	struct atom_context *ctx = adev->mode_info.atom_context;
 	int ret, i;
 
 	/* VCE */
@@ -1146,6 +1165,30 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	seq_printf(m, "RLC feature version: %u, firmware version: 0x%08x\n",
 		   fw_info.feature, fw_info.ver);
 
+	/* RLC SAVE RESTORE LIST CNTL */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLC feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLC SAVE RESTORE LIST GPM MEM */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLG feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLC SAVE RESTORE LIST SRM MEM */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLS feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
 	/* MEC */
 	query_fw.fw_type = AMDGPU_INFO_FW_GFX_MEC;
 	query_fw.index = 0;
@@ -1210,6 +1253,9 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	seq_printf(m, "VCN feature version: %u, firmware version: 0x%08x\n",
 		   fw_info.feature, fw_info.ver);
 
+
+	seq_printf(m, "VBIOS version: %s\n", ctx->vbios_version);
+
 	return 0;
 }
 

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c

@@ -36,12 +36,14 @@
 #include <drm/drm.h>
 
 #include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
 
 struct amdgpu_mn {
 	/* constant after initialisation */
 	struct amdgpu_device	*adev;
 	struct mm_struct	*mm;
 	struct mmu_notifier	mn;
+	enum amdgpu_mn_type	type;
 
 	/* only used on destruction */
 	struct work_struct	work;
@@ -185,7 +187,7 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
 }
 
 /**
- * amdgpu_mn_invalidate_range_start - callback to notify about mm change
+ * amdgpu_mn_invalidate_range_start_gfx - callback to notify about mm change
  *
  * @mn: our notifier
  * @mn: the mm this callback is about
@@ -195,10 +197,10 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
  * We block for all BOs between start and end to be idle and
  * unmap them by move them into system domain again.
  */
-static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn,
-					     struct mm_struct *mm,
-					     unsigned long start,
-					     unsigned long end)
+static void amdgpu_mn_invalidate_range_start_gfx(struct mmu_notifier *mn,
+						 struct mm_struct *mm,
+						 unsigned long start,
+						 unsigned long end)
 {
 	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
 	struct interval_tree_node *it;
@@ -219,6 +221,49 @@ static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn,
 	}
 }
 
+/**
+ * amdgpu_mn_invalidate_range_start_hsa - callback to notify about mm change
+ *
+ * @mn: our notifier
+ * @mn: the mm this callback is about
+ * @start: start of updated range
+ * @end: end of updated range
+ *
+ * We temporarily evict all BOs between start and end. This
+ * necessitates evicting all user-mode queues of the process. The BOs
+ * are restorted in amdgpu_mn_invalidate_range_end_hsa.
+ */
+static void amdgpu_mn_invalidate_range_start_hsa(struct mmu_notifier *mn,
+						 struct mm_struct *mm,
+						 unsigned long start,
+						 unsigned long end)
+{
+	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
+	struct interval_tree_node *it;
+
+	/* notification is exclusive, but interval is inclusive */
+	end -= 1;
+
+	amdgpu_mn_read_lock(rmn);
+
+	it = interval_tree_iter_first(&rmn->objects, start, end);
+	while (it) {
+		struct amdgpu_mn_node *node;
+		struct amdgpu_bo *bo;
+
+		node = container_of(it, struct amdgpu_mn_node, it);
+		it = interval_tree_iter_next(it, start, end);
+
+		list_for_each_entry(bo, &node->bos, mn_list) {
+			struct kgd_mem *mem = bo->kfd_bo;
+
+			if (amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm,
+							 start, end))
+				amdgpu_amdkfd_evict_userptr(mem, mm);
+		}
+	}
+}
+
 /**
  * amdgpu_mn_invalidate_range_end - callback to notify about mm change
  *
@@ -239,23 +284,39 @@ static void amdgpu_mn_invalidate_range_end(struct mmu_notifier *mn,
 	amdgpu_mn_read_unlock(rmn);
 }
 
-static const struct mmu_notifier_ops amdgpu_mn_ops = {
-	.release = amdgpu_mn_release,
-	.invalidate_range_start = amdgpu_mn_invalidate_range_start,
-	.invalidate_range_end = amdgpu_mn_invalidate_range_end,
+static const struct mmu_notifier_ops amdgpu_mn_ops[] = {
+	[AMDGPU_MN_TYPE_GFX] = {
+		.release = amdgpu_mn_release,
+		.invalidate_range_start = amdgpu_mn_invalidate_range_start_gfx,
+		.invalidate_range_end = amdgpu_mn_invalidate_range_end,
+	},
+	[AMDGPU_MN_TYPE_HSA] = {
+		.release = amdgpu_mn_release,
+		.invalidate_range_start = amdgpu_mn_invalidate_range_start_hsa,
+		.invalidate_range_end = amdgpu_mn_invalidate_range_end,
+	},
 };
 
+/* Low bits of any reasonable mm pointer will be unused due to struct
+ * alignment. Use these bits to make a unique key from the mm pointer
+ * and notifier type.
+ */
+#define AMDGPU_MN_KEY(mm, type) ((unsigned long)(mm) + (type))
+
 /**
  * amdgpu_mn_get - create notifier context
  *
  * @adev: amdgpu device pointer
+ * @type: type of MMU notifier context
  *
  * Creates a notifier context for current->mm.
  */
-struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
+struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev,
+				enum amdgpu_mn_type type)
 {
 	struct mm_struct *mm = current->mm;
 	struct amdgpu_mn *rmn;
+	unsigned long key = AMDGPU_MN_KEY(mm, type);
 	int r;
 
 	mutex_lock(&adev->mn_lock);
@@ -264,8 +325,8 @@ struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
 		return ERR_PTR(-EINTR);
 	}
 
-	hash_for_each_possible(adev->mn_hash, rmn, node, (unsigned long)mm)
-		if (rmn->mm == mm)
+	hash_for_each_possible(adev->mn_hash, rmn, node, key)
+		if (AMDGPU_MN_KEY(rmn->mm, rmn->type) == key)
 			goto release_locks;
 
 	rmn = kzalloc(sizeof(*rmn), GFP_KERNEL);
@@ -276,8 +337,9 @@ struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
 
 	rmn->adev = adev;
 	rmn->mm = mm;
-	rmn->mn.ops = &amdgpu_mn_ops;
 	init_rwsem(&rmn->lock);
+	rmn->type = type;
+	rmn->mn.ops = &amdgpu_mn_ops[type];
 	rmn->objects = RB_ROOT_CACHED;
 	mutex_init(&rmn->read_lock);
 	atomic_set(&rmn->recursion, 0);
@@ -286,7 +348,7 @@ struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
 	if (r)
 		goto free_rmn;
 
-	hash_add(adev->mn_hash, &rmn->node, (unsigned long)mm);
+	hash_add(adev->mn_hash, &rmn->node, AMDGPU_MN_KEY(mm, type));
 
 release_locks:
 	up_write(&mm->mmap_sem);
@@ -315,15 +377,21 @@ int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
 {
 	unsigned long end = addr + amdgpu_bo_size(bo) - 1;
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	enum amdgpu_mn_type type =
+		bo->kfd_bo ? AMDGPU_MN_TYPE_HSA : AMDGPU_MN_TYPE_GFX;
 	struct amdgpu_mn *rmn;
-	struct amdgpu_mn_node *node = NULL;
+	struct amdgpu_mn_node *node = NULL, *new_node;
 	struct list_head bos;
 	struct interval_tree_node *it;
 
-	rmn = amdgpu_mn_get(adev);
+	rmn = amdgpu_mn_get(adev, type);
 	if (IS_ERR(rmn))
 		return PTR_ERR(rmn);
 
+	new_node = kmalloc(sizeof(*new_node), GFP_KERNEL);
+	if (!new_node)
+		return -ENOMEM;
+
 	INIT_LIST_HEAD(&bos);
 
 	down_write(&rmn->lock);
@@ -337,13 +405,10 @@ int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
 		list_splice(&node->bos, &bos);
 	}
 
-	if (!node) {
-		node = kmalloc(sizeof(struct amdgpu_mn_node), GFP_KERNEL);
-		if (!node) {
-			up_write(&rmn->lock);
-			return -ENOMEM;
-		}
-	}
+	if (!node)
+		node = new_node;
+	else
+		kfree(new_node);
 
 	bo->mn = rmn;
 

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h

@@ -29,16 +29,23 @@
  */
 struct amdgpu_mn;
 
+enum amdgpu_mn_type {
+	AMDGPU_MN_TYPE_GFX,
+	AMDGPU_MN_TYPE_HSA,
+};
+
 #if defined(CONFIG_MMU_NOTIFIER)
 void amdgpu_mn_lock(struct amdgpu_mn *mn);
 void amdgpu_mn_unlock(struct amdgpu_mn *mn);
-struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev);
+struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev,
+				enum amdgpu_mn_type type);
 int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr);
 void amdgpu_mn_unregister(struct amdgpu_bo *bo);
 #else
 static inline void amdgpu_mn_lock(struct amdgpu_mn *mn) {}
 static inline void amdgpu_mn_unlock(struct amdgpu_mn *mn) {}
-static inline struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
+static inline struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev,
+					      enum amdgpu_mn_type type)
 {
 	return NULL;
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h

@@ -308,7 +308,6 @@ struct amdgpu_display_funcs {
 
 struct amdgpu_framebuffer {
 	struct drm_framebuffer base;
-	struct drm_gem_object *obj;
 
 	/* caching for later use */
 	uint64_t address;

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

@@ -191,14 +191,21 @@ int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
 			      u32 domain, struct amdgpu_bo **bo_ptr,
 			      u64 *gpu_addr, void **cpu_addr)
 {
+	struct amdgpu_bo_param bp;
 	bool free = false;
 	int r;
 
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = align;
+	bp.domain = domain;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+		AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+
 	if (!*bo_ptr) {
-		r = amdgpu_bo_create(adev, size, align, domain,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-				     ttm_bo_type_kernel, NULL, bo_ptr);
+		r = amdgpu_bo_create(adev, &bp, bo_ptr);
 		if (r) {
 			dev_err(adev->dev, "(%d) failed to allocate kernel bo\n",
 				r);
@@ -341,27 +348,25 @@ fail:
 	return false;
 }
 
-static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
-			       int byte_align, u32 domain,
-			       u64 flags, enum ttm_bo_type type,
-			       struct reservation_object *resv,
+static int amdgpu_bo_do_create(struct amdgpu_device *adev,
+			       struct amdgpu_bo_param *bp,
 			       struct amdgpu_bo **bo_ptr)
 {
 	struct ttm_operation_ctx ctx = {
-		.interruptible = (type != ttm_bo_type_kernel),
+		.interruptible = (bp->type != ttm_bo_type_kernel),
 		.no_wait_gpu = false,
-		.resv = resv,
+		.resv = bp->resv,
 		.flags = TTM_OPT_FLAG_ALLOW_RES_EVICT
 	};
 	struct amdgpu_bo *bo;
-	unsigned long page_align;
+	unsigned long page_align, size = bp->size;
 	size_t acc_size;
 	int r;
 
-	page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
+	page_align = roundup(bp->byte_align, PAGE_SIZE) >> PAGE_SHIFT;
 	size = ALIGN(size, PAGE_SIZE);
 
-	if (!amdgpu_bo_validate_size(adev, size, domain))
+	if (!amdgpu_bo_validate_size(adev, size, bp->domain))
 		return -ENOMEM;
 
 	*bo_ptr = NULL;
@@ -375,18 +380,14 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
 	drm_gem_private_object_init(adev->ddev, &bo->gem_base, size);
 	INIT_LIST_HEAD(&bo->shadow_list);
 	INIT_LIST_HEAD(&bo->va);
-	bo->preferred_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM |
-					 AMDGPU_GEM_DOMAIN_GTT |
-					 AMDGPU_GEM_DOMAIN_CPU |
-					 AMDGPU_GEM_DOMAIN_GDS |
-					 AMDGPU_GEM_DOMAIN_GWS |
-					 AMDGPU_GEM_DOMAIN_OA);
+	bo->preferred_domains = bp->preferred_domain ? bp->preferred_domain :
+		bp->domain;
 	bo->allowed_domains = bo->preferred_domains;
-	if (type != ttm_bo_type_kernel &&
+	if (bp->type != ttm_bo_type_kernel &&
 	    bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
 		bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
 
-	bo->flags = flags;
+	bo->flags = bp->flags;
 
 #ifdef CONFIG_X86_32
 	/* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit
@@ -417,11 +418,13 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
 #endif
 
 	bo->tbo.bdev = &adev->mman.bdev;
-	amdgpu_ttm_placement_from_domain(bo, domain);
+	amdgpu_ttm_placement_from_domain(bo, bp->domain);
+	if (bp->type == ttm_bo_type_kernel)
+		bo->tbo.priority = 1;
 
-	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, type,
+	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, bp->type,
 				 &bo->placement, page_align, &ctx, acc_size,
-				 NULL, resv, &amdgpu_ttm_bo_destroy);
+				 NULL, bp->resv, &amdgpu_ttm_bo_destroy);
 	if (unlikely(r != 0))
 		return r;
 
@@ -433,10 +436,7 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
 	else
 		amdgpu_cs_report_moved_bytes(adev, ctx.bytes_moved, 0);
 
-	if (type == ttm_bo_type_kernel)
-		bo->tbo.priority = 1;
-
-	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
+	if (bp->flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
 	    bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
 		struct dma_fence *fence;
 
@@ -449,20 +449,20 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
 		bo->tbo.moving = dma_fence_get(fence);
 		dma_fence_put(fence);
 	}
-	if (!resv)
+	if (!bp->resv)
 		amdgpu_bo_unreserve(bo);
 	*bo_ptr = bo;
 
 	trace_amdgpu_bo_create(bo);
 
 	/* Treat CPU_ACCESS_REQUIRED only as a hint if given by UMD */
-	if (type == ttm_bo_type_device)
+	if (bp->type == ttm_bo_type_device)
 		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
 
 	return 0;
 
 fail_unreserve:
-	if (!resv)
+	if (!bp->resv)
 		ww_mutex_unlock(&bo->tbo.resv->lock);
 	amdgpu_bo_unref(&bo);
 	return r;
@@ -472,16 +472,22 @@ static int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
 				   unsigned long size, int byte_align,
 				   struct amdgpu_bo *bo)
 {
+	struct amdgpu_bo_param bp;
 	int r;
 
 	if (bo->shadow)
 		return 0;
 
-	r = amdgpu_bo_do_create(adev, size, byte_align, AMDGPU_GEM_DOMAIN_GTT,
-				AMDGPU_GEM_CREATE_CPU_GTT_USWC |
-				AMDGPU_GEM_CREATE_SHADOW,
-				ttm_bo_type_kernel,
-				bo->tbo.resv, &bo->shadow);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = byte_align;
+	bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC |
+		AMDGPU_GEM_CREATE_SHADOW;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = bo->tbo.resv;
+
+	r = amdgpu_bo_do_create(adev, &bp, &bo->shadow);
 	if (!r) {
 		bo->shadow->parent = amdgpu_bo_ref(bo);
 		mutex_lock(&adev->shadow_list_lock);
@@ -492,28 +498,26 @@ static int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
 	return r;
 }
 
-int amdgpu_bo_create(struct amdgpu_device *adev, unsigned long size,
-		     int byte_align, u32 domain,
-		     u64 flags, enum ttm_bo_type type,
-		     struct reservation_object *resv,
+int amdgpu_bo_create(struct amdgpu_device *adev,
+		     struct amdgpu_bo_param *bp,
 		     struct amdgpu_bo **bo_ptr)
 {
-	uint64_t parent_flags = flags & ~AMDGPU_GEM_CREATE_SHADOW;
+	u64 flags = bp->flags;
 	int r;
 
-	r = amdgpu_bo_do_create(adev, size, byte_align, domain,
-				parent_flags, type, resv, bo_ptr);
+	bp->flags = bp->flags & ~AMDGPU_GEM_CREATE_SHADOW;
+	r = amdgpu_bo_do_create(adev, bp, bo_ptr);
 	if (r)
 		return r;
 
 	if ((flags & AMDGPU_GEM_CREATE_SHADOW) && amdgpu_need_backup(adev)) {
-		if (!resv)
+		if (!bp->resv)
 			WARN_ON(reservation_object_lock((*bo_ptr)->tbo.resv,
 							NULL));
 
-		r = amdgpu_bo_create_shadow(adev, size, byte_align, (*bo_ptr));
+		r = amdgpu_bo_create_shadow(adev, bp->size, bp->byte_align, (*bo_ptr));
 
-		if (!resv)
+		if (!bp->resv)
 			reservation_object_unlock((*bo_ptr)->tbo.resv);
 
 		if (r)
@@ -689,8 +693,21 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 		return -EINVAL;
 
 	/* A shared bo cannot be migrated to VRAM */
-	if (bo->prime_shared_count && (domain == AMDGPU_GEM_DOMAIN_VRAM))
-		return -EINVAL;
+	if (bo->prime_shared_count) {
+		if (domain & AMDGPU_GEM_DOMAIN_GTT)
+			domain = AMDGPU_GEM_DOMAIN_GTT;
+		else
+			return -EINVAL;
+	}
+
+	/* This assumes only APU display buffers are pinned with (VRAM|GTT).
+	 * See function amdgpu_display_supported_domains()
+	 */
+	if (domain == (AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT)) {
+		domain = AMDGPU_GEM_DOMAIN_VRAM;
+		if (adev->gmc.real_vram_size <= AMDGPU_SG_THRESHOLD)
+			domain = AMDGPU_GEM_DOMAIN_GTT;
+	}
 
 	if (bo->pin_count) {
 		uint32_t mem_type = bo->tbo.mem.mem_type;
@@ -838,6 +855,13 @@ int amdgpu_bo_init(struct amdgpu_device *adev)
 	return amdgpu_ttm_init(adev);
 }
 
+int amdgpu_bo_late_init(struct amdgpu_device *adev)
+{
+	amdgpu_ttm_late_init(adev);
+
+	return 0;
+}
+
 void amdgpu_bo_fini(struct amdgpu_device *adev)
 {
 	amdgpu_ttm_fini(adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_object.h

@@ -33,6 +33,16 @@
 
 #define AMDGPU_BO_INVALID_OFFSET	LONG_MAX
 
+struct amdgpu_bo_param {
+	unsigned long			size;
+	int				byte_align;
+	u32				domain;
+	u32				preferred_domain;
+	u64				flags;
+	enum ttm_bo_type		type;
+	struct reservation_object	*resv;
+};
+
 /* bo virtual addresses in a vm */
 struct amdgpu_bo_va_mapping {
 	struct amdgpu_bo_va		*bo_va;
@@ -195,6 +205,27 @@ static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
 	}
 }
 
+/**
+ * amdgpu_bo_in_cpu_visible_vram - check if BO is (partly) in visible VRAM
+ */
+static inline bool amdgpu_bo_in_cpu_visible_vram(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
+	struct drm_mm_node *node = bo->tbo.mem.mm_node;
+	unsigned long pages_left;
+
+	if (bo->tbo.mem.mem_type != TTM_PL_VRAM)
+		return false;
+
+	for (pages_left = bo->tbo.mem.num_pages; pages_left;
+	     pages_left -= node->size, node++)
+		if (node->start < fpfn)
+			return true;
+
+	return false;
+}
+
 /**
  * amdgpu_bo_explicit_sync - return whether the bo is explicitly synced
  */
@@ -203,10 +234,8 @@ static inline bool amdgpu_bo_explicit_sync(struct amdgpu_bo *bo)
 	return bo->flags & AMDGPU_GEM_CREATE_EXPLICIT_SYNC;
 }
 
-int amdgpu_bo_create(struct amdgpu_device *adev, unsigned long size,
-		     int byte_align, u32 domain,
-		     u64 flags, enum ttm_bo_type type,
-		     struct reservation_object *resv,
+int amdgpu_bo_create(struct amdgpu_device *adev,
+		     struct amdgpu_bo_param *bp,
 		     struct amdgpu_bo **bo_ptr);
 int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
 			      unsigned long size, int align,
@@ -230,6 +259,7 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 int amdgpu_bo_unpin(struct amdgpu_bo *bo);
 int amdgpu_bo_evict_vram(struct amdgpu_device *adev);
 int amdgpu_bo_init(struct amdgpu_device *adev);
+int amdgpu_bo_late_init(struct amdgpu_device *adev);
 void amdgpu_bo_fini(struct amdgpu_device *adev);
 int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
 				struct vm_area_struct *vma);

drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c

@@ -77,6 +77,37 @@ void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev)
 	}
 }
 
+/**
+ * DOC: power_dpm_state
+ *
+ * This is a legacy interface and is only provided for backwards compatibility.
+ * The amdgpu driver provides a sysfs API for adjusting certain power
+ * related parameters.  The file power_dpm_state is used for this.
+ * It accepts the following arguments:
+ * - battery
+ * - balanced
+ * - performance
+ *
+ * battery
+ *
+ * On older GPUs, the vbios provided a special power state for battery
+ * operation.  Selecting battery switched to this state.  This is no
+ * longer provided on newer GPUs so the option does nothing in that case.
+ *
+ * balanced
+ *
+ * On older GPUs, the vbios provided a special power state for balanced
+ * operation.  Selecting balanced switched to this state.  This is no
+ * longer provided on newer GPUs so the option does nothing in that case.
+ *
+ * performance
+ *
+ * On older GPUs, the vbios provided a special power state for performance
+ * operation.  Selecting performance switched to this state.  This is no
+ * longer provided on newer GPUs so the option does nothing in that case.
+ *
+ */
+
 static ssize_t amdgpu_get_dpm_state(struct device *dev,
 				    struct device_attribute *attr,
 				    char *buf)
@@ -131,6 +162,59 @@ fail:
 	return count;
 }
 
+
+/**
+ * DOC: power_dpm_force_performance_level
+ *
+ * The amdgpu driver provides a sysfs API for adjusting certain power
+ * related parameters.  The file power_dpm_force_performance_level is
+ * used for this.  It accepts the following arguments:
+ * - auto
+ * - low
+ * - high
+ * - manual
+ * - GPU fan
+ * - profile_standard
+ * - profile_min_sclk
+ * - profile_min_mclk
+ * - profile_peak
+ *
+ * auto
+ *
+ * When auto is selected, the driver will attempt to dynamically select
+ * the optimal power profile for current conditions in the driver.
+ *
+ * low
+ *
+ * When low is selected, the clocks are forced to the lowest power state.
+ *
+ * high
+ *
+ * When high is selected, the clocks are forced to the highest power state.
+ *
+ * manual
+ *
+ * When manual is selected, the user can manually adjust which power states
+ * are enabled for each clock domain via the sysfs pp_dpm_mclk, pp_dpm_sclk,
+ * and pp_dpm_pcie files and adjust the power state transition heuristics
+ * via the pp_power_profile_mode sysfs file.
+ *
+ * profile_standard
+ * profile_min_sclk
+ * profile_min_mclk
+ * profile_peak
+ *
+ * When the profiling modes are selected, clock and power gating are
+ * disabled and the clocks are set for different profiling cases. This
+ * mode is recommended for profiling specific work loads where you do
+ * not want clock or power gating for clock fluctuation to interfere
+ * with your results. profile_standard sets the clocks to a fixed clock
+ * level which varies from asic to asic.  profile_min_sclk forces the sclk
+ * to the lowest level.  profile_min_mclk forces the mclk to the lowest level.
+ * profile_peak sets all clocks (mclk, sclk, pcie) to the highest levels.
+ *
+ */
+
 static ssize_t amdgpu_get_dpm_forced_performance_level(struct device *dev,
 						struct device_attribute *attr,
 								char *buf)
@@ -324,6 +408,17 @@ fail:
 	return count;
 }
 
+/**
+ * DOC: pp_table
+ *
+ * The amdgpu driver provides a sysfs API for uploading new powerplay
+ * tables.  The file pp_table is used for this.  Reading the file
+ * will dump the current power play table.  Writing to the file
+ * will attempt to upload a new powerplay table and re-initialize
+ * powerplay using that new table.
+ *
+ */
+
 static ssize_t amdgpu_get_pp_table(struct device *dev,
 		struct device_attribute *attr,
 		char *buf)
@@ -360,6 +455,29 @@ static ssize_t amdgpu_set_pp_table(struct device *dev,
 	return count;
 }
 
+/**
+ * DOC: pp_od_clk_voltage
+ *
+ * The amdgpu driver provides a sysfs API for adjusting the clocks and voltages
+ * in each power level within a power state.  The pp_od_clk_voltage is used for
+ * this.
+ *
+ * Reading the file will display:
+ * - a list of engine clock levels and voltages labeled OD_SCLK
+ * - a list of memory clock levels and voltages labeled OD_MCLK
+ * - a list of valid ranges for sclk, mclk, and voltage labeled OD_RANGE
+ *
+ * To manually adjust these settings, first select manual using
+ * power_dpm_force_performance_level. Enter a new value for each
+ * level by writing a string that contains "s/m level clock voltage" to
+ * the file.  E.g., "s 1 500 820" will update sclk level 1 to be 500 MHz
+ * at 820 mV; "m 0 350 810" will update mclk level 0 to be 350 MHz at
+ * 810 mV.  When you have edited all of the states as needed, write
+ * "c" (commit) to the file to commit your changes.  If you want to reset to the
+ * default power levels, write "r" (reset) to the file to reset them.
+ *
+ */
+
 static ssize_t amdgpu_set_pp_od_clk_voltage(struct device *dev,
 		struct device_attribute *attr,
 		const char *buf,
@@ -437,6 +555,7 @@ static ssize_t amdgpu_get_pp_od_clk_voltage(struct device *dev,
 	if (adev->powerplay.pp_funcs->print_clock_levels) {
 		size = amdgpu_dpm_print_clock_levels(adev, OD_SCLK, buf);
 		size += amdgpu_dpm_print_clock_levels(adev, OD_MCLK, buf+size);
+		size += amdgpu_dpm_print_clock_levels(adev, OD_RANGE, buf+size);
 		return size;
 	} else {
 		return snprintf(buf, PAGE_SIZE, "\n");
@@ -444,6 +563,23 @@ static ssize_t amdgpu_get_pp_od_clk_voltage(struct device *dev,
 
 }
 
+/**
+ * DOC: pp_dpm_sclk pp_dpm_mclk pp_dpm_pcie
+ *
+ * The amdgpu driver provides a sysfs API for adjusting what power levels
+ * are enabled for a given power state.  The files pp_dpm_sclk, pp_dpm_mclk,
+ * and pp_dpm_pcie are used for this.
+ *
+ * Reading back the files will show you the available power levels within
+ * the power state and the clock information for those levels.
+ *
+ * To manually adjust these states, first select manual using
+ * power_dpm_force_performance_level.
+ * Secondly,Enter a new value for each level by inputing a string that
+ * contains " echo xx xx xx > pp_dpm_sclk/mclk/pcie"
+ * E.g., echo 4 5 6 to > pp_dpm_sclk will enable sclk levels 4, 5, and 6.
+ */
+
 static ssize_t amdgpu_get_pp_dpm_sclk(struct device *dev,
 		struct device_attribute *attr,
 		char *buf)
@@ -466,23 +602,27 @@ static ssize_t amdgpu_set_pp_dpm_sclk(struct device *dev,
 	struct amdgpu_device *adev = ddev->dev_private;
 	int ret;
 	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
+	uint32_t mask = 0;
+	char *sub_str = NULL;
+	char *tmp;
+	char buf_cpy[count];
+	const char delimiter[3] = {' ', '\n', '\0'};
 
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
+	memcpy(buf_cpy, buf, count+1);
+	tmp = buf_cpy;
+	while (tmp[0]) {
+		sub_str =  strsep(&tmp, delimiter);
+		if (strlen(sub_str)) {
+			ret = kstrtol(sub_str, 0, &level);
 
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
+			if (ret) {
+				count = -EINVAL;
+				goto fail;
+			}
+			mask |= 1 << level;
+		} else
+			break;
 	}
-
 	if (adev->powerplay.pp_funcs->force_clock_level)
 		amdgpu_dpm_force_clock_level(adev, PP_SCLK, mask);
 
@@ -512,21 +652,26 @@ static ssize_t amdgpu_set_pp_dpm_mclk(struct device *dev,
 	struct amdgpu_device *adev = ddev->dev_private;
 	int ret;
 	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
+	uint32_t mask = 0;
+	char *sub_str = NULL;
+	char *tmp;
+	char buf_cpy[count];
+	const char delimiter[3] = {' ', '\n', '\0'};
 
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
+	memcpy(buf_cpy, buf, count+1);
+	tmp = buf_cpy;
+	while (tmp[0]) {
+		sub_str =  strsep(&tmp, delimiter);
+		if (strlen(sub_str)) {
+			ret = kstrtol(sub_str, 0, &level);
 
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
+			if (ret) {
+				count = -EINVAL;
+				goto fail;
+			}
+			mask |= 1 << level;
+		} else
+			break;
 	}
 	if (adev->powerplay.pp_funcs->force_clock_level)
 		amdgpu_dpm_force_clock_level(adev, PP_MCLK, mask);
@@ -557,21 +702,27 @@ static ssize_t amdgpu_set_pp_dpm_pcie(struct device *dev,
 	struct amdgpu_device *adev = ddev->dev_private;
 	int ret;
 	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
+	uint32_t mask = 0;
+	char *sub_str = NULL;
+	char *tmp;
+	char buf_cpy[count];
+	const char delimiter[3] = {' ', '\n', '\0'};
 
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
+	memcpy(buf_cpy, buf, count+1);
+	tmp = buf_cpy;
 
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
+	while (tmp[0]) {
+		sub_str =  strsep(&tmp, delimiter);
+		if (strlen(sub_str)) {
+			ret = kstrtol(sub_str, 0, &level);
+
+			if (ret) {
+				count = -EINVAL;
+				goto fail;
+			}
+			mask |= 1 << level;
+		} else
+			break;
 	}
 	if (adev->powerplay.pp_funcs->force_clock_level)
 		amdgpu_dpm_force_clock_level(adev, PP_PCIE, mask);
@@ -668,6 +819,26 @@ fail:
 	return count;
 }
 
+/**
+ * DOC: pp_power_profile_mode
+ *
+ * The amdgpu driver provides a sysfs API for adjusting the heuristics
+ * related to switching between power levels in a power state.  The file
+ * pp_power_profile_mode is used for this.
+ *
+ * Reading this file outputs a list of all of the predefined power profiles
+ * and the relevant heuristics settings for that profile.
+ *
+ * To select a profile or create a custom profile, first select manual using
+ * power_dpm_force_performance_level.  Writing the number of a predefined
+ * profile to pp_power_profile_mode will enable those heuristics.  To
+ * create a custom set of heuristics, write a string of numbers to the file
+ * starting with the number of the custom profile along with a setting
+ * for each heuristic parameter.  Due to differences across asic families
+ * the heuristic parameters vary from family to family.
+ *
+ */
+
 static ssize_t amdgpu_get_pp_power_profile_mode(struct device *dev,
 		struct device_attribute *attr,
 		char *buf)
@@ -1020,8 +1191,8 @@ static ssize_t amdgpu_hwmon_show_power_avg(struct device *dev,
 {
 	struct amdgpu_device *adev = dev_get_drvdata(dev);
 	struct drm_device *ddev = adev->ddev;
-	struct pp_gpu_power query = {0};
-	int r, size = sizeof(query);
+	u32 query = 0;
+	int r, size = sizeof(u32);
 	unsigned uw;
 
 	/* Can't get power when the card is off */
@@ -1041,7 +1212,7 @@ static ssize_t amdgpu_hwmon_show_power_avg(struct device *dev,
 		return r;
 
 	/* convert to microwatts */
-	uw = (query.average_gpu_power >> 8) * 1000000;
+	uw = (query >> 8) * 1000000 + (query & 0xff) * 1000;
 
 	return snprintf(buf, PAGE_SIZE, "%u\n", uw);
 }
@@ -1109,6 +1280,46 @@ static ssize_t amdgpu_hwmon_set_power_cap(struct device *dev,
 	return count;
 }
 
+
+/**
+ * DOC: hwmon
+ *
+ * The amdgpu driver exposes the following sensor interfaces:
+ * - GPU temperature (via the on-die sensor)
+ * - GPU voltage
+ * - Northbridge voltage (APUs only)
+ * - GPU power
+ * - GPU fan
+ *
+ * hwmon interfaces for GPU temperature:
+ * - temp1_input: the on die GPU temperature in millidegrees Celsius
+ * - temp1_crit: temperature critical max value in millidegrees Celsius
+ * - temp1_crit_hyst: temperature hysteresis for critical limit in millidegrees Celsius
+ *
+ * hwmon interfaces for GPU voltage:
+ * - in0_input: the voltage on the GPU in millivolts
+ * - in1_input: the voltage on the Northbridge in millivolts
+ *
+ * hwmon interfaces for GPU power:
+ * - power1_average: average power used by the GPU in microWatts
+ * - power1_cap_min: minimum cap supported in microWatts
+ * - power1_cap_max: maximum cap supported in microWatts
+ * - power1_cap: selected power cap in microWatts
+ *
+ * hwmon interfaces for GPU fan:
+ * - pwm1: pulse width modulation fan level (0-255)
+ * - pwm1_enable: pulse width modulation fan control method
+ *                0: no fan speed control
+ *                1: manual fan speed control using pwm interface
+ *                2: automatic fan speed control
+ * - pwm1_min: pulse width modulation fan control minimum level (0)
+ * - pwm1_max: pulse width modulation fan control maximum level (255)
+ * - fan1_input: fan speed in RPM
+ *
+ * You can use hwmon tools like sensors to view this information on your system.
+ *
+ */
+
 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO, amdgpu_hwmon_show_temp, NULL, 0);
 static SENSOR_DEVICE_ATTR(temp1_crit, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 0);
 static SENSOR_DEVICE_ATTR(temp1_crit_hyst, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 1);
@@ -1153,19 +1364,14 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj,
 	struct amdgpu_device *adev = dev_get_drvdata(dev);
 	umode_t effective_mode = attr->mode;
 
-	/* handle non-powerplay limitations */
-	if (!adev->powerplay.pp_handle) {
-		/* Skip fan attributes if fan is not present */
-		if (adev->pm.no_fan &&
-		    (attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
-		     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
-		     attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
-		     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr))
-			return 0;
-		/* requires powerplay */
-		if (attr == &sensor_dev_attr_fan1_input.dev_attr.attr)
-			return 0;
-	}
+
+	/* Skip fan attributes if fan is not present */
+	if (adev->pm.no_fan && (attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
+	    attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
+	    attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
+	    attr == &sensor_dev_attr_pwm1_min.dev_attr.attr ||
+	    attr == &sensor_dev_attr_fan1_input.dev_attr.attr))
+		return 0;
 
 	/* Skip limit attributes if DPM is not enabled */
 	if (!adev->pm.dpm_enabled &&
@@ -1658,9 +1864,6 @@ void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev)
 
 void amdgpu_pm_compute_clocks(struct amdgpu_device *adev)
 {
-	struct drm_device *ddev = adev->ddev;
-	struct drm_crtc *crtc;
-	struct amdgpu_crtc *amdgpu_crtc;
 	int i = 0;
 
 	if (!adev->pm.dpm_enabled)
@@ -1676,21 +1879,25 @@ void amdgpu_pm_compute_clocks(struct amdgpu_device *adev)
 	}
 
 	if (adev->powerplay.pp_funcs->dispatch_tasks) {
+		if (!amdgpu_device_has_dc_support(adev)) {
+			mutex_lock(&adev->pm.mutex);
+			amdgpu_dpm_get_active_displays(adev);
+			adev->pm.pm_display_cfg.num_display = adev->pm.dpm.new_active_crtcs;
+			adev->pm.pm_display_cfg.vrefresh = amdgpu_dpm_get_vrefresh(adev);
+			adev->pm.pm_display_cfg.min_vblank_time = amdgpu_dpm_get_vblank_time(adev);
+			/* we have issues with mclk switching with refresh rates over 120 hz on the non-DC code. */
+			if (adev->pm.pm_display_cfg.vrefresh > 120)
+				adev->pm.pm_display_cfg.min_vblank_time = 0;
+			if (adev->powerplay.pp_funcs->display_configuration_change)
+				adev->powerplay.pp_funcs->display_configuration_change(
+								adev->powerplay.pp_handle,
+								&adev->pm.pm_display_cfg);
+			mutex_unlock(&adev->pm.mutex);
+		}
 		amdgpu_dpm_dispatch_task(adev, AMD_PP_TASK_DISPLAY_CONFIG_CHANGE, NULL);
 	} else {
 		mutex_lock(&adev->pm.mutex);
-		adev->pm.dpm.new_active_crtcs = 0;
-		adev->pm.dpm.new_active_crtc_count = 0;
-		if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-			list_for_each_entry(crtc,
-					    &ddev->mode_config.crtc_list, head) {
-				amdgpu_crtc = to_amdgpu_crtc(crtc);
-				if (amdgpu_crtc->enabled) {
-					adev->pm.dpm.new_active_crtcs |= (1 << amdgpu_crtc->crtc_id);
-					adev->pm.dpm.new_active_crtc_count++;
-				}
-			}
-		}
+		amdgpu_dpm_get_active_displays(adev);
 		/* update battery/ac status */
 		if (power_supply_is_system_supplied() > 0)
 			adev->pm.dpm.ac_power = true;
@@ -1711,7 +1918,7 @@ void amdgpu_pm_compute_clocks(struct amdgpu_device *adev)
 static int amdgpu_debugfs_pm_info_pp(struct seq_file *m, struct amdgpu_device *adev)
 {
 	uint32_t value;
-	struct pp_gpu_power query = {0};
+	uint32_t query = 0;
 	int size;
 
 	/* sanity check PP is enabled */
@@ -1734,17 +1941,9 @@ static int amdgpu_debugfs_pm_info_pp(struct seq_file *m, struct amdgpu_device *a
 		seq_printf(m, "\t%u mV (VDDGFX)\n", value);
 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VDDNB, (void *)&value, &size))
 		seq_printf(m, "\t%u mV (VDDNB)\n", value);
-	size = sizeof(query);
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_POWER, (void *)&query, &size)) {
-		seq_printf(m, "\t%u.%u W (VDDC)\n", query.vddc_power >> 8,
-				query.vddc_power & 0xff);
-		seq_printf(m, "\t%u.%u W (VDDCI)\n", query.vddci_power >> 8,
-				query.vddci_power & 0xff);
-		seq_printf(m, "\t%u.%u W (max GPU)\n", query.max_gpu_power >> 8,
-				query.max_gpu_power & 0xff);
-		seq_printf(m, "\t%u.%u W (average GPU)\n", query.average_gpu_power >> 8,
-				query.average_gpu_power & 0xff);
-	}
+	size = sizeof(uint32_t);
+	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_POWER, (void *)&query, &size))
+		seq_printf(m, "\t%u.%u W (average GPU)\n", query >> 8, query & 0xff);
 	size = sizeof(value);
 	seq_printf(m, "\n");
 

drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c

@@ -102,12 +102,18 @@ amdgpu_gem_prime_import_sg_table(struct drm_device *dev,
 	struct reservation_object *resv = attach->dmabuf->resv;
 	struct amdgpu_device *adev = dev->dev_private;
 	struct amdgpu_bo *bo;
+	struct amdgpu_bo_param bp;
 	int ret;
 
+	memset(&bp, 0, sizeof(bp));
+	bp.size = attach->dmabuf->size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_CPU;
+	bp.flags = 0;
+	bp.type = ttm_bo_type_sg;
+	bp.resv = resv;
 	ww_mutex_lock(&resv->lock, NULL);
-	ret = amdgpu_bo_create(adev, attach->dmabuf->size, PAGE_SIZE,
-			       AMDGPU_GEM_DOMAIN_CPU, 0, ttm_bo_type_sg,
-			       resv, &bo);
+	ret = amdgpu_bo_create(adev, &bp, &bo);
 	if (ret)
 		goto error;
 
@@ -209,7 +215,7 @@ static int amdgpu_gem_begin_cpu_access(struct dma_buf *dma_buf,
 	struct amdgpu_bo *bo = gem_to_amdgpu_bo(dma_buf->priv);
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
 	struct ttm_operation_ctx ctx = { true, false };
-	u32 domain = amdgpu_display_framebuffer_domains(adev);
+	u32 domain = amdgpu_display_supported_domains(adev);
 	int ret;
 	bool reads = (direction == DMA_BIDIRECTIONAL ||
 		      direction == DMA_FROM_DEVICE);

drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c

@@ -52,6 +52,7 @@ static int psp_sw_init(void *handle)
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 		psp_v3_1_set_psp_funcs(psp);
 		break;
 	case CHIP_RAVEN:

drivers/gpu/drm/amd/amdgpu/amdgpu_queue_mgr.c

@@ -66,6 +66,8 @@ static int amdgpu_identity_map(struct amdgpu_device *adev,
 			       u32 ring,
 			       struct amdgpu_ring **out_ring)
 {
+	u32 instance;
+
 	switch (mapper->hw_ip) {
 	case AMDGPU_HW_IP_GFX:
 		*out_ring = &adev->gfx.gfx_ring[ring];
@@ -77,13 +79,16 @@ static int amdgpu_identity_map(struct amdgpu_device *adev,
 		*out_ring = &adev->sdma.instance[ring].ring;
 		break;
 	case AMDGPU_HW_IP_UVD:
-		*out_ring = &adev->uvd.ring;
+		instance = ring;
+		*out_ring = &adev->uvd.inst[instance].ring;
 		break;
 	case AMDGPU_HW_IP_VCE:
 		*out_ring = &adev->vce.ring[ring];
 		break;
 	case AMDGPU_HW_IP_UVD_ENC:
-		*out_ring = &adev->uvd.ring_enc[ring];
+		instance = ring / adev->uvd.num_enc_rings;
+		*out_ring =
+		&adev->uvd.inst[instance].ring_enc[ring%adev->uvd.num_enc_rings];
 		break;
 	case AMDGPU_HW_IP_VCN_DEC:
 		*out_ring = &adev->vcn.ring_dec;
@@ -240,13 +245,14 @@ int amdgpu_queue_mgr_map(struct amdgpu_device *adev,
 		ip_num_rings = adev->sdma.num_instances;
 		break;
 	case AMDGPU_HW_IP_UVD:
-		ip_num_rings = 1;
+		ip_num_rings = adev->uvd.num_uvd_inst;
 		break;
 	case AMDGPU_HW_IP_VCE:
 		ip_num_rings = adev->vce.num_rings;
 		break;
 	case AMDGPU_HW_IP_UVD_ENC:
-		ip_num_rings = adev->uvd.num_enc_rings;
+		ip_num_rings =
+			adev->uvd.num_enc_rings * adev->uvd.num_uvd_inst;
 		break;
 	case AMDGPU_HW_IP_VCN_DEC:
 		ip_num_rings = 1;

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c

@@ -362,6 +362,7 @@ void amdgpu_ring_fini(struct amdgpu_ring *ring)
 
 	dma_fence_put(ring->vmid_wait);
 	ring->vmid_wait = NULL;
+	ring->me = 0;
 
 	ring->adev->rings[ring->idx] = NULL;
 }
@@ -459,6 +460,26 @@ void amdgpu_ring_lru_touch(struct amdgpu_device *adev, struct amdgpu_ring *ring)
 	spin_unlock(&adev->ring_lru_list_lock);
 }
 
+/**
+ * amdgpu_ring_emit_reg_write_reg_wait_helper - ring helper
+ *
+ * @adev: amdgpu_device pointer
+ * @reg0: register to write
+ * @reg1: register to wait on
+ * @ref: reference value to write/wait on
+ * @mask: mask to wait on
+ *
+ * Helper for rings that don't support write and wait in a
+ * single oneshot packet.
+ */
+void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
+						uint32_t reg0, uint32_t reg1,
+						uint32_t ref, uint32_t mask)
+{
+	amdgpu_ring_emit_wreg(ring, reg0, ref);
+	amdgpu_ring_emit_reg_wait(ring, reg1, mask, mask);
+}
+
 /*
  * Debugfs info
  */

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h

@@ -29,7 +29,7 @@
 #include <drm/drm_print.h>
 
 /* max number of rings */
-#define AMDGPU_MAX_RINGS		18
+#define AMDGPU_MAX_RINGS		21
 #define AMDGPU_MAX_GFX_RINGS		1
 #define AMDGPU_MAX_COMPUTE_RINGS	8
 #define AMDGPU_MAX_VCE_RINGS		3
@@ -42,6 +42,7 @@
 
 #define AMDGPU_FENCE_FLAG_64BIT         (1 << 0)
 #define AMDGPU_FENCE_FLAG_INT           (1 << 1)
+#define AMDGPU_FENCE_FLAG_TC_WB_ONLY    (1 << 2)
 
 enum amdgpu_ring_type {
 	AMDGPU_RING_TYPE_GFX,
@@ -90,7 +91,8 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 				   unsigned irq_type);
 void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
 void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
-int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence);
+int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence,
+		      unsigned flags);
 int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s);
 void amdgpu_fence_process(struct amdgpu_ring *ring);
 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
@@ -154,6 +156,9 @@ struct amdgpu_ring_funcs {
 	void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
 	void (*emit_reg_wait)(struct amdgpu_ring *ring, uint32_t reg,
 			      uint32_t val, uint32_t mask);
+	void (*emit_reg_write_reg_wait)(struct amdgpu_ring *ring,
+					uint32_t reg0, uint32_t reg1,
+					uint32_t ref, uint32_t mask);
 	void (*emit_tmz)(struct amdgpu_ring *ring, bool start);
 	/* priority functions */
 	void (*set_priority) (struct amdgpu_ring *ring,
@@ -228,6 +233,10 @@ int amdgpu_ring_lru_get(struct amdgpu_device *adev, int type,
 			int *blacklist, int num_blacklist,
 			bool lru_pipe_order, struct amdgpu_ring **ring);
 void amdgpu_ring_lru_touch(struct amdgpu_device *adev, struct amdgpu_ring *ring);
+void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
+						uint32_t reg0, uint32_t val0,
+						uint32_t reg1, uint32_t val1);
+
 static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
 {
 	int i = 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_test.c

@@ -33,6 +33,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
 	struct amdgpu_bo *vram_obj = NULL;
 	struct amdgpu_bo **gtt_obj = NULL;
+	struct amdgpu_bo_param bp;
 	uint64_t gart_addr, vram_addr;
 	unsigned n, size;
 	int i, r;
@@ -58,9 +59,15 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 		r = 1;
 		goto out_cleanup;
 	}
-
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, 0,
-			     ttm_bo_type_kernel, NULL, &vram_obj);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = 0;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+
+	r = amdgpu_bo_create(adev, &bp, &vram_obj);
 	if (r) {
 		DRM_ERROR("Failed to create VRAM object\n");
 		goto out_cleanup;
@@ -79,9 +86,8 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 		void **vram_start, **vram_end;
 		struct dma_fence *fence = NULL;
 
-		r = amdgpu_bo_create(adev, size, PAGE_SIZE,
-				     AMDGPU_GEM_DOMAIN_GTT, 0,
-				     ttm_bo_type_kernel, NULL, gtt_obj + i);
+		bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+		r = amdgpu_bo_create(adev, &bp, gtt_obj + i);
 		if (r) {
 			DRM_ERROR("Failed to create GTT object %d\n", i);
 			goto out_lclean;

drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h

@@ -275,7 +275,7 @@ TRACE_EVENT(amdgpu_vm_bo_unmap,
 			     ),
 
 	    TP_fast_assign(
-			   __entry->bo = bo_va->base.bo;
+			   __entry->bo = bo_va ? bo_va->base.bo : NULL;
 			   __entry->start = mapping->start;
 			   __entry->last = mapping->last;
 			   __entry->offset = mapping->offset;

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

@@ -63,16 +63,44 @@ static void amdgpu_ttm_debugfs_fini(struct amdgpu_device *adev);
 /*
  * Global memory.
  */
+
+/**
+ * amdgpu_ttm_mem_global_init - Initialize and acquire reference to
+ * memory object
+ *
+ * @ref: Object for initialization.
+ *
+ * This is called by drm_global_item_ref() when an object is being
+ * initialized.
+ */
 static int amdgpu_ttm_mem_global_init(struct drm_global_reference *ref)
 {
 	return ttm_mem_global_init(ref->object);
 }
 
+/**
+ * amdgpu_ttm_mem_global_release - Drop reference to a memory object
+ *
+ * @ref: Object being removed
+ *
+ * This is called by drm_global_item_unref() when an object is being
+ * released.
+ */
 static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref)
 {
 	ttm_mem_global_release(ref->object);
 }
 
+/**
+ * amdgpu_ttm_global_init - Initialize global TTM memory reference
+ * 							structures.
+ *
+ * @adev:  	AMDGPU device for which the global structures need to be
+ *			registered.
+ *
+ * This is called as part of the AMDGPU ttm init from amdgpu_ttm_init()
+ * during bring up.
+ */
 static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
 {
 	struct drm_global_reference *global_ref;
@@ -80,7 +108,9 @@ static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
 	struct drm_sched_rq *rq;
 	int r;
 
+	/* ensure reference is false in case init fails */
 	adev->mman.mem_global_referenced = false;
+
 	global_ref = &adev->mman.mem_global_ref;
 	global_ref->global_type = DRM_GLOBAL_TTM_MEM;
 	global_ref->size = sizeof(struct ttm_mem_global);
@@ -111,7 +141,7 @@ static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
 	ring = adev->mman.buffer_funcs_ring;
 	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
 	r = drm_sched_entity_init(&ring->sched, &adev->mman.entity,
-				  rq, amdgpu_sched_jobs, NULL);
+				  rq, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up TTM BO move run queue.\n");
 		goto error_entity;
@@ -146,6 +176,18 @@ static int amdgpu_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
 	return 0;
 }
 
+/**
+ * amdgpu_init_mem_type - 	Initialize a memory manager for a specific
+ * 							type of memory request.
+ *
+ * @bdev:	The TTM BO device object (contains a reference to
+ * 			amdgpu_device)
+ * @type:	The type of memory requested
+ * @man:
+ *
+ * This is called by ttm_bo_init_mm() when a buffer object is being
+ * initialized.
+ */
 static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 				struct ttm_mem_type_manager *man)
 {
@@ -161,6 +203,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		break;
 	case TTM_PL_TT:
+		/* GTT memory  */
 		man->func = &amdgpu_gtt_mgr_func;
 		man->gpu_offset = adev->gmc.gart_start;
 		man->available_caching = TTM_PL_MASK_CACHING;
@@ -193,6 +236,14 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	return 0;
 }
 
+/**
+ * amdgpu_evict_flags - Compute placement flags
+ *
+ * @bo: The buffer object to evict
+ * @placement: Possible destination(s) for evicted BO
+ *
+ * Fill in placement data when ttm_bo_evict() is called
+ */
 static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 				struct ttm_placement *placement)
 {
@@ -204,12 +255,14 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 		.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM
 	};
 
+	/* Don't handle scatter gather BOs */
 	if (bo->type == ttm_bo_type_sg) {
 		placement->num_placement = 0;
 		placement->num_busy_placement = 0;
 		return;
 	}
 
+	/* Object isn't an AMDGPU object so ignore */
 	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo)) {
 		placement->placement = &placements;
 		placement->busy_placement = &placements;
@@ -217,26 +270,16 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 		placement->num_busy_placement = 1;
 		return;
 	}
+
 	abo = ttm_to_amdgpu_bo(bo);
 	switch (bo->mem.mem_type) {
 	case TTM_PL_VRAM:
 		if (!adev->mman.buffer_funcs_enabled) {
+			/* Move to system memory */
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
 		} else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
-			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) {
-			unsigned fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
-			struct drm_mm_node *node = bo->mem.mm_node;
-			unsigned long pages_left;
-
-			for (pages_left = bo->mem.num_pages;
-			     pages_left;
-			     pages_left -= node->size, node++) {
-				if (node->start < fpfn)
-					break;
-			}
-
-			if (!pages_left)
-				goto gtt;
+			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
+			   amdgpu_bo_in_cpu_visible_vram(abo)) {
 
 			/* Try evicting to the CPU inaccessible part of VRAM
 			 * first, but only set GTT as busy placement, so this
@@ -245,12 +288,12 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
 			 */
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
 							 AMDGPU_GEM_DOMAIN_GTT);
-			abo->placements[0].fpfn = fpfn;
+			abo->placements[0].fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
 			abo->placements[0].lpfn = 0;
 			abo->placement.busy_placement = &abo->placements[1];
 			abo->placement.num_busy_placement = 1;
 		} else {
-gtt:
+			/* Move to GTT memory */
 			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
 		}
 		break;
@@ -261,6 +304,15 @@ gtt:
 	*placement = abo->placement;
 }
 
+/**
+ * amdgpu_verify_access - Verify access for a mmap call
+ *
+ * @bo:		The buffer object to map
+ * @filp:	The file pointer from the process performing the mmap
+ *
+ * This is called by ttm_bo_mmap() to verify whether a process
+ * has the right to mmap a BO to their process space.
+ */
 static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp)
 {
 	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
@@ -278,6 +330,15 @@ static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp)
 					  filp->private_data);
 }
 
+/**
+ * amdgpu_move_null - Register memory for a buffer object
+ *
+ * @bo:			The bo to assign the memory to
+ * @new_mem:	The memory to be assigned.
+ *
+ * Assign the memory from new_mem to the memory of the buffer object
+ * bo.
+ */
 static void amdgpu_move_null(struct ttm_buffer_object *bo,
 			     struct ttm_mem_reg *new_mem)
 {
@@ -288,6 +349,10 @@ static void amdgpu_move_null(struct ttm_buffer_object *bo,
 	new_mem->mm_node = NULL;
 }
 
+/**
+ * amdgpu_mm_node_addr -	Compute the GPU relative offset of a GTT
+ * 							buffer.
+ */
 static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
 				    struct drm_mm_node *mm_node,
 				    struct ttm_mem_reg *mem)
@@ -302,9 +367,10 @@ static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
 }
 
 /**
- * amdgpu_find_mm_node - Helper function finds the drm_mm_node
- *  corresponding to @offset. It also modifies the offset to be
- *  within the drm_mm_node returned
+ * amdgpu_find_mm_node -	Helper function finds the drm_mm_node
+ *  						corresponding to @offset. It also modifies
+ * 							the offset to be within the drm_mm_node
+ * 							returned
  */
 static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_mem_reg *mem,
 					       unsigned long *offset)
@@ -443,7 +509,12 @@ error:
 	return r;
 }
 
-
+/**
+ * amdgpu_move_blit - Copy an entire buffer to another buffer
+ *
+ * This is a helper called by amdgpu_bo_move() and
+ * amdgpu_move_vram_ram() to help move buffers to and from VRAM.
+ */
 static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 			    bool evict, bool no_wait_gpu,
 			    struct ttm_mem_reg *new_mem,
@@ -478,6 +549,11 @@ error:
 	return r;
 }
 
+/**
+ * amdgpu_move_vram_ram - Copy VRAM buffer to RAM buffer
+ *
+ * Called by amdgpu_bo_move().
+ */
 static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
 				struct ttm_operation_ctx *ctx,
 				struct ttm_mem_reg *new_mem)
@@ -490,6 +566,8 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
 	int r;
 
 	adev = amdgpu_ttm_adev(bo->bdev);
+
+	/* create space/pages for new_mem in GTT space */
 	tmp_mem = *new_mem;
 	tmp_mem.mm_node = NULL;
 	placement.num_placement = 1;
@@ -504,25 +582,36 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
 		return r;
 	}
 
+	/* set caching flags */
 	r = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
 
+	/* Bind the memory to the GTT space */
 	r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
+
+	/* blit VRAM to GTT */
 	r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, &tmp_mem, old_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
+
+	/* move BO (in tmp_mem) to new_mem */
 	r = ttm_bo_move_ttm(bo, ctx, new_mem);
 out_cleanup:
 	ttm_bo_mem_put(bo, &tmp_mem);
 	return r;
 }
 
+/**
+ * amdgpu_move_ram_vram - Copy buffer from RAM to VRAM
+ *
+ * Called by amdgpu_bo_move().
+ */
 static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
 				struct ttm_operation_ctx *ctx,
 				struct ttm_mem_reg *new_mem)
@@ -535,6 +624,8 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
 	int r;
 
 	adev = amdgpu_ttm_adev(bo->bdev);
+
+	/* make space in GTT for old_mem buffer */
 	tmp_mem = *new_mem;
 	tmp_mem.mm_node = NULL;
 	placement.num_placement = 1;
@@ -548,10 +639,14 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
 	if (unlikely(r)) {
 		return r;
 	}
+
+	/* move/bind old memory to GTT space */
 	r = ttm_bo_move_ttm(bo, ctx, &tmp_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
+
+	/* copy to VRAM */
 	r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, new_mem, old_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
@@ -561,6 +656,11 @@ out_cleanup:
 	return r;
 }
 
+/**
+ * amdgpu_bo_move - Move a buffer object to a new memory location
+ *
+ * Called by ttm_bo_handle_move_mem()
+ */
 static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict,
 			  struct ttm_operation_ctx *ctx,
 			  struct ttm_mem_reg *new_mem)
@@ -626,6 +726,11 @@ memcpy:
 	return 0;
 }
 
+/**
+ * amdgpu_ttm_io_mem_reserve - Reserve a block of memory during a fault
+ *
+ * Called by ttm_mem_io_reserve() ultimately via ttm_bo_vm_fault()
+ */
 static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 {
 	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
@@ -695,7 +800,7 @@ struct amdgpu_ttm_tt {
 	struct ttm_dma_tt	ttm;
 	u64			offset;
 	uint64_t		userptr;
-	struct mm_struct	*usermm;
+	struct task_struct	*usertask;
 	uint32_t		userflags;
 	spinlock_t              guptasklock;
 	struct list_head        guptasks;
@@ -703,17 +808,29 @@ struct amdgpu_ttm_tt {
 	uint32_t		last_set_pages;
 };
 
+/**
+ * amdgpu_ttm_tt_get_user_pages - 	Pin pages of memory pointed to
+ * 									by a USERPTR pointer to memory
+ *
+ * Called by amdgpu_gem_userptr_ioctl() and amdgpu_cs_parser_bos().
+ * This provides a wrapper around the get_user_pages() call to provide
+ * device accessible pages that back user memory.
+ */
 int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	struct mm_struct *mm = gtt->usertask->mm;
 	unsigned int flags = 0;
 	unsigned pinned = 0;
 	int r;
 
+	if (!mm) /* Happens during process shutdown */
+		return -ESRCH;
+
 	if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
 		flags |= FOLL_WRITE;
 
-	down_read(&current->mm->mmap_sem);
+	down_read(&mm->mmap_sem);
 
 	if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) {
 		/* check that we only use anonymous memory
@@ -721,13 +838,14 @@ int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
 		unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
 		struct vm_area_struct *vma;
 
-		vma = find_vma(gtt->usermm, gtt->userptr);
+		vma = find_vma(mm, gtt->userptr);
 		if (!vma || vma->vm_file || vma->vm_end < end) {
-			up_read(&current->mm->mmap_sem);
+			up_read(&mm->mmap_sem);
 			return -EPERM;
 		}
 	}
 
+	/* loop enough times using contiguous pages of memory */
 	do {
 		unsigned num_pages = ttm->num_pages - pinned;
 		uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
@@ -739,7 +857,12 @@ int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
 		list_add(&guptask.list, &gtt->guptasks);
 		spin_unlock(&gtt->guptasklock);
 
-		r = get_user_pages(userptr, num_pages, flags, p, NULL);
+		if (mm == current->mm)
+			r = get_user_pages(userptr, num_pages, flags, p, NULL);
+		else
+			r = get_user_pages_remote(gtt->usertask,
+					mm, userptr, num_pages,
+					flags, p, NULL, NULL);
 
 		spin_lock(&gtt->guptasklock);
 		list_del(&guptask.list);
@@ -752,15 +875,23 @@ int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
 
 	} while (pinned < ttm->num_pages);
 
-	up_read(&current->mm->mmap_sem);
+	up_read(&mm->mmap_sem);
 	return 0;
 
 release_pages:
 	release_pages(pages, pinned);
-	up_read(&current->mm->mmap_sem);
+	up_read(&mm->mmap_sem);
 	return r;
 }
 
+/**
+ * amdgpu_ttm_tt_set_user_pages - 	Copy pages in, putting old pages
+ * 									as necessary.
+ *
+ * Called by amdgpu_cs_list_validate().  This creates the page list
+ * that backs user memory and will ultimately be mapped into the device
+ * address space.
+ */
 void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -775,6 +906,11 @@ void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
 	}
 }
 
+/**
+ * amdgpu_ttm_tt_mark_user_page - Mark pages as dirty
+ *
+ * Called while unpinning userptr pages
+ */
 void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -793,7 +929,12 @@ void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
 	}
 }
 
-/* prepare the sg table with the user pages */
+/**
+ * amdgpu_ttm_tt_pin_userptr - 	prepare the sg table with the
+ * 								user pages
+ *
+ * Called by amdgpu_ttm_backend_bind()
+ **/
 static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
@@ -805,17 +946,20 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 	enum dma_data_direction direction = write ?
 		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
 
+	/* Allocate an SG array and squash pages into it */
 	r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
 				      ttm->num_pages << PAGE_SHIFT,
 				      GFP_KERNEL);
 	if (r)
 		goto release_sg;
 
+	/* Map SG to device */
 	r = -ENOMEM;
 	nents = dma_map_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
 	if (nents != ttm->sg->nents)
 		goto release_sg;
 
+	/* convert SG to linear array of pages and dma addresses */
 	drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
 					 gtt->ttm.dma_address, ttm->num_pages);
 
@@ -826,6 +970,9 @@ release_sg:
 	return r;
 }
 
+/**
+ * amdgpu_ttm_tt_unpin_userptr - Unpin and unmap userptr pages
+ */
 static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
@@ -839,14 +986,60 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
 	if (!ttm->sg->sgl)
 		return;
 
-	/* free the sg table and pages again */
+	/* unmap the pages mapped to the device */
 	dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
 
+	/* mark the pages as dirty */
 	amdgpu_ttm_tt_mark_user_pages(ttm);
 
 	sg_free_table(ttm->sg);
 }
 
+int amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
+				struct ttm_buffer_object *tbo,
+				uint64_t flags)
+{
+	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(tbo);
+	struct ttm_tt *ttm = tbo->ttm;
+	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	int r;
+
+	if (abo->flags & AMDGPU_GEM_CREATE_MQD_GFX9) {
+		uint64_t page_idx = 1;
+
+		r = amdgpu_gart_bind(adev, gtt->offset, page_idx,
+				ttm->pages, gtt->ttm.dma_address, flags);
+		if (r)
+			goto gart_bind_fail;
+
+		/* Patch mtype of the second part BO */
+		flags &=  ~AMDGPU_PTE_MTYPE_MASK;
+		flags |= AMDGPU_PTE_MTYPE(AMDGPU_MTYPE_NC);
+
+		r = amdgpu_gart_bind(adev,
+				gtt->offset + (page_idx << PAGE_SHIFT),
+				ttm->num_pages - page_idx,
+				&ttm->pages[page_idx],
+				&(gtt->ttm.dma_address[page_idx]), flags);
+	} else {
+		r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
+				     ttm->pages, gtt->ttm.dma_address, flags);
+	}
+
+gart_bind_fail:
+	if (r)
+		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
+			  ttm->num_pages, gtt->offset);
+
+	return r;
+}
+
+/**
+ * amdgpu_ttm_backend_bind - Bind GTT memory
+ *
+ * Called by ttm_tt_bind() on behalf of ttm_bo_handle_move_mem().
+ * This handles binding GTT memory to the device address space.
+ */
 static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
 				   struct ttm_mem_reg *bo_mem)
 {
@@ -877,7 +1070,10 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
 		return 0;
 	}
 
+	/* compute PTE flags relevant to this BO memory */
 	flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem);
+
+	/* bind pages into GART page tables */
 	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
 	r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
 		ttm->pages, gtt->ttm.dma_address, flags);
@@ -888,6 +1084,9 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
 	return r;
 }
 
+/**
+ * amdgpu_ttm_alloc_gart - Allocate GART memory for buffer object
+ */
 int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
@@ -903,6 +1102,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 	    amdgpu_gtt_mgr_has_gart_addr(&bo->mem))
 		return 0;
 
+	/* allocate GTT space */
 	tmp = bo->mem;
 	tmp.mm_node = NULL;
 	placement.num_placement = 1;
@@ -918,10 +1118,12 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 	if (unlikely(r))
 		return r;
 
+	/* compute PTE flags for this buffer object */
 	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, &tmp);
+
+	/* Bind pages */
 	gtt->offset = (u64)tmp.start << PAGE_SHIFT;
-	r = amdgpu_gart_bind(adev, gtt->offset, bo->ttm->num_pages,
-			     bo->ttm->pages, gtt->ttm.dma_address, flags);
+	r = amdgpu_ttm_gart_bind(adev, bo, flags);
 	if (unlikely(r)) {
 		ttm_bo_mem_put(bo, &tmp);
 		return r;
@@ -935,31 +1137,40 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 	return 0;
 }
 
+/**
+ * amdgpu_ttm_recover_gart - Rebind GTT pages
+ *
+ * Called by amdgpu_gtt_mgr_recover() from amdgpu_device_reset() to
+ * rebind GTT pages during a GPU reset.
+ */
 int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	struct amdgpu_ttm_tt *gtt = (void *)tbo->ttm;
 	uint64_t flags;
 	int r;
 
-	if (!gtt)
+	if (!tbo->ttm)
 		return 0;
 
-	flags = amdgpu_ttm_tt_pte_flags(adev, &gtt->ttm.ttm, &tbo->mem);
-	r = amdgpu_gart_bind(adev, gtt->offset, gtt->ttm.ttm.num_pages,
-			     gtt->ttm.ttm.pages, gtt->ttm.dma_address, flags);
-	if (r)
-		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
-			  gtt->ttm.ttm.num_pages, gtt->offset);
+	flags = amdgpu_ttm_tt_pte_flags(adev, tbo->ttm, &tbo->mem);
+	r = amdgpu_ttm_gart_bind(adev, tbo, flags);
+
 	return r;
 }
 
+/**
+ * amdgpu_ttm_backend_unbind - Unbind GTT mapped pages
+ *
+ * Called by ttm_tt_unbind() on behalf of ttm_bo_move_ttm() and
+ * ttm_tt_destroy().
+ */
 static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
 	int r;
 
+	/* if the pages have userptr pinning then clear that first */
 	if (gtt->userptr)
 		amdgpu_ttm_tt_unpin_userptr(ttm);
 
@@ -978,6 +1189,9 @@ static void amdgpu_ttm_backend_destroy(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
 
+	if (gtt->usertask)
+		put_task_struct(gtt->usertask);
+
 	ttm_dma_tt_fini(&gtt->ttm);
 	kfree(gtt);
 }
@@ -988,6 +1202,13 @@ static struct ttm_backend_func amdgpu_backend_func = {
 	.destroy = &amdgpu_ttm_backend_destroy,
 };
 
+/**
+ * amdgpu_ttm_tt_create - Create a ttm_tt object for a given BO
+ *
+ * @bo: The buffer object to create a GTT ttm_tt object around
+ *
+ * Called by ttm_tt_create().
+ */
 static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
 					   uint32_t page_flags)
 {
@@ -1001,6 +1222,8 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
 		return NULL;
 	}
 	gtt->ttm.ttm.func = &amdgpu_backend_func;
+
+	/* allocate space for the uninitialized page entries */
 	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags)) {
 		kfree(gtt);
 		return NULL;
@@ -1008,6 +1231,12 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
 	return &gtt->ttm.ttm;
 }
 
+/**
+ * amdgpu_ttm_tt_populate - Map GTT pages visible to the device
+ *
+ * Map the pages of a ttm_tt object to an address space visible
+ * to the underlying device.
+ */
 static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
 			struct ttm_operation_ctx *ctx)
 {
@@ -1015,6 +1244,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
 	bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
 
+	/* user pages are bound by amdgpu_ttm_tt_pin_userptr() */
 	if (gtt && gtt->userptr) {
 		ttm->sg = kzalloc(sizeof(struct sg_table), GFP_KERNEL);
 		if (!ttm->sg)
@@ -1039,9 +1269,17 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
 	}
 #endif
 
+	/* fall back to generic helper to populate the page array
+	 * and map them to the device */
 	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx);
 }
 
+/**
+ * amdgpu_ttm_tt_unpopulate - unmap GTT pages and unpopulate page arrays
+ *
+ * Unmaps pages of a ttm_tt object from the device address space and
+ * unpopulates the page array backing it.
+ */
 static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev;
@@ -1067,9 +1305,21 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
 	}
 #endif
 
+	/* fall back to generic helper to unmap and unpopulate array */
 	ttm_unmap_and_unpopulate_pages(adev->dev, &gtt->ttm);
 }
 
+/**
+ * amdgpu_ttm_tt_set_userptr -	Initialize userptr GTT ttm_tt
+ * 								for the current task
+ *
+ * @ttm: The ttm_tt object to bind this userptr object to
+ * @addr:  The address in the current tasks VM space to use
+ * @flags: Requirements of userptr object.
+ *
+ * Called by amdgpu_gem_userptr_ioctl() to bind userptr pages
+ * to current task
+ */
 int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 			      uint32_t flags)
 {
@@ -1079,8 +1329,13 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 		return -EINVAL;
 
 	gtt->userptr = addr;
-	gtt->usermm = current->mm;
 	gtt->userflags = flags;
+
+	if (gtt->usertask)
+		put_task_struct(gtt->usertask);
+	gtt->usertask = current->group_leader;
+	get_task_struct(gtt->usertask);
+
 	spin_lock_init(&gtt->guptasklock);
 	INIT_LIST_HEAD(&gtt->guptasks);
 	atomic_set(&gtt->mmu_invalidations, 0);
@@ -1089,6 +1344,9 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 	return 0;
 }
 
+/**
+ * amdgpu_ttm_tt_get_usermm - Return memory manager for ttm_tt object
+ */
 struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -1096,9 +1354,18 @@ struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
 	if (gtt == NULL)
 		return NULL;
 
-	return gtt->usermm;
+	if (gtt->usertask == NULL)
+		return NULL;
+
+	return gtt->usertask->mm;
 }
 
+/**
+ * amdgpu_ttm_tt_affect_userptr -	Determine if a ttm_tt object lays
+ * 									inside an address range for the
+ * 									current task.
+ *
+ */
 bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 				  unsigned long end)
 {
@@ -1109,10 +1376,16 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 	if (gtt == NULL || !gtt->userptr)
 		return false;
 
+	/* Return false if no part of the ttm_tt object lies within
+	 * the range
+	 */
 	size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE;
 	if (gtt->userptr > end || gtt->userptr + size <= start)
 		return false;
 
+	/* Search the lists of tasks that hold this mapping and see
+	 * if current is one of them.  If it is return false.
+	 */
 	spin_lock(&gtt->guptasklock);
 	list_for_each_entry(entry, &gtt->guptasks, list) {
 		if (entry->task == current) {
@@ -1127,6 +1400,10 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 	return true;
 }
 
+/**
+ * amdgpu_ttm_tt_userptr_invalidated -	Has the ttm_tt object been
+ * 										invalidated?
+ */
 bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
 				       int *last_invalidated)
 {
@@ -1137,6 +1414,12 @@ bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
 	return prev_invalidated != *last_invalidated;
 }
 
+/**
+ * amdgpu_ttm_tt_userptr_needs_pages -	Have the pages backing this
+ * 										ttm_tt object been invalidated
+ * 										since the last time they've
+ * 										been set?
+ */
 bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -1147,6 +1430,9 @@ bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
 	return atomic_read(&gtt->mmu_invalidations) != gtt->last_set_pages;
 }
 
+/**
+ * amdgpu_ttm_tt_is_readonly - Is the ttm_tt object read only?
+ */
 bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
@@ -1157,6 +1443,12 @@ bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
 	return !!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
 }
 
+/**
+ * amdgpu_ttm_tt_pte_flags - Compute PTE flags for ttm_tt object
+ *
+ * @ttm: The ttm_tt object to compute the flags for
+ * @mem: The memory registry backing this ttm_tt object
+ */
 uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 				 struct ttm_mem_reg *mem)
 {
@@ -1181,6 +1473,16 @@ uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 	return flags;
 }
 
+/**
+ * amdgpu_ttm_bo_eviction_valuable -	Check to see if we can evict
+ * 										a buffer object.
+ *
+ * Return true if eviction is sensible.  Called by
+ * ttm_mem_evict_first() on behalf of ttm_bo_mem_force_space()
+ * which tries to evict buffer objects until it can find space
+ * for a new object and by ttm_bo_force_list_clean() which is
+ * used to clean out a memory space.
+ */
 static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 					    const struct ttm_place *place)
 {
@@ -1227,6 +1529,19 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 	return ttm_bo_eviction_valuable(bo, place);
 }
 
+/**
+ * amdgpu_ttm_access_memory -	Read or Write memory that backs a
+ * 								buffer object.
+ *
+ * @bo:  The buffer object to read/write
+ * @offset:  Offset into buffer object
+ * @buf:  Secondary buffer to write/read from
+ * @len: Length in bytes of access
+ * @write:  true if writing
+ *
+ * This is used to access VRAM that backs a buffer object via MMIO
+ * access for debugging purposes.
+ */
 static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
 				    unsigned long offset,
 				    void *buf, int len, int write)
@@ -1329,6 +1644,7 @@ static void amdgpu_ttm_fw_reserve_vram_fini(struct amdgpu_device *adev)
 static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev)
 {
 	struct ttm_operation_ctx ctx = { false, false };
+	struct amdgpu_bo_param bp;
 	int r = 0;
 	int i;
 	u64 vram_size = adev->gmc.visible_vram_size;
@@ -1336,17 +1652,21 @@ static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev)
 	u64 size = adev->fw_vram_usage.size;
 	struct amdgpu_bo *bo;
 
+	memset(&bp, 0, sizeof(bp));
+	bp.size = adev->fw_vram_usage.size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+		AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
 	adev->fw_vram_usage.va = NULL;
 	adev->fw_vram_usage.reserved_bo = NULL;
 
 	if (adev->fw_vram_usage.size > 0 &&
 		adev->fw_vram_usage.size <= vram_size) {
 
-		r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, PAGE_SIZE,
-				     AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-				     ttm_bo_type_kernel, NULL,
+		r = amdgpu_bo_create(adev, &bp,
 				     &adev->fw_vram_usage.reserved_bo);
 		if (r)
 			goto error_create;
@@ -1398,13 +1718,22 @@ error_create:
 	adev->fw_vram_usage.reserved_bo = NULL;
 	return r;
 }
-
+/**
+ * amdgpu_ttm_init -	Init the memory management (ttm) as well as
+ * 						various gtt/vram related fields.
+ *
+ * This initializes all of the memory space pools that the TTM layer
+ * will need such as the GTT space (system memory mapped to the device),
+ * VRAM (on-board memory), and on-chip memories (GDS, GWS, OA) which
+ * can be mapped per VMID.
+ */
 int amdgpu_ttm_init(struct amdgpu_device *adev)
 {
 	uint64_t gtt_size;
 	int r;
 	u64 vis_vram_limit;
 
+	/* initialize global references for vram/gtt */
 	r = amdgpu_ttm_global_init(adev);
 	if (r) {
 		return r;
@@ -1425,6 +1754,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	/* We opt to avoid OOM on system pages allocations */
 	adev->mman.bdev.no_retry = true;
 
+	/* Initialize VRAM pool with all of VRAM divided into pages */
 	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
 				adev->gmc.real_vram_size >> PAGE_SHIFT);
 	if (r) {
@@ -1454,15 +1784,23 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		return r;
 	}
 
-	r = amdgpu_bo_create_kernel(adev, adev->gmc.stolen_size, PAGE_SIZE,
-				    AMDGPU_GEM_DOMAIN_VRAM,
-				    &adev->stolen_vga_memory,
-				    NULL, NULL);
-	if (r)
-		return r;
+	/* allocate memory as required for VGA
+	 * This is used for VGA emulation and pre-OS scanout buffers to
+	 * avoid display artifacts while transitioning between pre-OS
+	 * and driver.  */
+	if (adev->gmc.stolen_size) {
+		r = amdgpu_bo_create_kernel(adev, adev->gmc.stolen_size, PAGE_SIZE,
+					    AMDGPU_GEM_DOMAIN_VRAM,
+					    &adev->stolen_vga_memory,
+					    NULL, NULL);
+		if (r)
+			return r;
+	}
 	DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
 		 (unsigned) (adev->gmc.real_vram_size / (1024 * 1024)));
 
+	/* Compute GTT size, either bsaed on 3/4th the size of RAM size
+	 * or whatever the user passed on module init */
 	if (amdgpu_gtt_size == -1) {
 		struct sysinfo si;
 
@@ -1473,6 +1811,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	}
 	else
 		gtt_size = (uint64_t)amdgpu_gtt_size << 20;
+
+	/* Initialize GTT memory pool */
 	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_TT, gtt_size >> PAGE_SHIFT);
 	if (r) {
 		DRM_ERROR("Failed initializing GTT heap.\n");
@@ -1481,6 +1821,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	DRM_INFO("amdgpu: %uM of GTT memory ready.\n",
 		 (unsigned)(gtt_size / (1024 * 1024)));
 
+	/* Initialize various on-chip memory pools */
 	adev->gds.mem.total_size = adev->gds.mem.total_size << AMDGPU_GDS_SHIFT;
 	adev->gds.mem.gfx_partition_size = adev->gds.mem.gfx_partition_size << AMDGPU_GDS_SHIFT;
 	adev->gds.mem.cs_partition_size = adev->gds.mem.cs_partition_size << AMDGPU_GDS_SHIFT;
@@ -1520,6 +1861,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		}
 	}
 
+	/* Register debugfs entries for amdgpu_ttm */
 	r = amdgpu_ttm_debugfs_init(adev);
 	if (r) {
 		DRM_ERROR("Failed to init debugfs\n");
@@ -1528,13 +1870,25 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+/**
+ * amdgpu_ttm_late_init -	Handle any late initialization for
+ * 							amdgpu_ttm
+ */
+void amdgpu_ttm_late_init(struct amdgpu_device *adev)
+{
+	/* return the VGA stolen memory (if any) back to VRAM */
+	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
+}
+
+/**
+ * amdgpu_ttm_fini - De-initialize the TTM memory pools
+ */
 void amdgpu_ttm_fini(struct amdgpu_device *adev)
 {
 	if (!adev->mman.initialized)
 		return;
 
 	amdgpu_ttm_debugfs_fini(adev);
-	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
 	amdgpu_ttm_fw_reserve_vram_fini(adev);
 	if (adev->mman.aper_base_kaddr)
 		iounmap(adev->mman.aper_base_kaddr);
@@ -1856,6 +2210,11 @@ static const struct drm_info_list amdgpu_ttm_debugfs_list[] = {
 #endif
 };
 
+/**
+ * amdgpu_ttm_vram_read - Linear read access to VRAM
+ *
+ * Accesses VRAM via MMIO for debugging purposes.
+ */
 static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf,
 				    size_t size, loff_t *pos)
 {
@@ -1895,6 +2254,11 @@ static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_ttm_vram_write - Linear write access to VRAM
+ *
+ * Accesses VRAM via MMIO for debugging purposes.
+ */
 static ssize_t amdgpu_ttm_vram_write(struct file *f, const char __user *buf,
 				    size_t size, loff_t *pos)
 {
@@ -1943,6 +2307,9 @@ static const struct file_operations amdgpu_ttm_vram_fops = {
 
 #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
 
+/**
+ * amdgpu_ttm_gtt_read - Linear read access to GTT memory
+ */
 static ssize_t amdgpu_ttm_gtt_read(struct file *f, char __user *buf,
 				   size_t size, loff_t *pos)
 {
@@ -1990,6 +2357,13 @@ static const struct file_operations amdgpu_ttm_gtt_fops = {
 
 #endif
 
+/**
+ * amdgpu_iomem_read - Virtual read access to GPU mapped memory
+ *
+ * This function is used to read memory that has been mapped to the
+ * GPU and the known addresses are not physical addresses but instead
+ * bus addresses (e.g., what you'd put in an IB or ring buffer).
+ */
 static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
 				 size_t size, loff_t *pos)
 {
@@ -1998,6 +2372,7 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
 	ssize_t result = 0;
 	int r;
 
+	/* retrieve the IOMMU domain if any for this device */
 	dom = iommu_get_domain_for_dev(adev->dev);
 
 	while (size) {
@@ -2010,6 +2385,10 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
 
 		bytes = bytes < size ? bytes : size;
 
+		/* Translate the bus address to a physical address.  If
+		 * the domain is NULL it means there is no IOMMU active
+		 * and the address translation is the identity
+		 */
 		addr = dom ? iommu_iova_to_phys(dom, addr) : addr;
 
 		pfn = addr >> PAGE_SHIFT;
@@ -2034,6 +2413,13 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
 	return result;
 }
 
+/**
+ * amdgpu_iomem_write - Virtual write access to GPU mapped memory
+ *
+ * This function is used to write memory that has been mapped to the
+ * GPU and the known addresses are not physical addresses but instead
+ * bus addresses (e.g., what you'd put in an IB or ring buffer).
+ */
 static ssize_t amdgpu_iomem_write(struct file *f, const char __user *buf,
 				 size_t size, loff_t *pos)
 {

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h

@@ -77,6 +77,7 @@ uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man);
 uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man);
 
 int amdgpu_ttm_init(struct amdgpu_device *adev);
+void amdgpu_ttm_late_init(struct amdgpu_device *adev);
 void amdgpu_ttm_fini(struct amdgpu_device *adev);
 void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev,
 					bool enable);

drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c

@@ -161,8 +161,38 @@ void amdgpu_ucode_print_rlc_hdr(const struct common_firmware_header *hdr)
 			  le32_to_cpu(rlc_hdr->reg_list_format_separate_array_offset_bytes));
 		DRM_DEBUG("reg_list_separate_size_bytes: %u\n",
 			  le32_to_cpu(rlc_hdr->reg_list_separate_size_bytes));
-		DRM_DEBUG("reg_list_separate_size_bytes: %u\n",
-			  le32_to_cpu(rlc_hdr->reg_list_separate_size_bytes));
+		DRM_DEBUG("reg_list_separate_array_offset_bytes: %u\n",
+			  le32_to_cpu(rlc_hdr->reg_list_separate_array_offset_bytes));
+		if (version_minor == 1) {
+			const struct rlc_firmware_header_v2_1 *v2_1 =
+				container_of(rlc_hdr, struct rlc_firmware_header_v2_1, v2_0);
+			DRM_DEBUG("reg_list_format_direct_reg_list_length: %u\n",
+				  le32_to_cpu(v2_1->reg_list_format_direct_reg_list_length));
+			DRM_DEBUG("save_restore_list_cntl_ucode_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_cntl_ucode_ver));
+			DRM_DEBUG("save_restore_list_cntl_feature_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_cntl_feature_ver));
+			DRM_DEBUG("save_restore_list_cntl_size_bytes %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_cntl_size_bytes));
+			DRM_DEBUG("save_restore_list_cntl_offset_bytes: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_cntl_offset_bytes));
+			DRM_DEBUG("save_restore_list_gpm_ucode_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_gpm_ucode_ver));
+			DRM_DEBUG("save_restore_list_gpm_feature_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_gpm_feature_ver));
+			DRM_DEBUG("save_restore_list_gpm_size_bytes %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_gpm_size_bytes));
+			DRM_DEBUG("save_restore_list_gpm_offset_bytes: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_gpm_offset_bytes));
+			DRM_DEBUG("save_restore_list_srm_ucode_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_srm_ucode_ver));
+			DRM_DEBUG("save_restore_list_srm_feature_ver: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_srm_feature_ver));
+			DRM_DEBUG("save_restore_list_srm_size_bytes %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_srm_size_bytes));
+			DRM_DEBUG("save_restore_list_srm_offset_bytes: %u\n",
+				  le32_to_cpu(v2_1->save_restore_list_srm_offset_bytes));
+		}
 	} else {
 		DRM_ERROR("Unknown RLC ucode version: %u.%u\n", version_major, version_minor);
 	}
@@ -265,6 +295,7 @@ amdgpu_ucode_get_load_type(struct amdgpu_device *adev, int load_type)
 	case CHIP_POLARIS10:
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 		if (!load_type)
 			return AMDGPU_FW_LOAD_DIRECT;
 		else
@@ -276,6 +307,8 @@ amdgpu_ucode_get_load_type(struct amdgpu_device *adev, int load_type)
 			return AMDGPU_FW_LOAD_DIRECT;
 		else
 			return AMDGPU_FW_LOAD_PSP;
+	case CHIP_VEGA20:
+		return AMDGPU_FW_LOAD_DIRECT;
 	default:
 		DRM_ERROR("Unknown firmware load type\n");
 	}
@@ -307,7 +340,10 @@ static int amdgpu_ucode_init_single_fw(struct amdgpu_device *adev,
 	    (ucode->ucode_id != AMDGPU_UCODE_ID_CP_MEC1 &&
 	     ucode->ucode_id != AMDGPU_UCODE_ID_CP_MEC2 &&
 	     ucode->ucode_id != AMDGPU_UCODE_ID_CP_MEC1_JT &&
-	     ucode->ucode_id != AMDGPU_UCODE_ID_CP_MEC2_JT)) {
+	     ucode->ucode_id != AMDGPU_UCODE_ID_CP_MEC2_JT &&
+	     ucode->ucode_id != AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL &&
+	     ucode->ucode_id != AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM &&
+	     ucode->ucode_id != AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM)) {
 		ucode->ucode_size = le32_to_cpu(header->ucode_size_bytes);
 
 		memcpy(ucode->kaddr, (void *)((uint8_t *)ucode->fw->data +
@@ -329,6 +365,18 @@ static int amdgpu_ucode_init_single_fw(struct amdgpu_device *adev,
 					      le32_to_cpu(header->ucode_array_offset_bytes) +
 					      le32_to_cpu(cp_hdr->jt_offset) * 4),
 		       ucode->ucode_size);
+	} else if (ucode->ucode_id == AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL) {
+		ucode->ucode_size = adev->gfx.rlc.save_restore_list_cntl_size_bytes;
+		memcpy(ucode->kaddr, adev->gfx.rlc.save_restore_list_cntl,
+		       ucode->ucode_size);
+	} else if (ucode->ucode_id == AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM) {
+		ucode->ucode_size = adev->gfx.rlc.save_restore_list_gpm_size_bytes;
+		memcpy(ucode->kaddr, adev->gfx.rlc.save_restore_list_gpm,
+		       ucode->ucode_size);
+	} else if (ucode->ucode_id == AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM) {
+		ucode->ucode_size = adev->gfx.rlc.save_restore_list_srm_size_bytes;
+		memcpy(ucode->kaddr, adev->gfx.rlc.save_restore_list_srm,
+		       ucode->ucode_size);
 	}
 
 	return 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h

@@ -98,6 +98,24 @@ struct rlc_firmware_header_v2_0 {
 	uint32_t reg_list_separate_array_offset_bytes; /* payload offset from the start of the header */
 };
 
+/* version_major=2, version_minor=1 */
+struct rlc_firmware_header_v2_1 {
+	struct rlc_firmware_header_v2_0 v2_0;
+	uint32_t reg_list_format_direct_reg_list_length; /* length of direct reg list format array */
+	uint32_t save_restore_list_cntl_ucode_ver;
+	uint32_t save_restore_list_cntl_feature_ver;
+	uint32_t save_restore_list_cntl_size_bytes;
+	uint32_t save_restore_list_cntl_offset_bytes;
+	uint32_t save_restore_list_gpm_ucode_ver;
+	uint32_t save_restore_list_gpm_feature_ver;
+	uint32_t save_restore_list_gpm_size_bytes;
+	uint32_t save_restore_list_gpm_offset_bytes;
+	uint32_t save_restore_list_srm_ucode_ver;
+	uint32_t save_restore_list_srm_feature_ver;
+	uint32_t save_restore_list_srm_size_bytes;
+	uint32_t save_restore_list_srm_offset_bytes;
+};
+
 /* version_major=1, version_minor=0 */
 struct sdma_firmware_header_v1_0 {
 	struct common_firmware_header header;
@@ -148,6 +166,7 @@ union amdgpu_firmware_header {
 	struct gfx_firmware_header_v1_0 gfx;
 	struct rlc_firmware_header_v1_0 rlc;
 	struct rlc_firmware_header_v2_0 rlc_v2_0;
+	struct rlc_firmware_header_v2_1 rlc_v2_1;
 	struct sdma_firmware_header_v1_0 sdma;
 	struct sdma_firmware_header_v1_1 sdma_v1_1;
 	struct gpu_info_firmware_header_v1_0 gpu_info;
@@ -168,6 +187,9 @@ enum AMDGPU_UCODE_ID {
 	AMDGPU_UCODE_ID_CP_MEC2,
 	AMDGPU_UCODE_ID_CP_MEC2_JT,
 	AMDGPU_UCODE_ID_RLC_G,
+	AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL,
+	AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM,
+	AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM,
 	AMDGPU_UCODE_ID_STORAGE,
 	AMDGPU_UCODE_ID_SMC,
 	AMDGPU_UCODE_ID_UVD,

drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c

@@ -66,15 +66,18 @@
 #define FIRMWARE_POLARIS10	"amdgpu/polaris10_uvd.bin"
 #define FIRMWARE_POLARIS11	"amdgpu/polaris11_uvd.bin"
 #define FIRMWARE_POLARIS12	"amdgpu/polaris12_uvd.bin"
+#define FIRMWARE_VEGAM		"amdgpu/vegam_uvd.bin"
 
 #define FIRMWARE_VEGA10		"amdgpu/vega10_uvd.bin"
 #define FIRMWARE_VEGA12		"amdgpu/vega12_uvd.bin"
+#define FIRMWARE_VEGA20		"amdgpu/vega20_uvd.bin"
 
-#define mmUVD_GPCOM_VCPU_DATA0_VEGA10 (0x03c4 + 0x7e00)
-#define mmUVD_GPCOM_VCPU_DATA1_VEGA10 (0x03c5 + 0x7e00)
-#define mmUVD_GPCOM_VCPU_CMD_VEGA10 (0x03c3 + 0x7e00)
-#define mmUVD_NO_OP_VEGA10 (0x03ff + 0x7e00)
-#define mmUVD_ENGINE_CNTL_VEGA10 (0x03c6 + 0x7e00)
+/* These are common relative offsets for all asics, from uvd_7_0_offset.h,  */
+#define UVD_GPCOM_VCPU_CMD		0x03c3
+#define UVD_GPCOM_VCPU_DATA0	0x03c4
+#define UVD_GPCOM_VCPU_DATA1	0x03c5
+#define UVD_NO_OP				0x03ff
+#define UVD_BASE_SI				0x3800
 
 /**
  * amdgpu_uvd_cs_ctx - Command submission parser context
@@ -109,9 +112,11 @@ MODULE_FIRMWARE(FIRMWARE_STONEY);
 MODULE_FIRMWARE(FIRMWARE_POLARIS10);
 MODULE_FIRMWARE(FIRMWARE_POLARIS11);
 MODULE_FIRMWARE(FIRMWARE_POLARIS12);
+MODULE_FIRMWARE(FIRMWARE_VEGAM);
 
 MODULE_FIRMWARE(FIRMWARE_VEGA10);
 MODULE_FIRMWARE(FIRMWARE_VEGA12);
+MODULE_FIRMWARE(FIRMWARE_VEGA20);
 
 static void amdgpu_uvd_idle_work_handler(struct work_struct *work);
 
@@ -123,9 +128,9 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 	const char *fw_name;
 	const struct common_firmware_header *hdr;
 	unsigned version_major, version_minor, family_id;
-	int i, r;
+	int i, j, r;
 
-	INIT_DELAYED_WORK(&adev->uvd.idle_work, amdgpu_uvd_idle_work_handler);
+	INIT_DELAYED_WORK(&adev->uvd.inst->idle_work, amdgpu_uvd_idle_work_handler);
 
 	switch (adev->asic_type) {
 #ifdef CONFIG_DRM_AMDGPU_CIK
@@ -172,6 +177,12 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 	case CHIP_VEGA12:
 		fw_name = FIRMWARE_VEGA12;
 		break;
+	case CHIP_VEGAM:
+		fw_name = FIRMWARE_VEGAM;
+		break;
+	case CHIP_VEGA20:
+		fw_name = FIRMWARE_VEGA20;
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -226,28 +237,30 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
 
-	r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
-				    AMDGPU_GEM_DOMAIN_VRAM, &adev->uvd.vcpu_bo,
-				    &adev->uvd.gpu_addr, &adev->uvd.cpu_addr);
-	if (r) {
-		dev_err(adev->dev, "(%d) failed to allocate UVD bo\n", r);
-		return r;
-	}
+	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
 
-	ring = &adev->uvd.ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&ring->sched, &adev->uvd.entity,
-				  rq, amdgpu_sched_jobs, NULL);
-	if (r != 0) {
-		DRM_ERROR("Failed setting up UVD run queue.\n");
-		return r;
-	}
+		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
+					    AMDGPU_GEM_DOMAIN_VRAM, &adev->uvd.inst[j].vcpu_bo,
+					    &adev->uvd.inst[j].gpu_addr, &adev->uvd.inst[j].cpu_addr);
+		if (r) {
+			dev_err(adev->dev, "(%d) failed to allocate UVD bo\n", r);
+			return r;
+		}
 
-	for (i = 0; i < adev->uvd.max_handles; ++i) {
-		atomic_set(&adev->uvd.handles[i], 0);
-		adev->uvd.filp[i] = NULL;
-	}
+		ring = &adev->uvd.inst[j].ring;
+		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+		r = drm_sched_entity_init(&ring->sched, &adev->uvd.inst[j].entity,
+					  rq, NULL);
+		if (r != 0) {
+			DRM_ERROR("Failed setting up UVD(%d) run queue.\n", j);
+			return r;
+		}
 
+		for (i = 0; i < adev->uvd.max_handles; ++i) {
+			atomic_set(&adev->uvd.inst[j].handles[i], 0);
+			adev->uvd.inst[j].filp[i] = NULL;
+		}
+	}
 	/* from uvd v5.0 HW addressing capacity increased to 64 bits */
 	if (!amdgpu_device_ip_block_version_cmp(adev, AMD_IP_BLOCK_TYPE_UVD, 5, 0))
 		adev->uvd.address_64_bit = true;
@@ -274,20 +287,22 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 
 int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 {
-	int i;
-	kfree(adev->uvd.saved_bo);
+	int i, j;
 
-	drm_sched_entity_fini(&adev->uvd.ring.sched, &adev->uvd.entity);
+	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		kfree(adev->uvd.inst[j].saved_bo);
 
-	amdgpu_bo_free_kernel(&adev->uvd.vcpu_bo,
-			      &adev->uvd.gpu_addr,
-			      (void **)&adev->uvd.cpu_addr);
+		drm_sched_entity_fini(&adev->uvd.inst[j].ring.sched, &adev->uvd.inst[j].entity);
 
-	amdgpu_ring_fini(&adev->uvd.ring);
+		amdgpu_bo_free_kernel(&adev->uvd.inst[j].vcpu_bo,
+				      &adev->uvd.inst[j].gpu_addr,
+				      (void **)&adev->uvd.inst[j].cpu_addr);
 
-	for (i = 0; i < AMDGPU_MAX_UVD_ENC_RINGS; ++i)
-		amdgpu_ring_fini(&adev->uvd.ring_enc[i]);
+		amdgpu_ring_fini(&adev->uvd.inst[j].ring);
 
+		for (i = 0; i < AMDGPU_MAX_UVD_ENC_RINGS; ++i)
+			amdgpu_ring_fini(&adev->uvd.inst[j].ring_enc[i]);
+	}
 	release_firmware(adev->uvd.fw);
 
 	return 0;
@@ -297,32 +312,33 @@ int amdgpu_uvd_suspend(struct amdgpu_device *adev)
 {
 	unsigned size;
 	void *ptr;
-	int i;
-
-	if (adev->uvd.vcpu_bo == NULL)
-		return 0;
+	int i, j;
 
-	cancel_delayed_work_sync(&adev->uvd.idle_work);
+	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		if (adev->uvd.inst[j].vcpu_bo == NULL)
+			continue;
 
-	/* only valid for physical mode */
-	if (adev->asic_type < CHIP_POLARIS10) {
-		for (i = 0; i < adev->uvd.max_handles; ++i)
-			if (atomic_read(&adev->uvd.handles[i]))
-				break;
+		cancel_delayed_work_sync(&adev->uvd.inst[j].idle_work);
 
-		if (i == adev->uvd.max_handles)
-			return 0;
-	}
+		/* only valid for physical mode */
+		if (adev->asic_type < CHIP_POLARIS10) {
+			for (i = 0; i < adev->uvd.max_handles; ++i)
+				if (atomic_read(&adev->uvd.inst[j].handles[i]))
+					break;
 
-	size = amdgpu_bo_size(adev->uvd.vcpu_bo);
-	ptr = adev->uvd.cpu_addr;
+			if (i == adev->uvd.max_handles)
+				continue;
+		}
 
-	adev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
-	if (!adev->uvd.saved_bo)
-		return -ENOMEM;
+		size = amdgpu_bo_size(adev->uvd.inst[j].vcpu_bo);
+		ptr = adev->uvd.inst[j].cpu_addr;
 
-	memcpy_fromio(adev->uvd.saved_bo, ptr, size);
+		adev->uvd.inst[j].saved_bo = kmalloc(size, GFP_KERNEL);
+		if (!adev->uvd.inst[j].saved_bo)
+			return -ENOMEM;
 
+		memcpy_fromio(adev->uvd.inst[j].saved_bo, ptr, size);
+	}
 	return 0;
 }
 
@@ -330,59 +346,65 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev)
 {
 	unsigned size;
 	void *ptr;
+	int i;
 
-	if (adev->uvd.vcpu_bo == NULL)
-		return -EINVAL;
+	for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+		if (adev->uvd.inst[i].vcpu_bo == NULL)
+			return -EINVAL;
 
-	size = amdgpu_bo_size(adev->uvd.vcpu_bo);
-	ptr = adev->uvd.cpu_addr;
+		size = amdgpu_bo_size(adev->uvd.inst[i].vcpu_bo);
+		ptr = adev->uvd.inst[i].cpu_addr;
 
-	if (adev->uvd.saved_bo != NULL) {
-		memcpy_toio(ptr, adev->uvd.saved_bo, size);
-		kfree(adev->uvd.saved_bo);
-		adev->uvd.saved_bo = NULL;
-	} else {
-		const struct common_firmware_header *hdr;
-		unsigned offset;
-
-		hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
-		if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
-			offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
-			memcpy_toio(adev->uvd.cpu_addr, adev->uvd.fw->data + offset,
-				    le32_to_cpu(hdr->ucode_size_bytes));
-			size -= le32_to_cpu(hdr->ucode_size_bytes);
-			ptr += le32_to_cpu(hdr->ucode_size_bytes);
+		if (adev->uvd.inst[i].saved_bo != NULL) {
+			memcpy_toio(ptr, adev->uvd.inst[i].saved_bo, size);
+			kfree(adev->uvd.inst[i].saved_bo);
+			adev->uvd.inst[i].saved_bo = NULL;
+		} else {
+			const struct common_firmware_header *hdr;
+			unsigned offset;
+
+			hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
+			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
+				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
+				memcpy_toio(adev->uvd.inst[i].cpu_addr, adev->uvd.fw->data + offset,
+					    le32_to_cpu(hdr->ucode_size_bytes));
+				size -= le32_to_cpu(hdr->ucode_size_bytes);
+				ptr += le32_to_cpu(hdr->ucode_size_bytes);
+			}
+			memset_io(ptr, 0, size);
+			/* to restore uvd fence seq */
+			amdgpu_fence_driver_force_completion(&adev->uvd.inst[i].ring);
 		}
-		memset_io(ptr, 0, size);
-		/* to restore uvd fence seq */
-		amdgpu_fence_driver_force_completion(&adev->uvd.ring);
 	}
-
 	return 0;
 }
 
 void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 {
-	struct amdgpu_ring *ring = &adev->uvd.ring;
-	int i, r;
+	struct amdgpu_ring *ring;
+	int i, j, r;
 
-	for (i = 0; i < adev->uvd.max_handles; ++i) {
-		uint32_t handle = atomic_read(&adev->uvd.handles[i]);
-		if (handle != 0 && adev->uvd.filp[i] == filp) {
-			struct dma_fence *fence;
-
-			r = amdgpu_uvd_get_destroy_msg(ring, handle,
-						       false, &fence);
-			if (r) {
-				DRM_ERROR("Error destroying UVD (%d)!\n", r);
-				continue;
-			}
+	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
+		ring = &adev->uvd.inst[j].ring;
+
+		for (i = 0; i < adev->uvd.max_handles; ++i) {
+			uint32_t handle = atomic_read(&adev->uvd.inst[j].handles[i]);
+			if (handle != 0 && adev->uvd.inst[j].filp[i] == filp) {
+				struct dma_fence *fence;
+
+				r = amdgpu_uvd_get_destroy_msg(ring, handle,
+							       false, &fence);
+				if (r) {
+					DRM_ERROR("Error destroying UVD(%d) %d!\n", j, r);
+					continue;
+				}
 
-			dma_fence_wait(fence, false);
-			dma_fence_put(fence);
+				dma_fence_wait(fence, false);
+				dma_fence_put(fence);
 
-			adev->uvd.filp[i] = NULL;
-			atomic_set(&adev->uvd.handles[i], 0);
+				adev->uvd.inst[j].filp[i] = NULL;
+				atomic_set(&adev->uvd.inst[j].handles[i], 0);
+			}
 		}
 	}
 }
@@ -657,15 +679,16 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
 	void *ptr;
 	long r;
 	int i;
+	uint32_t ip_instance = ctx->parser->job->ring->me;
 
 	if (offset & 0x3F) {
-		DRM_ERROR("UVD messages must be 64 byte aligned!\n");
+		DRM_ERROR("UVD(%d) messages must be 64 byte aligned!\n", ip_instance);
 		return -EINVAL;
 	}
 
 	r = amdgpu_bo_kmap(bo, &ptr);
 	if (r) {
-		DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r);
+		DRM_ERROR("Failed mapping the UVD(%d) message (%ld)!\n", ip_instance, r);
 		return r;
 	}
 
@@ -675,7 +698,7 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
 	handle = msg[2];
 
 	if (handle == 0) {
-		DRM_ERROR("Invalid UVD handle!\n");
+		DRM_ERROR("Invalid UVD(%d) handle!\n", ip_instance);
 		return -EINVAL;
 	}
 
@@ -686,18 +709,18 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
 
 		/* try to alloc a new handle */
 		for (i = 0; i < adev->uvd.max_handles; ++i) {
-			if (atomic_read(&adev->uvd.handles[i]) == handle) {
-				DRM_ERROR("Handle 0x%x already in use!\n", handle);
+			if (atomic_read(&adev->uvd.inst[ip_instance].handles[i]) == handle) {
+				DRM_ERROR("(%d)Handle 0x%x already in use!\n", ip_instance, handle);
 				return -EINVAL;
 			}
 
-			if (!atomic_cmpxchg(&adev->uvd.handles[i], 0, handle)) {
-				adev->uvd.filp[i] = ctx->parser->filp;
+			if (!atomic_cmpxchg(&adev->uvd.inst[ip_instance].handles[i], 0, handle)) {
+				adev->uvd.inst[ip_instance].filp[i] = ctx->parser->filp;
 				return 0;
 			}
 		}
 
-		DRM_ERROR("No more free UVD handles!\n");
+		DRM_ERROR("No more free UVD(%d) handles!\n", ip_instance);
 		return -ENOSPC;
 
 	case 1:
@@ -709,27 +732,27 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
 
 		/* validate the handle */
 		for (i = 0; i < adev->uvd.max_handles; ++i) {
-			if (atomic_read(&adev->uvd.handles[i]) == handle) {
-				if (adev->uvd.filp[i] != ctx->parser->filp) {
-					DRM_ERROR("UVD handle collision detected!\n");
+			if (atomic_read(&adev->uvd.inst[ip_instance].handles[i]) == handle) {
+				if (adev->uvd.inst[ip_instance].filp[i] != ctx->parser->filp) {
+					DRM_ERROR("UVD(%d) handle collision detected!\n", ip_instance);
 					return -EINVAL;
 				}
 				return 0;
 			}
 		}
 
-		DRM_ERROR("Invalid UVD handle 0x%x!\n", handle);
+		DRM_ERROR("Invalid UVD(%d) handle 0x%x!\n", ip_instance, handle);
 		return -ENOENT;
 
 	case 2:
 		/* it's a destroy msg, free the handle */
 		for (i = 0; i < adev->uvd.max_handles; ++i)
-			atomic_cmpxchg(&adev->uvd.handles[i], handle, 0);
+			atomic_cmpxchg(&adev->uvd.inst[ip_instance].handles[i], handle, 0);
 		amdgpu_bo_kunmap(bo);
 		return 0;
 
 	default:
-		DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type);
+		DRM_ERROR("Illegal UVD(%d) message type (%d)!\n", ip_instance, msg_type);
 		return -EINVAL;
 	}
 	BUG();
@@ -800,7 +823,7 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
 		}
 
 		if ((cmd == 0 || cmd == 0x3) &&
-		    (start >> 28) != (ctx->parser->adev->uvd.gpu_addr >> 28)) {
+		    (start >> 28) != (ctx->parser->adev->uvd.inst->gpu_addr >> 28)) {
 			DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
 				  start, end);
 			return -EINVAL;
@@ -968,6 +991,8 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 	uint64_t addr;
 	long r;
 	int i;
+	unsigned offset_idx = 0;
+	unsigned offset[3] = { UVD_BASE_SI, 0, 0 };
 
 	amdgpu_bo_kunmap(bo);
 	amdgpu_bo_unpin(bo);
@@ -987,17 +1012,16 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 		goto err;
 
 	if (adev->asic_type >= CHIP_VEGA10) {
-		data[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0_VEGA10, 0);
-		data[1] = PACKET0(mmUVD_GPCOM_VCPU_DATA1_VEGA10, 0);
-		data[2] = PACKET0(mmUVD_GPCOM_VCPU_CMD_VEGA10, 0);
-		data[3] = PACKET0(mmUVD_NO_OP_VEGA10, 0);
-	} else {
-		data[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
-		data[1] = PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0);
-		data[2] = PACKET0(mmUVD_GPCOM_VCPU_CMD, 0);
-		data[3] = PACKET0(mmUVD_NO_OP, 0);
+		offset_idx = 1 + ring->me;
+		offset[1] = adev->reg_offset[UVD_HWIP][0][1];
+		offset[2] = adev->reg_offset[UVD_HWIP][1][1];
 	}
 
+	data[0] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_DATA0, 0);
+	data[1] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_DATA1, 0);
+	data[2] = PACKET0(offset[offset_idx] + UVD_GPCOM_VCPU_CMD, 0);
+	data[3] = PACKET0(offset[offset_idx] + UVD_NO_OP, 0);
+
 	ib = &job->ibs[0];
 	addr = amdgpu_bo_gpu_offset(bo);
 	ib->ptr[0] = data[0];
@@ -1033,7 +1057,7 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 		if (r)
 			goto err_free;
 
-		r = amdgpu_job_submit(job, ring, &adev->uvd.entity,
+		r = amdgpu_job_submit(job, ring, &adev->uvd.inst[ring->me].entity,
 				      AMDGPU_FENCE_OWNER_UNDEFINED, &f);
 		if (r)
 			goto err_free;
@@ -1121,8 +1145,15 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 static void amdgpu_uvd_idle_work_handler(struct work_struct *work)
 {
 	struct amdgpu_device *adev =
-		container_of(work, struct amdgpu_device, uvd.idle_work.work);
-	unsigned fences = amdgpu_fence_count_emitted(&adev->uvd.ring);
+		container_of(work, struct amdgpu_device, uvd.inst->idle_work.work);
+	unsigned fences = 0, i, j;
+
+	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring);
+		for (j = 0; j < adev->uvd.num_enc_rings; ++j) {
+			fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring_enc[j]);
+		}
+	}
 
 	if (fences == 0) {
 		if (adev->pm.dpm_enabled) {
@@ -1136,7 +1167,7 @@ static void amdgpu_uvd_idle_work_handler(struct work_struct *work)
 							       AMD_CG_STATE_GATE);
 		}
 	} else {
-		schedule_delayed_work(&adev->uvd.idle_work, UVD_IDLE_TIMEOUT);
+		schedule_delayed_work(&adev->uvd.inst->idle_work, UVD_IDLE_TIMEOUT);
 	}
 }
 
@@ -1148,7 +1179,7 @@ void amdgpu_uvd_ring_begin_use(struct amdgpu_ring *ring)
 	if (amdgpu_sriov_vf(adev))
 		return;
 
-	set_clocks = !cancel_delayed_work_sync(&adev->uvd.idle_work);
+	set_clocks = !cancel_delayed_work_sync(&adev->uvd.inst->idle_work);
 	if (set_clocks) {
 		if (adev->pm.dpm_enabled) {
 			amdgpu_dpm_enable_uvd(adev, true);
@@ -1165,7 +1196,7 @@ void amdgpu_uvd_ring_begin_use(struct amdgpu_ring *ring)
 void amdgpu_uvd_ring_end_use(struct amdgpu_ring *ring)
 {
 	if (!amdgpu_sriov_vf(ring->adev))
-		schedule_delayed_work(&ring->adev->uvd.idle_work, UVD_IDLE_TIMEOUT);
+		schedule_delayed_work(&ring->adev->uvd.inst->idle_work, UVD_IDLE_TIMEOUT);
 }
 
 /**
@@ -1179,27 +1210,28 @@ int amdgpu_uvd_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 {
 	struct dma_fence *fence;
 	long r;
+	uint32_t ip_instance = ring->me;
 
 	r = amdgpu_uvd_get_create_msg(ring, 1, NULL);
 	if (r) {
-		DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r);
+		DRM_ERROR("amdgpu: (%d)failed to get create msg (%ld).\n", ip_instance, r);
 		goto error;
 	}
 
 	r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence);
 	if (r) {
-		DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r);
+		DRM_ERROR("amdgpu: (%d)failed to get destroy ib (%ld).\n", ip_instance, r);
 		goto error;
 	}
 
 	r = dma_fence_wait_timeout(fence, false, timeout);
 	if (r == 0) {
-		DRM_ERROR("amdgpu: IB test timed out.\n");
+		DRM_ERROR("amdgpu: (%d)IB test timed out.\n", ip_instance);
 		r = -ETIMEDOUT;
 	} else if (r < 0) {
-		DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
+		DRM_ERROR("amdgpu: (%d)fence wait failed (%ld).\n", ip_instance, r);
 	} else {
-		DRM_DEBUG("ib test on ring %d succeeded\n",  ring->idx);
+		DRM_DEBUG("ib test on (%d)ring %d succeeded\n", ip_instance, ring->idx);
 		r = 0;
 	}
 
@@ -1227,7 +1259,7 @@ uint32_t amdgpu_uvd_used_handles(struct amdgpu_device *adev)
 		 * necessarily linear. So we need to count
 		 * all non-zero handles.
 		 */
-		if (atomic_read(&adev->uvd.handles[i]))
+		if (atomic_read(&adev->uvd.inst->handles[i]))
 			used_handles++;
 	}
 

drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h

@@ -31,30 +31,37 @@
 #define AMDGPU_UVD_SESSION_SIZE		(50*1024)
 #define AMDGPU_UVD_FIRMWARE_OFFSET	256
 
+#define AMDGPU_MAX_UVD_INSTANCES			2
+
 #define AMDGPU_UVD_FIRMWARE_SIZE(adev)    \
 	(AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(((const struct common_firmware_header *)(adev)->uvd.fw->data)->ucode_size_bytes) + \
 			       8) - AMDGPU_UVD_FIRMWARE_OFFSET)
 
-struct amdgpu_uvd {
+struct amdgpu_uvd_inst {
 	struct amdgpu_bo	*vcpu_bo;
 	void			*cpu_addr;
 	uint64_t		gpu_addr;
-	unsigned		fw_version;
 	void			*saved_bo;
-	unsigned		max_handles;
 	atomic_t		handles[AMDGPU_MAX_UVD_HANDLES];
 	struct drm_file		*filp[AMDGPU_MAX_UVD_HANDLES];
 	struct delayed_work	idle_work;
-	const struct firmware	*fw;	/* UVD firmware */
 	struct amdgpu_ring	ring;
 	struct amdgpu_ring	ring_enc[AMDGPU_MAX_UVD_ENC_RINGS];
 	struct amdgpu_irq_src	irq;
-	bool			address_64_bit;
-	bool			use_ctx_buf;
 	struct drm_sched_entity entity;
 	struct drm_sched_entity entity_enc;
 	uint32_t                srbm_soft_reset;
+};
+
+struct amdgpu_uvd {
+	const struct firmware	*fw;	/* UVD firmware */
+	unsigned		fw_version;
+	unsigned		max_handles;
 	unsigned		num_enc_rings;
+	uint8_t		num_uvd_inst;
+	bool			address_64_bit;
+	bool			use_ctx_buf;
+	struct amdgpu_uvd_inst		inst[AMDGPU_MAX_UVD_INSTANCES];
 };
 
 int amdgpu_uvd_sw_init(struct amdgpu_device *adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c

@@ -51,11 +51,13 @@
 #define FIRMWARE_FIJI		"amdgpu/fiji_vce.bin"
 #define FIRMWARE_STONEY		"amdgpu/stoney_vce.bin"
 #define FIRMWARE_POLARIS10	"amdgpu/polaris10_vce.bin"
-#define FIRMWARE_POLARIS11         "amdgpu/polaris11_vce.bin"
-#define FIRMWARE_POLARIS12         "amdgpu/polaris12_vce.bin"
+#define FIRMWARE_POLARIS11	"amdgpu/polaris11_vce.bin"
+#define FIRMWARE_POLARIS12	"amdgpu/polaris12_vce.bin"
+#define FIRMWARE_VEGAM		"amdgpu/vegam_vce.bin"
 
 #define FIRMWARE_VEGA10		"amdgpu/vega10_vce.bin"
 #define FIRMWARE_VEGA12		"amdgpu/vega12_vce.bin"
+#define FIRMWARE_VEGA20		"amdgpu/vega20_vce.bin"
 
 #ifdef CONFIG_DRM_AMDGPU_CIK
 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
@@ -71,9 +73,11 @@ MODULE_FIRMWARE(FIRMWARE_STONEY);
 MODULE_FIRMWARE(FIRMWARE_POLARIS10);
 MODULE_FIRMWARE(FIRMWARE_POLARIS11);
 MODULE_FIRMWARE(FIRMWARE_POLARIS12);
+MODULE_FIRMWARE(FIRMWARE_VEGAM);
 
 MODULE_FIRMWARE(FIRMWARE_VEGA10);
 MODULE_FIRMWARE(FIRMWARE_VEGA12);
+MODULE_FIRMWARE(FIRMWARE_VEGA20);
 
 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
 
@@ -132,12 +136,18 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
 	case CHIP_POLARIS12:
 		fw_name = FIRMWARE_POLARIS12;
 		break;
+	case CHIP_VEGAM:
+		fw_name = FIRMWARE_VEGAM;
+		break;
 	case CHIP_VEGA10:
 		fw_name = FIRMWARE_VEGA10;
 		break;
 	case CHIP_VEGA12:
 		fw_name = FIRMWARE_VEGA12;
 		break;
+	case CHIP_VEGA20:
+		fw_name = FIRMWARE_VEGA20;
+		break;
 
 	default:
 		return -EINVAL;
@@ -181,7 +191,7 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
 	ring = &adev->vce.ring[0];
 	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
 	r = drm_sched_entity_init(&ring->sched, &adev->vce.entity,
-				  rq, amdgpu_sched_jobs, NULL);
+				  rq, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
@@ -755,6 +765,18 @@ int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
 			if (r)
 				goto out;
 			break;
+
+		case 0x0500000d: /* MV buffer */
+			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 3,
+							idx + 2, 0, 0);
+			if (r)
+				goto out;
+
+			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 8,
+							idx + 7, 0, 0);
+			if (r)
+				goto out;
+			break;
 		}
 
 		idx += len / 4;
@@ -860,6 +882,18 @@ int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
 				goto out;
 			break;
 
+		case 0x0500000d: /* MV buffer */
+			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3,
+							idx + 2, *size, 0);
+			if (r)
+				goto out;
+
+			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 8,
+							idx + 7, *size / 12, 0);
+			if (r)
+				goto out;
+			break;
+
 		default:
 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
 			r = -EINVAL;

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c

@@ -105,7 +105,7 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 	ring = &adev->vcn.ring_dec;
 	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
 	r = drm_sched_entity_init(&ring->sched, &adev->vcn.entity_dec,
-				  rq, amdgpu_sched_jobs, NULL);
+				  rq, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCN dec run queue.\n");
 		return r;
@@ -114,7 +114,7 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 	ring = &adev->vcn.ring_enc[0];
 	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
 	r = drm_sched_entity_init(&ring->sched, &adev->vcn.entity_enc,
-				  rq, amdgpu_sched_jobs, NULL);
+				  rq, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCN enc run queue.\n");
 		return r;
@@ -205,13 +205,18 @@ static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
 	struct amdgpu_device *adev =
 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
 	unsigned fences = amdgpu_fence_count_emitted(&adev->vcn.ring_dec);
+	unsigned i;
+
+	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
+		fences += amdgpu_fence_count_emitted(&adev->vcn.ring_enc[i]);
+	}
 
 	if (fences == 0) {
-		if (adev->pm.dpm_enabled) {
-			/* might be used when with pg/cg
+		if (adev->pm.dpm_enabled)
 			amdgpu_dpm_enable_uvd(adev, false);
-			*/
-		}
+		else
+			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
+							       AMD_PG_STATE_GATE);
 	} else {
 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
 	}
@@ -223,9 +228,11 @@ void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
 	bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.idle_work);
 
 	if (set_clocks && adev->pm.dpm_enabled) {
-		/* might be used when with pg/cg
-		amdgpu_dpm_enable_uvd(adev, true);
-		*/
+		if (adev->pm.dpm_enabled)
+			amdgpu_dpm_enable_uvd(adev, true);
+		else
+			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
+							       AMD_PG_STATE_UNGATE);
 	}
 }
 

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h

@@ -45,6 +45,17 @@
 #define VCN_ENC_CMD_REG_WRITE		0x0000000b
 #define VCN_ENC_CMD_REG_WAIT		0x0000000c
 
+enum engine_status_constants {
+	UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON = 0x2AAAA0,
+	UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON = 0x00000002,
+	UVD_STATUS__UVD_BUSY = 0x00000004,
+	GB_ADDR_CONFIG_DEFAULT = 0x26010011,
+	UVD_STATUS__IDLE = 0x2,
+	UVD_STATUS__BUSY = 0x5,
+	UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF = 0x1,
+	UVD_STATUS__RBC_BUSY = 0x1,
+};
+
 struct amdgpu_vcn {
 	struct amdgpu_bo	*vcpu_bo;
 	void			*cpu_addr;

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -94,6 +94,34 @@ struct amdgpu_prt_cb {
 	struct dma_fence_cb cb;
 };
 
+static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+				   struct amdgpu_vm *vm,
+				   struct amdgpu_bo *bo)
+{
+	base->vm = vm;
+	base->bo = bo;
+	INIT_LIST_HEAD(&base->bo_list);
+	INIT_LIST_HEAD(&base->vm_status);
+
+	if (!bo)
+		return;
+	list_add_tail(&base->bo_list, &bo->va);
+
+	if (bo->tbo.resv != vm->root.base.bo->tbo.resv)
+		return;
+
+	if (bo->preferred_domains &
+	    amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type))
+		return;
+
+	/*
+	 * we checked all the prerequisites, but it looks like this per vm bo
+	 * is currently evicted. add the bo to the evicted list to make sure it
+	 * is validated on next vm use to avoid fault.
+	 * */
+	list_move_tail(&base->vm_status, &vm->evicted);
+}
+
 /**
  * amdgpu_vm_level_shift - return the addr shift for each level
  *
@@ -196,24 +224,16 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 			      void *param)
 {
 	struct ttm_bo_global *glob = adev->mman.bdev.glob;
-	int r;
+	struct amdgpu_vm_bo_base *bo_base, *tmp;
+	int r = 0;
 
-	spin_lock(&vm->status_lock);
-	while (!list_empty(&vm->evicted)) {
-		struct amdgpu_vm_bo_base *bo_base;
-		struct amdgpu_bo *bo;
+	list_for_each_entry_safe(bo_base, tmp, &vm->evicted, vm_status) {
+		struct amdgpu_bo *bo = bo_base->bo;
 
-		bo_base = list_first_entry(&vm->evicted,
-					   struct amdgpu_vm_bo_base,
-					   vm_status);
-		spin_unlock(&vm->status_lock);
-
-		bo = bo_base->bo;
-		BUG_ON(!bo);
 		if (bo->parent) {
 			r = validate(param, bo);
 			if (r)
-				return r;
+				break;
 
 			spin_lock(&glob->lru_lock);
 			ttm_bo_move_to_lru_tail(&bo->tbo);
@@ -222,22 +242,29 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 			spin_unlock(&glob->lru_lock);
 		}
 
-		if (bo->tbo.type == ttm_bo_type_kernel &&
-		    vm->use_cpu_for_update) {
-			r = amdgpu_bo_kmap(bo, NULL);
-			if (r)
-				return r;
-		}
-
-		spin_lock(&vm->status_lock);
-		if (bo->tbo.type != ttm_bo_type_kernel)
+		if (bo->tbo.type != ttm_bo_type_kernel) {
+			spin_lock(&vm->moved_lock);
 			list_move(&bo_base->vm_status, &vm->moved);
-		else
+			spin_unlock(&vm->moved_lock);
+		} else {
 			list_move(&bo_base->vm_status, &vm->relocated);
+		}
 	}
-	spin_unlock(&vm->status_lock);
 
-	return 0;
+	spin_lock(&glob->lru_lock);
+	list_for_each_entry(bo_base, &vm->idle, vm_status) {
+		struct amdgpu_bo *bo = bo_base->bo;
+
+		if (!bo->parent)
+			continue;
+
+		ttm_bo_move_to_lru_tail(&bo->tbo);
+		if (bo->shadow)
+			ttm_bo_move_to_lru_tail(&bo->shadow->tbo);
+	}
+	spin_unlock(&glob->lru_lock);
+
+	return r;
 }
 
 /**
@@ -249,13 +276,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
  */
 bool amdgpu_vm_ready(struct amdgpu_vm *vm)
 {
-	bool ready;
-
-	spin_lock(&vm->status_lock);
-	ready = list_empty(&vm->evicted);
-	spin_unlock(&vm->status_lock);
-
-	return ready;
+	return list_empty(&vm->evicted);
 }
 
 /**
@@ -412,11 +433,16 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 		struct amdgpu_bo *pt;
 
 		if (!entry->base.bo) {
-			r = amdgpu_bo_create(adev,
-					     amdgpu_vm_bo_size(adev, level),
-					     AMDGPU_GPU_PAGE_SIZE,
-					     AMDGPU_GEM_DOMAIN_VRAM, flags,
-					     ttm_bo_type_kernel, resv, &pt);
+			struct amdgpu_bo_param bp;
+
+			memset(&bp, 0, sizeof(bp));
+			bp.size = amdgpu_vm_bo_size(adev, level);
+			bp.byte_align = AMDGPU_GPU_PAGE_SIZE;
+			bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+			bp.flags = flags;
+			bp.type = ttm_bo_type_kernel;
+			bp.resv = resv;
+			r = amdgpu_bo_create(adev, &bp, &pt);
 			if (r)
 				return r;
 
@@ -441,12 +467,8 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 			*/
 			pt->parent = amdgpu_bo_ref(parent->base.bo);
 
-			entry->base.vm = vm;
-			entry->base.bo = pt;
-			list_add_tail(&entry->base.bo_list, &pt->va);
-			spin_lock(&vm->status_lock);
-			list_add(&entry->base.vm_status, &vm->relocated);
-			spin_unlock(&vm->status_lock);
+			amdgpu_vm_bo_base_init(&entry->base, vm, pt);
+			list_move(&entry->base.vm_status, &vm->relocated);
 		}
 
 		if (level < AMDGPU_VM_PTB) {
@@ -628,7 +650,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_
 		amdgpu_gmc_emit_pasid_mapping(ring, job->vmid, job->pasid);
 
 	if (vm_flush_needed || pasid_mapping_needed) {
-		r = amdgpu_fence_emit(ring, &fence);
+		r = amdgpu_fence_emit(ring, &fence, 0);
 		if (r)
 			return r;
 	}
@@ -893,10 +915,8 @@ static void amdgpu_vm_invalidate_level(struct amdgpu_device *adev,
 		if (!entry->base.bo)
 			continue;
 
-		spin_lock(&vm->status_lock);
-		if (list_empty(&entry->base.vm_status))
-			list_add(&entry->base.vm_status, &vm->relocated);
-		spin_unlock(&vm->status_lock);
+		if (!entry->base.moved)
+			list_move(&entry->base.vm_status, &vm->relocated);
 		amdgpu_vm_invalidate_level(adev, vm, entry, level + 1);
 	}
 }
@@ -926,6 +946,14 @@ restart:
 	params.adev = adev;
 
 	if (vm->use_cpu_for_update) {
+		struct amdgpu_vm_bo_base *bo_base;
+
+		list_for_each_entry(bo_base, &vm->relocated, vm_status) {
+			r = amdgpu_bo_kmap(bo_base->bo, NULL);
+			if (unlikely(r))
+				return r;
+		}
+
 		r = amdgpu_vm_wait_pd(adev, vm, AMDGPU_FENCE_OWNER_VM);
 		if (unlikely(r))
 			return r;
@@ -941,7 +969,6 @@ restart:
 		params.func = amdgpu_vm_do_set_ptes;
 	}
 
-	spin_lock(&vm->status_lock);
 	while (!list_empty(&vm->relocated)) {
 		struct amdgpu_vm_bo_base *bo_base, *parent;
 		struct amdgpu_vm_pt *pt, *entry;
@@ -950,14 +977,12 @@ restart:
 		bo_base = list_first_entry(&vm->relocated,
 					   struct amdgpu_vm_bo_base,
 					   vm_status);
-		list_del_init(&bo_base->vm_status);
-		spin_unlock(&vm->status_lock);
+		bo_base->moved = false;
+		list_move(&bo_base->vm_status, &vm->idle);
 
 		bo = bo_base->bo->parent;
-		if (!bo) {
-			spin_lock(&vm->status_lock);
+		if (!bo)
 			continue;
-		}
 
 		parent = list_first_entry(&bo->va, struct amdgpu_vm_bo_base,
 					  bo_list);
@@ -966,12 +991,10 @@ restart:
 
 		amdgpu_vm_update_pde(&params, vm, pt, entry);
 
-		spin_lock(&vm->status_lock);
 		if (!vm->use_cpu_for_update &&
 		    (ndw - params.ib->length_dw) < 32)
 			break;
 	}
-	spin_unlock(&vm->status_lock);
 
 	if (vm->use_cpu_for_update) {
 		/* Flush HDP */
@@ -1074,9 +1097,7 @@ static void amdgpu_vm_handle_huge_pages(struct amdgpu_pte_update_params *p,
 		if (entry->huge) {
 			/* Add the entry to the relocated list to update it. */
 			entry->huge = false;
-			spin_lock(&p->vm->status_lock);
 			list_move(&entry->base.vm_status, &p->vm->relocated);
-			spin_unlock(&p->vm->status_lock);
 		}
 		return;
 	}
@@ -1555,9 +1576,22 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 		amdgpu_asic_flush_hdp(adev, NULL);
 	}
 
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	list_del_init(&bo_va->base.vm_status);
-	spin_unlock(&vm->status_lock);
+	spin_unlock(&vm->moved_lock);
+
+	/* If the BO is not in its preferred location add it back to
+	 * the evicted list so that it gets validated again on the
+	 * next command submission.
+	 */
+	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
+		uint32_t mem_type = bo->tbo.mem.mem_type;
+
+		if (!(bo->preferred_domains & amdgpu_mem_type_to_domain(mem_type)))
+			list_add_tail(&bo_va->base.vm_status, &vm->evicted);
+		else
+			list_add(&bo_va->base.vm_status, &vm->idle);
+	}
 
 	list_splice_init(&bo_va->invalids, &bo_va->valids);
 	bo_va->cleared = clear;
@@ -1766,19 +1800,18 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 			   struct amdgpu_vm *vm)
 {
+	struct amdgpu_bo_va *bo_va, *tmp;
+	struct list_head moved;
 	bool clear;
-	int r = 0;
-
-	spin_lock(&vm->status_lock);
-	while (!list_empty(&vm->moved)) {
-		struct amdgpu_bo_va *bo_va;
-		struct reservation_object *resv;
+	int r;
 
-		bo_va = list_first_entry(&vm->moved,
-			struct amdgpu_bo_va, base.vm_status);
-		spin_unlock(&vm->status_lock);
+	INIT_LIST_HEAD(&moved);
+	spin_lock(&vm->moved_lock);
+	list_splice_init(&vm->moved, &moved);
+	spin_unlock(&vm->moved_lock);
 
-		resv = bo_va->base.bo->tbo.resv;
+	list_for_each_entry_safe(bo_va, tmp, &moved, base.vm_status) {
+		struct reservation_object *resv = bo_va->base.bo->tbo.resv;
 
 		/* Per VM BOs never need to bo cleared in the page tables */
 		if (resv == vm->root.base.bo->tbo.resv)
@@ -1791,17 +1824,19 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 			clear = true;
 
 		r = amdgpu_vm_bo_update(adev, bo_va, clear);
-		if (r)
+		if (r) {
+			spin_lock(&vm->moved_lock);
+			list_splice(&moved, &vm->moved);
+			spin_unlock(&vm->moved_lock);
 			return r;
+		}
 
 		if (!clear && resv != vm->root.base.bo->tbo.resv)
 			reservation_object_unlock(resv);
 
-		spin_lock(&vm->status_lock);
 	}
-	spin_unlock(&vm->status_lock);
 
-	return r;
+	return 0;
 }
 
 /**
@@ -1827,36 +1862,12 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 	if (bo_va == NULL) {
 		return NULL;
 	}
-	bo_va->base.vm = vm;
-	bo_va->base.bo = bo;
-	INIT_LIST_HEAD(&bo_va->base.bo_list);
-	INIT_LIST_HEAD(&bo_va->base.vm_status);
+	amdgpu_vm_bo_base_init(&bo_va->base, vm, bo);
 
 	bo_va->ref_count = 1;
 	INIT_LIST_HEAD(&bo_va->valids);
 	INIT_LIST_HEAD(&bo_va->invalids);
 
-	if (!bo)
-		return bo_va;
-
-	list_add_tail(&bo_va->base.bo_list, &bo->va);
-
-	if (bo->tbo.resv != vm->root.base.bo->tbo.resv)
-		return bo_va;
-
-	if (bo->preferred_domains &
-	    amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type))
-		return bo_va;
-
-	/*
-	 * We checked all the prerequisites, but it looks like this per VM BO
-	 * is currently evicted. add the BO to the evicted list to make sure it
-	 * is validated on next VM use to avoid fault.
-	 * */
-	spin_lock(&vm->status_lock);
-	list_move_tail(&bo_va->base.vm_status, &vm->evicted);
-	spin_unlock(&vm->status_lock);
-
 	return bo_va;
 }
 
@@ -1884,11 +1895,11 @@ static void amdgpu_vm_bo_insert_map(struct amdgpu_device *adev,
 	if (mapping->flags & AMDGPU_PTE_PRT)
 		amdgpu_vm_prt_get(adev);
 
-	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
-		spin_lock(&vm->status_lock);
-		if (list_empty(&bo_va->base.vm_status))
-			list_add(&bo_va->base.vm_status, &vm->moved);
-		spin_unlock(&vm->status_lock);
+	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv &&
+	    !bo_va->base.moved) {
+		spin_lock(&vm->moved_lock);
+		list_move(&bo_va->base.vm_status, &vm->moved);
+		spin_unlock(&vm->moved_lock);
 	}
 	trace_amdgpu_vm_bo_map(bo_va, mapping);
 }
@@ -2198,9 +2209,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 
 	list_del(&bo_va->base.bo_list);
 
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	list_del(&bo_va->base.vm_status);
-	spin_unlock(&vm->status_lock);
+	spin_unlock(&vm->moved_lock);
 
 	list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
 		list_del(&mapping->list);
@@ -2234,33 +2245,34 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 {
 	struct amdgpu_vm_bo_base *bo_base;
 
+	/* shadow bo doesn't have bo base, its validation needs its parent */
+	if (bo->parent && bo->parent->shadow == bo)
+		bo = bo->parent;
+
 	list_for_each_entry(bo_base, &bo->va, bo_list) {
 		struct amdgpu_vm *vm = bo_base->vm;
+		bool was_moved = bo_base->moved;
 
 		bo_base->moved = true;
 		if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
-			spin_lock(&bo_base->vm->status_lock);
 			if (bo->tbo.type == ttm_bo_type_kernel)
 				list_move(&bo_base->vm_status, &vm->evicted);
 			else
 				list_move_tail(&bo_base->vm_status,
 					       &vm->evicted);
-			spin_unlock(&bo_base->vm->status_lock);
 			continue;
 		}
 
-		if (bo->tbo.type == ttm_bo_type_kernel) {
-			spin_lock(&bo_base->vm->status_lock);
-			if (list_empty(&bo_base->vm_status))
-				list_add(&bo_base->vm_status, &vm->relocated);
-			spin_unlock(&bo_base->vm->status_lock);
+		if (was_moved)
 			continue;
-		}
 
-		spin_lock(&bo_base->vm->status_lock);
-		if (list_empty(&bo_base->vm_status))
-			list_add(&bo_base->vm_status, &vm->moved);
-		spin_unlock(&bo_base->vm->status_lock);
+		if (bo->tbo.type == ttm_bo_type_kernel) {
+			list_move(&bo_base->vm_status, &vm->relocated);
+		} else {
+			spin_lock(&bo_base->vm->moved_lock);
+			list_move(&bo_base->vm_status, &vm->moved);
+			spin_unlock(&bo_base->vm->moved_lock);
+		}
 	}
 }
 
@@ -2355,6 +2367,8 @@ void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t vm_size,
 int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		   int vm_context, unsigned int pasid)
 {
+	struct amdgpu_bo_param bp;
+	struct amdgpu_bo *root;
 	const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE,
 		AMDGPU_VM_PTE_COUNT(adev) * 8);
 	unsigned ring_instance;
@@ -2367,10 +2381,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	vm->va = RB_ROOT_CACHED;
 	for (i = 0; i < AMDGPU_MAX_VMHUBS; i++)
 		vm->reserved_vmid[i] = NULL;
-	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->evicted);
 	INIT_LIST_HEAD(&vm->relocated);
+	spin_lock_init(&vm->moved_lock);
 	INIT_LIST_HEAD(&vm->moved);
+	INIT_LIST_HEAD(&vm->idle);
 	INIT_LIST_HEAD(&vm->freed);
 
 	/* create scheduler entity for page table updates */
@@ -2380,7 +2395,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	ring = adev->vm_manager.vm_pte_rings[ring_instance];
 	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
 	r = drm_sched_entity_init(&ring->sched, &vm->entity,
-				  rq, amdgpu_sched_jobs, NULL);
+				  rq, NULL);
 	if (r)
 		return r;
 
@@ -2409,24 +2424,28 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		flags |= AMDGPU_GEM_CREATE_SHADOW;
 
 	size = amdgpu_vm_bo_size(adev, adev->vm_manager.root_level);
-	r = amdgpu_bo_create(adev, size, align, AMDGPU_GEM_DOMAIN_VRAM, flags,
-			     ttm_bo_type_kernel, NULL, &vm->root.base.bo);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = align;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = flags;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	r = amdgpu_bo_create(adev, &bp, &root);
 	if (r)
 		goto error_free_sched_entity;
 
-	r = amdgpu_bo_reserve(vm->root.base.bo, true);
+	r = amdgpu_bo_reserve(root, true);
 	if (r)
 		goto error_free_root;
 
-	r = amdgpu_vm_clear_bo(adev, vm, vm->root.base.bo,
+	r = amdgpu_vm_clear_bo(adev, vm, root,
 			       adev->vm_manager.root_level,
 			       vm->pte_support_ats);
 	if (r)
 		goto error_unreserve;
 
-	vm->root.base.vm = vm;
-	list_add_tail(&vm->root.base.bo_list, &vm->root.base.bo->va);
-	list_add_tail(&vm->root.base.vm_status, &vm->evicted);
+	amdgpu_vm_bo_base_init(&vm->root.base, vm, root);
 	amdgpu_bo_unreserve(vm->root.base.bo);
 
 	if (pasid) {

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h

@@ -75,11 +75,12 @@ struct amdgpu_bo_list_entry;
 /* PDE Block Fragment Size for VEGA10 */
 #define AMDGPU_PDE_BFS(a)	((uint64_t)a << 59)
 
-/* VEGA10 only */
+
+/* For GFX9 */
 #define AMDGPU_PTE_MTYPE(a)    ((uint64_t)a << 57)
 #define AMDGPU_PTE_MTYPE_MASK	AMDGPU_PTE_MTYPE(3ULL)
 
-/* For Raven */
+#define AMDGPU_MTYPE_NC 0
 #define AMDGPU_MTYPE_CC 2
 
 #define AMDGPU_PTE_DEFAULT_ATC  (AMDGPU_PTE_SYSTEM      \
@@ -167,9 +168,6 @@ struct amdgpu_vm {
 	/* tree of virtual addresses mapped */
 	struct rb_root_cached	va;
 
-	/* protecting invalidated */
-	spinlock_t		status_lock;
-
 	/* BOs who needs a validation */
 	struct list_head	evicted;
 
@@ -178,6 +176,10 @@ struct amdgpu_vm {
 
 	/* BOs moved, but not yet updated in the PT */
 	struct list_head	moved;
+	spinlock_t		moved_lock;
+
+	/* All BOs of this VM not currently in the state machine */
+	struct list_head	idle;
 
 	/* BO mappings freed, but not yet updated in the PT */
 	struct list_head	freed;
@@ -186,9 +188,6 @@ struct amdgpu_vm {
 	struct amdgpu_vm_pt     root;
 	struct dma_fence	*last_update;
 
-	/* protecting freed */
-	spinlock_t		freed_lock;
-
 	/* Scheduler entity for page table updates */
 	struct drm_sched_entity	entity;
 

drivers/gpu/drm/amd/amdgpu/ci_dpm.c

@@ -5903,7 +5903,7 @@ static int ci_dpm_init(struct amdgpu_device *adev)
 	pi->pcie_dpm_key_disabled = 0;
 	pi->thermal_sclk_dpm_enabled = 0;
 
-	if (amdgpu_pp_feature_mask & SCLK_DEEP_SLEEP_MASK)
+	if (adev->powerplay.pp_feature & PP_SCLK_DEEP_SLEEP_MASK)
 		pi->caps_sclk_ds = true;
 	else
 		pi->caps_sclk_ds = false;
@@ -6255,7 +6255,7 @@ static int ci_dpm_late_init(void *handle)
 	int ret;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	if (!amdgpu_dpm)
+	if (!adev->pm.dpm_enabled)
 		return 0;
 
 	/* init the sysfs and debugfs files late */

drivers/gpu/drm/amd/amdgpu/cik.c

@@ -1735,6 +1735,12 @@ static void cik_invalidate_hdp(struct amdgpu_device *adev,
 	}
 }
 
+static bool cik_need_full_reset(struct amdgpu_device *adev)
+{
+	/* change this when we support soft reset */
+	return true;
+}
+
 static const struct amdgpu_asic_funcs cik_asic_funcs =
 {
 	.read_disabled_bios = &cik_read_disabled_bios,
@@ -1748,6 +1754,7 @@ static const struct amdgpu_asic_funcs cik_asic_funcs =
 	.get_config_memsize = &cik_get_config_memsize,
 	.flush_hdp = &cik_flush_hdp,
 	.invalidate_hdp = &cik_invalidate_hdp,
+	.need_full_reset = &cik_need_full_reset,
 };
 
 static int cik_common_early_init(void *handle)

drivers/gpu/drm/amd/amdgpu/dce_v10_0.c

@@ -1823,7 +1823,6 @@ static int dce_v10_0_crtc_do_set_base(struct drm_crtc *crtc,
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
 	struct drm_device *dev = crtc->dev;
 	struct amdgpu_device *adev = dev->dev_private;
-	struct amdgpu_framebuffer *amdgpu_fb;
 	struct drm_framebuffer *target_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *abo;
@@ -1842,18 +1841,15 @@ static int dce_v10_0_crtc_do_set_base(struct drm_crtc *crtc,
 		return 0;
 	}
 
-	if (atomic) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
+	if (atomic)
 		target_fb = fb;
-	} else {
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
+	else
 		target_fb = crtc->primary->fb;
-	}
 
 	/* If atomic, assume fb object is pinned & idle & fenced and
 	 * just update base pointers
 	 */
-	obj = amdgpu_fb->obj;
+	obj = target_fb->obj[0];
 	abo = gem_to_amdgpu_bo(obj);
 	r = amdgpu_bo_reserve(abo, false);
 	if (unlikely(r != 0))
@@ -2043,8 +2039,7 @@ static int dce_v10_0_crtc_do_set_base(struct drm_crtc *crtc,
 	WREG32(mmMASTER_UPDATE_MODE + amdgpu_crtc->crtc_offset, 0);
 
 	if (!atomic && fb && fb != crtc->primary->fb) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r != 0))
 			return r;
@@ -2526,11 +2521,9 @@ static void dce_v10_0_crtc_disable(struct drm_crtc *crtc)
 	dce_v10_0_crtc_dpms(crtc, DRM_MODE_DPMS_OFF);
 	if (crtc->primary->fb) {
 		int r;
-		struct amdgpu_framebuffer *amdgpu_fb;
 		struct amdgpu_bo *abo;
 
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r))
 			DRM_ERROR("failed to reserve abo before unpin\n");

drivers/gpu/drm/amd/amdgpu/dce_v11_0.c

@@ -173,6 +173,7 @@ static void dce_v11_0_init_golden_registers(struct amdgpu_device *adev)
 							ARRAY_SIZE(polaris11_golden_settings_a11));
 		break;
 	case CHIP_POLARIS10:
+	case CHIP_VEGAM:
 		amdgpu_device_program_register_sequence(adev,
 							polaris10_golden_settings_a11,
 							ARRAY_SIZE(polaris10_golden_settings_a11));
@@ -473,6 +474,7 @@ static int dce_v11_0_get_num_crtc (struct amdgpu_device *adev)
 		num_crtc = 2;
 		break;
 	case CHIP_POLARIS10:
+	case CHIP_VEGAM:
 		num_crtc = 6;
 		break;
 	case CHIP_POLARIS11:
@@ -1445,6 +1447,7 @@ static int dce_v11_0_audio_init(struct amdgpu_device *adev)
 		adev->mode_info.audio.num_pins = 7;
 		break;
 	case CHIP_POLARIS10:
+	case CHIP_VEGAM:
 		adev->mode_info.audio.num_pins = 8;
 		break;
 	case CHIP_POLARIS11:
@@ -1862,7 +1865,6 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc *crtc,
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
 	struct drm_device *dev = crtc->dev;
 	struct amdgpu_device *adev = dev->dev_private;
-	struct amdgpu_framebuffer *amdgpu_fb;
 	struct drm_framebuffer *target_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *abo;
@@ -1881,18 +1883,15 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc *crtc,
 		return 0;
 	}
 
-	if (atomic) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
+	if (atomic)
 		target_fb = fb;
-	} else {
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
+	else
 		target_fb = crtc->primary->fb;
-	}
 
 	/* If atomic, assume fb object is pinned & idle & fenced and
 	 * just update base pointers
 	 */
-	obj = amdgpu_fb->obj;
+	obj = target_fb->obj[0];
 	abo = gem_to_amdgpu_bo(obj);
 	r = amdgpu_bo_reserve(abo, false);
 	if (unlikely(r != 0))
@@ -2082,8 +2081,7 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc *crtc,
 	WREG32(mmCRTC_MASTER_UPDATE_MODE + amdgpu_crtc->crtc_offset, 0);
 
 	if (!atomic && fb && fb != crtc->primary->fb) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r != 0))
 			return r;
@@ -2253,7 +2251,8 @@ static u32 dce_v11_0_pick_pll(struct drm_crtc *crtc)
 
 	if ((adev->asic_type == CHIP_POLARIS10) ||
 	    (adev->asic_type == CHIP_POLARIS11) ||
-	    (adev->asic_type == CHIP_POLARIS12)) {
+	    (adev->asic_type == CHIP_POLARIS12) ||
+	    (adev->asic_type == CHIP_VEGAM)) {
 		struct amdgpu_encoder *amdgpu_encoder =
 			to_amdgpu_encoder(amdgpu_crtc->encoder);
 		struct amdgpu_encoder_atom_dig *dig = amdgpu_encoder->enc_priv;
@@ -2601,11 +2600,9 @@ static void dce_v11_0_crtc_disable(struct drm_crtc *crtc)
 	dce_v11_0_crtc_dpms(crtc, DRM_MODE_DPMS_OFF);
 	if (crtc->primary->fb) {
 		int r;
-		struct amdgpu_framebuffer *amdgpu_fb;
 		struct amdgpu_bo *abo;
 
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r))
 			DRM_ERROR("failed to reserve abo before unpin\n");
@@ -2673,7 +2670,8 @@ static int dce_v11_0_crtc_mode_set(struct drm_crtc *crtc,
 
 	if ((adev->asic_type == CHIP_POLARIS10) ||
 	    (adev->asic_type == CHIP_POLARIS11) ||
-	    (adev->asic_type == CHIP_POLARIS12)) {
+	    (adev->asic_type == CHIP_POLARIS12) ||
+	    (adev->asic_type == CHIP_VEGAM)) {
 		struct amdgpu_encoder *amdgpu_encoder =
 			to_amdgpu_encoder(amdgpu_crtc->encoder);
 		int encoder_mode =
@@ -2830,6 +2828,7 @@ static int dce_v11_0_early_init(void *handle)
 		adev->mode_info.num_dig = 9;
 		break;
 	case CHIP_POLARIS10:
+	case CHIP_VEGAM:
 		adev->mode_info.num_hpd = 6;
 		adev->mode_info.num_dig = 6;
 		break;
@@ -2949,7 +2948,8 @@ static int dce_v11_0_hw_init(void *handle)
 	amdgpu_atombios_encoder_init_dig(adev);
 	if ((adev->asic_type == CHIP_POLARIS10) ||
 	    (adev->asic_type == CHIP_POLARIS11) ||
-	    (adev->asic_type == CHIP_POLARIS12)) {
+	    (adev->asic_type == CHIP_POLARIS12) ||
+	    (adev->asic_type == CHIP_VEGAM)) {
 		amdgpu_atombios_crtc_set_dce_clock(adev, adev->clock.default_dispclk,
 						   DCE_CLOCK_TYPE_DISPCLK, ATOM_GCK_DFS);
 		amdgpu_atombios_crtc_set_dce_clock(adev, 0,

drivers/gpu/drm/amd/amdgpu/dce_v6_0.c

@@ -1780,7 +1780,6 @@ static int dce_v6_0_crtc_do_set_base(struct drm_crtc *crtc,
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
 	struct drm_device *dev = crtc->dev;
 	struct amdgpu_device *adev = dev->dev_private;
-	struct amdgpu_framebuffer *amdgpu_fb;
 	struct drm_framebuffer *target_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *abo;
@@ -1798,18 +1797,15 @@ static int dce_v6_0_crtc_do_set_base(struct drm_crtc *crtc,
 		return 0;
 	}
 
-	if (atomic) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
+	if (atomic)
 		target_fb = fb;
-	} else {
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
+	else
 		target_fb = crtc->primary->fb;
-	}
 
 	/* If atomic, assume fb object is pinned & idle & fenced and
 	 * just update base pointers
 	 */
-	obj = amdgpu_fb->obj;
+	obj = target_fb->obj[0];
 	abo = gem_to_amdgpu_bo(obj);
 	r = amdgpu_bo_reserve(abo, false);
 	if (unlikely(r != 0))
@@ -1978,8 +1974,7 @@ static int dce_v6_0_crtc_do_set_base(struct drm_crtc *crtc,
 	WREG32(mmMASTER_UPDATE_MODE + amdgpu_crtc->crtc_offset, 0);
 
 	if (!atomic && fb && fb != crtc->primary->fb) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r != 0))
 			return r;
@@ -2414,11 +2409,9 @@ static void dce_v6_0_crtc_disable(struct drm_crtc *crtc)
 	dce_v6_0_crtc_dpms(crtc, DRM_MODE_DPMS_OFF);
 	if (crtc->primary->fb) {
 		int r;
-		struct amdgpu_framebuffer *amdgpu_fb;
 		struct amdgpu_bo *abo;
 
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r))
 			DRM_ERROR("failed to reserve abo before unpin\n");

drivers/gpu/drm/amd/amdgpu/dce_v8_0.c

@@ -1754,7 +1754,6 @@ static int dce_v8_0_crtc_do_set_base(struct drm_crtc *crtc,
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
 	struct drm_device *dev = crtc->dev;
 	struct amdgpu_device *adev = dev->dev_private;
-	struct amdgpu_framebuffer *amdgpu_fb;
 	struct drm_framebuffer *target_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *abo;
@@ -1773,18 +1772,15 @@ static int dce_v8_0_crtc_do_set_base(struct drm_crtc *crtc,
 		return 0;
 	}
 
-	if (atomic) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
+	if (atomic)
 		target_fb = fb;
-	} else {
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
+	else
 		target_fb = crtc->primary->fb;
-	}
 
 	/* If atomic, assume fb object is pinned & idle & fenced and
 	 * just update base pointers
 	 */
-	obj = amdgpu_fb->obj;
+	obj = target_fb->obj[0];
 	abo = gem_to_amdgpu_bo(obj);
 	r = amdgpu_bo_reserve(abo, false);
 	if (unlikely(r != 0))
@@ -1955,8 +1951,7 @@ static int dce_v8_0_crtc_do_set_base(struct drm_crtc *crtc,
 	WREG32(mmMASTER_UPDATE_MODE + amdgpu_crtc->crtc_offset, 0);
 
 	if (!atomic && fb && fb != crtc->primary->fb) {
-		amdgpu_fb = to_amdgpu_framebuffer(fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r != 0))
 			return r;
@@ -2430,11 +2425,9 @@ static void dce_v8_0_crtc_disable(struct drm_crtc *crtc)
 	dce_v8_0_crtc_dpms(crtc, DRM_MODE_DPMS_OFF);
 	if (crtc->primary->fb) {
 		int r;
-		struct amdgpu_framebuffer *amdgpu_fb;
 		struct amdgpu_bo *abo;
 
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r))
 			DRM_ERROR("failed to reserve abo before unpin\n");

drivers/gpu/drm/amd/amdgpu/dce_virtual.c

@@ -168,11 +168,9 @@ static void dce_virtual_crtc_disable(struct drm_crtc *crtc)
 	dce_virtual_crtc_dpms(crtc, DRM_MODE_DPMS_OFF);
 	if (crtc->primary->fb) {
 		int r;
-		struct amdgpu_framebuffer *amdgpu_fb;
 		struct amdgpu_bo *abo;
 
-		amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-		abo = gem_to_amdgpu_bo(amdgpu_fb->obj);
+		abo = gem_to_amdgpu_bo(crtc->primary->fb->obj[0]);
 		r = amdgpu_bo_reserve(abo, true);
 		if (unlikely(r))
 			DRM_ERROR("failed to reserve abo before unpin\n");
@@ -329,7 +327,7 @@ static int dce_virtual_get_modes(struct drm_connector *connector)
 	return 0;
 }
 
-static int dce_virtual_mode_valid(struct drm_connector *connector,
+static enum drm_mode_status dce_virtual_mode_valid(struct drm_connector *connector,
 				  struct drm_display_mode *mode)
 {
 	return MODE_OK;
@@ -462,8 +460,9 @@ static int dce_virtual_hw_init(void *handle)
 		break;
 	case CHIP_CARRIZO:
 	case CHIP_STONEY:
-	case CHIP_POLARIS11:
 	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
+	case CHIP_VEGAM:
 		dce_v11_0_disable_dce(adev);
 		break;
 	case CHIP_TOPAZ:
@@ -474,6 +473,7 @@ static int dce_virtual_hw_init(void *handle)
 		break;
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 		break;
 	default:
 		DRM_ERROR("Virtual display unsupported ASIC type: 0x%X\n", adev->asic_type);

drivers/gpu/drm/amd/amdgpu/df_v1_7.c

@@ -0,0 +1,120 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu.h"
+#include "df_v1_7.h"
+
+#include "df/df_1_7_default.h"
+#include "df/df_1_7_offset.h"
+#include "df/df_1_7_sh_mask.h"
+
+static u32 df_v1_7_channel_number[] = {1, 2, 0, 4, 0, 8, 0, 16, 2};
+
+static void df_v1_7_init (struct amdgpu_device *adev)
+{
+}
+
+static void df_v1_7_enable_broadcast_mode(struct amdgpu_device *adev,
+                                          bool enable)
+{
+	u32 tmp;
+
+	if (enable) {
+		tmp = RREG32_SOC15(DF, 0, mmFabricConfigAccessControl);
+		tmp &= ~FabricConfigAccessControl__CfgRegInstAccEn_MASK;
+		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl, tmp);
+	} else
+		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl,
+			     mmFabricConfigAccessControl_DEFAULT);
+}
+
+static u32 df_v1_7_get_fb_channel_number(struct amdgpu_device *adev)
+{
+	u32 tmp;
+
+	tmp = RREG32_SOC15(DF, 0, mmDF_CS_AON0_DramBaseAddress0);
+	tmp &= DF_CS_AON0_DramBaseAddress0__IntLvNumChan_MASK;
+	tmp >>= DF_CS_AON0_DramBaseAddress0__IntLvNumChan__SHIFT;
+
+	return tmp;
+}
+
+static u32 df_v1_7_get_hbm_channel_number(struct amdgpu_device *adev)
+{
+	int fb_channel_number;
+
+	fb_channel_number = adev->df_funcs->get_fb_channel_number(adev);
+
+	return df_v1_7_channel_number[fb_channel_number];
+}
+
+static void df_v1_7_update_medium_grain_clock_gating(struct amdgpu_device *adev,
+						     bool enable)
+{
+	u32 tmp;
+
+	/* Put DF on broadcast mode */
+	adev->df_funcs->enable_broadcast_mode(adev, true);
+
+	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DF_MGCG)) {
+		tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+		tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
+		tmp |= DF_V1_7_MGCG_ENABLE_15_CYCLE_DELAY;
+		WREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater, tmp);
+	} else {
+		tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+		tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
+		tmp |= DF_V1_7_MGCG_DISABLE;
+		WREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater, tmp);
+	}
+
+	/* Exit boradcast mode */
+	adev->df_funcs->enable_broadcast_mode(adev, false);
+}
+
+static void df_v1_7_get_clockgating_state(struct amdgpu_device *adev,
+					  u32 *flags)
+{
+	u32 tmp;
+
+	/* AMD_CG_SUPPORT_DF_MGCG */
+	tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+	if (tmp & DF_V1_7_MGCG_ENABLE_15_CYCLE_DELAY)
+		*flags |= AMD_CG_SUPPORT_DF_MGCG;
+}
+
+static void df_v1_7_enable_ecc_force_par_wr_rmw(struct amdgpu_device *adev,
+						bool enable)
+{
+	WREG32_FIELD15(DF, 0, DF_CS_AON0_CoherentSlaveModeCtrlA0,
+		       ForceParWrRMW, enable);
+}
+
+const struct amdgpu_df_funcs df_v1_7_funcs = {
+	.init = df_v1_7_init,
+	.enable_broadcast_mode = df_v1_7_enable_broadcast_mode,
+	.get_fb_channel_number = df_v1_7_get_fb_channel_number,
+	.get_hbm_channel_number = df_v1_7_get_hbm_channel_number,
+	.update_medium_grain_clock_gating = df_v1_7_update_medium_grain_clock_gating,
+	.get_clockgating_state = df_v1_7_get_clockgating_state,
+	.enable_ecc_force_par_wr_rmw = df_v1_7_enable_ecc_force_par_wr_rmw,
+};

drivers/gpu/drm/nouveau/nvkm/engine/disp/basegk104.c → drivers/gpu/drm/amd/amdgpu/df_v1_7.h

@@ -1,5 +1,5 @@
 /*
- * Copyright 2015 Red Hat Inc.
+ * Copyright 2018 Advanced Micro Devices, Inc.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -19,20 +19,22 @@
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  *
- * Authors: Ben Skeggs <bskeggs@redhat.com>
  */
-#include "dmacnv50.h"
-#include "rootnv50.h"
 
-#include <nvif/class.h>
+#ifndef __DF_V1_7_H__
+#define __DF_V1_7_H__
 
-const struct nv50_disp_dmac_oclass
-gk104_disp_base_oclass = {
-	.base.oclass = GK104_DISP_BASE_CHANNEL_DMA,
-	.base.minver = 0,
-	.base.maxver = 0,
-	.ctor = nv50_disp_base_new,
-	.func = &gf119_disp_dmac_func,
-	.mthd = &gf119_disp_base_chan_mthd,
-	.chid = 1,
+#include "soc15_common.h"
+enum DF_V1_7_MGCG
+{
+	DF_V1_7_MGCG_DISABLE = 0,
+	DF_V1_7_MGCG_ENABLE_00_CYCLE_DELAY =1,
+	DF_V1_7_MGCG_ENABLE_01_CYCLE_DELAY =2,
+	DF_V1_7_MGCG_ENABLE_15_CYCLE_DELAY =13,
+	DF_V1_7_MGCG_ENABLE_31_CYCLE_DELAY =14,
+	DF_V1_7_MGCG_ENABLE_63_CYCLE_DELAY =15
 };
+
+extern const struct amdgpu_df_funcs df_v1_7_funcs;
+
+#endif

drivers/gpu/drm/amd/amdgpu/df_v3_6.c

@@ -0,0 +1,116 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu.h"
+#include "df_v3_6.h"
+
+#include "df/df_3_6_default.h"
+#include "df/df_3_6_offset.h"
+#include "df/df_3_6_sh_mask.h"
+
+static u32 df_v3_6_channel_number[] = {1, 2, 0, 4, 0, 8, 0,
+				       16, 32, 0, 0, 0, 2, 4, 8};
+
+static void df_v3_6_init(struct amdgpu_device *adev)
+{
+}
+
+static void df_v3_6_enable_broadcast_mode(struct amdgpu_device *adev,
+					  bool enable)
+{
+	u32 tmp;
+
+	if (enable) {
+		tmp = RREG32_SOC15(DF, 0, mmFabricConfigAccessControl);
+		tmp &= ~FabricConfigAccessControl__CfgRegInstAccEn_MASK;
+		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl, tmp);
+	} else
+		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl,
+			     mmFabricConfigAccessControl_DEFAULT);
+}
+
+static u32 df_v3_6_get_fb_channel_number(struct amdgpu_device *adev)
+{
+	u32 tmp;
+
+	tmp = RREG32_SOC15(DF, 0, mmDF_CS_UMC_AON0_DramBaseAddress0);
+	tmp &= DF_CS_UMC_AON0_DramBaseAddress0__IntLvNumChan_MASK;
+	tmp >>= DF_CS_UMC_AON0_DramBaseAddress0__IntLvNumChan__SHIFT;
+
+	return tmp;
+}
+
+static u32 df_v3_6_get_hbm_channel_number(struct amdgpu_device *adev)
+{
+	int fb_channel_number;
+
+	fb_channel_number = adev->df_funcs->get_fb_channel_number(adev);
+	if (fb_channel_number > ARRAY_SIZE(df_v3_6_channel_number))
+		fb_channel_number = 0;
+
+	return df_v3_6_channel_number[fb_channel_number];
+}
+
+static void df_v3_6_update_medium_grain_clock_gating(struct amdgpu_device *adev,
+						     bool enable)
+{
+	u32 tmp;
+
+	/* Put DF on broadcast mode */
+	adev->df_funcs->enable_broadcast_mode(adev, true);
+
+	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DF_MGCG)) {
+		tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+		tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
+		tmp |= DF_V3_6_MGCG_ENABLE_15_CYCLE_DELAY;
+		WREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater, tmp);
+	} else {
+		tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+		tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
+		tmp |= DF_V3_6_MGCG_DISABLE;
+		WREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater, tmp);
+	}
+
+	/* Exit broadcast mode */
+	adev->df_funcs->enable_broadcast_mode(adev, false);
+}
+
+static void df_v3_6_get_clockgating_state(struct amdgpu_device *adev,
+					  u32 *flags)
+{
+	u32 tmp;
+
+	/* AMD_CG_SUPPORT_DF_MGCG */
+	tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
+	if (tmp & DF_V3_6_MGCG_ENABLE_15_CYCLE_DELAY)
+		*flags |= AMD_CG_SUPPORT_DF_MGCG;
+}
+
+const struct amdgpu_df_funcs df_v3_6_funcs = {
+	.init = df_v3_6_init,
+	.enable_broadcast_mode = df_v3_6_enable_broadcast_mode,
+	.get_fb_channel_number = df_v3_6_get_fb_channel_number,
+	.get_hbm_channel_number = df_v3_6_get_hbm_channel_number,
+	.update_medium_grain_clock_gating =
+			df_v3_6_update_medium_grain_clock_gating,
+	.get_clockgating_state = df_v3_6_get_clockgating_state,
+};

drivers/gpu/drm/nouveau/nvkm/engine/disp/basegk110.c → drivers/gpu/drm/amd/amdgpu/df_v3_6.h

@@ -1,5 +1,5 @@
 /*
- * Copyright 2015 Red Hat Inc.
+ * Copyright 2018 Advanced Micro Devices, Inc.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -19,20 +19,22 @@
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  *
- * Authors: Ben Skeggs <bskeggs@redhat.com>
  */
-#include "dmacnv50.h"
-#include "rootnv50.h"
 
-#include <nvif/class.h>
+#ifndef __DF_V3_6_H__
+#define __DF_V3_6_H__
 
-const struct nv50_disp_dmac_oclass
-gk110_disp_base_oclass = {
-	.base.oclass = GK110_DISP_BASE_CHANNEL_DMA,
-	.base.minver = 0,
-	.base.maxver = 0,
-	.ctor = nv50_disp_base_new,
-	.func = &gf119_disp_dmac_func,
-	.mthd = &gf119_disp_base_chan_mthd,
-	.chid = 1,
+#include "soc15_common.h"
+
+enum DF_V3_6_MGCG {
+	DF_V3_6_MGCG_DISABLE = 0,
+	DF_V3_6_MGCG_ENABLE_00_CYCLE_DELAY = 1,
+	DF_V3_6_MGCG_ENABLE_01_CYCLE_DELAY = 2,
+	DF_V3_6_MGCG_ENABLE_15_CYCLE_DELAY = 13,
+	DF_V3_6_MGCG_ENABLE_31_CYCLE_DELAY = 14,
+	DF_V3_6_MGCG_ENABLE_63_CYCLE_DELAY = 15
 };
+
+extern const struct amdgpu_df_funcs df_v3_6_funcs;
+
+#endif

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -125,18 +125,6 @@ MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
 MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
 MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
 
-MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
-MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
-
 MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
 MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
 MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
@@ -149,6 +137,18 @@ MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
 MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
 MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
 
+MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
+MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
+
 MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
@@ -161,6 +161,13 @@ MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
 
+MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
+MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
+MODULE_FIRMWARE("amdgpu/vegam_me.bin");
+MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
+MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
+MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
+
 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
 {
 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
@@ -292,6 +299,37 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 };
 
+static const u32 golden_settings_vegam_a11[] =
+{
+	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
+	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
+	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
+	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
+	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
+	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
+	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
+	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
+	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
+	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
+	mmSQ_CONFIG, 0x07f80000, 0x01180000,
+	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
+	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
+	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
+	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
+	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
+	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
+};
+
+static const u32 vegam_golden_common_all[] =
+{
+	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
+	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
+	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
+	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
+	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
+	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
+};
+
 static const u32 golden_settings_polaris11_a11[] =
 {
 	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
@@ -712,6 +750,14 @@ static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
 							tonga_golden_common_all,
 							ARRAY_SIZE(tonga_golden_common_all));
 		break;
+	case CHIP_VEGAM:
+		amdgpu_device_program_register_sequence(adev,
+							golden_settings_vegam_a11,
+							ARRAY_SIZE(golden_settings_vegam_a11));
+		amdgpu_device_program_register_sequence(adev,
+							vegam_golden_common_all,
+							ARRAY_SIZE(vegam_golden_common_all));
+		break;
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
 		amdgpu_device_program_register_sequence(adev,
@@ -918,17 +964,20 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_FIJI:
 		chip_name = "fiji";
 		break;
-	case CHIP_POLARIS11:
-		chip_name = "polaris11";
+	case CHIP_STONEY:
+		chip_name = "stoney";
 		break;
 	case CHIP_POLARIS10:
 		chip_name = "polaris10";
 		break;
+	case CHIP_POLARIS11:
+		chip_name = "polaris11";
+		break;
 	case CHIP_POLARIS12:
 		chip_name = "polaris12";
 		break;
-	case CHIP_STONEY:
-		chip_name = "stoney";
+	case CHIP_VEGAM:
+		chip_name = "vegam";
 		break;
 	default:
 		BUG();
@@ -1770,6 +1819,7 @@ static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
 		break;
 	case CHIP_POLARIS10:
+	case CHIP_VEGAM:
 		ret = amdgpu_atombios_get_gfx_info(adev);
 		if (ret)
 			return ret;
@@ -1957,12 +2007,13 @@ static int gfx_v8_0_sw_init(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	switch (adev->asic_type) {
-	case CHIP_FIJI:
 	case CHIP_TONGA:
+	case CHIP_CARRIZO:
+	case CHIP_FIJI:
+	case CHIP_POLARIS10:
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
-	case CHIP_POLARIS10:
-	case CHIP_CARRIZO:
+	case CHIP_VEGAM:
 		adev->gfx.mec.num_mec = 2;
 		break;
 	case CHIP_TOPAZ:
@@ -2323,6 +2374,7 @@ static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
 
 		break;
 	case CHIP_FIJI:
+	case CHIP_VEGAM:
 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
@@ -3504,6 +3556,7 @@ gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
 {
 	switch (adev->asic_type) {
 	case CHIP_FIJI:
+	case CHIP_VEGAM:
 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
 			  RB_XSEL2(1) | PKR_MAP(2) |
 			  PKR_XSEL(1) | PKR_YSEL(1) |
@@ -4071,7 +4124,8 @@ static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
 		gfx_v8_0_init_power_gating(adev);
 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
-		   (adev->asic_type == CHIP_POLARIS12)) {
+		   (adev->asic_type == CHIP_POLARIS12) ||
+		   (adev->asic_type == CHIP_VEGAM)) {
 		gfx_v8_0_init_csb(adev);
 		gfx_v8_0_init_save_restore_list(adev);
 		gfx_v8_0_enable_save_restore_machine(adev);
@@ -4146,7 +4200,8 @@ static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
 	if (adev->asic_type == CHIP_POLARIS11 ||
 	    adev->asic_type == CHIP_POLARIS10 ||
-	    adev->asic_type == CHIP_POLARIS12) {
+	    adev->asic_type == CHIP_POLARIS12 ||
+	    adev->asic_type == CHIP_VEGAM) {
 		tmp = RREG32(mmRLC_CGCG_CGLS_CTRL_3D);
 		tmp &= ~0x3;
 		WREG32(mmRLC_CGCG_CGLS_CTRL_3D, tmp);
@@ -5498,7 +5553,8 @@ static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *ade
 						       bool enable)
 {
 	if ((adev->asic_type == CHIP_POLARIS11) ||
-	    (adev->asic_type == CHIP_POLARIS12))
+	    (adev->asic_type == CHIP_POLARIS12) ||
+	    (adev->asic_type == CHIP_VEGAM))
 		/* Send msg to SMU via Powerplay */
 		amdgpu_device_ip_set_powergating_state(adev,
 						       AMD_IP_BLOCK_TYPE_SMC,
@@ -5588,6 +5644,7 @@ static int gfx_v8_0_set_powergating_state(void *handle,
 		break;
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
 		else
@@ -6154,6 +6211,7 @@ static int gfx_v8_0_set_clockgating_state(void *handle,
 	case CHIP_POLARIS10:
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
 		break;
 	default:

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

@@ -27,6 +27,7 @@
 #include "amdgpu_gfx.h"
 #include "soc15.h"
 #include "soc15d.h"
+#include "amdgpu_atomfirmware.h"
 
 #include "gc/gc_9_0_offset.h"
 #include "gc/gc_9_0_sh_mask.h"
@@ -41,7 +42,6 @@
 #define GFX9_MEC_HPD_SIZE 2048
 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
-#define GFX9_RLC_FORMAT_DIRECT_REG_LIST_LENGTH 34
 
 #define mmPWR_MISC_CNTL_STATUS					0x0183
 #define mmPWR_MISC_CNTL_STATUS_BASE_IDX				0
@@ -64,6 +64,13 @@ MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
 MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
 MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
 
+MODULE_FIRMWARE("amdgpu/vega20_ce.bin");
+MODULE_FIRMWARE("amdgpu/vega20_pfp.bin");
+MODULE_FIRMWARE("amdgpu/vega20_me.bin");
+MODULE_FIRMWARE("amdgpu/vega20_mec.bin");
+MODULE_FIRMWARE("amdgpu/vega20_mec2.bin");
+MODULE_FIRMWARE("amdgpu/vega20_rlc.bin");
+
 MODULE_FIRMWARE("amdgpu/raven_ce.bin");
 MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
 MODULE_FIRMWARE("amdgpu/raven_me.bin");
@@ -73,29 +80,22 @@ MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
 
 static const struct soc15_reg_golden golden_settings_gc_9_0[] =
 {
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
-	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
 };
 
 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
@@ -109,6 +109,20 @@ static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800)
 };
 
+static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
+{
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
+	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
+};
+
 static const struct soc15_reg_golden golden_settings_gc_9_1[] =
 {
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
@@ -185,6 +199,30 @@ static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000)
 };
 
+static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
+{
+	mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+	mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
+};
+
+static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
+{
+	mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+	mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
+};
+
 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
@@ -218,6 +256,14 @@ static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
 						golden_settings_gc_9_2_1_vg12,
 						ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
 		break;
+	case CHIP_VEGA20:
+		soc15_program_register_sequence(adev,
+						golden_settings_gc_9_0,
+						ARRAY_SIZE(golden_settings_gc_9_0));
+		soc15_program_register_sequence(adev,
+						golden_settings_gc_9_0_vg20,
+						ARRAY_SIZE(golden_settings_gc_9_0_vg20));
+		break;
 	case CHIP_RAVEN:
 		soc15_program_register_sequence(adev,
 						 golden_settings_gc_9_1,
@@ -401,6 +447,27 @@ static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
 	kfree(adev->gfx.rlc.register_list_format);
 }
 
+static void gfx_v9_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
+{
+	const struct rlc_firmware_header_v2_1 *rlc_hdr;
+
+	rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
+	adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
+	adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
+	adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
+	adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
+	adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
+	adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
+	adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
+	adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
+	adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
+	adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
+	adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
+	adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
+	adev->gfx.rlc.reg_list_format_direct_reg_list_length =
+			le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
+}
+
 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 {
 	const char *chip_name;
@@ -412,6 +479,8 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
 	unsigned int *tmp = NULL;
 	unsigned int i = 0;
+	uint16_t version_major;
+	uint16_t version_minor;
 
 	DRM_DEBUG("\n");
 
@@ -422,6 +491,9 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_VEGA12:
 		chip_name = "vega12";
 		break;
+	case CHIP_VEGA20:
+		chip_name = "vega20";
+		break;
 	case CHIP_RAVEN:
 		chip_name = "raven";
 		break;
@@ -468,6 +540,12 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 		goto out;
 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
+
+	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
+	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
+	if (version_major == 2 && version_minor == 1)
+		adev->gfx.rlc.is_rlc_v2_1 = true;
+
 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
 	adev->gfx.rlc.save_and_restore_offset =
@@ -508,6 +586,9 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 	for (i = 0 ; i < (rlc_hdr->reg_list_size_bytes >> 2); i++)
 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
 
+	if (adev->gfx.rlc.is_rlc_v2_1)
+		gfx_v9_0_init_rlc_ext_microcode(adev);
+
 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
 	if (err)
@@ -566,6 +647,26 @@ static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
 		adev->firmware.fw_size +=
 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
 
+		if (adev->gfx.rlc.is_rlc_v2_1) {
+			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL];
+			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL;
+			info->fw = adev->gfx.rlc_fw;
+			adev->firmware.fw_size +=
+				ALIGN(adev->gfx.rlc.save_restore_list_cntl_size_bytes, PAGE_SIZE);
+
+			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
+			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
+			info->fw = adev->gfx.rlc_fw;
+			adev->firmware.fw_size +=
+				ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
+
+			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
+			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
+			info->fw = adev->gfx.rlc_fw;
+			adev->firmware.fw_size +=
+				ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
+		}
+
 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
 		info->fw = adev->gfx.mec_fw;
@@ -1013,9 +1114,10 @@ static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
 	.select_me_pipe_q = &gfx_v9_0_select_me_pipe_q
 };
 
-static void gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
+static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
 {
 	u32 gb_addr_config;
+	int err;
 
 	adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
 
@@ -1037,6 +1139,20 @@ static void gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
 		gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
 		DRM_INFO("fix gfx.config for vega12\n");
 		break;
+	case CHIP_VEGA20:
+		adev->gfx.config.max_hw_contexts = 8;
+		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
+		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
+		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
+		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
+		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
+		gb_addr_config &= ~0xf3e777ff;
+		gb_addr_config |= 0x22014042;
+		/* check vbios table if gpu info is not available */
+		err = amdgpu_atomfirmware_get_gfx_info(adev);
+		if (err)
+			return err;
+		break;
 	case CHIP_RAVEN:
 		adev->gfx.config.max_hw_contexts = 8;
 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
@@ -1086,6 +1202,8 @@ static void gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
 					adev->gfx.config.gb_addr_config,
 					GB_ADDR_CONFIG,
 					PIPE_INTERLEAVE_SIZE));
+
+	return 0;
 }
 
 static int gfx_v9_0_ngg_create_buf(struct amdgpu_device *adev,
@@ -1319,6 +1437,7 @@ static int gfx_v9_0_sw_init(void *handle)
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		adev->gfx.mec.num_mec = 2;
 		break;
@@ -1446,7 +1565,9 @@ static int gfx_v9_0_sw_init(void *handle)
 
 	adev->gfx.ce_ram_size = 0x8000;
 
-	gfx_v9_0_gpu_early_init(adev);
+	r = gfx_v9_0_gpu_early_init(adev);
+	if (r)
+		return r;
 
 	r = gfx_v9_0_ngg_init(adev);
 	if (r)
@@ -1600,6 +1721,7 @@ static void gfx_v9_0_gpu_init(struct amdgpu_device *adev)
 
 	gfx_v9_0_setup_rb(adev);
 	gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
+	adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
 
 	/* XXX SH_MEM regs */
 	/* where to put LDS, scratch, GPUVM in FSA64 space */
@@ -1616,7 +1738,10 @@ static void gfx_v9_0_gpu_init(struct amdgpu_device *adev)
 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
 			WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, tmp);
-			tmp = adev->gmc.shared_aperture_start >> 48;
+			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
+				(adev->gmc.private_aperture_start >> 48));
+			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
+				(adev->gmc.shared_aperture_start >> 48));
 			WREG32_SOC15(GC, 0, mmSH_MEM_BASES, tmp);
 		}
 	}
@@ -1708,55 +1833,42 @@ static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
 			adev->gfx.rlc.clear_state_size);
 }
 
-static void gfx_v9_0_parse_ind_reg_list(int *register_list_format,
+static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
 				int indirect_offset,
 				int list_size,
 				int *unique_indirect_regs,
 				int *unique_indirect_reg_count,
-				int max_indirect_reg_count,
 				int *indirect_start_offsets,
-				int *indirect_start_offsets_count,
-				int max_indirect_start_offsets_count)
+				int *indirect_start_offsets_count)
 {
 	int idx;
-	bool new_entry = true;
 
 	for (; indirect_offset < list_size; indirect_offset++) {
+		indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
+		*indirect_start_offsets_count = *indirect_start_offsets_count + 1;
 
-		if (new_entry) {
-			new_entry = false;
-			indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
-			*indirect_start_offsets_count = *indirect_start_offsets_count + 1;
-			BUG_ON(*indirect_start_offsets_count >= max_indirect_start_offsets_count);
-		}
+		while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
+			indirect_offset += 2;
 
-		if (register_list_format[indirect_offset] == 0xFFFFFFFF) {
-			new_entry = true;
-			continue;
-		}
+			/* look for the matching indice */
+			for (idx = 0; idx < *unique_indirect_reg_count; idx++) {
+				if (unique_indirect_regs[idx] ==
+					register_list_format[indirect_offset] ||
+					!unique_indirect_regs[idx])
+					break;
+			}
 
-		indirect_offset += 2;
+			BUG_ON(idx >= *unique_indirect_reg_count);
 
-		/* look for the matching indice */
-		for (idx = 0; idx < *unique_indirect_reg_count; idx++) {
-			if (unique_indirect_regs[idx] ==
-				register_list_format[indirect_offset])
-				break;
-		}
+			if (!unique_indirect_regs[idx])
+				unique_indirect_regs[idx] = register_list_format[indirect_offset];
 
-		if (idx >= *unique_indirect_reg_count) {
-			unique_indirect_regs[*unique_indirect_reg_count] =
-				register_list_format[indirect_offset];
-			idx = *unique_indirect_reg_count;
-			*unique_indirect_reg_count = *unique_indirect_reg_count + 1;
-			BUG_ON(*unique_indirect_reg_count >= max_indirect_reg_count);
+			indirect_offset++;
 		}
-
-		register_list_format[indirect_offset] = idx;
 	}
 }
 
-static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
+static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
 {
 	int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
 	int unique_indirect_reg_count = 0;
@@ -1765,7 +1877,7 @@ static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
 	int indirect_start_offsets_count = 0;
 
 	int list_size = 0;
-	int i = 0;
+	int i = 0, j = 0;
 	u32 tmp = 0;
 
 	u32 *register_list_format =
@@ -1776,15 +1888,14 @@ static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
 		adev->gfx.rlc.reg_list_format_size_bytes);
 
 	/* setup unique_indirect_regs array and indirect_start_offsets array */
-	gfx_v9_0_parse_ind_reg_list(register_list_format,
-				GFX9_RLC_FORMAT_DIRECT_REG_LIST_LENGTH,
-				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
-				unique_indirect_regs,
-				&unique_indirect_reg_count,
-				ARRAY_SIZE(unique_indirect_regs),
-				indirect_start_offsets,
-				&indirect_start_offsets_count,
-				ARRAY_SIZE(indirect_start_offsets));
+	unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
+	gfx_v9_1_parse_ind_reg_list(register_list_format,
+				    adev->gfx.rlc.reg_list_format_direct_reg_list_length,
+				    adev->gfx.rlc.reg_list_format_size_bytes >> 2,
+				    unique_indirect_regs,
+				    &unique_indirect_reg_count,
+				    indirect_start_offsets,
+				    &indirect_start_offsets_count);
 
 	/* enable auto inc in case it is disabled */
 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
@@ -1798,19 +1909,37 @@ static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
 			adev->gfx.rlc.register_restore[i]);
 
-	/* load direct register */
-	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR), 0);
-	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
-		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
-			adev->gfx.rlc.register_restore[i]);
-
 	/* load indirect register */
 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
 		adev->gfx.rlc.reg_list_format_start);
-	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
+
+	/* direct register portion */
+	for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
 			register_list_format[i]);
 
+	/* indirect register portion */
+	while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
+		if (register_list_format[i] == 0xFFFFFFFF) {
+			WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
+			continue;
+		}
+
+		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
+		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
+
+		for (j = 0; j < unique_indirect_reg_count; j++) {
+			if (register_list_format[i] == unique_indirect_regs[j]) {
+				WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
+				break;
+			}
+		}
+
+		BUG_ON(j >= unique_indirect_reg_count);
+
+		i++;
+	}
+
 	/* set save/restore list size */
 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
 	list_size = list_size >> 1;
@@ -1823,14 +1952,19 @@ static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
 		adev->gfx.rlc.starting_offsets_start);
 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
-			indirect_start_offsets[i]);
+		       indirect_start_offsets[i]);
 
 	/* load unique indirect regs*/
 	for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
-		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0) + i,
-			unique_indirect_regs[i] & 0x3FFFF);
-		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0) + i,
-			unique_indirect_regs[i] >> 20);
+		if (unique_indirect_regs[i] != 0) {
+			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
+			       + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
+			       unique_indirect_regs[i] & 0x3FFFF);
+
+			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
+			       + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
+			       unique_indirect_regs[i] >> 20);
+		}
 	}
 
 	kfree(register_list_format);
@@ -2010,6 +2144,9 @@ static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *ad
 
 static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
 {
+	if (!adev->gfx.rlc.is_rlc_v2_1)
+		return;
+
 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
 			      AMD_PG_SUPPORT_GFX_SMG |
 			      AMD_PG_SUPPORT_GFX_DMG |
@@ -2017,27 +2154,12 @@ static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
 			      AMD_PG_SUPPORT_GDS |
 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
 		gfx_v9_0_init_csb(adev);
-		gfx_v9_0_init_rlc_save_restore_list(adev);
+		gfx_v9_1_init_rlc_save_restore_list(adev);
 		gfx_v9_0_enable_save_restore_machine(adev);
 
-		if (adev->asic_type == CHIP_RAVEN) {
-			WREG32(mmRLC_JUMP_TABLE_RESTORE,
-				adev->gfx.rlc.cp_table_gpu_addr >> 8);
-			gfx_v9_0_init_gfx_power_gating(adev);
-
-			if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
-				gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
-				gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
-			} else {
-				gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
-				gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
-			}
-
-			if (adev->pg_flags & AMD_PG_SUPPORT_CP)
-				gfx_v9_0_enable_cp_power_gating(adev, true);
-			else
-				gfx_v9_0_enable_cp_power_gating(adev, false);
-		}
+		WREG32(mmRLC_JUMP_TABLE_RESTORE,
+		       adev->gfx.rlc.cp_table_gpu_addr >> 8);
+		gfx_v9_0_init_gfx_power_gating(adev);
 	}
 }
 
@@ -3061,6 +3183,9 @@ static int gfx_v9_0_hw_fini(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 	int i;
 
+	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_GFX,
+					       AMD_PG_STATE_UNGATE);
+
 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
 
@@ -3279,6 +3404,11 @@ static int gfx_v9_0_late_init(void *handle)
 	if (r)
 		return r;
 
+	r = amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_GFX,
+						   AMD_PG_STATE_GATE);
+	if (r)
+		return r;
+
 	return 0;
 }
 
@@ -3339,8 +3469,7 @@ static void gfx_v9_0_exit_rlc_safe_mode(struct amdgpu_device *adev)
 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
 						bool enable)
 {
-	/* TODO: double check if we need to perform under safe mdoe */
-	/* gfx_v9_0_enter_rlc_safe_mode(adev); */
+	gfx_v9_0_enter_rlc_safe_mode(adev);
 
 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
 		gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
@@ -3351,7 +3480,7 @@ static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
 		gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
 	}
 
-	/* gfx_v9_0_exit_rlc_safe_mode(adev); */
+	gfx_v9_0_exit_rlc_safe_mode(adev);
 }
 
 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
@@ -3605,6 +3734,7 @@ static int gfx_v9_0_set_clockgating_state(void *handle,
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		gfx_v9_0_update_gfx_clock_gating(adev,
 						 state == AMD_CG_STATE_GATE ? true : false);
@@ -3742,7 +3872,7 @@ static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
 	}
 
 	amdgpu_ring_write(ring, header);
-BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
+	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
 	amdgpu_ring_write(ring,
 #ifdef __BIG_ENDIAN
 		(2 << 0) |
@@ -3774,13 +3904,16 @@ static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
 {
 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
+	bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
 
 	/* RELEASE_MEM - flush caches, send int */
 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
-	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
-				 EOP_TC_ACTION_EN |
-				 EOP_TC_WB_ACTION_EN |
-				 EOP_TC_MD_ACTION_EN |
+	amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN |
+					       EOP_TC_NC_ACTION_EN) :
+					      (EOP_TCL1_ACTION_EN |
+					       EOP_TC_ACTION_EN |
+					       EOP_TC_WB_ACTION_EN |
+					       EOP_TC_MD_ACTION_EN)) |
 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
 				 EVENT_INDEX(5)));
 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
@@ -4137,6 +4270,20 @@ static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
 	gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
 }
 
+static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
+						  uint32_t reg0, uint32_t reg1,
+						  uint32_t ref, uint32_t mask)
+{
+	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
+
+	if (amdgpu_sriov_vf(ring->adev))
+		gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
+				      ref, mask, 0x20);
+	else
+		amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
+							   ref, mask);
+}
+
 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
 						 enum amdgpu_interrupt_state state)
 {
@@ -4458,6 +4605,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
 	.emit_tmz = gfx_v9_0_ring_emit_tmz,
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
+	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
 };
 
 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
@@ -4492,6 +4640,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 	.set_priority = gfx_v9_0_ring_set_priority_compute,
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
+	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
 };
 
 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
@@ -4522,6 +4671,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
 	.emit_rreg = gfx_v9_0_ring_emit_rreg,
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
+	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
 };
 
 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
@@ -4577,6 +4727,7 @@ static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
 		break;
@@ -4686,6 +4837,7 @@ static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
 
 	cu_info->number = active_cu_number;
 	cu_info->ao_cu_mask = ao_cu_mask;
+	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
 
 	return 0;
 }

drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c

@@ -819,12 +819,33 @@ static int gmc_v6_0_late_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_bo_late_init(adev);
+
 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
 	else
 		return 0;
 }
 
+static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device *adev)
+{
+	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
+	unsigned size;
+
+	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
+		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
+	} else {
+		u32 viewport = RREG32(mmVIEWPORT_SIZE);
+		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
+			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
+			4);
+	}
+	/* return 0 if the pre-OS buffer uses up most of vram */
+	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
+		return 0;
+	return size;
+}
+
 static int gmc_v6_0_sw_init(void *handle)
 {
 	int r;
@@ -851,8 +872,6 @@ static int gmc_v6_0_sw_init(void *handle)
 
 	adev->gmc.mc_mask = 0xffffffffffULL;
 
-	adev->gmc.stolen_size = 256 * 1024;
-
 	adev->need_dma32 = false;
 	dma_bits = adev->need_dma32 ? 32 : 40;
 	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
@@ -878,6 +897,8 @@ static int gmc_v6_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	adev->gmc.stolen_size = gmc_v6_0_get_vbios_fb_size(adev);
+
 	r = amdgpu_bo_init(adev);
 	if (r)
 		return r;

drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c

@@ -958,12 +958,33 @@ static int gmc_v7_0_late_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_bo_late_init(adev);
+
 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
 	else
 		return 0;
 }
 
+static unsigned gmc_v7_0_get_vbios_fb_size(struct amdgpu_device *adev)
+{
+	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
+	unsigned size;
+
+	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
+		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
+	} else {
+		u32 viewport = RREG32(mmVIEWPORT_SIZE);
+		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
+			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
+			4);
+	}
+	/* return 0 if the pre-OS buffer uses up most of vram */
+	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
+		return 0;
+	return size;
+}
+
 static int gmc_v7_0_sw_init(void *handle)
 {
 	int r;
@@ -998,8 +1019,6 @@ static int gmc_v7_0_sw_init(void *handle)
 	 */
 	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
 
-	adev->gmc.stolen_size = 256 * 1024;
-
 	/* set DMA mask + need_dma32 flags.
 	 * PCIE - can handle 40-bits.
 	 * IGP - can handle 40-bits
@@ -1030,6 +1049,8 @@ static int gmc_v7_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	adev->gmc.stolen_size = gmc_v7_0_get_vbios_fb_size(adev);
+
 	/* Memory manager */
 	r = amdgpu_bo_init(adev);
 	if (r)

drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c

@@ -138,6 +138,7 @@ static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
 		break;
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 		amdgpu_device_program_register_sequence(adev,
 							golden_settings_polaris11_a11,
 							ARRAY_SIZE(golden_settings_polaris11_a11));
@@ -231,6 +232,7 @@ static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_FIJI:
 	case CHIP_CARRIZO:
 	case CHIP_STONEY:
+	case CHIP_VEGAM:
 		return 0;
 	default: BUG();
 	}
@@ -567,9 +569,10 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 	/* set the gart size */
 	if (amdgpu_gart_size == -1) {
 		switch (adev->asic_type) {
-		case CHIP_POLARIS11: /* all engines support GPUVM */
 		case CHIP_POLARIS10: /* all engines support GPUVM */
+		case CHIP_POLARIS11: /* all engines support GPUVM */
 		case CHIP_POLARIS12: /* all engines support GPUVM */
+		case CHIP_VEGAM:     /* all engines support GPUVM */
 		default:
 			adev->gmc.gart_size = 256ULL << 20;
 			break;
@@ -1049,12 +1052,33 @@ static int gmc_v8_0_late_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_bo_late_init(adev);
+
 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
 	else
 		return 0;
 }
 
+static unsigned gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
+{
+	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
+	unsigned size;
+
+	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
+		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
+	} else {
+		u32 viewport = RREG32(mmVIEWPORT_SIZE);
+		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
+			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
+			4);
+	}
+	/* return 0 if the pre-OS buffer uses up most of vram */
+	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
+		return 0;
+	return size;
+}
+
 #define mmMC_SEQ_MISC0_FIJI 0xA71
 
 static int gmc_v8_0_sw_init(void *handle)
@@ -1068,7 +1092,8 @@ static int gmc_v8_0_sw_init(void *handle)
 	} else {
 		u32 tmp;
 
-		if (adev->asic_type == CHIP_FIJI)
+		if ((adev->asic_type == CHIP_FIJI) ||
+		    (adev->asic_type == CHIP_VEGAM))
 			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
 		else
 			tmp = RREG32(mmMC_SEQ_MISC0);
@@ -1096,8 +1121,6 @@ static int gmc_v8_0_sw_init(void *handle)
 	 */
 	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
 
-	adev->gmc.stolen_size = 256 * 1024;
-
 	/* set DMA mask + need_dma32 flags.
 	 * PCIE - can handle 40-bits.
 	 * IGP - can handle 40-bits
@@ -1128,6 +1151,8 @@ static int gmc_v8_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	adev->gmc.stolen_size = gmc_v8_0_get_vbios_fb_size(adev);
+
 	/* Memory manager */
 	r = amdgpu_bo_init(adev);
 	if (r)

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

@@ -43,19 +43,13 @@
 #include "gfxhub_v1_0.h"
 #include "mmhub_v1_0.h"
 
-#define mmDF_CS_AON0_DramBaseAddress0                                                                  0x0044
-#define mmDF_CS_AON0_DramBaseAddress0_BASE_IDX                                                         0
-//DF_CS_AON0_DramBaseAddress0
-#define DF_CS_AON0_DramBaseAddress0__AddrRngVal__SHIFT                                                        0x0
-#define DF_CS_AON0_DramBaseAddress0__LgcyMmioHoleEn__SHIFT                                                    0x1
-#define DF_CS_AON0_DramBaseAddress0__IntLvNumChan__SHIFT                                                      0x4
-#define DF_CS_AON0_DramBaseAddress0__IntLvAddrSel__SHIFT                                                      0x8
-#define DF_CS_AON0_DramBaseAddress0__DramBaseAddr__SHIFT                                                      0xc
-#define DF_CS_AON0_DramBaseAddress0__AddrRngVal_MASK                                                          0x00000001L
-#define DF_CS_AON0_DramBaseAddress0__LgcyMmioHoleEn_MASK                                                      0x00000002L
-#define DF_CS_AON0_DramBaseAddress0__IntLvNumChan_MASK                                                        0x000000F0L
-#define DF_CS_AON0_DramBaseAddress0__IntLvAddrSel_MASK                                                        0x00000700L
-#define DF_CS_AON0_DramBaseAddress0__DramBaseAddr_MASK                                                        0xFFFFF000L
+/* add these here since we already include dce12 headers and these are for DCN */
+#define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION                                                          0x055d
+#define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION_BASE_IDX                                                 2
+#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_WIDTH__SHIFT                                        0x0
+#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_HEIGHT__SHIFT                                       0x10
+#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_WIDTH_MASK                                          0x00003FFFL
+#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_HEIGHT_MASK                                         0x3FFF0000L
 
 /* XXX Move this macro to VEGA10 header file, which is like vid.h for VI.*/
 #define AMDGPU_NUM_OF_VMIDS			8
@@ -385,11 +379,9 @@ static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_hi32 + (2 * vmid),
 			      upper_32_bits(pd_addr));
 
-	amdgpu_ring_emit_wreg(ring, hub->vm_inv_eng0_req + eng, req);
-
-	/* wait for the invalidate to complete */
-	amdgpu_ring_emit_reg_wait(ring, hub->vm_inv_eng0_ack + eng,
-				  1 << vmid, 1 << vmid);
+	amdgpu_ring_emit_reg_write_reg_wait(ring, hub->vm_inv_eng0_req + eng,
+					    hub->vm_inv_eng0_ack + eng,
+					    req, 1 << vmid);
 
 	return pd_addr;
 }
@@ -556,8 +548,7 @@ static int gmc_v9_0_early_init(void *handle)
 	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
 	adev->gmc.shared_aperture_end =
 		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
-	adev->gmc.private_aperture_start =
-		adev->gmc.shared_aperture_end + 1;
+	adev->gmc.private_aperture_start = 0x1000000000000000ULL;
 	adev->gmc.private_aperture_end =
 		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
 
@@ -659,6 +650,11 @@ static int gmc_v9_0_late_init(void *handle)
 	unsigned i;
 	int r;
 
+	/*
+	 * TODO - Uncomment once GART corruption issue is fixed.
+	 */
+	/* amdgpu_bo_late_init(adev); */
+
 	for(i = 0; i < adev->num_rings; ++i) {
 		struct amdgpu_ring *ring = adev->rings[i];
 		unsigned vmhub = ring->funcs->vmhub;
@@ -679,6 +675,7 @@ static int gmc_v9_0_late_init(void *handle)
 			DRM_INFO("ECC is active.\n");
 		} else if (r == 0) {
 			DRM_INFO("ECC is not present.\n");
+			adev->df_funcs->enable_ecc_force_par_wr_rmw(adev, false);
 		} else {
 			DRM_ERROR("gmc_v9_0_ecc_available() failed. r: %d\n", r);
 			return r;
@@ -697,10 +694,7 @@ static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev,
 	amdgpu_device_vram_location(adev, &adev->gmc, base);
 	amdgpu_device_gart_location(adev, mc);
 	/* base offset of vram pages */
-	if (adev->flags & AMD_IS_APU)
-		adev->vm_manager.vram_base_offset = gfxhub_v1_0_get_mc_fb_offset(adev);
-	else
-		adev->vm_manager.vram_base_offset = 0;
+	adev->vm_manager.vram_base_offset = gfxhub_v1_0_get_mc_fb_offset(adev);
 }
 
 /**
@@ -714,7 +708,6 @@ static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev,
  */
 static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 {
-	u32 tmp;
 	int chansize, numchan;
 	int r;
 
@@ -727,39 +720,7 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 		else
 			chansize = 128;
 
-		tmp = RREG32_SOC15(DF, 0, mmDF_CS_AON0_DramBaseAddress0);
-		tmp &= DF_CS_AON0_DramBaseAddress0__IntLvNumChan_MASK;
-		tmp >>= DF_CS_AON0_DramBaseAddress0__IntLvNumChan__SHIFT;
-		switch (tmp) {
-		case 0:
-		default:
-			numchan = 1;
-			break;
-		case 1:
-			numchan = 2;
-			break;
-		case 2:
-			numchan = 0;
-			break;
-		case 3:
-			numchan = 4;
-			break;
-		case 4:
-			numchan = 0;
-			break;
-		case 5:
-			numchan = 8;
-			break;
-		case 6:
-			numchan = 0;
-			break;
-		case 7:
-			numchan = 16;
-			break;
-		case 8:
-			numchan = 2;
-			break;
-		}
+		numchan = adev->df_funcs->get_hbm_channel_number(adev);
 		adev->gmc.vram_width = numchan * chansize;
 	}
 
@@ -792,6 +753,7 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 		switch (adev->asic_type) {
 		case CHIP_VEGA10:  /* all engines support GPUVM */
 		case CHIP_VEGA12:  /* all engines support GPUVM */
+		case CHIP_VEGA20:
 		default:
 			adev->gmc.gart_size = 512ULL << 20;
 			break;
@@ -826,6 +788,52 @@ static int gmc_v9_0_gart_init(struct amdgpu_device *adev)
 	return amdgpu_gart_table_vram_alloc(adev);
 }
 
+static unsigned gmc_v9_0_get_vbios_fb_size(struct amdgpu_device *adev)
+{
+#if 0
+	u32 d1vga_control = RREG32_SOC15(DCE, 0, mmD1VGA_CONTROL);
+#endif
+	unsigned size;
+
+	/*
+	 * TODO Remove once GART corruption is resolved
+	 * Check related code in gmc_v9_0_sw_fini
+	 * */
+	size = 9 * 1024 * 1024;
+
+#if 0
+	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
+		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
+	} else {
+		u32 viewport;
+
+		switch (adev->asic_type) {
+		case CHIP_RAVEN:
+			viewport = RREG32_SOC15(DCE, 0, mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION);
+			size = (REG_GET_FIELD(viewport,
+					      HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION, PRI_VIEWPORT_HEIGHT) *
+				REG_GET_FIELD(viewport,
+					      HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION, PRI_VIEWPORT_WIDTH) *
+				4);
+			break;
+		case CHIP_VEGA10:
+		case CHIP_VEGA12:
+		default:
+			viewport = RREG32_SOC15(DCE, 0, mmSCL0_VIEWPORT_SIZE);
+			size = (REG_GET_FIELD(viewport, SCL0_VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
+				REG_GET_FIELD(viewport, SCL0_VIEWPORT_SIZE, VIEWPORT_WIDTH) *
+				4);
+			break;
+		}
+	}
+	/* return 0 if the pre-OS buffer uses up most of vram */
+	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
+		return 0;
+
+#endif
+	return size;
+}
+
 static int gmc_v9_0_sw_init(void *handle)
 {
 	int r;
@@ -851,6 +859,7 @@ static int gmc_v9_0_sw_init(void *handle)
 		break;
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 		/*
 		 * To fulfill 4-level page support,
 		 * vm size is 256TB (48bit), maximum size of Vega10,
@@ -877,12 +886,6 @@ static int gmc_v9_0_sw_init(void *handle)
 	 */
 	adev->gmc.mc_mask = 0xffffffffffffULL; /* 48 bit MC */
 
-	/*
-	 * It needs to reserve 8M stolen memory for vega10
-	 * TODO: Figure out how to avoid that...
-	 */
-	adev->gmc.stolen_size = 8 * 1024 * 1024;
-
 	/* set DMA mask + need_dma32 flags.
 	 * PCIE - can handle 44-bits.
 	 * IGP - can handle 44-bits
@@ -907,6 +910,8 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	adev->gmc.stolen_size = gmc_v9_0_get_vbios_fb_size(adev);
+
 	/* Memory manager */
 	r = amdgpu_bo_init(adev);
 	if (r)
@@ -950,6 +955,18 @@ static int gmc_v9_0_sw_fini(void *handle)
 	amdgpu_gem_force_release(adev);
 	amdgpu_vm_manager_fini(adev);
 	gmc_v9_0_gart_fini(adev);
+
+	/*
+	* TODO:
+	* Currently there is a bug where some memory client outside
+	* of the driver writes to first 8M of VRAM on S3 resume,
+	* this overrides GART which by default gets placed in first 8M and
+	* causes VM_FAULTS once GTT is accessed.
+	* Keep the stolen memory reservation until the while this is not solved.
+	* Also check code in gmc_v9_0_get_vbios_fb_size and gmc_v9_0_late_init
+	*/
+	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
+
 	amdgpu_bo_fini(adev);
 
 	return 0;
@@ -960,6 +977,7 @@ static void gmc_v9_0_init_golden_registers(struct amdgpu_device *adev)
 
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
+	case CHIP_VEGA20:
 		soc15_program_register_sequence(adev,
 						golden_settings_mmhub_1_0_0,
 						ARRAY_SIZE(golden_settings_mmhub_1_0_0));

drivers/gpu/drm/amd/amdgpu/kv_dpm.c

@@ -2817,7 +2817,7 @@ static int kv_dpm_init(struct amdgpu_device *adev)
 		pi->caps_tcp_ramping = true;
 	}
 
-	if (amdgpu_pp_feature_mask & SCLK_DEEP_SLEEP_MASK)
+	if (adev->powerplay.pp_feature & PP_SCLK_DEEP_SLEEP_MASK)
 		pi->caps_sclk_ds = true;
 	else
 		pi->caps_sclk_ds = false;
@@ -2974,7 +2974,7 @@ static int kv_dpm_late_init(void *handle)
 	/* powerdown unused blocks for now */
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	if (!amdgpu_dpm)
+	if (!adev->pm.dpm_enabled)
 		return 0;
 
 	kv_dpm_powergate_acp(adev, true);

drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c

@@ -734,6 +734,7 @@ int mmhub_v1_0_set_clockgating(struct amdgpu_device *adev,
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		mmhub_v1_0_update_medium_grain_clock_gating(adev,
 				state == AMD_CG_STATE_GATE ? true : false);

drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c

@@ -260,8 +260,10 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 	} while (timeout > 1);
 
 flr_done:
-	if (locked)
+	if (locked) {
+		adev->in_gpu_reset = 0;
 		mutex_unlock(&adev->lock_reset);
+	}
 
 	/* Trigger recovery for world switch failure if no TDR */
 	if (amdgpu_lockup_timeout == 0)

drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c

@@ -34,10 +34,19 @@
 #define smnCPM_CONTROL                                                                                  0x11180460
 #define smnPCIE_CNTL2                                                                                   0x11180070
 
+/* vega20 */
+#define mmRCC_DEV0_EPF0_STRAP0_VG20                                                                         0x0011
+#define mmRCC_DEV0_EPF0_STRAP0_VG20_BASE_IDX                                                                2
+
 static u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev)
 {
         u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
 
+	if (adev->asic_type == CHIP_VEGA20)
+		tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0_VG20);
+	else
+		tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
+
 	tmp &= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0_MASK;
 	tmp >>= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0__SHIFT;
 
@@ -75,10 +84,14 @@ static void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instan
 			SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA1_DOORBELL_RANGE);
 
 	u32 doorbell_range = RREG32(reg);
+	u32 range = 2;
+
+	if (adev->asic_type == CHIP_VEGA20)
+		range = 8;
 
 	if (use_doorbell) {
 		doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, OFFSET, doorbell_index);
-		doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 2);
+		doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, range);
 	} else
 		doorbell_range = REG_SET_FIELD(doorbell_range, BIF_SDMA0_DOORBELL_RANGE, SIZE, 0);
 
@@ -133,6 +146,9 @@ static void nbio_v7_0_update_medium_grain_clock_gating(struct amdgpu_device *ade
 {
 	uint32_t def, data;
 
+	if (adev->asic_type == CHIP_VEGA20)
+		return;
+
 	/* NBIF_MGCG_CTRL_LCLK */
 	def = data = RREG32_PCIE(smnNBIF_MGCG_CTRL_LCLK);
 

drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h

@@ -40,11 +40,20 @@ enum psp_gfx_crtl_cmd_id
     GFX_CTRL_CMD_ID_INIT_GPCOM_RING = 0x00020000,   /* initialize GPCOM ring */
     GFX_CTRL_CMD_ID_DESTROY_RINGS   = 0x00030000,   /* destroy rings */
     GFX_CTRL_CMD_ID_CAN_INIT_RINGS  = 0x00040000,   /* is it allowed to initialized the rings */
+    GFX_CTRL_CMD_ID_ENABLE_INT      = 0x00050000,   /* enable PSP-to-Gfx interrupt */
+    GFX_CTRL_CMD_ID_DISABLE_INT     = 0x00060000,   /* disable PSP-to-Gfx interrupt */
+    GFX_CTRL_CMD_ID_MODE1_RST       = 0x00070000,   /* trigger the Mode 1 reset */
 
     GFX_CTRL_CMD_ID_MAX             = 0x000F0000,   /* max command ID */
 };
 
 
+/*-----------------------------------------------------------------------------
+    NOTE:   All physical addresses used in this interface are actually
+            GPU Virtual Addresses.
+*/
+
+
 /* Control registers of the TEE Gfx interface. These are located in
 *  SRBM-to-PSP mailbox registers (total 8 registers).
 */
@@ -55,8 +64,8 @@ struct psp_gfx_ctrl
     volatile uint32_t   rbi_rptr;         /* +8   Read pointer (index) of RBI ring */
     volatile uint32_t   gpcom_wptr;       /* +12  Write pointer (index) of GPCOM ring */
     volatile uint32_t   gpcom_rptr;       /* +16  Read pointer (index) of GPCOM ring */
-    volatile uint32_t   ring_addr_lo;     /* +20  bits [31:0] of physical address of ring buffer */
-    volatile uint32_t   ring_addr_hi;     /* +24  bits [63:32] of physical address of ring buffer */
+    volatile uint32_t   ring_addr_lo;     /* +20  bits [31:0] of GPU Virtual of ring buffer (VMID=0)*/
+    volatile uint32_t   ring_addr_hi;     /* +24  bits [63:32] of GPU Virtual of ring buffer (VMID=0) */
     volatile uint32_t   ring_buf_size;    /* +28  Ring buffer size (in bytes) */
 
 };
@@ -78,6 +87,8 @@ enum psp_gfx_cmd_id
     GFX_CMD_ID_LOAD_ASD     = 0x00000004,   /* load ASD Driver */
     GFX_CMD_ID_SETUP_TMR    = 0x00000005,   /* setup TMR region */
     GFX_CMD_ID_LOAD_IP_FW   = 0x00000006,   /* load HW IP FW */
+    GFX_CMD_ID_DESTROY_TMR  = 0x00000007,   /* destroy TMR region */
+    GFX_CMD_ID_SAVE_RESTORE = 0x00000008,   /* save/restore HW IP FW */
 
 };
 
@@ -85,11 +96,11 @@ enum psp_gfx_cmd_id
 /* Command to load Trusted Application binary into PSP OS. */
 struct psp_gfx_cmd_load_ta
 {
-    uint32_t        app_phy_addr_lo;        /* bits [31:0] of the physical address of the TA binary (must be 4 KB aligned) */
-    uint32_t        app_phy_addr_hi;        /* bits [63:32] of the physical address of the TA binary */
+    uint32_t        app_phy_addr_lo;        /* bits [31:0] of the GPU Virtual address of the TA binary (must be 4 KB aligned) */
+    uint32_t        app_phy_addr_hi;        /* bits [63:32] of the GPU Virtual address of the TA binary */
     uint32_t        app_len;                /* length of the TA binary in bytes */
-    uint32_t        cmd_buf_phy_addr_lo;    /* bits [31:0] of the physical address of CMD buffer (must be 4 KB aligned) */
-    uint32_t        cmd_buf_phy_addr_hi;    /* bits [63:32] of the physical address of CMD buffer */
+    uint32_t        cmd_buf_phy_addr_lo;    /* bits [31:0] of the GPU Virtual address of CMD buffer (must be 4 KB aligned) */
+    uint32_t        cmd_buf_phy_addr_hi;    /* bits [63:32] of the GPU Virtual address of CMD buffer */
     uint32_t        cmd_buf_len;            /* length of the CMD buffer in bytes; must be multiple of 4 KB */
 
     /* Note: CmdBufLen can be set to 0. In this case no persistent CMD buffer is provided
@@ -111,8 +122,8 @@ struct psp_gfx_cmd_unload_ta
 */
 struct psp_gfx_buf_desc
 {
-    uint32_t        buf_phy_addr_lo;       /* bits [31:0] of physical address of the buffer (must be 4 KB aligned) */
-    uint32_t        buf_phy_addr_hi;       /* bits [63:32] of physical address of the buffer */
+    uint32_t        buf_phy_addr_lo;       /* bits [31:0] of GPU Virtual address of the buffer (must be 4 KB aligned) */
+    uint32_t        buf_phy_addr_hi;       /* bits [63:32] of GPU Virtual address of the buffer */
     uint32_t        buf_size;              /* buffer size in bytes (must be multiple of 4 KB and no bigger than 64 MB) */
 
 };
@@ -145,8 +156,8 @@ struct psp_gfx_cmd_invoke_cmd
 /* Command to setup TMR region. */
 struct psp_gfx_cmd_setup_tmr
 {
-    uint32_t        buf_phy_addr_lo;       /* bits [31:0] of physical address of TMR buffer (must be 4 KB aligned) */
-    uint32_t        buf_phy_addr_hi;       /* bits [63:32] of physical address of TMR buffer */
+    uint32_t        buf_phy_addr_lo;       /* bits [31:0] of GPU Virtual address of TMR buffer (must be 4 KB aligned) */
+    uint32_t        buf_phy_addr_hi;       /* bits [63:32] of GPU Virtual address of TMR buffer */
     uint32_t        buf_size;              /* buffer size in bytes (must be multiple of 4 KB) */
 
 };
@@ -174,18 +185,32 @@ enum psp_gfx_fw_type
     GFX_FW_TYPE_ISP         = 16,
     GFX_FW_TYPE_ACP         = 17,
     GFX_FW_TYPE_SMU         = 18,
+    GFX_FW_TYPE_MMSCH       = 19,
+    GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM        = 20,
+    GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM        = 21,
+    GFX_FW_TYPE_RLC_RESTORE_LIST_CNTL           = 22,
+    GFX_FW_TYPE_MAX         = 23
 };
 
 /* Command to load HW IP FW. */
 struct psp_gfx_cmd_load_ip_fw
 {
-    uint32_t                fw_phy_addr_lo;    /* bits [31:0] of physical address of FW location (must be 4 KB aligned) */
-    uint32_t                fw_phy_addr_hi;    /* bits [63:32] of physical address of FW location */
+    uint32_t                fw_phy_addr_lo;    /* bits [31:0] of GPU Virtual address of FW location (must be 4 KB aligned) */
+    uint32_t                fw_phy_addr_hi;    /* bits [63:32] of GPU Virtual address of FW location */
     uint32_t                fw_size;           /* FW buffer size in bytes */
     enum psp_gfx_fw_type    fw_type;           /* FW type */
 
 };
 
+/* Command to save/restore HW IP FW. */
+struct psp_gfx_cmd_save_restore_ip_fw
+{
+    uint32_t                save_fw;              /* if set, command is used for saving fw otherwise for resetoring*/
+    uint32_t                save_restore_addr_lo; /* bits [31:0] of FB address of GART memory used as save/restore buffer (must be 4 KB aligned) */
+    uint32_t                save_restore_addr_hi; /* bits [63:32] of FB address of GART memory used as save/restore buffer */
+    uint32_t                buf_size;             /* Size of the save/restore buffer in bytes */
+    enum psp_gfx_fw_type    fw_type;              /* FW type */
+};
 
 /* All GFX ring buffer commands. */
 union psp_gfx_commands
@@ -195,7 +220,7 @@ union psp_gfx_commands
     struct psp_gfx_cmd_invoke_cmd       cmd_invoke_cmd;
     struct psp_gfx_cmd_setup_tmr        cmd_setup_tmr;
     struct psp_gfx_cmd_load_ip_fw       cmd_load_ip_fw;
-
+    struct psp_gfx_cmd_save_restore_ip_fw cmd_save_restore_ip_fw;
 };
 
 
@@ -226,8 +251,8 @@ struct psp_gfx_cmd_resp
 
     /* These fields are used for RBI only. They are all 0 in GPCOM commands
     */
-    uint32_t        resp_buf_addr_lo;   /* +12 bits [31:0] of physical address of response buffer (must be 4 KB aligned) */
-    uint32_t        resp_buf_addr_hi;   /* +16 bits [63:32] of physical address of response buffer */
+    uint32_t        resp_buf_addr_lo;   /* +12 bits [31:0] of GPU Virtual address of response buffer (must be 4 KB aligned) */
+    uint32_t        resp_buf_addr_hi;   /* +16 bits [63:32] of GPU Virtual address of response buffer */
     uint32_t        resp_offset;        /* +20 offset within response buffer */
     uint32_t        resp_buf_size;      /* +24 total size of the response buffer in bytes */
 
@@ -251,19 +276,19 @@ struct psp_gfx_cmd_resp
 /* Structure of the Ring Buffer Frame */
 struct psp_gfx_rb_frame
 {
-    uint32_t    cmd_buf_addr_lo;    /* +0  bits [31:0] of physical address of command buffer (must be 4 KB aligned) */
-    uint32_t    cmd_buf_addr_hi;    /* +4  bits [63:32] of physical address of command buffer */
+    uint32_t    cmd_buf_addr_lo;    /* +0  bits [31:0] of GPU Virtual address of command buffer (must be 4 KB aligned) */
+    uint32_t    cmd_buf_addr_hi;    /* +4  bits [63:32] of GPU Virtual address of command buffer */
     uint32_t    cmd_buf_size;       /* +8  command buffer size in bytes */
-    uint32_t    fence_addr_lo;      /* +12 bits [31:0] of physical address of Fence for this frame */
-    uint32_t    fence_addr_hi;      /* +16 bits [63:32] of physical address of Fence for this frame */
+    uint32_t    fence_addr_lo;      /* +12 bits [31:0] of GPU Virtual address of Fence for this frame */
+    uint32_t    fence_addr_hi;      /* +16 bits [63:32] of GPU Virtual address of Fence for this frame */
     uint32_t    fence_value;        /* +20 Fence value */
     uint32_t    sid_lo;             /* +24 bits [31:0] of SID value (used only for RBI frames) */
     uint32_t    sid_hi;             /* +28 bits [63:32] of SID value (used only for RBI frames) */
     uint8_t     vmid;               /* +32 VMID value used for mapping of all addresses for this frame */
     uint8_t     frame_type;         /* +33 1: destory context frame, 0: all other frames; used only for RBI frames */
     uint8_t     reserved1[2];       /* +34 reserved, must be 0 */
-    uint32_t    reserved2[7];       /* +40 reserved, must be 0 */
-    /* total 64 bytes */
+    uint32_t    reserved2[7];       /* +36 reserved, must be 0 */
+                /* total 64 bytes */
 };
 
 #endif /* _PSP_TEE_GFX_IF_H_ */

drivers/gpu/drm/amd/amdgpu/psp_v10_0.c

@@ -70,6 +70,15 @@ psp_v10_0_get_fw_type(struct amdgpu_firmware_info *ucode, enum psp_gfx_fw_type *
 	case AMDGPU_UCODE_ID_RLC_G:
 		*type = GFX_FW_TYPE_RLC_G;
 		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_CNTL;
+		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM;
+		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM;
+		break;
 	case AMDGPU_UCODE_ID_SMC:
 		*type = GFX_FW_TYPE_SMU;
 		break;

drivers/gpu/drm/amd/amdgpu/psp_v3_1.c

@@ -41,6 +41,9 @@ MODULE_FIRMWARE("amdgpu/vega10_sos.bin");
 MODULE_FIRMWARE("amdgpu/vega10_asd.bin");
 MODULE_FIRMWARE("amdgpu/vega12_sos.bin");
 MODULE_FIRMWARE("amdgpu/vega12_asd.bin");
+MODULE_FIRMWARE("amdgpu/vega20_sos.bin");
+MODULE_FIRMWARE("amdgpu/vega20_asd.bin");
+
 
 #define smnMP1_FIRMWARE_FLAGS 0x3010028
 

drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

@@ -62,6 +62,8 @@ MODULE_FIRMWARE("amdgpu/polaris11_sdma.bin");
 MODULE_FIRMWARE("amdgpu/polaris11_sdma1.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_sdma.bin");
 MODULE_FIRMWARE("amdgpu/polaris12_sdma1.bin");
+MODULE_FIRMWARE("amdgpu/vegam_sdma.bin");
+MODULE_FIRMWARE("amdgpu/vegam_sdma1.bin");
 
 
 static const u32 sdma_offsets[SDMA_MAX_INSTANCE] =
@@ -209,6 +211,7 @@ static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev)
 		break;
 	case CHIP_POLARIS11:
 	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
 		amdgpu_device_program_register_sequence(adev,
 							golden_settings_polaris11_a11,
 							ARRAY_SIZE(golden_settings_polaris11_a11));
@@ -275,15 +278,18 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_FIJI:
 		chip_name = "fiji";
 		break;
-	case CHIP_POLARIS11:
-		chip_name = "polaris11";
-		break;
 	case CHIP_POLARIS10:
 		chip_name = "polaris10";
 		break;
+	case CHIP_POLARIS11:
+		chip_name = "polaris11";
+		break;
 	case CHIP_POLARIS12:
 		chip_name = "polaris12";
 		break;
+	case CHIP_VEGAM:
+		chip_name = "vegam";
+		break;
 	case CHIP_CARRIZO:
 		chip_name = "carrizo";
 		break;

drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c

@@ -42,6 +42,8 @@ MODULE_FIRMWARE("amdgpu/vega10_sdma.bin");
 MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin");
 MODULE_FIRMWARE("amdgpu/vega12_sdma.bin");
 MODULE_FIRMWARE("amdgpu/vega12_sdma1.bin");
+MODULE_FIRMWARE("amdgpu/vega20_sdma.bin");
+MODULE_FIRMWARE("amdgpu/vega20_sdma1.bin");
 MODULE_FIRMWARE("amdgpu/raven_sdma.bin");
 
 #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK  0x000000F8L
@@ -107,6 +109,28 @@ static const struct soc15_reg_golden golden_settings_sdma_4_1[] =
 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0)
 };
 
+static const struct soc15_reg_golden golden_settings_sdma_4_2[] =
+{
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
+	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0)
+};
+
 static const struct soc15_reg_golden golden_settings_sdma_rv1[] =
 {
 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
@@ -139,6 +163,11 @@ static void sdma_v4_0_init_golden_registers(struct amdgpu_device *adev)
 						golden_settings_sdma_vg12,
 						ARRAY_SIZE(golden_settings_sdma_vg12));
 		break;
+	case CHIP_VEGA20:
+		soc15_program_register_sequence(adev,
+						golden_settings_sdma_4_2,
+						ARRAY_SIZE(golden_settings_sdma_4_2));
+		break;
 	case CHIP_RAVEN:
 		soc15_program_register_sequence(adev,
 						 golden_settings_sdma_4_1,
@@ -182,6 +211,9 @@ static int sdma_v4_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_VEGA12:
 		chip_name = "vega12";
 		break;
+	case CHIP_VEGA20:
+		chip_name = "vega20";
+		break;
 	case CHIP_RAVEN:
 		chip_name = "raven";
 		break;
@@ -360,6 +392,31 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
 
 }
 
+static void sdma_v4_0_wait_reg_mem(struct amdgpu_ring *ring,
+				   int mem_space, int hdp,
+				   uint32_t addr0, uint32_t addr1,
+				   uint32_t ref, uint32_t mask,
+				   uint32_t inv)
+{
+	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
+			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(hdp) |
+			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(mem_space) |
+			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
+	if (mem_space) {
+		/* memory */
+		amdgpu_ring_write(ring, addr0);
+		amdgpu_ring_write(ring, addr1);
+	} else {
+		/* registers */
+		amdgpu_ring_write(ring, addr0 << 2);
+		amdgpu_ring_write(ring, addr1 << 2);
+	}
+	amdgpu_ring_write(ring, ref); /* reference */
+	amdgpu_ring_write(ring, mask); /* mask */
+	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
+			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(inv)); /* retry count, poll interval */
+}
+
 /**
  * sdma_v4_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
  *
@@ -378,15 +435,10 @@ static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 	else
 		ref_and_mask = nbio_hf_reg->ref_and_mask_sdma1;
 
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(1) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
-	amdgpu_ring_write(ring, (adev->nbio_funcs->get_hdp_flush_done_offset(adev)) << 2);
-	amdgpu_ring_write(ring, (adev->nbio_funcs->get_hdp_flush_req_offset(adev)) << 2);
-	amdgpu_ring_write(ring, ref_and_mask); /* reference */
-	amdgpu_ring_write(ring, ref_and_mask); /* mask */
-	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
-			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */
+	sdma_v4_0_wait_reg_mem(ring, 0, 1,
+			       adev->nbio_funcs->get_hdp_flush_done_offset(adev),
+			       adev->nbio_funcs->get_hdp_flush_req_offset(adev),
+			       ref_and_mask, ref_and_mask, 10);
 }
 
 /**
@@ -1114,16 +1166,10 @@ static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
 	uint64_t addr = ring->fence_drv.gpu_addr;
 
 	/* wait for idle */
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3) | /* equal */
-			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(1));
-	amdgpu_ring_write(ring, addr & 0xfffffffc);
-	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
-	amdgpu_ring_write(ring, seq); /* reference */
-	amdgpu_ring_write(ring, 0xffffffff); /* mask */
-	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
-			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(4)); /* retry count, poll interval */
+	sdma_v4_0_wait_reg_mem(ring, 1, 0,
+			       addr & 0xfffffffc,
+			       upper_32_bits(addr) & 0xffffffff,
+			       seq, 0xffffffff, 4);
 }
 
 
@@ -1154,15 +1200,7 @@ static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
 					 uint32_t val, uint32_t mask)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) |
-			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* equal */
-	amdgpu_ring_write(ring, reg << 2);
-	amdgpu_ring_write(ring, 0);
-	amdgpu_ring_write(ring, val); /* reference */
-	amdgpu_ring_write(ring, mask); /* mask */
-	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
-			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10));
+	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
 }
 
 static int sdma_v4_0_early_init(void *handle)
@@ -1510,6 +1548,7 @@ static int sdma_v4_0_set_clockgating_state(void *handle,
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		sdma_v4_0_update_medium_grain_clock_gating(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
@@ -1605,6 +1644,7 @@ static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
 	.pad_ib = sdma_v4_0_ring_pad_ib,
 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
+	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
 };
 
 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/si.c

@@ -1252,6 +1252,12 @@ static void si_invalidate_hdp(struct amdgpu_device *adev,
 	}
 }
 
+static bool si_need_full_reset(struct amdgpu_device *adev)
+{
+	/* change this when we support soft reset */
+	return true;
+}
+
 static int si_get_pcie_lanes(struct amdgpu_device *adev)
 {
 	u32 link_width_cntl;
@@ -1332,6 +1338,7 @@ static const struct amdgpu_asic_funcs si_asic_funcs =
 	.get_config_memsize = &si_get_config_memsize,
 	.flush_hdp = &si_flush_hdp,
 	.invalidate_hdp = &si_invalidate_hdp,
+	.need_full_reset = &si_need_full_reset,
 };
 
 static uint32_t si_get_rev_id(struct amdgpu_device *adev)