Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "This is the main pull request for the drm for 4.3.  Nouveau is
  probably the biggest amount of changes in here, since it missed 4.2.
  Highlights below, along with the usual bunch of fixes.

  All stuff outside drm should have applicable acks.

  Highlights:

   - new drivers:
        freescale dcu kms driver

   - core:
        more atomic fixes
        disable some dri1 interfaces on kms drivers
        drop fb panic handling, this was just getting more broken, as more locking was required.
        new core fbdev Kconfig support - instead of each driver enable/disabling it
        struct_mutex cleanups

   - panel:
        more new panels
        cleanup Kconfig

   - i915:
        Skylake support enabled by default
        legacy modesetting using atomic infrastructure
        Skylake fixes
        GEN9 workarounds

   - amdgpu:
        Fiji support
        CGS support for amdgpu
        Initial GPU scheduler - off by default
        Lots of bug fixes and optimisations.

   - radeon:
        DP fixes
        misc fixes

   - amdkfd:
        Add Carrizo support for amdkfd using amdgpu.

   - nouveau:
        long pending cleanup to complete driver,
        fully bisectable which makes it larger,
        perfmon work
        more reclocking improvements
        maxwell displayport fixes

   - vmwgfx:
        new DX device support, supports OpenGL 3.3
        screen targets support

   - mgag200:
        G200eW support
        G200e new revision support

   - msm:
        dragonboard 410c support, msm8x94 support, msm8x74v1 support
        yuv format support
        dma plane support
        mdp5 rotation
        initial hdcp

   - sti:
        atomic support

   - exynos:
        lots of cleanups
        atomic modesetting/pageflipping support
        render node support

   - tegra:
        tegra210 support (dc, dsi, dp/hdmi)
        dpms with atomic modesetting support

   - atmel:
        support for 3 more atmel SoCs
        new input formats, PRIME support.

   - dwhdmi:
        preparing to add audio support

   - rockchip:
        yuv plane support"

* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (1369 commits)
  drm/amdgpu: rename gmc_v8_0_init_compute_vmid
  drm/amdgpu: fix vce3 instance handling
  drm/amdgpu: remove ib test for the second VCE Ring
  drm/amdgpu: properly enable VM fault interrupts
  drm/amdgpu: fix warning in scheduler
  drm/amdgpu: fix buffer placement under memory pressure
  drm/amdgpu/cz: fix cz_dpm_update_low_memory_pstate logic
  drm/amdgpu: fix typo in dce11 watermark setup
  drm/amdgpu: fix typo in dce10 watermark setup
  drm/amdgpu: use top down allocation for non-CPU accessible vram
  drm/amdgpu: be explicit about cpu vram access for driver BOs (v2)
  drm/amdgpu: set MEC doorbell range for Fiji
  drm/amdgpu: implement burst NOP for SDMA
  drm/amdgpu: add insert_nop ring func and default implementation
  drm/amdgpu: add amdgpu_get_sdma_instance helper function
  drm/amdgpu: add AMDGPU_MAX_SDMA_INSTANCES
  drm/amdgpu: add burst_nop flag for sdma
  drm/amdgpu: add count field for the SDMA NOP packet v2
  drm/amdgpu: use PT for VM sync on unmap
  drm/amdgpu: make wait_event uninterruptible in push_job
  ...

Documentation/DocBook/drm.tmpl

@@ -3982,7 +3982,6 @@ int num_ioctls;</synopsis>
         <title>Interrupt Handling</title>
 !Pdrivers/gpu/drm/i915/i915_irq.c interrupt handling
 !Fdrivers/gpu/drm/i915/i915_irq.c intel_irq_init intel_irq_init_hw intel_hpd_init
-!Fdrivers/gpu/drm/i915/i915_irq.c intel_irq_fini
 !Fdrivers/gpu/drm/i915/i915_irq.c intel_runtime_pm_disable_interrupts
 !Fdrivers/gpu/drm/i915/i915_irq.c intel_runtime_pm_enable_interrupts
       </sect2>
@@ -4012,7 +4011,6 @@ int num_ioctls;</synopsis>
         <title>Frontbuffer Tracking</title>
 !Pdrivers/gpu/drm/i915/intel_frontbuffer.c frontbuffer tracking
 !Idrivers/gpu/drm/i915/intel_frontbuffer.c
-!Fdrivers/gpu/drm/i915/intel_drv.h intel_frontbuffer_flip
 !Fdrivers/gpu/drm/i915/i915_gem.c i915_gem_track_fb
       </sect2>
       <sect2>
@@ -4044,6 +4042,11 @@ int num_ioctls;</synopsis>
 	  probing, so those sections fully apply.
         </para>
       </sect2>
+      <sect2>
+        <title>Hotplug</title>
+!Pdrivers/gpu/drm/i915/intel_hotplug.c Hotplug
+!Idrivers/gpu/drm/i915/intel_hotplug.c
+      </sect2>
       <sect2>
 	<title>High Definition Audio</title>
 !Pdrivers/gpu/drm/i915/intel_audio.c High Definition Audio over HDMI and Display Port
@@ -4193,6 +4196,23 @@ int num_ioctls;</synopsis>
         <title>Global GTT views</title>
 !Pdrivers/gpu/drm/i915/i915_gem_gtt.c Global GTT views
 !Idrivers/gpu/drm/i915/i915_gem_gtt.c
+      </sect2>
+      <sect2>
+        <title>GTT Fences and Swizzling</title>
+!Idrivers/gpu/drm/i915/i915_gem_fence.c
+        <sect3>
+          <title>Global GTT Fence Handling</title>
+!Pdrivers/gpu/drm/i915/i915_gem_fence.c fence register handling
+        </sect3>
+        <sect3>
+          <title>Hardware Tiling and Swizzling Details</title>
+!Pdrivers/gpu/drm/i915/i915_gem_fence.c tiling swizzling details
+        </sect3>
+      </sect2>
+      <sect2>
+        <title>Object Tiling IOCTLs</title>
+!Idrivers/gpu/drm/i915/i915_gem_tiling.c
+!Pdrivers/gpu/drm/i915/i915_gem_tiling.c buffer object tiling
       </sect2>
       <sect2>
         <title>Buffer Object Eviction</title>

Documentation/devicetree/bindings/drm/msm/dsi.txt

@@ -30,20 +30,27 @@ Optional properties:
 - panel@0: Node of panel connected to this DSI controller.
   See files in Documentation/devicetree/bindings/panel/ for each supported
   panel.
-- qcom,dual-panel-mode: Boolean value indicating if the DSI controller is
+- qcom,dual-dsi-mode: Boolean value indicating if the DSI controller is
   driving a panel which needs 2 DSI links.
-- qcom,master-panel: Boolean value indicating if the DSI controller is driving
+- qcom,master-dsi: Boolean value indicating if the DSI controller is driving
   the master link of the 2-DSI panel.
-- qcom,sync-dual-panel: Boolean value indicating if the DSI controller is
+- qcom,sync-dual-dsi: Boolean value indicating if the DSI controller is
   driving a 2-DSI panel whose 2 links need receive command simultaneously.
 - interrupt-parent: phandle to the MDP block if the interrupt signal is routed
   through MDP block
+- pinctrl-names: the pin control state names; should contain "default"
+- pinctrl-0: the default pinctrl state (active)
+- pinctrl-n: the "sleep" pinctrl state
+- port: DSI controller output port. This contains one endpoint subnode, with its
+  remote-endpoint set to the phandle of the connected panel's endpoint.
+  See Documentation/devicetree/bindings/graph.txt for device graph info.
 
 DSI PHY:
 Required properties:
 - compatible: Could be the following
   * "qcom,dsi-phy-28nm-hpm"
   * "qcom,dsi-phy-28nm-lp"
+  * "qcom,dsi-phy-20nm"
 - reg: Physical base address and length of the registers of PLL, PHY and PHY
   regulator
 - reg-names: The names of register regions. The following regions are required:
@@ -59,6 +66,10 @@ Required properties:
   * "iface_clk"
 - vddio-supply: phandle to vdd-io regulator device node
 
+Optional properties:
+- qcom,dsi-phy-regulator-ldo-mode: Boolean value indicating if the LDO mode PHY
+  regulator is wanted.
+
 Example:
 	mdss_dsi0: qcom,mdss_dsi@fd922800 {
 		compatible = "qcom,mdss-dsi-ctrl";
@@ -90,9 +101,13 @@ Example:
 
 		qcom,dsi-phy = <&mdss_dsi_phy0>;
 
-		qcom,dual-panel-mode;
-		qcom,master-panel;
-		qcom,sync-dual-panel;
+		qcom,dual-dsi-mode;
+		qcom,master-dsi;
+		qcom,sync-dual-dsi;
+
+		pinctrl-names = "default", "sleep";
+		pinctrl-0 = <&mdss_dsi_active>;
+		pinctrl-1 = <&mdss_dsi_suspend>;
 
 		panel: panel@0 {
 			compatible = "sharp,lq101r1sx01";
@@ -101,6 +116,18 @@ Example:
 
 			power-supply = <...>;
 			backlight = <...>;
+
+			port {
+				panel_in: endpoint {
+					remote-endpoint = <&dsi0_out>;
+				};
+			};
+		};
+
+		port {
+			dsi0_out: endpoint {
+				remote-endpoint = <&panel_in>;
+			};
 		};
 	};
 
@@ -117,4 +144,6 @@ Example:
 		clock-names = "iface_clk";
 		clocks = <&mmcc MDSS_AHB_CLK>;
 		vddio-supply = <&pma8084_l12>;
+
+		qcom,dsi-phy-regulator-ldo-mode;
 	};

Documentation/devicetree/bindings/drm/msm/hdmi.txt

@@ -2,8 +2,9 @@ Qualcomm adreno/snapdragon hdmi output
 
 Required properties:
 - compatible: one of the following
+   * "qcom,hdmi-tx-8994"
    * "qcom,hdmi-tx-8084"
-   * "qcom,hdmi-tx-8074"
+   * "qcom,hdmi-tx-8974"
    * "qcom,hdmi-tx-8660"
    * "qcom,hdmi-tx-8960"
 - reg: Physical base address and length of the controller's registers

Documentation/devicetree/bindings/gpu/nvidia,tegra20-host1x.txt

@@ -197,9 +197,11 @@ of the following host1x client modules:
 - sor: serial output resource
 
   Required properties:
-  - compatible: For Tegra124, must contain "nvidia,tegra124-sor".  Otherwise,
-    must contain '"nvidia,<chip>-sor", "nvidia,tegra124-sor"', where <chip>
-    is tegra132.
+  - compatible: Should be:
+    - "nvidia,tegra124-sor": for Tegra124 and Tegra132
+    - "nvidia,tegra132-sor": for Tegra132
+    - "nvidia,tegra210-sor": for Tegra210
+    - "nvidia,tegra210-sor1": for Tegra210
   - reg: Physical base address and length of the controller's registers.
   - interrupts: The interrupt outputs from the controller.
   - clocks: Must contain an entry for each entry in clock-names.

Documentation/devicetree/bindings/gpu/st,stih4xx.txt

@@ -52,10 +52,9 @@ STMicroelectronics stih4xx platforms
     See ../reset/reset.txt for details.
   - reset-names: names of the resets listed in resets property in the same
     order.
-  - ranges: to allow probing of subdevices
 
 - sti-hdmi: hdmi output block
-  must be a child of sti-tvout
+  must be a child of sti-display-subsystem
   Required properties:
   - compatible: "st,stih<chip>-hdmi";
   - reg: Physical base address of the IP registers and length of memory mapped region.
@@ -72,7 +71,7 @@ STMicroelectronics stih4xx platforms
 
 sti-hda:
   Required properties:
-  must be a child of sti-tvout
+  must be a child of sti-display-subsystem
   - compatible: "st,stih<chip>-hda"
   - reg: Physical base address of the IP registers and length of memory mapped region.
   - reg-names: names of the mapped memory regions listed in regs property in
@@ -85,7 +84,7 @@ sti-hda:
 
 sti-dvo:
   Required properties:
-  must be a child of sti-tvout
+  must be a child of sti-display-subsystem
   - compatible: "st,stih<chip>-dvo"
   - reg: Physical base address of the IP registers and length of memory mapped region.
   - reg-names: names of the mapped memory regions listed in regs property in
@@ -195,38 +194,37 @@ Example:
 			reg-names	= "tvout-reg", "hda-reg", "syscfg";
 			reset-names     = "tvout";
 			resets          = <&softreset STIH416_HDTVOUT_SOFTRESET>;
-			ranges;
-
-			sti-hdmi@fe85c000 {
-				compatible	= "st,stih416-hdmi";
-				reg		= <0xfe85c000 0x1000>, <0xfe830000 0x10000>;
-				reg-names	= "hdmi-reg", "syscfg";
-				interrupts	= <GIC_SPI 173 IRQ_TYPE_NONE>;
-				interrupt-names	= "irq";
-				clock-names	= "pix", "tmds", "phy", "audio";
-				clocks          = <&clockgen_c_vcc CLK_S_PIX_HDMI>, <&clockgen_c_vcc CLK_S_TMDS_HDMI>, <&clockgen_c_vcc CLK_S_HDMI_REJECT_PLL>, <&clockgen_b1 CLK_S_PCM_0>;
-			};
-
-			sti-hda@fe85a000 {
-				compatible	= "st,stih416-hda";
-				reg		= <0xfe85a000 0x400>, <0xfe83085c 0x4>;
-				reg-names	= "hda-reg", "video-dacs-ctrl";
-				clock-names	= "pix", "hddac";
-				clocks          = <&clockgen_c_vcc CLK_S_PIX_HD>, <&clockgen_c_vcc CLK_S_HDDAC>;
-			};
-
-			sti-dvo@8d00400 {
-				compatible	= "st,stih407-dvo";
-				reg		= <0x8d00400 0x200>;
-				reg-names	= "dvo-reg";
-				clock-names	= "dvo_pix", "dvo",
-						  "main_parent", "aux_parent";
-				clocks		= <&clk_s_d2_flexgen CLK_PIX_DVO>, <&clk_s_d2_flexgen CLK_DVO>,
-						  <&clk_s_d2_quadfs 0>, <&clk_s_d2_quadfs 1>;
-				pinctrl-names	= "default";
-				pinctrl-0	= <&pinctrl_dvo>;
-				sti,panel	= <&panel_dvo>;
-			};
+		};
+
+		sti-hdmi@fe85c000 {
+			compatible	= "st,stih416-hdmi";
+			reg		= <0xfe85c000 0x1000>, <0xfe830000 0x10000>;
+			reg-names	= "hdmi-reg", "syscfg";
+			interrupts	= <GIC_SPI 173 IRQ_TYPE_NONE>;
+			interrupt-names	= "irq";
+			clock-names	= "pix", "tmds", "phy", "audio";
+			clocks          = <&clockgen_c_vcc CLK_S_PIX_HDMI>, <&clockgen_c_vcc CLK_S_TMDS_HDMI>, <&clockgen_c_vcc CLK_S_HDMI_REJECT_PLL>, <&clockgen_b1 CLK_S_PCM_0>;
+		};
+
+		sti-hda@fe85a000 {
+			compatible	= "st,stih416-hda";
+			reg		= <0xfe85a000 0x400>, <0xfe83085c 0x4>;
+			reg-names	= "hda-reg", "video-dacs-ctrl";
+			clock-names	= "pix", "hddac";
+			clocks          = <&clockgen_c_vcc CLK_S_PIX_HD>, <&clockgen_c_vcc CLK_S_HDDAC>;
+		};
+
+		sti-dvo@8d00400 {
+			compatible	= "st,stih407-dvo";
+			reg		= <0x8d00400 0x200>;
+			reg-names	= "dvo-reg";
+			clock-names	= "dvo_pix", "dvo",
+					  "main_parent", "aux_parent";
+			clocks		= <&clk_s_d2_flexgen CLK_PIX_DVO>, <&clk_s_d2_flexgen CLK_DVO>,
+					  <&clk_s_d2_quadfs 0>, <&clk_s_d2_quadfs 1>;
+			pinctrl-names	= "default";
+			pinctrl-0	= <&pinctrl_dvo>;
+			sti,panel	= <&panel_dvo>;
 		};
 
 		sti-hqvdp@9c000000 {
@@ -237,7 +235,7 @@ Example:
 				reset-names     = "hqvdp";
 				resets          = <&softreset STIH407_HDQVDP_SOFTRESET>;
 				st,vtg		= <&vtg_main>;
-			};
+		};
 	};
 	...
 };

Documentation/devicetree/bindings/panel/auo,b080uan01.txt

@@ -0,0 +1,7 @@
+AU Optronics Corporation 8.0" WUXGA TFT LCD panel
+
+Required properties:
+- compatible: should be "auo,b101ean01"
+
+This binding is compatible with the simple-panel binding, which is specified
+in simple-panel.txt in this directory.

Documentation/devicetree/bindings/panel/lg,lg4573.txt

@@ -0,0 +1,19 @@
+LG LG4573 TFT Liquid Crystal Display with SPI control bus
+
+Required properties:
+  - compatible: "lg,lg4573"
+  - reg: address of the panel on the SPI bus
+
+The panel must obey rules for SPI slave device specified in document [1].
+
+[1]: Documentation/devicetree/bindings/spi/spi-bus.txt
+
+Example:
+
+	lcd_panel: display@0 {
+		#address-cells = <1>;
+		#size-cells = <1>;
+		compatible = "lg,lg4573";
+		spi-max-frequency = <10000000>;
+		reg = <0>;
+	};

Documentation/devicetree/bindings/panel/nec,nl4827hc19-05b.txt

@@ -0,0 +1,7 @@
+NEC LCD Technologies,Ltd. WQVGA TFT LCD panel
+
+Required properties:
+- compatible: should be "nec,nl4827hc19-05b"
+
+This binding is compatible with the simple-panel binding, which is specified
+in simple-panel.txt in this directory.

Documentation/devicetree/bindings/panel/okaya,rs800480t-7x0gp.txt

@@ -0,0 +1,7 @@
+OKAYA Electric America, Inc. RS800480T-7X0GP 7" WVGA LCD panel
+
+Required properties:
+- compatible: should be "okaya,rs800480t-7x0gp"
+
+This binding is compatible with the simple-panel binding, which is specified
+in simple-panel.txt in this directory.

Documentation/devicetree/bindings/vendor-prefixes.txt

@@ -145,6 +145,7 @@ mundoreader	Mundo Reader S.L.
 murata	Murata Manufacturing Co., Ltd.
 mxicy	Macronix International Co., Ltd.
 national	National Semiconductor
+nec	NEC LCD Technologies, Ltd.
 neonode		Neonode Inc.
 netgear	NETGEAR
 netlogic	Broadcom Corporation (formerly NetLogic Microsystems)
@@ -155,6 +156,7 @@ nokia	Nokia
 nuvoton	Nuvoton Technology Corporation
 nvidia	NVIDIA
 nxp	NXP Semiconductors
+okaya	Okaya Electric America, Inc.
 onnn	ON Semiconductor Corp.
 opencores	OpenCores.org
 option	Option NV

Documentation/devicetree/bindings/video/fsl,dcu.txt

@@ -0,0 +1,22 @@
+Device Tree bindings for Freescale DCU DRM Driver
+
+Required properties:
+- compatible:		Should be one of
+	* "fsl,ls1021a-dcu".
+	* "fsl,vf610-dcu".
+
+- reg:			Address and length of the register set for dcu.
+- clocks:		From common clock binding: handle to dcu clock.
+- clock-names:		From common clock binding: Shall be "dcu".
+- big-endian		Boolean property, LS1021A DCU registers are big-endian.
+- fsl,panel:		The phandle to panel node.
+
+Examples:
+dcu: dcu@2ce0000 {
+	compatible = "fsl,ls1021a-dcu";
+	reg = <0x0 0x2ce0000 0x0 0x10000>;
+	clocks = <&platform_clk 0>;
+	clock-names = "dcu";
+	big-endian;
+	fsl,panel = <&panel>;
+};

MAINTAINERS

@@ -643,9 +643,14 @@ M:	Oded Gabbay <oded.gabbay@gmail.com>
 L:	dri-devel@lists.freedesktop.org
 T:	git git://people.freedesktop.org/~gabbayo/linux.git
 S:	Supported
+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/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/radeon/radeon_kfd.c
 F:	drivers/gpu/drm/radeon/radeon_kfd.h
 F:	include/uapi/linux/kfd_ioctl.h
@@ -3577,6 +3582,15 @@ F:	drivers/gpu/drm/exynos/
 F:	include/drm/exynos*
 F:	include/uapi/drm/exynos*
 
+DRM DRIVERS FOR FREESCALE DCU
+M:	Jianwei Wang <jianwei.wang.chn@gmail.com>
+M:	Alison Wang <alison.wang@freescale.com>
+L:	dri-devel@lists.freedesktop.org
+S:	Supported
+F:	drivers/gpu/drm/fsl-dcu/
+F:	Documentation/devicetree/bindings/video/fsl,dcu.txt
+F:	Documentation/devicetree/bindings/panel/nec,nl4827hc19_05b.txt
+
 DRM DRIVERS FOR FREESCALE IMX
 M:	Philipp Zabel <p.zabel@pengutronix.de>
 L:	dri-devel@lists.freedesktop.org

arch/arm/configs/exynos_defconfig

@@ -124,14 +124,14 @@ CONFIG_REGULATOR_S2MPS11=y
 CONFIG_REGULATOR_S5M8767=y
 CONFIG_REGULATOR_TPS65090=y
 CONFIG_DRM=y
-CONFIG_DRM_PTN3460=y
-CONFIG_DRM_PS8622=y
+CONFIG_DRM_NXP_PTN3460=y
+CONFIG_DRM_PARADE_PS8622=y
 CONFIG_DRM_EXYNOS=y
 CONFIG_DRM_EXYNOS_FIMD=y
 CONFIG_DRM_EXYNOS_DSI=y
 CONFIG_DRM_EXYNOS_HDMI=y
 CONFIG_DRM_PANEL_SIMPLE=y
-CONFIG_DRM_PANEL_S6E8AA0=y
+CONFIG_DRM_PANEL_SAMSUNG_S6E8AA0=y
 CONFIG_FB_SIMPLE=y
 CONFIG_EXYNOS_VIDEO=y
 CONFIG_EXYNOS_MIPI_DSI=y

arch/arm/configs/multi_v7_defconfig

@@ -440,8 +440,8 @@ CONFIG_VIDEO_ML86V7667=m
 CONFIG_DRM=y
 # CONFIG_DRM_I2C_CH7006 is not set
 # CONFIG_DRM_I2C_SIL164 is not set
-CONFIG_DRM_PTN3460=m
-CONFIG_DRM_PS8622=m
+CONFIG_DRM_NXP_PTN3460=m
+CONFIG_DRM_PARADE_PS8622=m
 CONFIG_DRM_NOUVEAU=m
 CONFIG_DRM_EXYNOS=m
 CONFIG_DRM_EXYNOS_DSI=y
@@ -449,7 +449,7 @@ CONFIG_DRM_EXYNOS_FIMD=y
 CONFIG_DRM_EXYNOS_HDMI=y
 CONFIG_DRM_RCAR_DU=m
 CONFIG_DRM_TEGRA=y
-CONFIG_DRM_PANEL_S6E8AA0=m
+CONFIG_DRM_PANEL_SAMSUNG_S6E8AA0=m
 CONFIG_DRM_PANEL_SIMPLE=y
 CONFIG_FB_ARMCLCD=y
 CONFIG_FB_WM8505=y

arch/x86/configs/x86_64_defconfig

@@ -208,7 +208,6 @@ CONFIG_AGP_AMD64=y
 CONFIG_AGP_INTEL=y
 CONFIG_DRM=y
 CONFIG_DRM_I915=y
-CONFIG_DRM_I915_KMS=y
 CONFIG_FB_MODE_HELPERS=y
 CONFIG_FB_TILEBLITTING=y
 CONFIG_FB_EFI=y

drivers/char/agp/intel-gtt.c

@@ -1408,8 +1408,8 @@ int intel_gmch_probe(struct pci_dev *bridge_pdev, struct pci_dev *gpu_pdev,
 }
 EXPORT_SYMBOL(intel_gmch_probe);
 
-void intel_gtt_get(size_t *gtt_total, size_t *stolen_size,
-		   phys_addr_t *mappable_base, unsigned long *mappable_end)
+void intel_gtt_get(u64 *gtt_total, size_t *stolen_size,
+		   phys_addr_t *mappable_base, u64 *mappable_end)
 {
 	*gtt_total = intel_private.gtt_total_entries << PAGE_SHIFT;
 	*stolen_size = intel_private.stolen_size;

drivers/gpio/gpiolib.c

@@ -1686,6 +1686,19 @@ void gpiod_add_lookup_table(struct gpiod_lookup_table *table)
 	mutex_unlock(&gpio_lookup_lock);
 }
 
+/**
+ * gpiod_remove_lookup_table() - unregister GPIO device consumers
+ * @table: table of consumers to unregister
+ */
+void gpiod_remove_lookup_table(struct gpiod_lookup_table *table)
+{
+	mutex_lock(&gpio_lookup_lock);
+
+	list_del(&table->list);
+
+	mutex_unlock(&gpio_lookup_lock);
+}
+
 static struct gpio_desc *of_find_gpio(struct device *dev, const char *con_id,
 				      unsigned int idx,
 				      enum gpio_lookup_flags *flags)

drivers/gpu/drm/Kconfig

@@ -37,9 +37,29 @@ config DRM_KMS_FB_HELPER
 	select FB
 	select FRAMEBUFFER_CONSOLE if !EXPERT
 	select FRAMEBUFFER_CONSOLE_DETECT_PRIMARY if FRAMEBUFFER_CONSOLE
+	select FB_SYS_FOPS
+	select FB_SYS_FILLRECT
+	select FB_SYS_COPYAREA
+	select FB_SYS_IMAGEBLIT
+	select FB_CFB_FILLRECT
+	select FB_CFB_COPYAREA
+	select FB_CFB_IMAGEBLIT
 	help
 	  FBDEV helpers for KMS drivers.
 
+config DRM_FBDEV_EMULATION
+	bool "Enable legacy fbdev support for your modesetting driver"
+	depends on DRM
+	select DRM_KMS_HELPER
+	select DRM_KMS_FB_HELPER
+	default y
+	help
+	  Choose this option if you have a need for the legacy fbdev
+	  support. Note that this support also provides the linux console
+	  support on top of your modesetting driver.
+
+	  If in doubt, say "Y".
+
 config DRM_LOAD_EDID_FIRMWARE
 	bool "Allow to specify an EDID data set instead of probing for it"
 	depends on DRM_KMS_HELPER
@@ -79,8 +99,6 @@ config DRM_KMS_CMA_HELPER
 
 source "drivers/gpu/drm/i2c/Kconfig"
 
-source "drivers/gpu/drm/bridge/Kconfig"
-
 config DRM_TDFX
 	tristate "3dfx Banshee/Voodoo3+"
 	depends on DRM && PCI
@@ -110,6 +128,7 @@ config DRM_RADEON
 	select POWER_SUPPLY
 	select HWMON
 	select BACKLIGHT_CLASS_DEVICE
+	select BACKLIGHT_LCD_SUPPORT
 	select INTERVAL_TREE
 	help
 	  Choose this option if you have an ATI Radeon graphics card.  There
@@ -133,6 +152,7 @@ config DRM_AMDGPU
 	select POWER_SUPPLY
 	select HWMON
 	select BACKLIGHT_CLASS_DEVICE
+	select BACKLIGHT_LCD_SUPPORT
 	select INTERVAL_TREE
 	help
 	  Choose this option if you have a recent AMD Radeon graphics card.
@@ -231,10 +251,14 @@ source "drivers/gpu/drm/virtio/Kconfig"
 
 source "drivers/gpu/drm/msm/Kconfig"
 
+source "drivers/gpu/drm/fsl-dcu/Kconfig"
+
 source "drivers/gpu/drm/tegra/Kconfig"
 
 source "drivers/gpu/drm/panel/Kconfig"
 
+source "drivers/gpu/drm/bridge/Kconfig"
+
 source "drivers/gpu/drm/sti/Kconfig"
 
 source "drivers/gpu/drm/amd/amdkfd/Kconfig"

drivers/gpu/drm/Makefile

@@ -23,7 +23,7 @@ drm-$(CONFIG_OF) += drm_of.o
 drm_kms_helper-y := drm_crtc_helper.o drm_dp_helper.o drm_probe_helper.o \
 		drm_plane_helper.o drm_dp_mst_topology.o drm_atomic_helper.o
 drm_kms_helper-$(CONFIG_DRM_LOAD_EDID_FIRMWARE) += drm_edid_load.o
-drm_kms_helper-$(CONFIG_DRM_KMS_FB_HELPER) += drm_fb_helper.o
+drm_kms_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fb_helper.o
 drm_kms_helper-$(CONFIG_DRM_KMS_CMA_HELPER) += drm_fb_cma_helper.o
 
 obj-$(CONFIG_DRM_KMS_HELPER) += drm_kms_helper.o
@@ -70,3 +70,4 @@ obj-$(CONFIG_DRM_IMX) += imx/
 obj-y			+= i2c/
 obj-y			+= panel/
 obj-y			+= bridge/
+obj-$(CONFIG_DRM_FSL_DCU) += fsl-dcu/

drivers/gpu/drm/amd/amdgpu/Makefile

@@ -3,7 +3,9 @@
 # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
 
 ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/asic_reg \
-	-Idrivers/gpu/drm/amd/include
+	-Idrivers/gpu/drm/amd/include \
+	-Idrivers/gpu/drm/amd/amdgpu \
+	-Idrivers/gpu/drm/amd/scheduler
 
 amdgpu-y := amdgpu_drv.o
 
@@ -21,7 +23,8 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 
 # add asic specific block
 amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o gmc_v7_0.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
+	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
 
 amdgpu-y += \
 	vi.o
@@ -43,6 +46,7 @@ amdgpu-y += \
 	amdgpu_dpm.o \
 	cz_smc.o cz_dpm.o \
 	tonga_smc.o tonga_dpm.o \
+	fiji_smc.o fiji_dpm.o \
 	iceland_smc.o iceland_dpm.o
 
 # add DCE block
@@ -71,6 +75,20 @@ amdgpu-y += \
 	amdgpu_vce.o \
 	vce_v3_0.o
 
+# add amdkfd interfaces
+amdgpu-y += \
+	 amdgpu_amdkfd.o \
+	 amdgpu_amdkfd_gfx_v8.o
+
+# add cgs
+amdgpu-y += amdgpu_cgs.o
+
+# GPU scheduler
+amdgpu-y += \
+	../scheduler/gpu_scheduler.o \
+	../scheduler/sched_fence.o \
+	amdgpu_sched.o
+
 amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o
 amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o
 amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o

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

@@ -42,17 +42,19 @@
 #include <ttm/ttm_module.h>
 #include <ttm/ttm_execbuf_util.h>
 
+#include <drm/drmP.h>
 #include <drm/drm_gem.h>
 #include <drm/amdgpu_drm.h>
 
 #include "amd_shared.h"
-#include "amdgpu_family.h"
 #include "amdgpu_mode.h"
 #include "amdgpu_ih.h"
 #include "amdgpu_irq.h"
 #include "amdgpu_ucode.h"
 #include "amdgpu_gds.h"
 
+#include "gpu_scheduler.h"
+
 /*
  * Modules parameters.
  */
@@ -77,7 +79,11 @@ extern int amdgpu_bapm;
 extern int amdgpu_deep_color;
 extern int amdgpu_vm_size;
 extern int amdgpu_vm_block_size;
+extern int amdgpu_enable_scheduler;
+extern int amdgpu_sched_jobs;
+extern int amdgpu_sched_hw_submission;
 
+#define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS	        3000
 #define AMDGPU_MAX_USEC_TIMEOUT			100000	/* 100 ms */
 #define AMDGPU_FENCE_JIFFIES_TIMEOUT		(HZ / 2)
 /* AMDGPU_IB_POOL_SIZE must be a power of 2 */
@@ -92,6 +98,9 @@ extern int amdgpu_vm_block_size;
 #define AMDGPU_MAX_COMPUTE_RINGS		8
 #define AMDGPU_MAX_VCE_RINGS			2
 
+/* max number of IP instances */
+#define AMDGPU_MAX_SDMA_INSTANCES		2
+
 /* number of hw syncs before falling back on blocking */
 #define AMDGPU_NUM_SYNCS			4
 
@@ -177,7 +186,9 @@ struct amdgpu_vm;
 struct amdgpu_ring;
 struct amdgpu_semaphore;
 struct amdgpu_cs_parser;
+struct amdgpu_job;
 struct amdgpu_irq_src;
+struct amdgpu_fpriv;
 
 enum amdgpu_cp_irq {
 	AMDGPU_CP_IRQ_GFX_EOP = 0,
@@ -239,7 +250,7 @@ struct amdgpu_buffer_funcs {
 	unsigned	copy_num_dw;
 
 	/* used for buffer migration */
-	void (*emit_copy_buffer)(struct amdgpu_ring *ring,
+	void (*emit_copy_buffer)(struct amdgpu_ib *ib,
 				 /* src addr in bytes */
 				 uint64_t src_offset,
 				 /* dst addr in bytes */
@@ -254,7 +265,7 @@ struct amdgpu_buffer_funcs {
 	unsigned	fill_num_dw;
 
 	/* used for buffer clearing */
-	void (*emit_fill_buffer)(struct amdgpu_ring *ring,
+	void (*emit_fill_buffer)(struct amdgpu_ib *ib,
 				 /* value to write to memory */
 				 uint32_t src_data,
 				 /* dst addr in bytes */
@@ -332,6 +343,8 @@ struct amdgpu_ring_funcs {
 	int (*test_ring)(struct amdgpu_ring *ring);
 	int (*test_ib)(struct amdgpu_ring *ring);
 	bool (*is_lockup)(struct amdgpu_ring *ring);
+	/* insert NOP packets */
+	void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
 };
 
 /*
@@ -381,10 +394,10 @@ struct amdgpu_fence_driver {
 	uint64_t			sync_seq[AMDGPU_MAX_RINGS];
 	atomic64_t			last_seq;
 	bool				initialized;
-	bool				delayed_irq;
 	struct amdgpu_irq_src		*irq_src;
 	unsigned			irq_type;
 	struct delayed_work             lockup_work;
+	wait_queue_head_t		fence_queue;
 };
 
 /* some special values for the owner field */
@@ -423,20 +436,20 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring);
 int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 				   struct amdgpu_irq_src *irq_src,
 				   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, void *owner,
 		      struct amdgpu_fence **fence);
-int amdgpu_fence_recreate(struct amdgpu_ring *ring, void *owner,
-			  uint64_t seq, struct amdgpu_fence **fence);
 void amdgpu_fence_process(struct amdgpu_ring *ring);
 int amdgpu_fence_wait_next(struct amdgpu_ring *ring);
 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
 
-bool amdgpu_fence_signaled(struct amdgpu_fence *fence);
-int amdgpu_fence_wait(struct amdgpu_fence *fence, bool interruptible);
-int amdgpu_fence_wait_any(struct amdgpu_device *adev,
-			  struct amdgpu_fence **fences,
-			  bool intr);
+signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
+				  struct fence **array,
+				  uint32_t count,
+				  bool intr,
+				  signed long t);
 struct amdgpu_fence *amdgpu_fence_ref(struct amdgpu_fence *fence);
 void amdgpu_fence_unref(struct amdgpu_fence **fence);
 
@@ -481,7 +494,7 @@ static inline bool amdgpu_fence_is_earlier(struct amdgpu_fence *a,
 	return a->seq < b->seq;
 }
 
-int amdgpu_user_fence_emit(struct amdgpu_ring *ring, struct amdgpu_user_fence *user, 
+int amdgpu_user_fence_emit(struct amdgpu_ring *ring, struct amdgpu_user_fence *user,
 			   void *owner, struct amdgpu_fence **fence);
 
 /*
@@ -509,7 +522,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 		       uint64_t dst_offset,
 		       uint32_t byte_count,
 		       struct reservation_object *resv,
-		       struct amdgpu_fence **fence);
+		       struct fence **fence);
 int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
 
 struct amdgpu_bo_list_entry {
@@ -532,14 +545,16 @@ struct amdgpu_bo_va_mapping {
 struct amdgpu_bo_va {
 	/* protected by bo being reserved */
 	struct list_head		bo_list;
-	uint64_t			addr;
-	struct amdgpu_fence		*last_pt_update;
+	struct fence		        *last_pt_update;
 	unsigned			ref_count;
 
-	/* protected by vm mutex */
-	struct list_head		mappings;
+	/* protected by vm mutex and spinlock */
 	struct list_head		vm_status;
 
+	/* mappings for this bo_va */
+	struct list_head		invalids;
+	struct list_head		valids;
+
 	/* constant after initialization */
 	struct amdgpu_vm		*vm;
 	struct amdgpu_bo		*bo;
@@ -643,7 +658,7 @@ struct amdgpu_sa_bo {
 	struct amdgpu_sa_manager	*manager;
 	unsigned			soffset;
 	unsigned			eoffset;
-	struct amdgpu_fence		*fence;
+	struct fence		        *fence;
 };
 
 /*
@@ -685,7 +700,7 @@ bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring,
 				struct amdgpu_semaphore *semaphore);
 void amdgpu_semaphore_free(struct amdgpu_device *adev,
 			   struct amdgpu_semaphore **semaphore,
-			   struct amdgpu_fence *fence);
+			   struct fence *fence);
 
 /*
  * Synchronization
@@ -693,20 +708,23 @@ void amdgpu_semaphore_free(struct amdgpu_device *adev,
 struct amdgpu_sync {
 	struct amdgpu_semaphore *semaphores[AMDGPU_NUM_SYNCS];
 	struct amdgpu_fence	*sync_to[AMDGPU_MAX_RINGS];
-	struct amdgpu_fence	*last_vm_update;
+	DECLARE_HASHTABLE(fences, 4);
+	struct fence	        *last_vm_update;
 };
 
 void amdgpu_sync_create(struct amdgpu_sync *sync);
-void amdgpu_sync_fence(struct amdgpu_sync *sync,
-		       struct amdgpu_fence *fence);
+int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
+		      struct fence *f);
 int amdgpu_sync_resv(struct amdgpu_device *adev,
 		     struct amdgpu_sync *sync,
 		     struct reservation_object *resv,
 		     void *owner);
 int amdgpu_sync_rings(struct amdgpu_sync *sync,
 		      struct amdgpu_ring *ring);
+struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync);
+int amdgpu_sync_wait(struct amdgpu_sync *sync);
 void amdgpu_sync_free(struct amdgpu_device *adev, struct amdgpu_sync *sync,
-		      struct amdgpu_fence *fence);
+		      struct fence *fence);
 
 /*
  * GART structures, functions & helpers
@@ -821,7 +839,9 @@ struct amdgpu_flip_work {
 	uint64_t			base;
 	struct drm_pending_vblank_event *event;
 	struct amdgpu_bo		*old_rbo;
-	struct fence			*fence;
+	struct fence			*excl;
+	unsigned			shared_count;
+	struct fence			**shared;
 };
 
 
@@ -844,6 +864,8 @@ struct amdgpu_ib {
 	uint32_t			gws_base, gws_size;
 	uint32_t			oa_base, oa_size;
 	uint32_t			flags;
+	/* resulting sequence number */
+	uint64_t			sequence;
 };
 
 enum amdgpu_ring_type {
@@ -854,11 +876,23 @@ enum amdgpu_ring_type {
 	AMDGPU_RING_TYPE_VCE
 };
 
+extern struct amd_sched_backend_ops amdgpu_sched_ops;
+
+int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
+					 struct amdgpu_ring *ring,
+					 struct amdgpu_ib *ibs,
+					 unsigned num_ibs,
+					 int (*free_job)(struct amdgpu_job *),
+					 void *owner,
+					 struct fence **fence);
+
 struct amdgpu_ring {
 	struct amdgpu_device		*adev;
 	const struct amdgpu_ring_funcs	*funcs;
 	struct amdgpu_fence_driver	fence_drv;
+	struct amd_gpu_scheduler 	*scheduler;
 
+	spinlock_t              fence_lock;
 	struct mutex		*ring_lock;
 	struct amdgpu_bo	*ring_obj;
 	volatile uint32_t	*ring;
@@ -892,6 +926,7 @@ struct amdgpu_ring {
 	struct amdgpu_ctx	*current_ctx;
 	enum amdgpu_ring_type	type;
 	char			name[16];
+	bool                    is_pte_ring;
 };
 
 /*
@@ -933,7 +968,7 @@ struct amdgpu_vm_id {
 	unsigned		id;
 	uint64_t		pd_gpu_addr;
 	/* last flushed PD/PT update */
-	struct amdgpu_fence	*flushed_updates;
+	struct fence	        *flushed_updates;
 	/* last use of vmid */
 	struct amdgpu_fence	*last_id_use;
 };
@@ -943,18 +978,22 @@ struct amdgpu_vm {
 
 	struct rb_root		va;
 
-	/* protecting invalidated and freed */
+	/* protecting invalidated */
 	spinlock_t		status_lock;
 
 	/* BOs moved, but not yet updated in the PT */
 	struct list_head	invalidated;
 
-	/* BOs freed, but not yet updated in the PT */
+	/* BOs cleared in the PT because of a move */
+	struct list_head	cleared;
+
+	/* BO mappings freed, but not yet updated in the PT */
 	struct list_head	freed;
 
 	/* contains the page directory */
 	struct amdgpu_bo	*page_directory;
 	unsigned		max_pde_used;
+	struct fence		*page_directory_fence;
 
 	/* array of page tables, one for each page directory entry */
 	struct amdgpu_vm_pt	*page_tables;
@@ -983,27 +1022,47 @@ struct amdgpu_vm_manager {
  * context related structures
  */
 
-struct amdgpu_ctx_state {
-	uint64_t flags;
-	uint32_t hangs;
+#define AMDGPU_CTX_MAX_CS_PENDING	16
+
+struct amdgpu_ctx_ring {
+	uint64_t		sequence;
+	struct fence		*fences[AMDGPU_CTX_MAX_CS_PENDING];
+	struct amd_sched_entity	entity;
 };
 
 struct amdgpu_ctx {
-	/* call kref_get()before CS start and kref_put() after CS fence signaled */
-	struct kref refcount;
-	struct amdgpu_fpriv *fpriv;
-	struct amdgpu_ctx_state state;
-	uint32_t id;
-	unsigned reset_counter;
+	struct kref		refcount;
+	struct amdgpu_device    *adev;
+	unsigned		reset_counter;
+	spinlock_t		ring_lock;
+	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
 };
 
 struct amdgpu_ctx_mgr {
-	struct amdgpu_device *adev;
-	struct idr ctx_handles;
-	/* lock for IDR system */
-	struct mutex lock;
+	struct amdgpu_device	*adev;
+	struct mutex		lock;
+	/* protected by lock */
+	struct idr		ctx_handles;
 };
 
+int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel,
+		    struct amdgpu_ctx *ctx);
+void amdgpu_ctx_fini(struct amdgpu_ctx *ctx);
+
+struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
+int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
+
+uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
+			      struct fence *fence);
+struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
+				   struct amdgpu_ring *ring, uint64_t seq);
+
+int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
+		     struct drm_file *filp);
+
+void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
+void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
+
 /*
  * file private structure
  */
@@ -1012,7 +1071,7 @@ struct amdgpu_fpriv {
 	struct amdgpu_vm	vm;
 	struct mutex		bo_list_lock;
 	struct idr		bo_list_handles;
-	struct amdgpu_ctx_mgr ctx_mgr;
+	struct amdgpu_ctx_mgr	ctx_mgr;
 };
 
 /*
@@ -1160,6 +1219,7 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev);
 void amdgpu_ring_free_size(struct amdgpu_ring *ring);
 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
 int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw);
+void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
 void amdgpu_ring_commit(struct amdgpu_ring *ring);
 void amdgpu_ring_unlock_commit(struct amdgpu_ring *ring);
 void amdgpu_ring_undo(struct amdgpu_ring *ring);
@@ -1207,6 +1267,16 @@ struct amdgpu_cs_parser {
 	struct amdgpu_user_fence uf;
 };
 
+struct amdgpu_job {
+	struct amd_sched_job    base;
+	struct amdgpu_device	*adev;
+	struct amdgpu_ib	*ibs;
+	uint32_t		num_ibs;
+	struct mutex            job_lock;
+	struct amdgpu_user_fence uf;
+	int (*free_job)(struct amdgpu_job *sched_job);
+};
+
 static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p, uint32_t ib_idx, int idx)
 {
 	return p->ibs[ib_idx].ptr[idx];
@@ -1601,7 +1671,6 @@ struct amdgpu_uvd {
 	struct amdgpu_bo	*vcpu_bo;
 	void			*cpu_addr;
 	uint64_t		gpu_addr;
-	void			*saved_bo;
 	atomic_t		handles[AMDGPU_MAX_UVD_HANDLES];
 	struct drm_file		*filp[AMDGPU_MAX_UVD_HANDLES];
 	struct delayed_work	idle_work;
@@ -1645,6 +1714,7 @@ struct amdgpu_sdma {
 	uint32_t		feature_version;
 
 	struct amdgpu_ring	ring;
+	bool			burst_nop;
 };
 
 /*
@@ -1849,17 +1919,12 @@ struct amdgpu_atcs {
 	struct amdgpu_atcs_functions functions;
 };
 
-int amdgpu_ctx_alloc(struct amdgpu_device *adev,struct amdgpu_fpriv *fpriv,
-							uint32_t *id,uint32_t flags);
-int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv,
-						  uint32_t id);
-
-void amdgpu_ctx_fini(struct amdgpu_fpriv *fpriv);
-struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
-int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
+/*
+ * CGS
+ */
+void *amdgpu_cgs_create_device(struct amdgpu_device *adev);
+void amdgpu_cgs_destroy_device(void *cgs_device);
 
-extern int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
-						 struct drm_file *filp);
 
 /*
  * Core structure, functions and helpers.
@@ -1883,7 +1948,7 @@ struct amdgpu_device {
 	struct rw_semaphore		exclusive_lock;
 
 	/* ASIC */
-	enum amdgpu_asic_type           asic_type;
+	enum amd_asic_type		asic_type;
 	uint32_t			family;
 	uint32_t			rev_id;
 	uint32_t			external_rev_id;
@@ -1976,7 +2041,6 @@ struct amdgpu_device {
 	struct amdgpu_irq_src		hpd_irq;
 
 	/* rings */
-	wait_queue_head_t		fence_queue;
 	unsigned			fence_context;
 	struct mutex			ring_lock;
 	unsigned			num_rings;
@@ -1999,7 +2063,7 @@ struct amdgpu_device {
 	struct amdgpu_gfx		gfx;
 
 	/* sdma */
-	struct amdgpu_sdma		sdma[2];
+	struct amdgpu_sdma		sdma[AMDGPU_MAX_SDMA_INSTANCES];
 	struct amdgpu_irq_src		sdma_trap_irq;
 	struct amdgpu_irq_src		sdma_illegal_inst_irq;
 
@@ -2025,6 +2089,12 @@ struct amdgpu_device {
 	/* tracking pinned memory */
 	u64 vram_pin_size;
 	u64 gart_pin_size;
+
+	/* amdkfd interface */
+	struct kfd_dev          *kfd;
+
+	/* kernel conext for IB submission */
+	struct amdgpu_ctx	kernel_ctx;
 };
 
 bool amdgpu_device_is_px(struct drm_device *dev);
@@ -2132,6 +2202,21 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
 	ring->ring_free_dw--;
 }
 
+static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	int i;
+
+	for (i = 0; i < AMDGPU_MAX_SDMA_INSTANCES; i++)
+		if (&adev->sdma[i].ring == ring)
+			break;
+
+	if (i < AMDGPU_MAX_SDMA_INSTANCES)
+		return &adev->sdma[i];
+	else
+		return NULL;
+}
+
 /*
  * ASICs macro.
  */
@@ -2183,8 +2268,8 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
 #define amdgpu_display_add_connector(adev, ci, sd, ct, ib, coi, h, r) (adev)->mode_info.funcs->add_connector((adev), (ci), (sd), (ct), (ib), (coi), (h), (r))
 #define amdgpu_display_stop_mc_access(adev, s) (adev)->mode_info.funcs->stop_mc_access((adev), (s))
 #define amdgpu_display_resume_mc_access(adev, s) (adev)->mode_info.funcs->resume_mc_access((adev), (s))
-#define amdgpu_emit_copy_buffer(adev, r, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((r), (s), (d), (b))
-#define amdgpu_emit_fill_buffer(adev, r, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((r), (s), (d), (b))
+#define amdgpu_emit_copy_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((ib),  (s), (d), (b))
+#define amdgpu_emit_fill_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((ib), (s), (d), (b))
 #define amdgpu_dpm_get_temperature(adev) (adev)->pm.funcs->get_temperature((adev))
 #define amdgpu_dpm_pre_set_power_state(adev) (adev)->pm.funcs->pre_set_power_state((adev))
 #define amdgpu_dpm_set_power_state(adev) (adev)->pm.funcs->set_power_state((adev))
@@ -2212,6 +2297,12 @@ void amdgpu_pci_config_reset(struct amdgpu_device *adev);
 bool amdgpu_card_posted(struct amdgpu_device *adev);
 void amdgpu_update_display_priority(struct amdgpu_device *adev);
 bool amdgpu_boot_test_post_card(struct amdgpu_device *adev);
+struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
+						 struct drm_file *filp,
+						 struct amdgpu_ctx *ctx,
+						 struct amdgpu_ib *ibs,
+						 uint32_t num_ibs);
+
 int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data);
 int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 		       u32 ip_instance, u32 ring,
@@ -2275,11 +2366,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
 					  struct amdgpu_vm *vm,
 					  struct list_head *head);
-struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
-				       struct amdgpu_vm *vm);
+int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
+		      struct amdgpu_sync *sync);
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
 		     struct amdgpu_vm *vm,
-		     struct amdgpu_fence *updates);
+		     struct fence *updates);
 void amdgpu_vm_fence(struct amdgpu_device *adev,
 		     struct amdgpu_vm *vm,
 		     struct amdgpu_fence *fence);
@@ -2309,7 +2400,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 		       uint64_t addr);
 void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		      struct amdgpu_bo_va *bo_va);
-
+int amdgpu_vm_free_job(struct amdgpu_job *job);
 /*
  * functions used by amdgpu_encoder.c
  */

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

@@ -0,0 +1,269 @@
+/*
+ * Copyright 2014 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_amdkfd.h"
+#include "amd_shared.h"
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include <linux/module.h>
+
+const struct kfd2kgd_calls *kfd2kgd;
+const struct kgd2kfd_calls *kgd2kfd;
+bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+
+bool amdgpu_amdkfd_init(void)
+{
+#if defined(CONFIG_HSA_AMD_MODULE)
+	bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+
+	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+
+	if (kgd2kfd_init_p == NULL)
+		return false;
+#endif
+	return true;
+}
+
+bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev)
+{
+#if defined(CONFIG_HSA_AMD_MODULE)
+	bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+#endif
+
+	switch (rdev->asic_type) {
+#ifdef CONFIG_DRM_AMDGPU_CIK
+	case CHIP_KAVERI:
+		kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions();
+		break;
+#endif
+	case CHIP_CARRIZO:
+		kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
+		break;
+	default:
+		return false;
+	}
+
+#if defined(CONFIG_HSA_AMD_MODULE)
+	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+
+	if (kgd2kfd_init_p == NULL) {
+		kfd2kgd = NULL;
+		return false;
+	}
+
+	if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) {
+		symbol_put(kgd2kfd_init);
+		kfd2kgd = NULL;
+		kgd2kfd = NULL;
+
+		return false;
+	}
+
+	return true;
+#elif defined(CONFIG_HSA_AMD)
+	if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) {
+		kfd2kgd = NULL;
+		kgd2kfd = NULL;
+		return false;
+	}
+
+	return true;
+#else
+	kfd2kgd = NULL;
+	return false;
+#endif
+}
+
+void amdgpu_amdkfd_fini(void)
+{
+	if (kgd2kfd) {
+		kgd2kfd->exit();
+		symbol_put(kgd2kfd_init);
+	}
+}
+
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev)
+{
+	if (kgd2kfd)
+		rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev,
+					rdev->pdev, kfd2kgd);
+}
+
+void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd) {
+		struct kgd2kfd_shared_resources gpu_resources = {
+			.compute_vmid_bitmap = 0xFF00,
+
+			.first_compute_pipe = 1,
+			.compute_pipe_count = 4 - 1,
+		};
+
+		amdgpu_doorbell_get_kfd_info(rdev,
+				&gpu_resources.doorbell_physical_address,
+				&gpu_resources.doorbell_aperture_size,
+				&gpu_resources.doorbell_start_offset);
+
+		kgd2kfd->device_init(rdev->kfd, &gpu_resources);
+	}
+}
+
+void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd) {
+		kgd2kfd->device_exit(rdev->kfd);
+		rdev->kfd = NULL;
+	}
+}
+
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+		const void *ih_ring_entry)
+{
+	if (rdev->kfd)
+		kgd2kfd->interrupt(rdev->kfd, ih_ring_entry);
+}
+
+void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev)
+{
+	if (rdev->kfd)
+		kgd2kfd->suspend(rdev->kfd);
+}
+
+int amdgpu_amdkfd_resume(struct amdgpu_device *rdev)
+{
+	int r = 0;
+
+	if (rdev->kfd)
+		r = kgd2kfd->resume(rdev->kfd);
+
+	return r;
+}
+
+u32 pool_to_domain(enum kgd_memory_pool p)
+{
+	switch (p) {
+	case KGD_POOL_FRAMEBUFFER: return AMDGPU_GEM_DOMAIN_VRAM;
+	default: return AMDGPU_GEM_DOMAIN_GTT;
+	}
+}
+
+int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
+			void **mem_obj, uint64_t *gpu_addr,
+			void **cpu_ptr)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+	struct kgd_mem **mem = (struct kgd_mem **) mem_obj;
+	int r;
+
+	BUG_ON(kgd == NULL);
+	BUG_ON(gpu_addr == NULL);
+	BUG_ON(cpu_ptr == NULL);
+
+	*mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if ((*mem) == NULL)
+		return -ENOMEM;
+
+	r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT,
+			AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, &(*mem)->bo);
+	if (r) {
+		dev_err(rdev->dev,
+			"failed to allocate BO for amdkfd (%d)\n", r);
+		return r;
+	}
+
+	/* map the buffer */
+	r = amdgpu_bo_reserve((*mem)->bo, true);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r);
+		goto allocate_mem_reserve_bo_failed;
+	}
+
+	r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT,
+				&(*mem)->gpu_addr);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r);
+		goto allocate_mem_pin_bo_failed;
+	}
+	*gpu_addr = (*mem)->gpu_addr;
+
+	r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr);
+	if (r) {
+		dev_err(rdev->dev,
+			"(%d) failed to map bo to kernel for amdkfd\n", r);
+		goto allocate_mem_kmap_bo_failed;
+	}
+	*cpu_ptr = (*mem)->cpu_ptr;
+
+	amdgpu_bo_unreserve((*mem)->bo);
+
+	return 0;
+
+allocate_mem_kmap_bo_failed:
+	amdgpu_bo_unpin((*mem)->bo);
+allocate_mem_pin_bo_failed:
+	amdgpu_bo_unreserve((*mem)->bo);
+allocate_mem_reserve_bo_failed:
+	amdgpu_bo_unref(&(*mem)->bo);
+
+	return r;
+}
+
+void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj)
+{
+	struct kgd_mem *mem = (struct kgd_mem *) mem_obj;
+
+	BUG_ON(mem == NULL);
+
+	amdgpu_bo_reserve(mem->bo, true);
+	amdgpu_bo_kunmap(mem->bo);
+	amdgpu_bo_unpin(mem->bo);
+	amdgpu_bo_unreserve(mem->bo);
+	amdgpu_bo_unref(&(mem->bo));
+	kfree(mem);
+}
+
+uint64_t get_vmem_size(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev =
+		(struct amdgpu_device *)kgd;
+
+	BUG_ON(kgd == NULL);
+
+	return rdev->mc.real_vram_size;
+}
+
+uint64_t get_gpu_clock_counter(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+
+	if (rdev->asic_funcs->get_gpu_clock_counter)
+		return rdev->asic_funcs->get_gpu_clock_counter(rdev);
+	return 0;
+}
+
+uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+
+	/* The sclk is in quantas of 10kHz */
+	return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+}

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

@@ -0,0 +1,65 @@
+/*
+ * Copyright 2014 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.
+ */
+
+/* amdgpu_amdkfd.h defines the private interface between amdgpu and amdkfd. */
+
+#ifndef AMDGPU_AMDKFD_H_INCLUDED
+#define AMDGPU_AMDKFD_H_INCLUDED
+
+#include <linux/types.h>
+#include <kgd_kfd_interface.h>
+
+struct amdgpu_device;
+
+struct kgd_mem {
+	struct amdgpu_bo *bo;
+	uint64_t gpu_addr;
+	void *cpu_ptr;
+};
+
+bool amdgpu_amdkfd_init(void);
+void amdgpu_amdkfd_fini(void);
+
+bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev);
+
+void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev);
+int amdgpu_amdkfd_resume(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+			const void *ih_ring_entry);
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev);
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
+
+/* Shared API */
+int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
+			void **mem_obj, uint64_t *gpu_addr,
+			void **cpu_ptr);
+void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj);
+uint64_t get_vmem_size(struct kgd_dev *kgd);
+uint64_t get_gpu_clock_counter(struct kgd_dev *kgd);
+
+uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd);
+
+#endif /* AMDGPU_AMDKFD_H_INCLUDED */

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

@@ -0,0 +1,670 @@
+/*
+ * Copyright 2014 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 <linux/fdtable.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "cikd.h"
+#include "cik_sdma.h"
+#include "amdgpu_ucode.h"
+#include "gca/gfx_7_2_d.h"
+#include "gca/gfx_7_2_enum.h"
+#include "gca/gfx_7_2_sh_mask.h"
+#include "oss/oss_2_0_d.h"
+#include "oss/oss_2_0_sh_mask.h"
+#include "gmc/gmc_7_1_d.h"
+#include "gmc/gmc_7_1_sh_mask.h"
+#include "cik_structs.h"
+
+#define CIK_PIPE_PER_MEC	(4)
+
+enum {
+	MAX_TRAPID = 8,		/* 3 bits in the bitfield. */
+	MAX_WATCH_ADDRESSES = 4
+};
+
+enum {
+	ADDRESS_WATCH_REG_ADDR_HI = 0,
+	ADDRESS_WATCH_REG_ADDR_LO,
+	ADDRESS_WATCH_REG_CNTL,
+	ADDRESS_WATCH_REG_MAX
+};
+
+/*  not defined in the CI/KV reg file  */
+enum {
+	ADDRESS_WATCH_REG_CNTL_ATC_BIT = 0x10000000UL,
+	ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK = 0x00FFFFFF,
+	ADDRESS_WATCH_REG_ADDLOW_MASK_EXTENSION = 0x03000000,
+	/* extend the mask to 26 bits to match the low address field */
+	ADDRESS_WATCH_REG_ADDLOW_SHIFT = 6,
+	ADDRESS_WATCH_REG_ADDHIGH_MASK = 0xFFFF
+};
+
+static const uint32_t watchRegs[MAX_WATCH_ADDRESSES * ADDRESS_WATCH_REG_MAX] = {
+	mmTCP_WATCH0_ADDR_H, mmTCP_WATCH0_ADDR_L, mmTCP_WATCH0_CNTL,
+	mmTCP_WATCH1_ADDR_H, mmTCP_WATCH1_ADDR_L, mmTCP_WATCH1_CNTL,
+	mmTCP_WATCH2_ADDR_H, mmTCP_WATCH2_ADDR_L, mmTCP_WATCH2_CNTL,
+	mmTCP_WATCH3_ADDR_H, mmTCP_WATCH3_ADDR_L, mmTCP_WATCH3_CNTL
+};
+
+union TCP_WATCH_CNTL_BITS {
+	struct {
+		uint32_t mask:24;
+		uint32_t vmid:4;
+		uint32_t atc:1;
+		uint32_t mode:2;
+		uint32_t valid:1;
+	} bitfields, bits;
+	uint32_t u32All;
+	signed int i32All;
+	float f32All;
+};
+
+/*
+ * 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_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);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id);
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, 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_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout);
+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 write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_gtt_mem_allocation = alloc_gtt_mem,
+	.free_gtt_mem = free_gtt_mem,
+	.get_vmem_size = get_vmem_size,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.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,
+	.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,
+	.write_vmid_invalidate_request = write_vmid_invalidate_request,
+	.get_fw_version = get_fw_version
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions()
+{
+	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);
+	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+	mutex_lock(&adev->srbm_mutex);
+	WREG32(mmSRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	WREG32(mmSRBM_GFX_CNTL, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+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(mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
+	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+	WREG32(mmSH_MEM_BASES, sh_mem_bases);
+
+	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;
+
+	WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping);
+
+	while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid)))
+		cpu_relax();
+	WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping);
+
+	return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+	WREG32(mmCP_HPD_EOP_BASE_ADDR, lower_32_bits(hpd_gpu_addr >> 8));
+	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(hpd_gpu_addr >> 8));
+	WREG32(mmCP_HPD_EOP_VMID, 0);
+	WREG32(mmCP_HPD_EOP_CONTROL, hpd_size);
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+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 / CIK_PIPE_PER_MEC) + 1;
+	pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+
+	WREG32(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 inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+{
+	uint32_t retval;
+
+	retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET +
+			m->sdma_queue_id * KFD_CIK_SDMA_QUEUE_OFFSET;
+
+	pr_debug("kfd: sdma base address: 0x%x\n", retval);
+
+	return retval;
+}
+
+static inline struct cik_mqd *get_mqd(void *mqd)
+{
+	return (struct cik_mqd *)mqd;
+}
+
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	return (struct cik_sdma_rlc_registers *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t wptr_shadow, is_wptr_shadow_valid;
+	struct cik_mqd *m;
+
+	m = get_mqd(mqd);
+
+	is_wptr_shadow_valid = !get_user(wptr_shadow, wptr);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+
+	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+
+	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+	WREG32(mmCP_HQD_IB_BASE_ADDR, m->cp_hqd_ib_base_addr_lo);
+	WREG32(mmCP_HQD_IB_BASE_ADDR_HI, m->cp_hqd_ib_base_addr_hi);
+
+	WREG32(mmCP_HQD_IB_RPTR, m->cp_hqd_ib_rptr);
+
+	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
+	WREG32(mmCP_HQD_SEMA_CMD, m->cp_hqd_sema_cmd);
+	WREG32(mmCP_HQD_MSG_TYPE, m->cp_hqd_msg_type);
+
+	WREG32(mmCP_HQD_ATOMIC0_PREOP_LO, m->cp_hqd_atomic0_preop_lo);
+	WREG32(mmCP_HQD_ATOMIC0_PREOP_HI, m->cp_hqd_atomic0_preop_hi);
+	WREG32(mmCP_HQD_ATOMIC1_PREOP_LO, m->cp_hqd_atomic1_preop_lo);
+	WREG32(mmCP_HQD_ATOMIC1_PREOP_HI, m->cp_hqd_atomic1_preop_hi);
+
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+			m->cp_hqd_pq_rptr_report_addr_hi);
+
+	WREG32(mmCP_HQD_PQ_RPTR, m->cp_hqd_pq_rptr);
+
+	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, m->cp_hqd_pq_wptr_poll_addr_lo);
+	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, m->cp_hqd_pq_wptr_poll_addr_hi);
+
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+
+	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+
+	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+
+	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+
+	WREG32(mmCP_HQD_IQ_RPTR, m->cp_hqd_iq_rptr);
+
+	if (is_wptr_shadow_valid)
+		WREG32(mmCP_HQD_PQ_WPTR, wptr_shadow);
+
+	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+	release_queue(kgd);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
+			m->sdma_rlc_virtual_addr);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE,
+			m->sdma_rlc_rb_base);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdma_rlc_rb_base_hi);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdma_rlc_rb_rptr_addr_lo);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdma_rlc_rb_rptr_addr_hi);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
+			m->sdma_rlc_doorbell);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+			m->sdma_rlc_rb_cntl);
+
+	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(mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(mmCP_HQD_PQ_BASE) &&
+				high == RREG32(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 cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	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, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t temp;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+
+	while (true) {
+		temp = RREG32(mmCP_HQD_ACTIVE);
+		if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+			break;
+		if (timeout == 0) {
+			pr_err("kfd: cp queue preemption time out (%dms)\n",
+				temp);
+			release_queue(kgd);
+			return -ETIME;
+		}
+		msleep(20);
+		timeout -= 20;
+	}
+
+	release_queue(kgd);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t temp;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	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_STATUS_REG__RB_CMD_IDLE__SHIFT)
+			break;
+		if (timeout == 0)
+			return -ETIME;
+		msleep(20);
+		timeout -= 20;
+	}
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+
+	return 0;
+}
+
+static int kgd_address_watch_disable(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	union TCP_WATCH_CNTL_BITS cntl;
+	unsigned int i;
+
+	cntl.u32All = 0;
+
+	cntl.bitfields.valid = 0;
+	cntl.bitfields.mask = ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK;
+	cntl.bitfields.atc = 1;
+
+	/* Turning off this address until we set all the registers */
+	for (i = 0; i < MAX_WATCH_ADDRESSES; i++)
+		WREG32(watchRegs[i * ADDRESS_WATCH_REG_MAX +
+			ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	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)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	union TCP_WATCH_CNTL_BITS cntl;
+
+	cntl.u32All = cntl_val;
+
+	/* Turning off this watch point until we set all the registers */
+	cntl.bitfields.valid = 0;
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_ADDR_HI], addr_hi);
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_ADDR_LO], addr_lo);
+
+	/* Enable the watch point */
+	cntl.bitfields.valid = 1;
+
+	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
+		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+
+	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;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32(mmSQ_CMD, sq_cmd);
+
+	/*  Restore the GRBM_GFX_INDEX register  */
+
+	data = GRBM_GFX_INDEX__INSTANCE_BROADCAST_WRITES_MASK |
+		GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
+		GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
+
+	WREG32(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 watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + reg_offset];
+}
+
+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(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(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+}
+
+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;
+
+	BUG_ON(kgd == NULL);
+
+	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[0].fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[1].fw->data;
+		break;
+
+	default:
+		return 0;
+	}
+
+	if (hdr == NULL)
+		return 0;
+
+	/* Only 12 bit in use*/
+	return hdr->common.ucode_version;
+}
+

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

@@ -0,0 +1,543 @@
+/*
+ * Copyright 2014 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 <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 "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_d.h"
+#include "gca/gfx_8_0_enum.h"
+#include "oss/oss_3_0_sh_mask.h"
+#include "oss/oss_3_0_d.h"
+#include "gmc/gmc_8_1_sh_mask.h"
+#include "gmc/gmc_8_1_d.h"
+#include "vi_structs.h"
+#include "vid.h"
+
+#define VI_PIPE_PER_MEC	(4)
+
+struct cik_sdma_rlc_registers;
+
+/*
+ * 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_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);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+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, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id);
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+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 write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_gtt_mem_allocation = alloc_gtt_mem,
+	.free_gtt_mem = free_gtt_mem,
+	.get_vmem_size = get_vmem_size,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.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,
+	.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,
+	.write_vmid_invalidate_request = write_vmid_invalidate_request,
+	.get_fw_version = get_fw_version
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions()
+{
+	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);
+	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+	mutex_lock(&adev->srbm_mutex);
+	WREG32(mmSRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	WREG32(mmSRBM_GFX_CNTL, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+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(mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
+	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+	WREG32(mmSH_MEM_BASES, sh_mem_bases);
+
+	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;
+
+	WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping);
+
+	while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid)))
+		cpu_relax();
+	WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping);
+
+	return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+	return 0;
+}
+
+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 / VI_PIPE_PER_MEC) + 1;
+	pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+
+	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK);
+
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+{
+	return 0;
+}
+
+static inline struct vi_mqd *get_mqd(void *mqd)
+{
+	return (struct vi_mqd *)mqd;
+}
+
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	return (struct cik_sdma_rlc_registers *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	struct vi_mqd *m;
+	uint32_t shadow_wptr, valid_wptr;
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	m = get_mqd(mqd);
+
+	valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr));
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+
+	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
+	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
+	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+			m->cp_hqd_pq_rptr_report_addr_hi);
+
+	if (valid_wptr > 0)
+		WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr);
+
+	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+
+	WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo);
+	WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi);
+	WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control);
+	WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
+	WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
+	WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events);
+
+	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo);
+	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi);
+	WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control);
+	WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset);
+	WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size);
+	WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset);
+	WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size);
+
+	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request);
+	WREG32(mmCP_HQD_ERROR, m->cp_hqd_error);
+	WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
+	WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones);
+
+	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+
+	release_queue(kgd);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+{
+	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(mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(mmCP_HQD_PQ_BASE) &&
+				high == RREG32(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 cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	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, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t temp;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+
+	while (true) {
+		temp = RREG32(mmCP_HQD_ACTIVE);
+		if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+			break;
+		if (timeout == 0) {
+			pr_err("kfd: cp queue preemption time out (%dms)\n",
+				temp);
+			release_queue(kgd);
+			return -ETIME;
+		}
+		msleep(20);
+		timeout -= 20;
+	}
+
+	release_queue(kgd);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+				unsigned int timeout)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct cik_sdma_rlc_registers *m;
+	uint32_t sdma_base_addr;
+	uint32_t temp;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+
+	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_STATUS_REG__RB_CMD_IDLE__SHIFT)
+			break;
+		if (timeout == 0)
+			return -ETIME;
+		msleep(20);
+		timeout -= 20;
+	}
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+
+	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(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(mmATC_VMID0_PASID_MAPPING + vmid);
+	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+}
+
+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(mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32(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(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 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;
+
+	BUG_ON(kgd == NULL);
+
+	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[0].fw->data;
+		break;
+
+	case KGD_ENGINE_SDMA2:
+		hdr = (const union amdgpu_firmware_header *)
+							adev->sdma[1].fw->data;
+		break;
+
+	default:
+		return 0;
+	}
+
+	if (hdr == NULL)
+		return 0;
+
+	/* Only 12 bit in use*/
+	return hdr->common.ucode_version;
+}

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

@@ -897,7 +897,7 @@ bool amdgpu_atombios_get_asic_ss_info(struct amdgpu_device *adev,
 					if ((id == ASIC_INTERNAL_ENGINE_SS) ||
 					    (id == ASIC_INTERNAL_MEMORY_SS))
 						ss->rate /= 100;
-					if (adev->flags & AMDGPU_IS_APU)
+					if (adev->flags & AMD_IS_APU)
 						amdgpu_atombios_get_igp_ss_overrides(adev, ss, id);
 					return true;
 				}
@@ -1058,7 +1058,7 @@ void amdgpu_atombios_set_memory_clock(struct amdgpu_device *adev,
 	SET_MEMORY_CLOCK_PS_ALLOCATION args;
 	int index = GetIndexIntoMasterTable(COMMAND, SetMemoryClock);
 
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return;
 
 	args.ulTargetMemoryClock = cpu_to_le32(mem_clock);	/* 10 khz */

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

@@ -33,7 +33,7 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
 {
 	unsigned long start_jiffies;
 	unsigned long end_jiffies;
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int i, r;
 
 	start_jiffies = jiffies;
@@ -42,17 +42,17 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
 		r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence);
 		if (r)
 			goto exit_do_move;
-		r = amdgpu_fence_wait(fence, false);
+		r = fence_wait(fence, false);
 		if (r)
 			goto exit_do_move;
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 	}
 	end_jiffies = jiffies;
 	r = jiffies_to_msecs(end_jiffies - start_jiffies);
 
 exit_do_move:
 	if (fence)
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 	return r;
 }
 

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

@@ -48,7 +48,7 @@ static bool igp_read_bios_from_vram(struct amdgpu_device *adev)
 	resource_size_t vram_base;
 	resource_size_t size = 256 * 1024; /* ??? */
 
-	if (!(adev->flags & AMDGPU_IS_APU))
+	if (!(adev->flags & AMD_IS_APU))
 		if (!amdgpu_card_posted(adev))
 			return false;
 
@@ -184,7 +184,7 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 	bool found = false;
 
 	/* ATRM is for the discrete card only */
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return false;
 
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
@@ -246,7 +246,7 @@ static inline bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 
 static bool amdgpu_read_disabled_bios(struct amdgpu_device *adev)
 {
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return igp_read_bios_from_vram(adev);
 	else
 		return amdgpu_asic_read_disabled_bios(adev);

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

@@ -0,0 +1,838 @@
+/*
+ * Copyright 2015 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 <linux/list.h>
+#include <linux/slab.h>
+#include <linux/pci.h>
+#include <drm/drmP.h>
+#include <linux/firmware.h>
+#include <drm/amdgpu_drm.h>
+#include "amdgpu.h"
+#include "cgs_linux.h"
+#include "atom.h"
+#include "amdgpu_ucode.h"
+
+
+struct amdgpu_cgs_device {
+	struct cgs_device base;
+	struct amdgpu_device *adev;
+};
+
+#define CGS_FUNC_ADEV							\
+	struct amdgpu_device *adev =					\
+		((struct amdgpu_cgs_device *)cgs_device)->adev
+
+static int amdgpu_cgs_gpu_mem_info(void *cgs_device, enum cgs_gpu_mem_type type,
+				   uint64_t *mc_start, uint64_t *mc_size,
+				   uint64_t *mem_size)
+{
+	CGS_FUNC_ADEV;
+	switch(type) {
+	case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
+	case CGS_GPU_MEM_TYPE__VISIBLE_FB:
+		*mc_start = 0;
+		*mc_size = adev->mc.visible_vram_size;
+		*mem_size = adev->mc.visible_vram_size - adev->vram_pin_size;
+		break;
+	case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
+	case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
+		*mc_start = adev->mc.visible_vram_size;
+		*mc_size = adev->mc.real_vram_size - adev->mc.visible_vram_size;
+		*mem_size = *mc_size;
+		break;
+	case CGS_GPU_MEM_TYPE__GART_CACHEABLE:
+	case CGS_GPU_MEM_TYPE__GART_WRITECOMBINE:
+		*mc_start = adev->mc.gtt_start;
+		*mc_size = adev->mc.gtt_size;
+		*mem_size = adev->mc.gtt_size - adev->gart_pin_size;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cgs_gmap_kmem(void *cgs_device, void *kmem,
+				uint64_t size,
+				uint64_t min_offset, uint64_t max_offset,
+				cgs_handle_t *kmem_handle, uint64_t *mcaddr)
+{
+	CGS_FUNC_ADEV;
+	int ret;
+	struct amdgpu_bo *bo;
+	struct page *kmem_page = vmalloc_to_page(kmem);
+	int npages = ALIGN(size, PAGE_SIZE) >> PAGE_SHIFT;
+
+	struct sg_table *sg = drm_prime_pages_to_sg(&kmem_page, npages);
+	ret = amdgpu_bo_create(adev, size, PAGE_SIZE, false,
+			       AMDGPU_GEM_DOMAIN_GTT, 0, sg, &bo);
+	if (ret)
+		return ret;
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret != 0))
+		return ret;
+
+	/* pin buffer into GTT */
+	ret = amdgpu_bo_pin_restricted(bo, AMDGPU_GEM_DOMAIN_GTT,
+				       min_offset, max_offset, mcaddr);
+	amdgpu_bo_unreserve(bo);
+
+	*kmem_handle = (cgs_handle_t)bo;
+	return ret;
+}
+
+static int amdgpu_cgs_gunmap_kmem(void *cgs_device, cgs_handle_t kmem_handle)
+{
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)kmem_handle;
+
+	if (obj) {
+		int r = amdgpu_bo_reserve(obj, false);
+		if (likely(r == 0)) {
+			amdgpu_bo_unpin(obj);
+			amdgpu_bo_unreserve(obj);
+		}
+		amdgpu_bo_unref(&obj);
+
+	}
+	return 0;
+}
+
+static int amdgpu_cgs_alloc_gpu_mem(void *cgs_device,
+				    enum cgs_gpu_mem_type type,
+				    uint64_t size, uint64_t align,
+				    uint64_t min_offset, uint64_t max_offset,
+				    cgs_handle_t *handle)
+{
+	CGS_FUNC_ADEV;
+	uint16_t flags = 0;
+	int ret = 0;
+	uint32_t domain = 0;
+	struct amdgpu_bo *obj;
+	struct ttm_placement placement;
+	struct ttm_place place;
+
+	if (min_offset > max_offset) {
+		BUG_ON(1);
+		return -EINVAL;
+	}
+
+	/* fail if the alignment is not a power of 2 */
+	if (((align != 1) && (align & (align - 1)))
+	    || size == 0 || align == 0)
+		return -EINVAL;
+
+
+	switch(type) {
+	case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
+	case CGS_GPU_MEM_TYPE__VISIBLE_FB:
+		flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+		domain = AMDGPU_GEM_DOMAIN_VRAM;
+		if (max_offset > adev->mc.real_vram_size)
+			return -EINVAL;
+		place.fpfn = min_offset >> PAGE_SHIFT;
+		place.lpfn = max_offset >> PAGE_SHIFT;
+		place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
+			TTM_PL_FLAG_VRAM;
+		break;
+	case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
+	case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
+		flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+		domain = AMDGPU_GEM_DOMAIN_VRAM;
+		if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
+			place.fpfn =
+				max(min_offset, adev->mc.visible_vram_size) >> PAGE_SHIFT;
+			place.lpfn =
+				min(max_offset, adev->mc.real_vram_size) >> PAGE_SHIFT;
+			place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
+				TTM_PL_FLAG_VRAM;
+		}
+
+		break;
+	case CGS_GPU_MEM_TYPE__GART_CACHEABLE:
+		domain = AMDGPU_GEM_DOMAIN_GTT;
+		place.fpfn = min_offset >> PAGE_SHIFT;
+		place.lpfn = max_offset >> PAGE_SHIFT;
+		place.flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
+		break;
+	case CGS_GPU_MEM_TYPE__GART_WRITECOMBINE:
+		flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+		domain = AMDGPU_GEM_DOMAIN_GTT;
+		place.fpfn = min_offset >> PAGE_SHIFT;
+		place.lpfn = max_offset >> PAGE_SHIFT;
+		place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT |
+			TTM_PL_FLAG_UNCACHED;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+
+	*handle = 0;
+
+	placement.placement = &place;
+	placement.num_placement = 1;
+	placement.busy_placement = &place;
+	placement.num_busy_placement = 1;
+
+	ret = amdgpu_bo_create_restricted(adev, size, PAGE_SIZE,
+					  true, domain, flags,
+					  NULL, &placement, &obj);
+	if (ret) {
+		DRM_ERROR("(%d) bo create failed\n", ret);
+		return ret;
+	}
+	*handle = (cgs_handle_t)obj;
+
+	return ret;
+}
+
+static int amdgpu_cgs_import_gpu_mem(void *cgs_device, int dmabuf_fd,
+				     cgs_handle_t *handle)
+{
+	CGS_FUNC_ADEV;
+	int r;
+	uint32_t dma_handle;
+	struct drm_gem_object *obj;
+	struct amdgpu_bo *bo;
+	struct drm_device *dev = adev->ddev;
+	struct drm_file *file_priv = NULL, *priv;
+
+	mutex_lock(&dev->struct_mutex);
+	list_for_each_entry(priv, &dev->filelist, lhead) {
+		rcu_read_lock();
+		if (priv->pid == get_pid(task_pid(current)))
+			file_priv = priv;
+		rcu_read_unlock();
+		if (file_priv)
+			break;
+	}
+	mutex_unlock(&dev->struct_mutex);
+	r = dev->driver->prime_fd_to_handle(dev,
+					    file_priv, dmabuf_fd,
+					    &dma_handle);
+	spin_lock(&file_priv->table_lock);
+
+	/* Check if we currently have a reference on the object */
+	obj = idr_find(&file_priv->object_idr, dma_handle);
+	if (obj == NULL) {
+		spin_unlock(&file_priv->table_lock);
+		return -EINVAL;
+	}
+	spin_unlock(&file_priv->table_lock);
+	bo = gem_to_amdgpu_bo(obj);
+	*handle = (cgs_handle_t)bo;
+	return 0;
+}
+
+static int amdgpu_cgs_free_gpu_mem(void *cgs_device, cgs_handle_t handle)
+{
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
+
+	if (obj) {
+		int r = amdgpu_bo_reserve(obj, false);
+		if (likely(r == 0)) {
+			amdgpu_bo_kunmap(obj);
+			amdgpu_bo_unpin(obj);
+			amdgpu_bo_unreserve(obj);
+		}
+		amdgpu_bo_unref(&obj);
+
+	}
+	return 0;
+}
+
+static int amdgpu_cgs_gmap_gpu_mem(void *cgs_device, cgs_handle_t handle,
+				   uint64_t *mcaddr)
+{
+	int r;
+	u64 min_offset, max_offset;
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
+
+	WARN_ON_ONCE(obj->placement.num_placement > 1);
+
+	min_offset = obj->placements[0].fpfn << PAGE_SHIFT;
+	max_offset = obj->placements[0].lpfn << PAGE_SHIFT;
+
+	r = amdgpu_bo_reserve(obj, false);
+	if (unlikely(r != 0))
+		return r;
+	r = amdgpu_bo_pin_restricted(obj, AMDGPU_GEM_DOMAIN_GTT,
+				     min_offset, max_offset, mcaddr);
+	amdgpu_bo_unreserve(obj);
+	return r;
+}
+
+static int amdgpu_cgs_gunmap_gpu_mem(void *cgs_device, cgs_handle_t handle)
+{
+	int r;
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
+	r = amdgpu_bo_reserve(obj, false);
+	if (unlikely(r != 0))
+		return r;
+	r = amdgpu_bo_unpin(obj);
+	amdgpu_bo_unreserve(obj);
+	return r;
+}
+
+static int amdgpu_cgs_kmap_gpu_mem(void *cgs_device, cgs_handle_t handle,
+				   void **map)
+{
+	int r;
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
+	r = amdgpu_bo_reserve(obj, false);
+	if (unlikely(r != 0))
+		return r;
+	r = amdgpu_bo_kmap(obj, map);
+	amdgpu_bo_unreserve(obj);
+	return r;
+}
+
+static int amdgpu_cgs_kunmap_gpu_mem(void *cgs_device, cgs_handle_t handle)
+{
+	int r;
+	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
+	r = amdgpu_bo_reserve(obj, false);
+	if (unlikely(r != 0))
+		return r;
+	amdgpu_bo_kunmap(obj);
+	amdgpu_bo_unreserve(obj);
+	return r;
+}
+
+static uint32_t amdgpu_cgs_read_register(void *cgs_device, unsigned offset)
+{
+	CGS_FUNC_ADEV;
+	return RREG32(offset);
+}
+
+static void amdgpu_cgs_write_register(void *cgs_device, unsigned offset,
+				      uint32_t value)
+{
+	CGS_FUNC_ADEV;
+	WREG32(offset, value);
+}
+
+static uint32_t amdgpu_cgs_read_ind_register(void *cgs_device,
+					     enum cgs_ind_reg space,
+					     unsigned index)
+{
+	CGS_FUNC_ADEV;
+	switch (space) {
+	case CGS_IND_REG__MMIO:
+		return RREG32_IDX(index);
+	case CGS_IND_REG__PCIE:
+		return RREG32_PCIE(index);
+	case CGS_IND_REG__SMC:
+		return RREG32_SMC(index);
+	case CGS_IND_REG__UVD_CTX:
+		return RREG32_UVD_CTX(index);
+	case CGS_IND_REG__DIDT:
+		return RREG32_DIDT(index);
+	case CGS_IND_REG__AUDIO_ENDPT:
+		DRM_ERROR("audio endpt register access not implemented.\n");
+		return 0;
+	}
+	WARN(1, "Invalid indirect register space");
+	return 0;
+}
+
+static void amdgpu_cgs_write_ind_register(void *cgs_device,
+					  enum cgs_ind_reg space,
+					  unsigned index, uint32_t value)
+{
+	CGS_FUNC_ADEV;
+	switch (space) {
+	case CGS_IND_REG__MMIO:
+		return WREG32_IDX(index, value);
+	case CGS_IND_REG__PCIE:
+		return WREG32_PCIE(index, value);
+	case CGS_IND_REG__SMC:
+		return WREG32_SMC(index, value);
+	case CGS_IND_REG__UVD_CTX:
+		return WREG32_UVD_CTX(index, value);
+	case CGS_IND_REG__DIDT:
+		return WREG32_DIDT(index, value);
+	case CGS_IND_REG__AUDIO_ENDPT:
+		DRM_ERROR("audio endpt register access not implemented.\n");
+		return;
+	}
+	WARN(1, "Invalid indirect register space");
+}
+
+static uint8_t amdgpu_cgs_read_pci_config_byte(void *cgs_device, unsigned addr)
+{
+	CGS_FUNC_ADEV;
+	uint8_t val;
+	int ret = pci_read_config_byte(adev->pdev, addr, &val);
+	if (WARN(ret, "pci_read_config_byte error"))
+		return 0;
+	return val;
+}
+
+static uint16_t amdgpu_cgs_read_pci_config_word(void *cgs_device, unsigned addr)
+{
+	CGS_FUNC_ADEV;
+	uint16_t val;
+	int ret = pci_read_config_word(adev->pdev, addr, &val);
+	if (WARN(ret, "pci_read_config_word error"))
+		return 0;
+	return val;
+}
+
+static uint32_t amdgpu_cgs_read_pci_config_dword(void *cgs_device,
+						 unsigned addr)
+{
+	CGS_FUNC_ADEV;
+	uint32_t val;
+	int ret = pci_read_config_dword(adev->pdev, addr, &val);
+	if (WARN(ret, "pci_read_config_dword error"))
+		return 0;
+	return val;
+}
+
+static void amdgpu_cgs_write_pci_config_byte(void *cgs_device, unsigned addr,
+					     uint8_t value)
+{
+	CGS_FUNC_ADEV;
+	int ret = pci_write_config_byte(adev->pdev, addr, value);
+	WARN(ret, "pci_write_config_byte error");
+}
+
+static void amdgpu_cgs_write_pci_config_word(void *cgs_device, unsigned addr,
+					     uint16_t value)
+{
+	CGS_FUNC_ADEV;
+	int ret = pci_write_config_word(adev->pdev, addr, value);
+	WARN(ret, "pci_write_config_word error");
+}
+
+static void amdgpu_cgs_write_pci_config_dword(void *cgs_device, unsigned addr,
+					      uint32_t value)
+{
+	CGS_FUNC_ADEV;
+	int ret = pci_write_config_dword(adev->pdev, addr, value);
+	WARN(ret, "pci_write_config_dword error");
+}
+
+static const void *amdgpu_cgs_atom_get_data_table(void *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(void *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(void *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_create_pm_request(void *cgs_device, cgs_handle_t *request)
+{
+	/* TODO */
+	return 0;
+}
+
+static int amdgpu_cgs_destroy_pm_request(void *cgs_device, cgs_handle_t request)
+{
+	/* TODO */
+	return 0;
+}
+
+static int amdgpu_cgs_set_pm_request(void *cgs_device, cgs_handle_t request,
+				     int active)
+{
+	/* TODO */
+	return 0;
+}
+
+static int amdgpu_cgs_pm_request_clock(void *cgs_device, cgs_handle_t request,
+				       enum cgs_clock clock, unsigned freq)
+{
+	/* TODO */
+	return 0;
+}
+
+static int amdgpu_cgs_pm_request_engine(void *cgs_device, cgs_handle_t request,
+					enum cgs_engine engine, int powered)
+{
+	/* TODO */
+	return 0;
+}
+
+
+
+static int amdgpu_cgs_pm_query_clock_limits(void *cgs_device,
+					    enum cgs_clock clock,
+					    struct cgs_clock_limits *limits)
+{
+	/* TODO */
+	return 0;
+}
+
+static int amdgpu_cgs_set_camera_voltages(void *cgs_device, uint32_t mask,
+					  const uint32_t *voltages)
+{
+	DRM_ERROR("not implemented");
+	return -EPERM;
+}
+
+struct cgs_irq_params {
+	unsigned src_id;
+	cgs_irq_source_set_func_t set;
+	cgs_irq_handler_func_t handler;
+	void *private_data;
+};
+
+static int cgs_set_irq_state(struct amdgpu_device *adev,
+			     struct amdgpu_irq_src *src,
+			     unsigned type,
+			     enum amdgpu_interrupt_state state)
+{
+	struct cgs_irq_params *irq_params =
+		(struct cgs_irq_params *)src->data;
+	if (!irq_params)
+		return -EINVAL;
+	if (!irq_params->set)
+		return -EINVAL;
+	return irq_params->set(irq_params->private_data,
+			       irq_params->src_id,
+			       type,
+			       (int)state);
+}
+
+static int cgs_process_irq(struct amdgpu_device *adev,
+			   struct amdgpu_irq_src *source,
+			   struct amdgpu_iv_entry *entry)
+{
+	struct cgs_irq_params *irq_params =
+		(struct cgs_irq_params *)source->data;
+	if (!irq_params)
+		return -EINVAL;
+	if (!irq_params->handler)
+		return -EINVAL;
+	return irq_params->handler(irq_params->private_data,
+				   irq_params->src_id,
+				   entry->iv_entry);
+}
+
+static const struct amdgpu_irq_src_funcs cgs_irq_funcs = {
+	.set = cgs_set_irq_state,
+	.process = cgs_process_irq,
+};
+
+static int amdgpu_cgs_add_irq_source(void *cgs_device, unsigned src_id,
+				     unsigned num_types,
+				     cgs_irq_source_set_func_t set,
+				     cgs_irq_handler_func_t handler,
+				     void *private_data)
+{
+	CGS_FUNC_ADEV;
+	int ret = 0;
+	struct cgs_irq_params *irq_params;
+	struct amdgpu_irq_src *source =
+		kzalloc(sizeof(struct amdgpu_irq_src), GFP_KERNEL);
+	if (!source)
+		return -ENOMEM;
+	irq_params =
+		kzalloc(sizeof(struct cgs_irq_params), GFP_KERNEL);
+	if (!irq_params) {
+		kfree(source);
+		return -ENOMEM;
+	}
+	source->num_types = num_types;
+	source->funcs = &cgs_irq_funcs;
+	irq_params->src_id = src_id;
+	irq_params->set = set;
+	irq_params->handler = handler;
+	irq_params->private_data = private_data;
+	source->data = (void *)irq_params;
+	ret = amdgpu_irq_add_id(adev, src_id, source);
+	if (ret) {
+		kfree(irq_params);
+		kfree(source);
+	}
+
+	return ret;
+}
+
+static int amdgpu_cgs_irq_get(void *cgs_device, unsigned src_id, unsigned type)
+{
+	CGS_FUNC_ADEV;
+	return amdgpu_irq_get(adev, adev->irq.sources[src_id], type);
+}
+
+static int amdgpu_cgs_irq_put(void *cgs_device, unsigned src_id, unsigned type)
+{
+	CGS_FUNC_ADEV;
+	return amdgpu_irq_put(adev, adev->irq.sources[src_id], type);
+}
+
+int amdgpu_cgs_set_clockgating_state(void *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_block_status[i].valid)
+			continue;
+
+		if (adev->ip_blocks[i].type == block_type) {
+			r = adev->ip_blocks[i].funcs->set_clockgating_state(
+								(void *)adev,
+									state);
+			break;
+		}
+	}
+	return r;
+}
+
+int amdgpu_cgs_set_powergating_state(void *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_block_status[i].valid)
+			continue;
+
+		if (adev->ip_blocks[i].type == block_type) {
+			r = adev->ip_blocks[i].funcs->set_powergating_state(
+								(void *)adev,
+									state);
+			break;
+		}
+	}
+	return r;
+}
+
+
+static uint32_t fw_type_convert(void *cgs_device, uint32_t fw_type)
+{
+	CGS_FUNC_ADEV;
+	enum AMDGPU_UCODE_ID result = AMDGPU_UCODE_ID_MAXIMUM;
+
+	switch (fw_type) {
+	case CGS_UCODE_ID_SDMA0:
+		result = AMDGPU_UCODE_ID_SDMA0;
+		break;
+	case CGS_UCODE_ID_SDMA1:
+		result = AMDGPU_UCODE_ID_SDMA1;
+		break;
+	case CGS_UCODE_ID_CP_CE:
+		result = AMDGPU_UCODE_ID_CP_CE;
+		break;
+	case CGS_UCODE_ID_CP_PFP:
+		result = AMDGPU_UCODE_ID_CP_PFP;
+		break;
+	case CGS_UCODE_ID_CP_ME:
+		result = AMDGPU_UCODE_ID_CP_ME;
+		break;
+	case CGS_UCODE_ID_CP_MEC:
+	case CGS_UCODE_ID_CP_MEC_JT1:
+		result = AMDGPU_UCODE_ID_CP_MEC1;
+		break;
+	case CGS_UCODE_ID_CP_MEC_JT2:
+		if (adev->asic_type == CHIP_TONGA)
+			result = AMDGPU_UCODE_ID_CP_MEC2;
+		else if (adev->asic_type == CHIP_CARRIZO)
+			result = AMDGPU_UCODE_ID_CP_MEC1;
+		break;
+	case CGS_UCODE_ID_RLC_G:
+		result = AMDGPU_UCODE_ID_RLC_G;
+		break;
+	default:
+		DRM_ERROR("Firmware type not supported\n");
+	}
+	return result;
+}
+
+static int amdgpu_cgs_get_firmware_info(void *cgs_device,
+					enum cgs_ucode_id type,
+					struct cgs_firmware_info *info)
+{
+	CGS_FUNC_ADEV;
+
+	if (CGS_UCODE_ID_SMU != type) {
+		uint64_t gpu_addr;
+		uint32_t data_size;
+		const struct gfx_firmware_header_v1_0 *header;
+		enum AMDGPU_UCODE_ID id;
+		struct amdgpu_firmware_info *ucode;
+
+		id = fw_type_convert(cgs_device, type);
+		ucode = &adev->firmware.ucode[id];
+		if (ucode->fw == NULL)
+			return -EINVAL;
+
+		gpu_addr  = ucode->mc_addr;
+		header = (const struct gfx_firmware_header_v1_0 *)ucode->fw->data;
+		data_size = le32_to_cpu(header->header.ucode_size_bytes);
+
+		if ((type == CGS_UCODE_ID_CP_MEC_JT1) ||
+		    (type == CGS_UCODE_ID_CP_MEC_JT2)) {
+			gpu_addr += le32_to_cpu(header->jt_offset) << 2;
+			data_size = le32_to_cpu(header->jt_size) << 2;
+		}
+		info->mc_addr = gpu_addr;
+		info->image_size = data_size;
+		info->version = (uint16_t)le32_to_cpu(header->header.ucode_version);
+		info->feature_version = (uint16_t)le32_to_cpu(header->ucode_feature_version);
+	} else {
+		char fw_name[30] = {0};
+		int err = 0;
+		uint32_t ucode_size;
+		uint32_t ucode_start_address;
+		const uint8_t *src;
+		const struct smc_firmware_header_v1_0 *hdr;
+
+		switch (adev->asic_type) {
+		case CHIP_TONGA:
+			strcpy(fw_name, "amdgpu/tonga_smc.bin");
+			break;
+		default:
+			DRM_ERROR("SMC firmware not supported\n");
+			return -EINVAL;
+		}
+
+		err = request_firmware(&adev->pm.fw, fw_name, adev->dev);
+		if (err) {
+			DRM_ERROR("Failed to request firmware\n");
+			return err;
+		}
+
+		err = amdgpu_ucode_validate(adev->pm.fw);
+		if (err) {
+			DRM_ERROR("Failed to load firmware \"%s\"", fw_name);
+			release_firmware(adev->pm.fw);
+			adev->pm.fw = NULL;
+			return err;
+		}
+
+		hdr = (const struct smc_firmware_header_v1_0 *)	adev->pm.fw->data;
+		adev->pm.fw_version = le32_to_cpu(hdr->header.ucode_version);
+		ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes);
+		ucode_start_address = le32_to_cpu(hdr->ucode_start_addr);
+		src = (const uint8_t *)(adev->pm.fw->data +
+		       le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+
+		info->version = adev->pm.fw_version;
+		info->image_size = ucode_size;
+		info->kptr = (void *)src;
+	}
+	return 0;
+}
+
+static const struct cgs_ops amdgpu_cgs_ops = {
+	amdgpu_cgs_gpu_mem_info,
+	amdgpu_cgs_gmap_kmem,
+	amdgpu_cgs_gunmap_kmem,
+	amdgpu_cgs_alloc_gpu_mem,
+	amdgpu_cgs_free_gpu_mem,
+	amdgpu_cgs_gmap_gpu_mem,
+	amdgpu_cgs_gunmap_gpu_mem,
+	amdgpu_cgs_kmap_gpu_mem,
+	amdgpu_cgs_kunmap_gpu_mem,
+	amdgpu_cgs_read_register,
+	amdgpu_cgs_write_register,
+	amdgpu_cgs_read_ind_register,
+	amdgpu_cgs_write_ind_register,
+	amdgpu_cgs_read_pci_config_byte,
+	amdgpu_cgs_read_pci_config_word,
+	amdgpu_cgs_read_pci_config_dword,
+	amdgpu_cgs_write_pci_config_byte,
+	amdgpu_cgs_write_pci_config_word,
+	amdgpu_cgs_write_pci_config_dword,
+	amdgpu_cgs_atom_get_data_table,
+	amdgpu_cgs_atom_get_cmd_table_revs,
+	amdgpu_cgs_atom_exec_cmd_table,
+	amdgpu_cgs_create_pm_request,
+	amdgpu_cgs_destroy_pm_request,
+	amdgpu_cgs_set_pm_request,
+	amdgpu_cgs_pm_request_clock,
+	amdgpu_cgs_pm_request_engine,
+	amdgpu_cgs_pm_query_clock_limits,
+	amdgpu_cgs_set_camera_voltages,
+	amdgpu_cgs_get_firmware_info,
+	amdgpu_cgs_set_powergating_state,
+	amdgpu_cgs_set_clockgating_state
+};
+
+static const struct cgs_os_ops amdgpu_cgs_os_ops = {
+	amdgpu_cgs_import_gpu_mem,
+	amdgpu_cgs_add_irq_source,
+	amdgpu_cgs_irq_get,
+	amdgpu_cgs_irq_put
+};
+
+void *amdgpu_cgs_create_device(struct amdgpu_device *adev)
+{
+	struct amdgpu_cgs_device *cgs_device =
+		kmalloc(sizeof(*cgs_device), GFP_KERNEL);
+
+	if (!cgs_device) {
+		DRM_ERROR("Couldn't allocate CGS device structure\n");
+		return NULL;
+	}
+
+	cgs_device->base.ops = &amdgpu_cgs_ops;
+	cgs_device->base.os_ops = &amdgpu_cgs_os_ops;
+	cgs_device->adev = adev;
+
+	return cgs_device;
+}
+
+void amdgpu_cgs_destroy_device(void *cgs_device)
+{
+	kfree(cgs_device);
+}

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

@@ -75,6 +75,11 @@ void amdgpu_connector_hotplug(struct drm_connector *connector)
 			if (!amdgpu_display_hpd_sense(adev, amdgpu_connector->hpd.hpd)) {
 				drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
 			} else if (amdgpu_atombios_dp_needs_link_train(amdgpu_connector)) {
+				/* Don't try to start link training before we
+				 * have the dpcd */
+				if (!amdgpu_atombios_dp_get_dpcd(amdgpu_connector))
+					return;
+
 				/* set it to OFF so that drm_helper_connector_dpms()
 				 * won't return immediately since the current state
 				 * is ON at this point.

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

@@ -126,6 +126,30 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 	return 0;
 }
 
+struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
+                                               struct drm_file *filp,
+                                               struct amdgpu_ctx *ctx,
+                                               struct amdgpu_ib *ibs,
+                                               uint32_t num_ibs)
+{
+	struct amdgpu_cs_parser *parser;
+	int i;
+
+	parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL);
+	if (!parser)
+		return NULL;
+
+	parser->adev = adev;
+	parser->filp = filp;
+	parser->ctx = ctx;
+	parser->ibs = ibs;
+	parser->num_ibs = num_ibs;
+	for (i = 0; i < num_ibs; i++)
+		ibs[i].ctx = ctx;
+
+	return parser;
+}
+
 int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 {
 	union drm_amdgpu_cs *cs = data;
@@ -147,13 +171,13 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 
 	/* get chunks */
 	INIT_LIST_HEAD(&p->validated);
-	chunk_array = kcalloc(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
+	chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
 	if (chunk_array == NULL) {
 		r = -ENOMEM;
 		goto out;
 	}
 
-	chunk_array_user = (uint64_t *)(unsigned long)(cs->in.chunks);
+	chunk_array_user = (uint64_t __user *)(cs->in.chunks);
 	if (copy_from_user(chunk_array, chunk_array_user,
 			   sizeof(uint64_t)*cs->in.num_chunks)) {
 		r = -EFAULT;
@@ -161,7 +185,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 	}
 
 	p->nchunks = cs->in.num_chunks;
-	p->chunks = kcalloc(p->nchunks, sizeof(struct amdgpu_cs_chunk),
+	p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
 			    GFP_KERNEL);
 	if (p->chunks == NULL) {
 		r = -ENOMEM;
@@ -173,7 +197,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		struct drm_amdgpu_cs_chunk user_chunk;
 		uint32_t __user *cdata;
 
-		chunk_ptr = (void __user *)(unsigned long)chunk_array[i];
+		chunk_ptr = (void __user *)chunk_array[i];
 		if (copy_from_user(&user_chunk, chunk_ptr,
 				       sizeof(struct drm_amdgpu_cs_chunk))) {
 			r = -EFAULT;
@@ -183,7 +207,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		p->chunks[i].length_dw = user_chunk.length_dw;
 
 		size = p->chunks[i].length_dw;
-		cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
+		cdata = (void __user *)user_chunk.chunk_data;
 		p->chunks[i].user_ptr = cdata;
 
 		p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
@@ -235,11 +259,10 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		}
 	}
 
+
 	p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!p->ibs) {
+	if (!p->ibs)
 		r = -ENOMEM;
-		goto out;
-	}
 
 out:
 	kfree(chunk_array);
@@ -331,7 +354,7 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p)
 			 * into account. We don't want to disallow buffer moves
 			 * completely.
 			 */
-			if (current_domain != AMDGPU_GEM_DOMAIN_CPU &&
+			if ((lobj->allowed_domains & current_domain) != 0 &&
 			    (domain & current_domain) == 0 && /* will be moved */
 			    bytes_moved > bytes_moved_threshold) {
 				/* don't move it */
@@ -415,18 +438,8 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a,
 	return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
 }
 
-/**
- * cs_parser_fini() - clean parser states
- * @parser:	parser structure holding parsing context.
- * @error:	error number
- *
- * If error is set than unvalidate buffer, otherwise just free memory
- * used by parsing context.
- **/
-static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
+static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff)
 {
-	unsigned i;
-
 	if (!error) {
 		/* Sort the buffer list from the smallest to largest buffer,
 		 * which affects the order of buffers in the LRU list.
@@ -447,21 +460,45 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 		ttm_eu_backoff_reservation(&parser->ticket,
 					   &parser->validated);
 	}
+}
 
+static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
+{
+	unsigned i;
 	if (parser->ctx)
 		amdgpu_ctx_put(parser->ctx);
 	if (parser->bo_list)
 		amdgpu_bo_list_put(parser->bo_list);
+
 	drm_free_large(parser->vm_bos);
 	for (i = 0; i < parser->nchunks; i++)
 		drm_free_large(parser->chunks[i].kdata);
 	kfree(parser->chunks);
-	if (parser->ibs)
-		for (i = 0; i < parser->num_ibs; i++)
-			amdgpu_ib_free(parser->adev, &parser->ibs[i]);
-	kfree(parser->ibs);
-	if (parser->uf.bo)
-		drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+	if (!amdgpu_enable_scheduler)
+	{
+		if (parser->ibs)
+			for (i = 0; i < parser->num_ibs; i++)
+				amdgpu_ib_free(parser->adev, &parser->ibs[i]);
+		kfree(parser->ibs);
+		if (parser->uf.bo)
+			drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+	}
+
+	kfree(parser);
+}
+
+/**
+ * cs_parser_fini() - clean parser states
+ * @parser:	parser structure holding parsing context.
+ * @error:	error number
+ *
+ * If error is set than unvalidate buffer, otherwise just free memory
+ * used by parsing context.
+ **/
+static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
+{
+       amdgpu_cs_parser_fini_early(parser, error, backoff);
+       amdgpu_cs_parser_fini_late(parser);
 }
 
 static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
@@ -476,12 +513,18 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 	if (r)
 		return r;
 
+	r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
+	if (r)
+		return r;
+
 	r = amdgpu_vm_clear_freed(adev, vm);
 	if (r)
 		return r;
 
 	if (p->bo_list) {
 		for (i = 0; i < p->bo_list->num_entries; i++) {
+			struct fence *f;
+
 			/* ignore duplicates */
 			bo = p->bo_list->array[i].robj;
 			if (!bo)
@@ -495,7 +538,10 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 			if (r)
 				return r;
 
-			amdgpu_sync_fence(&p->ibs[0].sync, bo_va->last_pt_update);
+			f = bo_va->last_pt_update;
+			r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
+			if (r)
+				return r;
 		}
 	}
 
@@ -529,9 +575,9 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
 		goto out;
 	}
 	amdgpu_cs_sync_rings(parser);
-
-	r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
-			       parser->filp);
+	if (!amdgpu_enable_scheduler)
+		r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
+				       parser->filp);
 
 out:
 	mutex_unlock(&vm->mutex);
@@ -650,7 +696,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 			ib->oa_size = amdgpu_bo_size(oa);
 		}
 	}
-
 	/* wrap the last IB with user fence */
 	if (parser->uf.bo) {
 		struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
@@ -693,9 +738,9 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 			sizeof(struct drm_amdgpu_cs_chunk_dep);
 
 		for (j = 0; j < num_deps; ++j) {
-			struct amdgpu_fence *fence;
 			struct amdgpu_ring *ring;
 			struct amdgpu_ctx *ctx;
+			struct fence *fence;
 
 			r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
 					       deps[j].ip_instance,
@@ -707,85 +752,137 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 			if (ctx == NULL)
 				return -EINVAL;
 
-			r = amdgpu_fence_recreate(ring, p->filp,
-						  deps[j].handle,
-						  &fence);
-			if (r) {
+			fence = amdgpu_ctx_get_fence(ctx, ring,
+						     deps[j].handle);
+			if (IS_ERR(fence)) {
+				r = PTR_ERR(fence);
 				amdgpu_ctx_put(ctx);
 				return r;
-			}
 
-			amdgpu_sync_fence(&ib->sync, fence);
-			amdgpu_fence_unref(&fence);
-			amdgpu_ctx_put(ctx);
+			} else if (fence) {
+				r = amdgpu_sync_fence(adev, &ib->sync, fence);
+				fence_put(fence);
+				amdgpu_ctx_put(ctx);
+				if (r)
+					return r;
+			}
 		}
 	}
 
 	return 0;
 }
 
+static int amdgpu_cs_free_job(struct amdgpu_job *sched_job)
+{
+	int i;
+	if (sched_job->ibs)
+		for (i = 0; i < sched_job->num_ibs; i++)
+			amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
+	kfree(sched_job->ibs);
+	if (sched_job->uf.bo)
+		drm_gem_object_unreference_unlocked(&sched_job->uf.bo->gem_base);
+	return 0;
+}
+
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
 	union drm_amdgpu_cs *cs = data;
-	struct amdgpu_cs_parser parser;
-	int r, i;
+	struct amdgpu_cs_parser *parser;
 	bool reserved_buffers = false;
+	int i, r;
 
 	down_read(&adev->exclusive_lock);
 	if (!adev->accel_working) {
 		up_read(&adev->exclusive_lock);
 		return -EBUSY;
 	}
-	/* initialize parser */
-	memset(&parser, 0, sizeof(struct amdgpu_cs_parser));
-	parser.filp = filp;
-	parser.adev = adev;
-	r = amdgpu_cs_parser_init(&parser, data);
+
+	parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
+	if (!parser)
+		return -ENOMEM;
+	r = amdgpu_cs_parser_init(parser, data);
 	if (r) {
 		DRM_ERROR("Failed to initialize parser !\n");
-		amdgpu_cs_parser_fini(&parser, r, false);
+		amdgpu_cs_parser_fini(parser, r, false);
 		up_read(&adev->exclusive_lock);
 		r = amdgpu_cs_handle_lockup(adev, r);
 		return r;
 	}
 
-	r = amdgpu_cs_parser_relocs(&parser);
-	if (r) {
-		if (r != -ERESTARTSYS) {
-			if (r == -ENOMEM)
-				DRM_ERROR("Not enough memory for command submission!\n");
-			else
-				DRM_ERROR("Failed to process the buffer list %d!\n", r);
-		}
+	r = amdgpu_cs_parser_relocs(parser);
+	if (r == -ENOMEM)
+		DRM_ERROR("Not enough memory for command submission!\n");
+	else if (r && r != -ERESTARTSYS)
+		DRM_ERROR("Failed to process the buffer list %d!\n", r);
+	else if (!r) {
+		reserved_buffers = true;
+		r = amdgpu_cs_ib_fill(adev, parser);
 	}
 
 	if (!r) {
-		reserved_buffers = true;
-		r = amdgpu_cs_ib_fill(adev, &parser);
+		r = amdgpu_cs_dependencies(adev, parser);
+		if (r)
+			DRM_ERROR("Failed in the dependencies handling %d!\n", r);
 	}
 
-	if (!r)
-		r = amdgpu_cs_dependencies(adev, &parser);
-
-	if (r) {
-		amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-		up_read(&adev->exclusive_lock);
-		r = amdgpu_cs_handle_lockup(adev, r);
-		return r;
-	}
+	if (r)
+		goto out;
 
-	for (i = 0; i < parser.num_ibs; i++)
-		trace_amdgpu_cs(&parser, i);
+	for (i = 0; i < parser->num_ibs; i++)
+		trace_amdgpu_cs(parser, i);
 
-	r = amdgpu_cs_ib_vm_chunk(adev, &parser);
-	if (r) {
+	r = amdgpu_cs_ib_vm_chunk(adev, parser);
+	if (r)
 		goto out;
+
+	if (amdgpu_enable_scheduler && parser->num_ibs) {
+		struct amdgpu_job *job;
+		struct amdgpu_ring * ring =  parser->ibs->ring;
+		job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+		if (!job)
+			return -ENOMEM;
+		job->base.sched = ring->scheduler;
+		job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
+		job->adev = parser->adev;
+		job->ibs = parser->ibs;
+		job->num_ibs = parser->num_ibs;
+		job->base.owner = parser->filp;
+		mutex_init(&job->job_lock);
+		if (job->ibs[job->num_ibs - 1].user) {
+			memcpy(&job->uf,  &parser->uf,
+			       sizeof(struct amdgpu_user_fence));
+			job->ibs[job->num_ibs - 1].user = &job->uf;
+		}
+
+		job->free_job = amdgpu_cs_free_job;
+		mutex_lock(&job->job_lock);
+		r = amd_sched_entity_push_job((struct amd_sched_job *)job);
+		if (r) {
+			mutex_unlock(&job->job_lock);
+			amdgpu_cs_free_job(job);
+			kfree(job);
+			goto out;
+		}
+		cs->out.handle =
+			amdgpu_ctx_add_fence(parser->ctx, ring,
+					     &job->base.s_fence->base);
+		parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle;
+
+		list_sort(NULL, &parser->validated, cmp_size_smaller_first);
+		ttm_eu_fence_buffer_objects(&parser->ticket,
+				&parser->validated,
+				&job->base.s_fence->base);
+
+		mutex_unlock(&job->job_lock);
+		amdgpu_cs_parser_fini_late(parser);
+		up_read(&adev->exclusive_lock);
+		return 0;
 	}
 
-	cs->out.handle = parser.ibs[parser.num_ibs - 1].fence->seq;
+	cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
 out:
-	amdgpu_cs_parser_fini(&parser, r, true);
+	amdgpu_cs_parser_fini(parser, r, reserved_buffers);
 	up_read(&adev->exclusive_lock);
 	r = amdgpu_cs_handle_lockup(adev, r);
 	return r;
@@ -806,30 +903,29 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
 	union drm_amdgpu_wait_cs *wait = data;
 	struct amdgpu_device *adev = dev->dev_private;
 	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
-	struct amdgpu_fence *fence = NULL;
 	struct amdgpu_ring *ring = NULL;
 	struct amdgpu_ctx *ctx;
+	struct fence *fence;
 	long r;
 
-	ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
-	if (ctx == NULL)
-		return -EINVAL;
-
 	r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
 			       wait->in.ring, &ring);
-	if (r) {
-		amdgpu_ctx_put(ctx);
+	if (r)
 		return r;
-	}
 
-	r = amdgpu_fence_recreate(ring, filp, wait->in.handle, &fence);
-	if (r) {
-		amdgpu_ctx_put(ctx);
-		return r;
-	}
+	ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
+	if (ctx == NULL)
+		return -EINVAL;
+
+	fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
+	if (IS_ERR(fence))
+		r = PTR_ERR(fence);
+	else if (fence) {
+		r = fence_wait_timeout(fence, true, timeout);
+		fence_put(fence);
+	} else
+		r = 1;
 
-	r = fence_wait_timeout(&fence->base, true, timeout);
-	amdgpu_fence_unref(&fence);
 	amdgpu_ctx_put(ctx);
 	if (r < 0)
 		return r;
@@ -864,7 +960,16 @@ amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
 		if (!reloc->bo_va)
 			continue;
 
-		list_for_each_entry(mapping, &reloc->bo_va->mappings, list) {
+		list_for_each_entry(mapping, &reloc->bo_va->valids, list) {
+			if (mapping->it.start > addr ||
+			    addr > mapping->it.last)
+				continue;
+
+			*bo = reloc->bo_va->bo;
+			return mapping;
+		}
+
+		list_for_each_entry(mapping, &reloc->bo_va->invalids, list) {
 			if (mapping->it.start > addr ||
 			    addr > mapping->it.last)
 				continue;

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

@@ -25,54 +25,107 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 
-static void amdgpu_ctx_do_release(struct kref *ref)
+int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel,
+		    struct amdgpu_ctx *ctx)
 {
-	struct amdgpu_ctx *ctx;
-	struct amdgpu_ctx_mgr *mgr;
+	unsigned i, j;
+	int r;
 
-	ctx = container_of(ref, struct amdgpu_ctx, refcount);
-	mgr = &ctx->fpriv->ctx_mgr;
+	memset(ctx, 0, sizeof(*ctx));
+	ctx->adev = adev;
+	kref_init(&ctx->refcount);
+	spin_lock_init(&ctx->ring_lock);
+	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
+		ctx->rings[i].sequence = 1;
 
-	idr_remove(&mgr->ctx_handles, ctx->id);
-	kfree(ctx);
+	if (amdgpu_enable_scheduler) {
+		/* create context entity for each ring */
+		for (i = 0; i < adev->num_rings; i++) {
+			struct amd_sched_rq *rq;
+			if (kernel)
+				rq = &adev->rings[i]->scheduler->kernel_rq;
+			else
+				rq = &adev->rings[i]->scheduler->sched_rq;
+			r = amd_sched_entity_init(adev->rings[i]->scheduler,
+						  &ctx->rings[i].entity,
+						  rq, amdgpu_sched_jobs);
+			if (r)
+				break;
+		}
+
+		if (i < adev->num_rings) {
+			for (j = 0; j < i; j++)
+				amd_sched_entity_fini(adev->rings[j]->scheduler,
+						      &ctx->rings[j].entity);
+			kfree(ctx);
+			return r;
+		}
+	}
+	return 0;
 }
 
-int amdgpu_ctx_alloc(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t *id, uint32_t flags)
+void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
+{
+	struct amdgpu_device *adev = ctx->adev;
+	unsigned i, j;
+
+	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
+		for (j = 0; j < AMDGPU_CTX_MAX_CS_PENDING; ++j)
+			fence_put(ctx->rings[i].fences[j]);
+
+	if (amdgpu_enable_scheduler) {
+		for (i = 0; i < adev->num_rings; i++)
+			amd_sched_entity_fini(adev->rings[i]->scheduler,
+					      &ctx->rings[i].entity);
+	}
+}
+
+static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
+			    struct amdgpu_fpriv *fpriv,
+			    uint32_t *id)
 {
-	int r;
-	struct amdgpu_ctx *ctx;
 	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
+	struct amdgpu_ctx *ctx;
+	int r;
 
 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 	if (!ctx)
 		return -ENOMEM;
 
 	mutex_lock(&mgr->lock);
-	r = idr_alloc(&mgr->ctx_handles, ctx, 0, 0, GFP_KERNEL);
+	r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL);
 	if (r < 0) {
 		mutex_unlock(&mgr->lock);
 		kfree(ctx);
 		return r;
 	}
 	*id = (uint32_t)r;
-
-	memset(ctx, 0, sizeof(*ctx));
-	ctx->id = *id;
-	ctx->fpriv = fpriv;
-	kref_init(&ctx->refcount);
+	r = amdgpu_ctx_init(adev, false, ctx);
 	mutex_unlock(&mgr->lock);
 
-	return 0;
+	return r;
 }
 
-int amdgpu_ctx_free(struct amdgpu_device *adev, struct amdgpu_fpriv *fpriv, uint32_t id)
+static void amdgpu_ctx_do_release(struct kref *ref)
 {
 	struct amdgpu_ctx *ctx;
+
+	ctx = container_of(ref, struct amdgpu_ctx, refcount);
+
+	amdgpu_ctx_fini(ctx);
+
+	kfree(ctx);
+}
+
+static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id)
+{
 	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
+	struct amdgpu_ctx *ctx;
 
 	mutex_lock(&mgr->lock);
 	ctx = idr_find(&mgr->ctx_handles, id);
 	if (ctx) {
+		idr_remove(&mgr->ctx_handles, id);
 		kref_put(&ctx->refcount, amdgpu_ctx_do_release);
 		mutex_unlock(&mgr->lock);
 		return 0;
@@ -86,9 +139,13 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 			    union drm_amdgpu_ctx_out *out)
 {
 	struct amdgpu_ctx *ctx;
-	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
+	struct amdgpu_ctx_mgr *mgr;
 	unsigned reset_counter;
 
+	if (!fpriv)
+		return -EINVAL;
+
+	mgr = &fpriv->ctx_mgr;
 	mutex_lock(&mgr->lock);
 	ctx = idr_find(&mgr->ctx_handles, id);
 	if (!ctx) {
@@ -97,8 +154,8 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 	}
 
 	/* TODO: these two are always zero */
-	out->state.flags = ctx->state.flags;
-	out->state.hangs = ctx->state.hangs;
+	out->state.flags = 0x0;
+	out->state.hangs = 0x0;
 
 	/* determine if a GPU reset has occured since the last call */
 	reset_counter = atomic_read(&adev->gpu_reset_counter);
@@ -113,28 +170,11 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 	return 0;
 }
 
-void amdgpu_ctx_fini(struct amdgpu_fpriv *fpriv)
-{
-	struct idr *idp;
-	struct amdgpu_ctx *ctx;
-	uint32_t id;
-	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
-	idp = &mgr->ctx_handles;
-
-	idr_for_each_entry(idp,ctx,id) {
-		if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
-			DRM_ERROR("ctx (id=%ul) is still alive\n",ctx->id);
-	}
-
-	mutex_destroy(&mgr->lock);
-}
-
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp)
 {
 	int r;
 	uint32_t id;
-	uint32_t flags;
 
 	union drm_amdgpu_ctx *args = data;
 	struct amdgpu_device *adev = dev->dev_private;
@@ -142,15 +182,14 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 
 	r = 0;
 	id = args->in.ctx_id;
-	flags = args->in.flags;
 
 	switch (args->in.op) {
 		case AMDGPU_CTX_OP_ALLOC_CTX:
-			r = amdgpu_ctx_alloc(adev, fpriv, &id, flags);
+			r = amdgpu_ctx_alloc(adev, fpriv, &id);
 			args->out.alloc.ctx_id = id;
 			break;
 		case AMDGPU_CTX_OP_FREE_CTX:
-			r = amdgpu_ctx_free(adev, fpriv, id);
+			r = amdgpu_ctx_free(fpriv, id);
 			break;
 		case AMDGPU_CTX_OP_QUERY_STATE:
 			r = amdgpu_ctx_query(adev, fpriv, id, &args->out);
@@ -165,7 +204,12 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id)
 {
 	struct amdgpu_ctx *ctx;
-	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
+	struct amdgpu_ctx_mgr *mgr;
+
+	if (!fpriv)
+		return NULL;
+
+	mgr = &fpriv->ctx_mgr;
 
 	mutex_lock(&mgr->lock);
 	ctx = idr_find(&mgr->ctx_handles, id);
@@ -177,17 +221,86 @@ struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id)
 
 int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
 {
-	struct amdgpu_fpriv *fpriv;
-	struct amdgpu_ctx_mgr *mgr;
-
 	if (ctx == NULL)
 		return -EINVAL;
 
-	fpriv = ctx->fpriv;
-	mgr = &fpriv->ctx_mgr;
-	mutex_lock(&mgr->lock);
 	kref_put(&ctx->refcount, amdgpu_ctx_do_release);
-	mutex_unlock(&mgr->lock);
-
 	return 0;
 }
+
+uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
+			      struct fence *fence)
+{
+	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
+	uint64_t seq = cring->sequence;
+	unsigned idx = 0;
+	struct fence *other = NULL;
+
+	idx = seq % AMDGPU_CTX_MAX_CS_PENDING;
+	other = cring->fences[idx];
+	if (other) {
+		signed long r;
+		r = fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT);
+		if (r < 0)
+			DRM_ERROR("Error (%ld) waiting for fence!\n", r);
+	}
+
+	fence_get(fence);
+
+	spin_lock(&ctx->ring_lock);
+	cring->fences[idx] = fence;
+	cring->sequence++;
+	spin_unlock(&ctx->ring_lock);
+
+	fence_put(other);
+
+	return seq;
+}
+
+struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
+				   struct amdgpu_ring *ring, uint64_t seq)
+{
+	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
+	struct fence *fence;
+
+	spin_lock(&ctx->ring_lock);
+
+	if (seq >= cring->sequence) {
+		spin_unlock(&ctx->ring_lock);
+		return ERR_PTR(-EINVAL);
+	}
+
+
+	if (seq + AMDGPU_CTX_MAX_CS_PENDING < cring->sequence) {
+		spin_unlock(&ctx->ring_lock);
+		return NULL;
+	}
+
+	fence = fence_get(cring->fences[seq % AMDGPU_CTX_MAX_CS_PENDING]);
+	spin_unlock(&ctx->ring_lock);
+
+	return fence;
+}
+
+void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
+{
+	mutex_init(&mgr->lock);
+	idr_init(&mgr->ctx_handles);
+}
+
+void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
+{
+	struct amdgpu_ctx *ctx;
+	struct idr *idp;
+	uint32_t id;
+
+	idp = &mgr->ctx_handles;
+
+	idr_for_each_entry(idp, ctx, id) {
+		if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
+			DRM_ERROR("ctx %p is still alive\n", ctx);
+	}
+
+	idr_destroy(&mgr->ctx_handles);
+	mutex_destroy(&mgr->lock);
+}

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

@@ -55,6 +55,7 @@ static const char *amdgpu_asic_name[] = {
 	"MULLINS",
 	"TOPAZ",
 	"TONGA",
+	"FIJI",
 	"CARRIZO",
 	"LAST",
 };
@@ -63,7 +64,7 @@ bool amdgpu_device_is_px(struct drm_device *dev)
 {
 	struct amdgpu_device *adev = dev->dev_private;
 
-	if (adev->flags & AMDGPU_IS_PX)
+	if (adev->flags & AMD_IS_PX)
 		return true;
 	return false;
 }
@@ -243,7 +244,8 @@ static int amdgpu_vram_scratch_init(struct amdgpu_device *adev)
 
 	if (adev->vram_scratch.robj == NULL) {
 		r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE,
-				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, 0,
+				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
+				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
 				     NULL, &adev->vram_scratch.robj);
 		if (r) {
 			return r;
@@ -1160,6 +1162,7 @@ static int amdgpu_early_init(struct amdgpu_device *adev)
 	switch (adev->asic_type) {
 	case CHIP_TOPAZ:
 	case CHIP_TONGA:
+	case CHIP_FIJI:
 	case CHIP_CARRIZO:
 		if (adev->asic_type == CHIP_CARRIZO)
 			adev->family = AMDGPU_FAMILY_CZ;
@@ -1377,7 +1380,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->ddev = ddev;
 	adev->pdev = pdev;
 	adev->flags = flags;
-	adev->asic_type = flags & AMDGPU_ASIC_MASK;
+	adev->asic_type = flags & AMD_ASIC_MASK;
 	adev->is_atom_bios = false;
 	adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
 	adev->mc.gtt_size = 512 * 1024 * 1024;
@@ -1523,6 +1526,11 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		return r;
 	}
 
+	r = amdgpu_ctx_init(adev, true, &adev->kernel_ctx);
+	if (r) {
+		dev_err(adev->dev, "failed to create kernel context (%d).\n", r);
+		return r;
+	}
 	r = amdgpu_ib_ring_tests(adev);
 	if (r)
 		DRM_ERROR("ib ring test failed (%d).\n", r);
@@ -1584,6 +1592,7 @@ void amdgpu_device_fini(struct amdgpu_device *adev)
 	adev->shutdown = true;
 	/* evict vram memory */
 	amdgpu_bo_evict_vram(adev);
+	amdgpu_ctx_fini(&adev->kernel_ctx);
 	amdgpu_ib_pool_fini(adev);
 	amdgpu_fence_driver_fini(adev);
 	amdgpu_fbdev_fini(adev);
@@ -1627,8 +1636,7 @@ int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
 	struct amdgpu_device *adev;
 	struct drm_crtc *crtc;
 	struct drm_connector *connector;
-	int i, r;
-	bool force_completion = false;
+	int r;
 
 	if (dev == NULL || dev->dev_private == NULL) {
 		return -ENODEV;
@@ -1667,21 +1675,7 @@ int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon)
 	/* evict vram memory */
 	amdgpu_bo_evict_vram(adev);
 
-	/* wait for gpu to finish processing current batch */
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring)
-			continue;
-
-		r = amdgpu_fence_wait_empty(ring);
-		if (r) {
-			/* delay GPU reset to resume */
-			force_completion = true;
-		}
-	}
-	if (force_completion) {
-		amdgpu_fence_driver_force_completion(adev);
-	}
+	amdgpu_fence_driver_suspend(adev);
 
 	r = amdgpu_suspend(adev);
 
@@ -1739,6 +1733,8 @@ int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon)
 
 	r = amdgpu_resume(adev);
 
+	amdgpu_fence_driver_resume(adev);
+
 	r = amdgpu_ib_ring_tests(adev);
 	if (r)
 		DRM_ERROR("ib ring test failed (%d).\n", r);

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

@@ -35,6 +35,36 @@
 #include <drm/drm_crtc_helper.h>
 #include <drm/drm_edid.h>
 
+static void amdgpu_flip_wait_fence(struct amdgpu_device *adev,
+				   struct fence **f)
+{
+	struct amdgpu_fence *fence;
+	long r;
+
+	if (*f == NULL)
+		return;
+
+	fence = to_amdgpu_fence(*f);
+	if (fence) {
+		r = fence_wait(&fence->base, false);
+		if (r == -EDEADLK) {
+			up_read(&adev->exclusive_lock);
+			r = amdgpu_gpu_reset(adev);
+			down_read(&adev->exclusive_lock);
+		}
+	} else
+		r = fence_wait(*f, false);
+
+	if (r)
+		DRM_ERROR("failed to wait on page flip fence (%ld)!\n", r);
+
+	/* We continue with the page flip even if we failed to wait on
+	 * the fence, otherwise the DRM core and userspace will be
+	 * confused about which BO the CRTC is scanning out
+	 */
+	fence_put(*f);
+	*f = NULL;
+}
 
 static void amdgpu_flip_work_func(struct work_struct *__work)
 {
@@ -44,34 +74,13 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
 	struct amdgpu_crtc *amdgpuCrtc = adev->mode_info.crtcs[work->crtc_id];
 
 	struct drm_crtc *crtc = &amdgpuCrtc->base;
-	struct amdgpu_fence *fence;
 	unsigned long flags;
-	int r;
+	unsigned i;
 
 	down_read(&adev->exclusive_lock);
-	if (work->fence) {
-		fence = to_amdgpu_fence(work->fence);
-		if (fence) {
-			r = amdgpu_fence_wait(fence, false);
-			if (r == -EDEADLK) {
-				up_read(&adev->exclusive_lock);
-				r = amdgpu_gpu_reset(adev);
-				down_read(&adev->exclusive_lock);
-			}
-		} else
-			r = fence_wait(work->fence, false);
-
-		if (r)
-			DRM_ERROR("failed to wait on page flip fence (%d)!\n", r);
-
-		/* We continue with the page flip even if we failed to wait on
-		 * the fence, otherwise the DRM core and userspace will be
-		 * confused about which BO the CRTC is scanning out
-		 */
-
-		fence_put(work->fence);
-		work->fence = NULL;
-	}
+	amdgpu_flip_wait_fence(adev, &work->excl);
+	for (i = 0; i < work->shared_count; ++i)
+		amdgpu_flip_wait_fence(adev, &work->shared[i]);
 
 	/* We borrow the event spin lock for protecting flip_status */
 	spin_lock_irqsave(&crtc->dev->event_lock, flags);
@@ -108,6 +117,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work)
 		DRM_ERROR("failed to reserve buffer after flip\n");
 
 	drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
+	kfree(work->shared);
 	kfree(work);
 }
 
@@ -127,7 +137,7 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc,
 	unsigned long flags;
 	u64 tiling_flags;
 	u64 base;
-	int r;
+	int i, r;
 
 	work = kzalloc(sizeof *work, GFP_KERNEL);
 	if (work == NULL)
@@ -167,7 +177,19 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc,
 		goto cleanup;
 	}
 
-	work->fence = fence_get(reservation_object_get_excl(new_rbo->tbo.resv));
+	r = reservation_object_get_fences_rcu(new_rbo->tbo.resv, &work->excl,
+					      &work->shared_count,
+					      &work->shared);
+	if (unlikely(r != 0)) {
+		amdgpu_bo_unreserve(new_rbo);
+		DRM_ERROR("failed to get fences for buffer\n");
+		goto cleanup;
+	}
+
+	fence_get(work->excl);
+	for (i = 0; i < work->shared_count; ++i)
+		fence_get(work->shared[i]);
+
 	amdgpu_bo_get_tiling_flags(new_rbo, &tiling_flags);
 	amdgpu_bo_unreserve(new_rbo);
 
@@ -212,7 +234,10 @@ pflip_cleanup:
 
 cleanup:
 	drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base);
-	fence_put(work->fence);
+	fence_put(work->excl);
+	for (i = 0; i < work->shared_count; ++i)
+		fence_put(work->shared[i]);
+	kfree(work->shared);
 	kfree(work);
 
 	return r;

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

@@ -44,12 +44,15 @@
 #include "amdgpu.h"
 #include "amdgpu_irq.h"
 
+#include "amdgpu_amdkfd.h"
+
 /*
  * KMS wrapper.
  * - 3.0.0 - initial driver
+ * - 3.1.0 - allow reading more status registers (GRBM, SRBM, SDMA, CP)
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	0
+#define KMS_DRIVER_MINOR	1
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit = 0;
@@ -61,7 +64,7 @@ int amdgpu_disp_priority = 0;
 int amdgpu_hw_i2c = 0;
 int amdgpu_pcie_gen2 = -1;
 int amdgpu_msi = -1;
-int amdgpu_lockup_timeout = 10000;
+int amdgpu_lockup_timeout = 0;
 int amdgpu_dpm = -1;
 int amdgpu_smc_load_fw = 1;
 int amdgpu_aspm = -1;
@@ -73,6 +76,9 @@ int amdgpu_deep_color = 0;
 int amdgpu_vm_size = 8;
 int amdgpu_vm_block_size = -1;
 int amdgpu_exp_hw_support = 0;
+int amdgpu_enable_scheduler = 0;
+int amdgpu_sched_jobs = 16;
+int amdgpu_sched_hw_submission = 2;
 
 MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
 module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);
@@ -101,7 +107,7 @@ module_param_named(pcie_gen2, amdgpu_pcie_gen2, int, 0444);
 MODULE_PARM_DESC(msi, "MSI support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(msi, amdgpu_msi, int, 0444);
 
-MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (defaul 10000 = 10 seconds, 0 = disable)");
+MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (default 0 = disable)");
 module_param_named(lockup_timeout, amdgpu_lockup_timeout, int, 0444);
 
 MODULE_PARM_DESC(dpm, "DPM support (1 = enable, 0 = disable, -1 = auto)");
@@ -137,36 +143,45 @@ module_param_named(vm_block_size, amdgpu_vm_block_size, int, 0444);
 MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
 module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
 
+MODULE_PARM_DESC(enable_scheduler, "enable SW GPU scheduler (1 = enable, 0 = disable ((default))");
+module_param_named(enable_scheduler, amdgpu_enable_scheduler, int, 0444);
+
+MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 16)");
+module_param_named(sched_jobs, amdgpu_sched_jobs, int, 0444);
+
+MODULE_PARM_DESC(sched_hw_submission, "the max number of HW submissions (default 2)");
+module_param_named(sched_hw_submission, amdgpu_sched_hw_submission, int, 0444);
+
 static struct pci_device_id pciidlist[] = {
 #ifdef CONFIG_DRM_AMDGPU_CIK
 	/* Kaveri */
-	{0x1002, 0x1304, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x1305, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1306, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x1307, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1309, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x130F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1310, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1311, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1312, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1313, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1315, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1316, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x1317, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x1318, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x131B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x131C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
-	{0x1002, 0x131D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMDGPU_IS_APU},
+	{0x1002, 0x1304, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x1305, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1306, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x1307, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1309, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x130F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1310, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1311, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1312, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1313, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1315, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1316, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x1317, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x1318, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x131B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x131C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
+	{0x1002, 0x131D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
 	/* Bonaire */
-	{0x1002, 0x6640, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
-	{0x1002, 0x6641, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
-	{0x1002, 0x6646, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
-	{0x1002, 0x6647, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMDGPU_IS_MOBILITY},
+	{0x1002, 0x6640, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
+	{0x1002, 0x6641, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
+	{0x1002, 0x6646, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
+	{0x1002, 0x6647, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE|AMD_IS_MOBILITY},
 	{0x1002, 0x6649, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
 	{0x1002, 0x6650, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
 	{0x1002, 0x6651, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BONAIRE},
@@ -188,39 +203,39 @@ static struct pci_device_id pciidlist[] = {
 	{0x1002, 0x67BA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAWAII},
 	{0x1002, 0x67BE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAWAII},
 	/* Kabini */
-	{0x1002, 0x9830, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9831, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x9832, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9833, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x9834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x9836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9837, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x9838, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9839, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x983a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x983b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x983c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x983d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x983e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
-	{0x1002, 0x983f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMDGPU_IS_APU},
+	{0x1002, 0x9830, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9831, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x9832, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9833, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x9834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x9836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9837, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x9838, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9839, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x983a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x983b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x983c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x983d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x983e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
+	{0x1002, 0x983f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KABINI|AMD_IS_APU},
 	/* mullins */
-	{0x1002, 0x9850, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9851, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9852, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9853, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9854, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9855, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9856, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9857, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9858, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x9859, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
-	{0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMDGPU_IS_MOBILITY|AMDGPU_IS_APU},
+	{0x1002, 0x9850, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9851, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9852, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9853, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9854, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9855, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9856, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9857, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9858, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x9859, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
+	{0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
 #endif
 	/* topaz */
 	{0x1002, 0x6900, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ},
@@ -238,12 +253,14 @@ static struct pci_device_id pciidlist[] = {
 	{0x1002, 0x6930, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
 	{0x1002, 0x6938, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
 	{0x1002, 0x6939, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA},
+	/* fiji */
+	{0x1002, 0x7300, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_FIJI},
 	/* carrizo */
-	{0x1002, 0x9870, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
-	{0x1002, 0x9874, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
-	{0x1002, 0x9875, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
-	{0x1002, 0x9876, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
-	{0x1002, 0x9877, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMDGPU_IS_APU},
+	{0x1002, 0x9870, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
+	{0x1002, 0x9874, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
+	{0x1002, 0x9875, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
+	{0x1002, 0x9876, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
+	{0x1002, 0x9877, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU},
 
 	{0, 0, 0}
 };
@@ -279,7 +296,7 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
 	unsigned long flags = ent->driver_data;
 	int ret;
 
-	if ((flags & AMDGPU_EXP_HW_SUPPORT) && !amdgpu_exp_hw_support) {
+	if ((flags & AMD_EXP_HW_SUPPORT) && !amdgpu_exp_hw_support) {
 		DRM_INFO("This hardware requires experimental hardware support.\n"
 			 "See modparam exp_hw_support\n");
 		return -ENODEV;
@@ -527,12 +544,15 @@ static int __init amdgpu_init(void)
 	driver->num_ioctls = amdgpu_max_kms_ioctl;
 	amdgpu_register_atpx_handler();
 
+	amdgpu_amdkfd_init();
+
 	/* let modprobe override vga console setting */
 	return drm_pci_init(driver, pdriver);
 }
 
 static void __exit amdgpu_exit(void)
 {
+	amdgpu_amdkfd_fini();
 	drm_pci_exit(driver, pdriver);
 	amdgpu_unregister_atpx_handler();
 }

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

@@ -31,7 +31,7 @@
 #include <linux/firmware.h>
 #include <linux/platform_device.h>
 
-#include "amdgpu_family.h"
+#include "amd_shared.h"
 
 /* General customization:
  */

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

@@ -53,9 +53,9 @@ static struct fb_ops amdgpufb_ops = {
 	.owner = THIS_MODULE,
 	.fb_check_var = drm_fb_helper_check_var,
 	.fb_set_par = drm_fb_helper_set_par,
-	.fb_fillrect = cfb_fillrect,
-	.fb_copyarea = cfb_copyarea,
-	.fb_imageblit = cfb_imageblit,
+	.fb_fillrect = drm_fb_helper_cfb_fillrect,
+	.fb_copyarea = drm_fb_helper_cfb_copyarea,
+	.fb_imageblit = drm_fb_helper_cfb_imageblit,
 	.fb_pan_display = drm_fb_helper_pan_display,
 	.fb_blank = drm_fb_helper_blank,
 	.fb_setcmap = drm_fb_helper_setcmap,
@@ -126,8 +126,8 @@ static int amdgpufb_create_pinned_object(struct amdgpu_fbdev *rfbdev,
 	aligned_size = ALIGN(size, PAGE_SIZE);
 	ret = amdgpu_gem_object_create(adev, aligned_size, 0,
 				       AMDGPU_GEM_DOMAIN_VRAM,
-				       0, true,
-				       &gobj);
+				       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+				       true, &gobj);
 	if (ret) {
 		printk(KERN_ERR "failed to allocate framebuffer (%d)\n",
 		       aligned_size);
@@ -179,7 +179,6 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 	struct drm_mode_fb_cmd2 mode_cmd;
 	struct drm_gem_object *gobj = NULL;
 	struct amdgpu_bo *rbo = NULL;
-	struct device *device = &adev->pdev->dev;
 	int ret;
 	unsigned long tmp;
 
@@ -201,9 +200,9 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 	rbo = gem_to_amdgpu_bo(gobj);
 
 	/* okay we have an object now allocate the framebuffer */
-	info = framebuffer_alloc(0, device);
-	if (info == NULL) {
-		ret = -ENOMEM;
+	info = drm_fb_helper_alloc_fbi(helper);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
 		goto out_unref;
 	}
 
@@ -212,14 +211,13 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 	ret = amdgpu_framebuffer_init(adev->ddev, &rfbdev->rfb, &mode_cmd, gobj);
 	if (ret) {
 		DRM_ERROR("failed to initialize framebuffer %d\n", ret);
-		goto out_unref;
+		goto out_destroy_fbi;
 	}
 
 	fb = &rfbdev->rfb.base;
 
 	/* setup helper */
 	rfbdev->helper.fb = fb;
-	rfbdev->helper.fbdev = info;
 
 	memset_io(rbo->kptr, 0x0, amdgpu_bo_size(rbo));
 
@@ -239,11 +237,6 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 	drm_fb_helper_fill_var(info, &rfbdev->helper, sizes->fb_width, sizes->fb_height);
 
 	/* setup aperture base/size for vesafb takeover */
-	info->apertures = alloc_apertures(1);
-	if (!info->apertures) {
-		ret = -ENOMEM;
-		goto out_unref;
-	}
 	info->apertures->ranges[0].base = adev->ddev->mode_config.fb_base;
 	info->apertures->ranges[0].size = adev->mc.aper_size;
 
@@ -251,13 +244,7 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 
 	if (info->screen_base == NULL) {
 		ret = -ENOSPC;
-		goto out_unref;
-	}
-
-	ret = fb_alloc_cmap(&info->cmap, 256, 0);
-	if (ret) {
-		ret = -ENOMEM;
-		goto out_unref;
+		goto out_destroy_fbi;
 	}
 
 	DRM_INFO("fb mappable at 0x%lX\n",  info->fix.smem_start);
@@ -269,6 +256,8 @@ static int amdgpufb_create(struct drm_fb_helper *helper,
 	vga_switcheroo_client_fb_set(adev->ddev->pdev, info);
 	return 0;
 
+out_destroy_fbi:
+	drm_fb_helper_release_fbi(helper);
 out_unref:
 	if (rbo) {
 
@@ -290,17 +279,10 @@ void amdgpu_fb_output_poll_changed(struct amdgpu_device *adev)
 
 static int amdgpu_fbdev_destroy(struct drm_device *dev, struct amdgpu_fbdev *rfbdev)
 {
-	struct fb_info *info;
 	struct amdgpu_framebuffer *rfb = &rfbdev->rfb;
 
-	if (rfbdev->helper.fbdev) {
-		info = rfbdev->helper.fbdev;
-
-		unregister_framebuffer(info);
-		if (info->cmap.len)
-			fb_dealloc_cmap(&info->cmap);
-		framebuffer_release(info);
-	}
+	drm_fb_helper_unregister_fbi(&rfbdev->helper);
+	drm_fb_helper_release_fbi(&rfbdev->helper);
 
 	if (rfb->obj) {
 		amdgpufb_destroy_pinned_object(rfb->obj);
@@ -395,7 +377,8 @@ void amdgpu_fbdev_fini(struct amdgpu_device *adev)
 void amdgpu_fbdev_set_suspend(struct amdgpu_device *adev, int state)
 {
 	if (adev->mode_info.rfbdev)
-		fb_set_suspend(adev->mode_info.rfbdev->helper.fbdev, state);
+		drm_fb_helper_set_suspend(&adev->mode_info.rfbdev->helper,
+			state);
 }
 
 int amdgpu_fbdev_total_size(struct amdgpu_device *adev)

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

@@ -126,7 +126,8 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
 	(*fence)->ring = ring;
 	(*fence)->owner = owner;
 	fence_init(&(*fence)->base, &amdgpu_fence_ops,
-		&adev->fence_queue.lock, adev->fence_context + ring->idx,
+		&ring->fence_drv.fence_queue.lock,
+		adev->fence_context + ring->idx,
 		(*fence)->seq);
 	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
 			       (*fence)->seq,
@@ -135,38 +136,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
 	return 0;
 }
 
-/**
- * amdgpu_fence_recreate - recreate a fence from an user fence
- *
- * @ring: ring the fence is associated with
- * @owner: creator of the fence
- * @seq: user fence sequence number
- * @fence: resulting amdgpu fence object
- *
- * Recreates a fence command from the user fence sequence number (all asics).
- * Returns 0 on success, -ENOMEM on failure.
- */
-int amdgpu_fence_recreate(struct amdgpu_ring *ring, void *owner,
-			  uint64_t seq, struct amdgpu_fence **fence)
-{
-	struct amdgpu_device *adev = ring->adev;
-
-	if (seq > ring->fence_drv.sync_seq[ring->idx])
-		return -EINVAL;
-
-	*fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
-	if ((*fence) == NULL)
-		return -ENOMEM;
-
-	(*fence)->seq = seq;
-	(*fence)->ring = ring;
-	(*fence)->owner = owner;
-	fence_init(&(*fence)->base, &amdgpu_fence_ops,
-		&adev->fence_queue.lock, adev->fence_context + ring->idx,
-		(*fence)->seq);
-	return 0;
-}
-
 /**
  * amdgpu_fence_check_signaled - callback from fence_queue
  *
@@ -196,9 +165,7 @@ static int amdgpu_fence_check_signaled(wait_queue_t *wait, unsigned mode, int fl
 		else
 			FENCE_TRACE(&fence->base, "was already signaled\n");
 
-		amdgpu_irq_put(adev, fence->ring->fence_drv.irq_src,
-				fence->ring->fence_drv.irq_type);
-		__remove_wait_queue(&adev->fence_queue, &fence->fence_wake);
+		__remove_wait_queue(&fence->ring->fence_drv.fence_queue, &fence->fence_wake);
 		fence_put(&fence->base);
 	} else
 		FENCE_TRACE(&fence->base, "pending\n");
@@ -299,14 +266,9 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
 		return;
 	}
 
-	if (fence_drv->delayed_irq && ring->adev->ddev->irq_enabled) {
-		fence_drv->delayed_irq = false;
-		amdgpu_irq_update(ring->adev, fence_drv->irq_src,
-				fence_drv->irq_type);
+	if (amdgpu_fence_activity(ring)) {
+		wake_up_all(&ring->fence_drv.fence_queue);
 	}
-
-	if (amdgpu_fence_activity(ring))
-		wake_up_all(&ring->adev->fence_queue);
 	else if (amdgpu_ring_is_lockup(ring)) {
 		/* good news we believe it's a lockup */
 		dev_warn(ring->adev->dev, "GPU lockup (current fence id "
@@ -316,7 +278,7 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
 
 		/* remember that we need an reset */
 		ring->adev->needs_reset = true;
-		wake_up_all(&ring->adev->fence_queue);
+		wake_up_all(&ring->fence_drv.fence_queue);
 	}
 	up_read(&ring->adev->exclusive_lock);
 }
@@ -332,62 +294,8 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
  */
 void amdgpu_fence_process(struct amdgpu_ring *ring)
 {
-	uint64_t seq, last_seq, last_emitted;
-	unsigned count_loop = 0;
-	bool wake = false;
-
-	/* Note there is a scenario here for an infinite loop but it's
-	 * very unlikely to happen. For it to happen, the current polling
-	 * process need to be interrupted by another process and another
-	 * process needs to update the last_seq btw the atomic read and
-	 * xchg of the current process.
-	 *
-	 * More over for this to go in infinite loop there need to be
-	 * continuously new fence signaled ie amdgpu_fence_read needs
-	 * to return a different value each time for both the currently
-	 * polling process and the other process that xchg the last_seq
-	 * btw atomic read and xchg of the current process. And the
-	 * value the other process set as last seq must be higher than
-	 * the seq value we just read. Which means that current process
-	 * need to be interrupted after amdgpu_fence_read and before
-	 * atomic xchg.
-	 *
-	 * To be even more safe we count the number of time we loop and
-	 * we bail after 10 loop just accepting the fact that we might
-	 * have temporarly set the last_seq not to the true real last
-	 * seq but to an older one.
-	 */
-	last_seq = atomic64_read(&ring->fence_drv.last_seq);
-	do {
-		last_emitted = ring->fence_drv.sync_seq[ring->idx];
-		seq = amdgpu_fence_read(ring);
-		seq |= last_seq & 0xffffffff00000000LL;
-		if (seq < last_seq) {
-			seq &= 0xffffffff;
-			seq |= last_emitted & 0xffffffff00000000LL;
-		}
-
-		if (seq <= last_seq || seq > last_emitted) {
-			break;
-		}
-		/* If we loop over we don't want to return without
-		 * checking if a fence is signaled as it means that the
-		 * seq we just read is different from the previous on.
-		 */
-		wake = true;
-		last_seq = seq;
-		if ((count_loop++) > 10) {
-			/* We looped over too many time leave with the
-			 * fact that we might have set an older fence
-			 * seq then the current real last seq as signaled
-			 * by the hw.
-			 */
-			break;
-		}
-	} while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq);
-
-	if (wake)
-		wake_up_all(&ring->adev->fence_queue);
+	if (amdgpu_fence_activity(ring))
+		wake_up_all(&ring->fence_drv.fence_queue);
 }
 
 /**
@@ -447,284 +355,49 @@ static bool amdgpu_fence_enable_signaling(struct fence *f)
 {
 	struct amdgpu_fence *fence = to_amdgpu_fence(f);
 	struct amdgpu_ring *ring = fence->ring;
-	struct amdgpu_device *adev = ring->adev;
 
 	if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
 		return false;
 
-	if (down_read_trylock(&adev->exclusive_lock)) {
-		amdgpu_irq_get(adev, ring->fence_drv.irq_src,
-			ring->fence_drv.irq_type);
-		if (amdgpu_fence_activity(ring))
-			wake_up_all_locked(&adev->fence_queue);
-
-		/* did fence get signaled after we enabled the sw irq? */
-		if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) {
-			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
-				ring->fence_drv.irq_type);
-			up_read(&adev->exclusive_lock);
-			return false;
-		}
-
-		up_read(&adev->exclusive_lock);
-	} else {
-		/* we're probably in a lockup, lets not fiddle too much */
-		if (amdgpu_irq_get_delayed(adev, ring->fence_drv.irq_src,
-			ring->fence_drv.irq_type))
-			ring->fence_drv.delayed_irq = true;
-		amdgpu_fence_schedule_check(ring);
-	}
-
 	fence->fence_wake.flags = 0;
 	fence->fence_wake.private = NULL;
 	fence->fence_wake.func = amdgpu_fence_check_signaled;
-	__add_wait_queue(&adev->fence_queue, &fence->fence_wake);
+	__add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake);
 	fence_get(f);
 	FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
 	return true;
 }
 
-/**
- * amdgpu_fence_signaled - check if a fence has signaled
- *
- * @fence: amdgpu fence object
- *
- * Check if the requested fence has signaled (all asics).
- * Returns true if the fence has signaled or false if it has not.
- */
-bool amdgpu_fence_signaled(struct amdgpu_fence *fence)
-{
-	if (!fence)
-		return true;
-
-	if (amdgpu_fence_seq_signaled(fence->ring, fence->seq)) {
-		if (!fence_signal(&fence->base))
-			FENCE_TRACE(&fence->base, "signaled from amdgpu_fence_signaled\n");
-		return true;
-	}
-
-	return false;
-}
-
-/**
- * amdgpu_fence_any_seq_signaled - check if any sequence number is signaled
- *
- * @adev: amdgpu device pointer
- * @seq: sequence numbers
- *
- * Check if the last signaled fence sequnce number is >= the requested
- * sequence number (all asics).
- * Returns true if any has signaled (current value is >= requested value)
- * or false if it has not. Helper function for amdgpu_fence_wait_seq.
- */
-static bool amdgpu_fence_any_seq_signaled(struct amdgpu_device *adev, u64 *seq)
-{
-	unsigned i;
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		if (!adev->rings[i] || !seq[i])
-			continue;
-
-		if (amdgpu_fence_seq_signaled(adev->rings[i], seq[i]))
-			return true;
-	}
-
-	return false;
-}
-
-/**
- * amdgpu_fence_wait_seq_timeout - wait for a specific sequence numbers
- *
- * @adev: amdgpu device pointer
- * @target_seq: sequence number(s) we want to wait for
- * @intr: use interruptable sleep
- * @timeout: maximum time to wait, or MAX_SCHEDULE_TIMEOUT for infinite wait
- *
- * Wait for the requested sequence number(s) to be written by any ring
- * (all asics).  Sequnce number array is indexed by ring id.
- * @intr selects whether to use interruptable (true) or non-interruptable
- * (false) sleep when waiting for the sequence number.  Helper function
- * for amdgpu_fence_wait_*().
- * Returns remaining time if the sequence number has passed, 0 when
- * the wait timeout, or an error for all other cases.
- * -EDEADLK is returned when a GPU lockup has been detected.
- */
-static long amdgpu_fence_wait_seq_timeout(struct amdgpu_device *adev,
-					  u64 *target_seq, bool intr,
-					  long timeout)
-{
-	uint64_t last_seq[AMDGPU_MAX_RINGS];
-	bool signaled;
-	int i;
-	long r;
-
-	if (timeout == 0) {
-		return amdgpu_fence_any_seq_signaled(adev, target_seq);
-	}
-
-	while (!amdgpu_fence_any_seq_signaled(adev, target_seq)) {
-
-		/* Save current sequence values, used to check for GPU lockups */
-		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			struct amdgpu_ring *ring = adev->rings[i];
-
-			if (!ring || !target_seq[i])
-				continue;
-
-			last_seq[i] = atomic64_read(&ring->fence_drv.last_seq);
-			trace_amdgpu_fence_wait_begin(adev->ddev, i, target_seq[i]);
-			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
-				       ring->fence_drv.irq_type);
-		}
-
-		if (intr) {
-			r = wait_event_interruptible_timeout(adev->fence_queue, (
-				(signaled = amdgpu_fence_any_seq_signaled(adev, target_seq))
-				 || adev->needs_reset), AMDGPU_FENCE_JIFFIES_TIMEOUT);
-		} else {
-			r = wait_event_timeout(adev->fence_queue, (
-				(signaled = amdgpu_fence_any_seq_signaled(adev, target_seq))
-				 || adev->needs_reset), AMDGPU_FENCE_JIFFIES_TIMEOUT);
-		}
-
-		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			struct amdgpu_ring *ring = adev->rings[i];
-
-			if (!ring || !target_seq[i])
-				continue;
-
-			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
-				       ring->fence_drv.irq_type);
-			trace_amdgpu_fence_wait_end(adev->ddev, i, target_seq[i]);
-		}
-
-		if (unlikely(r < 0))
-			return r;
-
-		if (unlikely(!signaled)) {
-
-			if (adev->needs_reset)
-				return -EDEADLK;
-
-			/* we were interrupted for some reason and fence
-			 * isn't signaled yet, resume waiting */
-			if (r)
-				continue;
-
-			for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-				struct amdgpu_ring *ring = adev->rings[i];
-
-				if (!ring || !target_seq[i])
-					continue;
-
-				if (last_seq[i] != atomic64_read(&ring->fence_drv.last_seq))
-					break;
-			}
-
-			if (i != AMDGPU_MAX_RINGS)
-				continue;
-
-			for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-				if (!adev->rings[i] || !target_seq[i])
-					continue;
-
-				if (amdgpu_ring_is_lockup(adev->rings[i]))
-					break;
-			}
-
-			if (i < AMDGPU_MAX_RINGS) {
-				/* good news we believe it's a lockup */
-				dev_warn(adev->dev, "GPU lockup (waiting for "
-					 "0x%016llx last fence id 0x%016llx on"
-					 " ring %d)\n",
-					 target_seq[i], last_seq[i], i);
-
-				/* remember that we need an reset */
-				adev->needs_reset = true;
-				wake_up_all(&adev->fence_queue);
-				return -EDEADLK;
-			}
-
-			if (timeout < MAX_SCHEDULE_TIMEOUT) {
-				timeout -= AMDGPU_FENCE_JIFFIES_TIMEOUT;
-				if (timeout <= 0) {
-					return 0;
-				}
-			}
-		}
-	}
-	return timeout;
-}
-
-/**
- * amdgpu_fence_wait - wait for a fence to signal
- *
- * @fence: amdgpu fence object
- * @intr: use interruptable sleep
- *
- * Wait for the requested fence to signal (all asics).
- * @intr selects whether to use interruptable (true) or non-interruptable
- * (false) sleep when waiting for the fence.
- * Returns 0 if the fence has passed, error for all other cases.
- */
-int amdgpu_fence_wait(struct amdgpu_fence *fence, bool intr)
-{
-	uint64_t seq[AMDGPU_MAX_RINGS] = {};
-	long r;
-
-	seq[fence->ring->idx] = fence->seq;
-	r = amdgpu_fence_wait_seq_timeout(fence->ring->adev, seq, intr, MAX_SCHEDULE_TIMEOUT);
-	if (r < 0) {
-		return r;
-	}
-
-	r = fence_signal(&fence->base);
-	if (!r)
-		FENCE_TRACE(&fence->base, "signaled from fence_wait\n");
-	return 0;
-}
-
-/**
- * amdgpu_fence_wait_any - wait for a fence to signal on any ring
- *
- * @adev: amdgpu device pointer
- * @fences: amdgpu fence object(s)
- * @intr: use interruptable sleep
+/*
+ * amdgpu_ring_wait_seq_timeout - wait for seq of the specific ring to signal
+ * @ring: ring to wait on for the seq number
+ * @seq: seq number wait for
  *
- * Wait for any requested fence to signal (all asics).  Fence
- * array is indexed by ring id.  @intr selects whether to use
- * interruptable (true) or non-interruptable (false) sleep when
- * waiting for the fences. Used by the suballocator.
- * Returns 0 if any fence has passed, error for all other cases.
+ * return value:
+ * 0: seq signaled, and gpu not hang
+ * -EDEADL: GPU hang detected
+ * -EINVAL: some paramter is not valid
  */
-int amdgpu_fence_wait_any(struct amdgpu_device *adev,
-			  struct amdgpu_fence **fences,
-			  bool intr)
+static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq)
 {
-	uint64_t seq[AMDGPU_MAX_RINGS];
-	unsigned i, num_rings = 0;
-	long r;
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		seq[i] = 0;
+	struct amdgpu_device *adev = ring->adev;
+	bool signaled = false;
 
-		if (!fences[i]) {
-			continue;
-		}
+	BUG_ON(!ring);
+	if (seq > ring->fence_drv.sync_seq[ring->idx])
+		return -EINVAL;
 
-		seq[i] = fences[i]->seq;
-		++num_rings;
-	}
+	if (atomic64_read(&ring->fence_drv.last_seq) >= seq)
+		return 0;
 
-	/* nothing to wait for ? */
-	if (num_rings == 0)
-		return -ENOENT;
+	wait_event(ring->fence_drv.fence_queue, (
+		   (signaled = amdgpu_fence_seq_signaled(ring, seq))
+		   || adev->needs_reset));
 
-	r = amdgpu_fence_wait_seq_timeout(adev, seq, intr, MAX_SCHEDULE_TIMEOUT);
-	if (r < 0) {
-		return r;
-	}
-	return 0;
+	if (signaled)
+		return 0;
+	else
+		return -EDEADLK;
 }
 
 /**
@@ -739,19 +412,12 @@ int amdgpu_fence_wait_any(struct amdgpu_device *adev,
  */
 int amdgpu_fence_wait_next(struct amdgpu_ring *ring)
 {
-	uint64_t seq[AMDGPU_MAX_RINGS] = {};
-	long r;
+	uint64_t seq = atomic64_read(&ring->fence_drv.last_seq) + 1ULL;
 
-	seq[ring->idx] = atomic64_read(&ring->fence_drv.last_seq) + 1ULL;
-	if (seq[ring->idx] >= ring->fence_drv.sync_seq[ring->idx]) {
-		/* nothing to wait for, last_seq is
-		   already the last emited fence */
+	if (seq >= ring->fence_drv.sync_seq[ring->idx])
 		return -ENOENT;
-	}
-	r = amdgpu_fence_wait_seq_timeout(ring->adev, seq, false, MAX_SCHEDULE_TIMEOUT);
-	if (r < 0)
-		return r;
-	return 0;
+
+	return amdgpu_fence_ring_wait_seq(ring, seq);
 }
 
 /**
@@ -766,23 +432,12 @@ int amdgpu_fence_wait_next(struct amdgpu_ring *ring)
  */
 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
 {
-	struct amdgpu_device *adev = ring->adev;
-	uint64_t seq[AMDGPU_MAX_RINGS] = {};
-	long r;
+	uint64_t seq = ring->fence_drv.sync_seq[ring->idx];
 
-	seq[ring->idx] = ring->fence_drv.sync_seq[ring->idx];
-	if (!seq[ring->idx])
+	if (!seq)
 		return 0;
 
-	r = amdgpu_fence_wait_seq_timeout(adev, seq, false, MAX_SCHEDULE_TIMEOUT);
-	if (r < 0) {
-		if (r == -EDEADLK)
-			return -EDEADLK;
-
-		dev_err(adev->dev, "error waiting for ring[%d] to become idle (%ld)\n",
-			ring->idx, r);
-	}
-	return 0;
+	return amdgpu_fence_ring_wait_seq(ring, seq);
 }
 
 /**
@@ -933,9 +588,12 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 		ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index;
 	}
 	amdgpu_fence_write(ring, atomic64_read(&ring->fence_drv.last_seq));
-	ring->fence_drv.initialized = true;
+	amdgpu_irq_get(adev, irq_src, irq_type);
+
 	ring->fence_drv.irq_src = irq_src;
 	ring->fence_drv.irq_type = irq_type;
+	ring->fence_drv.initialized = true;
+
 	dev_info(adev->dev, "fence driver on ring %d use gpu addr 0x%016llx, "
 		 "cpu addr 0x%p\n", ring->idx,
 		 ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr);
@@ -966,6 +624,16 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
 	INIT_DELAYED_WORK(&ring->fence_drv.lockup_work,
 			amdgpu_fence_check_lockup);
 	ring->fence_drv.ring = ring;
+
+	if (amdgpu_enable_scheduler) {
+		ring->scheduler = amd_sched_create(&amdgpu_sched_ops,
+						   ring->idx,
+						   amdgpu_sched_hw_submission,
+						   (void *)ring->adev);
+		if (!ring->scheduler)
+			DRM_ERROR("Failed to create scheduler on ring %d.\n",
+				  ring->idx);
+	}
 }
 
 /**
@@ -982,7 +650,6 @@ void amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
  */
 int amdgpu_fence_driver_init(struct amdgpu_device *adev)
 {
-	init_waitqueue_head(&adev->fence_queue);
 	if (amdgpu_debugfs_fence_init(adev))
 		dev_err(adev->dev, "fence debugfs file creation failed\n");
 
@@ -1011,12 +678,77 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
 			/* no need to trigger GPU reset as we are unloading */
 			amdgpu_fence_driver_force_completion(adev);
 		}
-		wake_up_all(&adev->fence_queue);
+		wake_up_all(&ring->fence_drv.fence_queue);
+		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
+			       ring->fence_drv.irq_type);
+		if (ring->scheduler)
+			amd_sched_destroy(ring->scheduler);
 		ring->fence_drv.initialized = false;
 	}
 	mutex_unlock(&adev->ring_lock);
 }
 
+/**
+ * amdgpu_fence_driver_suspend - suspend the fence driver
+ * for all possible rings.
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Suspend the fence driver for all possible rings (all asics).
+ */
+void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
+{
+	int i, r;
+
+	mutex_lock(&adev->ring_lock);
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+		if (!ring || !ring->fence_drv.initialized)
+			continue;
+
+		/* wait for gpu to finish processing current batch */
+		r = amdgpu_fence_wait_empty(ring);
+		if (r) {
+			/* delay GPU reset to resume */
+			amdgpu_fence_driver_force_completion(adev);
+		}
+
+		/* disable the interrupt */
+		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
+			       ring->fence_drv.irq_type);
+	}
+	mutex_unlock(&adev->ring_lock);
+}
+
+/**
+ * amdgpu_fence_driver_resume - resume the fence driver
+ * for all possible rings.
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Resume the fence driver for all possible rings (all asics).
+ * Not all asics have all rings, so each asic will only
+ * start the fence driver on the rings it has using
+ * amdgpu_fence_driver_start_ring().
+ * Returns 0 for success.
+ */
+void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
+{
+	int i;
+
+	mutex_lock(&adev->ring_lock);
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+		if (!ring || !ring->fence_drv.initialized)
+			continue;
+
+		/* enable the interrupt */
+		amdgpu_irq_get(adev, ring->fence_drv.irq_src,
+			       ring->fence_drv.irq_type);
+	}
+	mutex_unlock(&adev->ring_lock);
+}
+
 /**
  * amdgpu_fence_driver_force_completion - force all fence waiter to complete
  *
@@ -1104,6 +836,21 @@ static inline bool amdgpu_test_signaled(struct amdgpu_fence *fence)
 	return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
 }
 
+static bool amdgpu_test_signaled_any(struct fence **fences, uint32_t count)
+{
+	int idx;
+	struct fence *fence;
+
+	for (idx = 0; idx < count; ++idx) {
+		fence = fences[idx];
+		if (fence) {
+			if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->flags))
+				return true;
+		}
+	}
+	return false;
+}
+
 struct amdgpu_wait_cb {
 	struct fence_cb base;
 	struct task_struct *task;
@@ -1121,12 +868,48 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
 {
 	struct amdgpu_fence *fence = to_amdgpu_fence(f);
 	struct amdgpu_device *adev = fence->ring->adev;
-	struct amdgpu_wait_cb cb;
 
-	cb.task = current;
+	return amdgpu_fence_wait_any(adev, &f, 1, intr, t);
+}
 
-	if (fence_add_callback(f, &cb.base, amdgpu_fence_wait_cb))
-		return t;
+/**
+ * Wait the fence array with timeout
+ *
+ * @adev:     amdgpu device
+ * @array:    the fence array with amdgpu fence pointer
+ * @count:    the number of the fence array
+ * @intr:     when sleep, set the current task interruptable or not
+ * @t:        timeout to wait
+ *
+ * It will return when any fence is signaled or timeout.
+ */
+signed long amdgpu_fence_wait_any(struct amdgpu_device *adev,
+				  struct fence **array, uint32_t count,
+				  bool intr, signed long t)
+{
+	struct amdgpu_wait_cb *cb;
+	struct fence *fence;
+	unsigned idx;
+
+	BUG_ON(!array);
+
+	cb = kcalloc(count, sizeof(struct amdgpu_wait_cb), GFP_KERNEL);
+	if (cb == NULL) {
+		t = -ENOMEM;
+		goto err_free_cb;
+	}
+
+	for (idx = 0; idx < count; ++idx) {
+		fence = array[idx];
+		if (fence) {
+			cb[idx].task = current;
+			if (fence_add_callback(fence,
+					&cb[idx].base, amdgpu_fence_wait_cb)) {
+				/* The fence is already signaled */
+				goto fence_rm_cb;
+			}
+		}
+	}
 
 	while (t > 0) {
 		if (intr)
@@ -1135,10 +918,10 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
 			set_current_state(TASK_UNINTERRUPTIBLE);
 
 		/*
-		 * amdgpu_test_signaled must be called after
+		 * amdgpu_test_signaled_any must be called after
 		 * set_current_state to prevent a race with wake_up_process
 		 */
-		if (amdgpu_test_signaled(fence))
+		if (amdgpu_test_signaled_any(array, count))
 			break;
 
 		if (adev->needs_reset) {
@@ -1153,7 +936,16 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
 	}
 
 	__set_current_state(TASK_RUNNING);
-	fence_remove_callback(f, &cb.base);
+
+fence_rm_cb:
+	for (idx = 0; idx < count; ++idx) {
+		fence = array[idx];
+		if (fence && cb[idx].base.func)
+			fence_remove_callback(fence, &cb[idx].base);
+	}
+
+err_free_cb:
+	kfree(cb);
 
 	return t;
 }

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

@@ -125,7 +125,8 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 
 	if (adev->gart.robj == NULL) {
 		r = amdgpu_bo_create(adev, adev->gart.table_size,
-				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, 0,
+				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
+				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
 				     NULL, &adev->gart.robj);
 		if (r) {
 			return r;

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

@@ -615,6 +615,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 		info.alignment = robj->tbo.mem.page_alignment << PAGE_SHIFT;
 		info.domains = robj->initial_domain;
 		info.domain_flags = robj->flags;
+		amdgpu_bo_unreserve(robj);
 		if (copy_to_user(out, &info, sizeof(info)))
 			r = -EFAULT;
 		break;
@@ -622,17 +623,19 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 	case AMDGPU_GEM_OP_SET_PLACEMENT:
 		if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm)) {
 			r = -EPERM;
+			amdgpu_bo_unreserve(robj);
 			break;
 		}
 		robj->initial_domain = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
 						      AMDGPU_GEM_DOMAIN_GTT |
 						      AMDGPU_GEM_DOMAIN_CPU);
+		amdgpu_bo_unreserve(robj);
 		break;
 	default:
+		amdgpu_bo_unreserve(robj);
 		r = -EINVAL;
 	}
 
-	amdgpu_bo_unreserve(robj);
 out:
 	drm_gem_object_unreference_unlocked(gobj);
 	return r;
@@ -653,7 +656,8 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv,
 
 	r = amdgpu_gem_object_create(adev, args->size, 0,
 				     AMDGPU_GEM_DOMAIN_VRAM,
-				     0, ttm_bo_type_device,
+				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+				     ttm_bo_type_device,
 				     &gobj);
 	if (r)
 		return -ENOMEM;

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

@@ -73,28 +73,12 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
 
 		if (!vm)
 			ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
-		else
-			ib->gpu_addr = 0;
-
-	} else {
-		ib->sa_bo = NULL;
-		ib->ptr = NULL;
-		ib->gpu_addr = 0;
 	}
 
 	amdgpu_sync_create(&ib->sync);
 
 	ib->ring = ring;
-	ib->fence = NULL;
-	ib->user = NULL;
 	ib->vm = vm;
-	ib->gds_base = 0;
-	ib->gds_size = 0;
-	ib->gws_base = 0;
-	ib->gws_size = 0;
-	ib->oa_base = 0;
-	ib->oa_size = 0;
-	ib->flags = 0;
 
 	return 0;
 }
@@ -109,8 +93,8 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
  */
 void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
 {
-	amdgpu_sync_free(adev, &ib->sync, ib->fence);
-	amdgpu_sa_bo_free(adev, &ib->sa_bo, ib->fence);
+	amdgpu_sync_free(adev, &ib->sync, &ib->fence->base);
+	amdgpu_sa_bo_free(adev, &ib->sa_bo, &ib->fence->base);
 	amdgpu_fence_unref(&ib->fence);
 }
 
@@ -156,7 +140,11 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 		dev_err(adev->dev, "couldn't schedule ib\n");
 		return -EINVAL;
 	}
-
+	r = amdgpu_sync_wait(&ibs->sync);
+	if (r) {
+		dev_err(adev->dev, "IB sync failed (%d).\n", r);
+		return r;
+	}
 	r = amdgpu_ring_lock(ring, (256 + AMDGPU_NUM_SYNCS * 8) * num_ibs);
 	if (r) {
 		dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
@@ -165,9 +153,11 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 
 	if (vm) {
 		/* grab a vm id if necessary */
-		struct amdgpu_fence *vm_id_fence = NULL;
-		vm_id_fence = amdgpu_vm_grab_id(ibs->ring, ibs->vm);
-		amdgpu_sync_fence(&ibs->sync, vm_id_fence);
+		r = amdgpu_vm_grab_id(ibs->vm, ibs->ring, &ibs->sync);
+		if (r) {
+			amdgpu_ring_unlock_undo(ring);
+			return r;
+		}
 	}
 
 	r = amdgpu_sync_rings(&ibs->sync, ring);
@@ -212,11 +202,15 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 		return r;
 	}
 
+	if (!amdgpu_enable_scheduler && ib->ctx)
+		ib->sequence = amdgpu_ctx_add_fence(ib->ctx, ring,
+						    &ib->fence->base);
+
 	/* wrap the last IB with fence */
 	if (ib->user) {
 		uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo);
 		addr += ib->user->offset;
-		amdgpu_ring_emit_fence(ring, addr, ib->fence->seq,
+		amdgpu_ring_emit_fence(ring, addr, ib->sequence,
 				       AMDGPU_FENCE_FLAG_64BIT);
 	}
 

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

@@ -24,6 +24,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_ih.h"
+#include "amdgpu_amdkfd.h"
 
 /**
  * amdgpu_ih_ring_alloc - allocate memory for the IH ring
@@ -97,18 +98,12 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
 			/* add 8 bytes for the rptr/wptr shadows and
 			 * add them to the end of the ring allocation.
 			 */
-			adev->irq.ih.ring = kzalloc(adev->irq.ih.ring_size + 8, GFP_KERNEL);
+			adev->irq.ih.ring = pci_alloc_consistent(adev->pdev,
+								 adev->irq.ih.ring_size + 8,
+								 &adev->irq.ih.rb_dma_addr);
 			if (adev->irq.ih.ring == NULL)
 				return -ENOMEM;
-			adev->irq.ih.rb_dma_addr = pci_map_single(adev->pdev,
-								  (void *)adev->irq.ih.ring,
-								  adev->irq.ih.ring_size,
-								  PCI_DMA_BIDIRECTIONAL);
-			if (pci_dma_mapping_error(adev->pdev, adev->irq.ih.rb_dma_addr)) {
-				dev_err(&adev->pdev->dev, "Failed to DMA MAP the IH RB page\n");
-				kfree((void *)adev->irq.ih.ring);
-				return -ENOMEM;
-			}
+			memset((void *)adev->irq.ih.ring, 0, adev->irq.ih.ring_size + 8);
 			adev->irq.ih.wptr_offs = (adev->irq.ih.ring_size / 4) + 0;
 			adev->irq.ih.rptr_offs = (adev->irq.ih.ring_size / 4) + 1;
 		}
@@ -148,9 +143,9 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev)
 			/* add 8 bytes for the rptr/wptr shadows and
 			 * add them to the end of the ring allocation.
 			 */
-			pci_unmap_single(adev->pdev, adev->irq.ih.rb_dma_addr,
-					 adev->irq.ih.ring_size + 8, PCI_DMA_BIDIRECTIONAL);
-			kfree((void *)adev->irq.ih.ring);
+			pci_free_consistent(adev->pdev, adev->irq.ih.ring_size + 8,
+					    (void *)adev->irq.ih.ring,
+					    adev->irq.ih.rb_dma_addr);
 			adev->irq.ih.ring = NULL;
 		}
 	} else {
@@ -199,6 +194,14 @@ restart_ih:
 	rmb();
 
 	while (adev->irq.ih.rptr != wptr) {
+		u32 ring_index = adev->irq.ih.rptr >> 2;
+
+		/* Before dispatching irq to IP blocks, send it to amdkfd */
+		amdgpu_amdkfd_interrupt(adev,
+				(const void *) &adev->irq.ih.ring[ring_index]);
+
+		entry.iv_entry = (const uint32_t *)
+			&adev->irq.ih.ring[ring_index];
 		amdgpu_ih_decode_iv(adev, &entry);
 		adev->irq.ih.rptr &= adev->irq.ih.ptr_mask;
 

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

@@ -52,6 +52,7 @@ struct amdgpu_iv_entry {
 	unsigned ring_id;
 	unsigned vm_id;
 	unsigned pas_id;
+	const uint32_t *iv_entry;
 };
 
 int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,

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

@@ -272,6 +272,11 @@ void amdgpu_irq_fini(struct amdgpu_device *adev)
 
 		kfree(src->enabled_types);
 		src->enabled_types = NULL;
+		if (src->data) {
+			kfree(src->data);
+			kfree(src);
+			adev->irq.sources[i] = NULL;
+		}
 	}
 }
 

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

@@ -40,6 +40,7 @@ struct amdgpu_irq_src {
 	unsigned				num_types;
 	atomic_t				*enabled_types;
 	const struct amdgpu_irq_src_funcs	*funcs;
+	void *data;
 };
 
 /* provided by interrupt generating IP blocks */

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

@@ -34,6 +34,7 @@
 #include <linux/vga_switcheroo.h>
 #include <linux/slab.h>
 #include <linux/pm_runtime.h>
+#include "amdgpu_amdkfd.h"
 
 #if defined(CONFIG_VGA_SWITCHEROO)
 bool amdgpu_has_atpx(void);
@@ -61,6 +62,8 @@ int amdgpu_driver_unload_kms(struct drm_device *dev)
 
 	pm_runtime_get_sync(dev->dev);
 
+	amdgpu_amdkfd_device_fini(adev);
+
 	amdgpu_acpi_fini(adev);
 
 	amdgpu_device_fini(adev);
@@ -93,8 +96,8 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
 
 	if ((amdgpu_runtime_pm != 0) &&
 	    amdgpu_has_atpx() &&
-	    ((flags & AMDGPU_IS_APU) == 0))
-		flags |= AMDGPU_IS_PX;
+	    ((flags & AMD_IS_APU) == 0))
+		flags |= AMD_IS_PX;
 
 	/* amdgpu_device_init should report only fatal error
 	 * like memory allocation failure or iomapping failure,
@@ -118,6 +121,10 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
 				"Error during ACPI methods call\n");
 	}
 
+	amdgpu_amdkfd_load_interface(adev);
+	amdgpu_amdkfd_device_probe(adev);
+	amdgpu_amdkfd_device_init(adev);
+
 	if (amdgpu_device_is_px(dev)) {
 		pm_runtime_use_autosuspend(dev->dev);
 		pm_runtime_set_autosuspend_delay(dev->dev, 5000);
@@ -444,11 +451,11 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		dev_info.num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts;
 		dev_info._pad = 0;
 		dev_info.ids_flags = 0;
-		if (adev->flags & AMDGPU_IS_APU)
+		if (adev->flags & AMD_IS_APU)
 			dev_info.ids_flags |= AMDGPU_IDS_FLAGS_FUSION;
 		dev_info.virtual_address_offset = AMDGPU_VA_RESERVED_SIZE;
 		dev_info.virtual_address_max = (uint64_t)adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
-		dev_info.virtual_address_alignment = max(PAGE_SIZE, 0x10000UL);
+		dev_info.virtual_address_alignment = max((int)PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
 		dev_info.pte_fragment_size = (1 << AMDGPU_LOG2_PAGES_PER_FRAG) *
 					     AMDGPU_GPU_PAGE_SIZE;
 		dev_info.gart_page_size = AMDGPU_GPU_PAGE_SIZE;
@@ -520,10 +527,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 	mutex_init(&fpriv->bo_list_lock);
 	idr_init(&fpriv->bo_list_handles);
 
-	/* init context manager */
-	mutex_init(&fpriv->ctx_mgr.lock);
-	idr_init(&fpriv->ctx_mgr.ctx_handles);
-	fpriv->ctx_mgr.adev = adev;
+	amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
 
 	file_priv->driver_priv = fpriv;
 
@@ -556,6 +560,8 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 	if (!fpriv)
 		return;
 
+	amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
+
 	amdgpu_vm_fini(adev, &fpriv->vm);
 
 	idr_for_each_entry(&fpriv->bo_list_handles, list, handle)
@@ -564,9 +570,6 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 	idr_destroy(&fpriv->bo_list_handles);
 	mutex_destroy(&fpriv->bo_list_lock);
 
-	/* release context */
-	amdgpu_ctx_fini(fpriv);
-
 	kfree(fpriv);
 	file_priv->driver_priv = NULL;
 }

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

@@ -127,7 +127,7 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
 			placements[c].fpfn =
 				adev->mc.visible_vram_size >> PAGE_SHIFT;
 			placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
-				TTM_PL_FLAG_VRAM;
+				TTM_PL_FLAG_VRAM | TTM_PL_FLAG_TOPDOWN;
 		}
 		placements[c].fpfn = 0;
 		placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
@@ -223,18 +223,6 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 	size_t acc_size;
 	int r;
 
-	/* VI has a hw bug where VM PTEs have to be allocated in groups of 8.
-	 * do this as a temporary workaround
-	 */
-	if (!(domain & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA))) {
-		if (adev->asic_type >= CHIP_TOPAZ) {
-			if (byte_align & 0x7fff)
-				byte_align = ALIGN(byte_align, 0x8000);
-			if (size & 0x7fff)
-				size = ALIGN(size, 0x8000);
-		}
-	}
-
 	page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
 	size = ALIGN(size, PAGE_SIZE);
 
@@ -462,7 +450,7 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo)
 int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
 {
 	/* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
-	if (0 && (adev->flags & AMDGPU_IS_APU)) {
+	if (0 && (adev->flags & AMD_IS_APU)) {
 		/* Useless to evict on IGP chips */
 		return 0;
 	}
@@ -478,7 +466,6 @@ void amdgpu_bo_force_delete(struct amdgpu_device *adev)
 	}
 	dev_err(adev->dev, "Userspace still has active objects !\n");
 	list_for_each_entry_safe(bo, n, &adev->gem.objects, list) {
-		mutex_lock(&adev->ddev->struct_mutex);
 		dev_err(adev->dev, "%p %p %lu %lu force free\n",
 			&bo->gem_base, bo, (unsigned long)bo->gem_base.size,
 			*((unsigned long *)&bo->gem_base.refcount));
@@ -486,8 +473,7 @@ void amdgpu_bo_force_delete(struct amdgpu_device *adev)
 		list_del_init(&bo->list);
 		mutex_unlock(&bo->adev->gem.mutex);
 		/* this should unref the ttm bo */
-		drm_gem_object_unreference(&bo->gem_base);
-		mutex_unlock(&adev->ddev->struct_mutex);
+		drm_gem_object_unreference_unlocked(&bo->gem_base);
 	}
 }
 
@@ -658,13 +644,13 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
  * @shared: true if fence should be added shared
  *
  */
-void amdgpu_bo_fence(struct amdgpu_bo *bo, struct amdgpu_fence *fence,
+void amdgpu_bo_fence(struct amdgpu_bo *bo, struct fence *fence,
 		     bool shared)
 {
 	struct reservation_object *resv = bo->tbo.resv;
 
 	if (shared)
-		reservation_object_add_shared_fence(resv, &fence->base);
+		reservation_object_add_shared_fence(resv, fence);
 	else
-		reservation_object_add_excl_fence(resv, &fence->base);
+		reservation_object_add_excl_fence(resv, fence);
 }

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

@@ -161,7 +161,7 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
 void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 				  struct ttm_mem_reg *new_mem);
 int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
-void amdgpu_bo_fence(struct amdgpu_bo *bo, struct amdgpu_fence *fence,
+void amdgpu_bo_fence(struct amdgpu_bo *bo, struct fence *fence,
 		     bool shared);
 
 /*
@@ -193,7 +193,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 			    unsigned size, unsigned align);
 void amdgpu_sa_bo_free(struct amdgpu_device *adev,
 			      struct amdgpu_sa_bo **sa_bo,
-			      struct amdgpu_fence *fence);
+			      struct fence *fence);
 #if defined(CONFIG_DEBUG_FS)
 void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager,
 					 struct seq_file *m);

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

@@ -82,7 +82,7 @@ static ssize_t amdgpu_set_dpm_state(struct device *dev,
 	mutex_unlock(&adev->pm.mutex);
 
 	/* Can't set dpm state when the card is off */
-	if (!(adev->flags & AMDGPU_IS_PX) ||
+	if (!(adev->flags & AMD_IS_PX) ||
 	    (ddev->switch_power_state == DRM_SWITCH_POWER_ON))
 		amdgpu_pm_compute_clocks(adev);
 fail:
@@ -538,7 +538,7 @@ static void amdgpu_dpm_change_power_state_locked(struct amdgpu_device *adev)
 		/* vce just modifies an existing state so force a change */
 		if (ps->vce_active != adev->pm.dpm.vce_active)
 			goto force;
-		if (adev->flags & AMDGPU_IS_APU) {
+		if (adev->flags & AMD_IS_APU) {
 			/* for APUs if the num crtcs changed but state is the same,
 			 * all we need to do is update the display configuration.
 			 */
@@ -580,7 +580,6 @@ force:
 		amdgpu_dpm_print_power_state(adev, adev->pm.dpm.requested_ps);
 	}
 
-	mutex_lock(&adev->ddev->struct_mutex);
 	mutex_lock(&adev->ring_lock);
 
 	/* update whether vce is active */
@@ -628,7 +627,6 @@ force:
 
 done:
 	mutex_unlock(&adev->ring_lock);
-	mutex_unlock(&adev->ddev->struct_mutex);
 }
 
 void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable)

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

@@ -131,6 +131,21 @@ int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw)
 	return 0;
 }
 
+/** amdgpu_ring_insert_nop - insert NOP packets
+ *
+ * @ring: amdgpu_ring structure holding ring information
+ * @count: the number of NOP packets to insert
+ *
+ * This is the generic insert_nop function for rings except SDMA
+ */
+void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	int i;
+
+	for (i = 0; i < count; i++)
+		amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * amdgpu_ring_commit - tell the GPU to execute the new
  * commands on the ring buffer
@@ -143,10 +158,13 @@ int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw)
  */
 void amdgpu_ring_commit(struct amdgpu_ring *ring)
 {
+	uint32_t count;
+
 	/* We pad to match fetch size */
-	while (ring->wptr & ring->align_mask) {
-		amdgpu_ring_write(ring, ring->nop);
-	}
+	count = ring->align_mask + 1 - (ring->wptr & ring->align_mask);
+	count %= ring->align_mask + 1;
+	ring->funcs->insert_nop(ring, count);
+
 	mb();
 	amdgpu_ring_set_wptr(ring);
 }
@@ -342,6 +360,8 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
 		amdgpu_fence_driver_init_ring(ring);
 	}
 
+	init_waitqueue_head(&ring->fence_drv.fence_queue);
+
 	r = amdgpu_wb_get(adev, &ring->rptr_offs);
 	if (r) {
 		dev_err(adev->dev, "(%d) ring rptr_offs wb alloc failed\n", r);
@@ -367,7 +387,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
 	}
 	ring->next_rptr_gpu_addr = adev->wb.gpu_addr + (ring->next_rptr_offs * 4);
 	ring->next_rptr_cpu_addr = &adev->wb.wb[ring->next_rptr_offs];
-
+	spin_lock_init(&ring->fence_lock);
 	r = amdgpu_fence_driver_start_ring(ring, irq_src, irq_type);
 	if (r) {
 		dev_err(adev->dev, "failed initializing fences (%d).\n", r);

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

@@ -139,6 +139,20 @@ int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev,
 	return r;
 }
 
+static uint32_t amdgpu_sa_get_ring_from_fence(struct fence *f)
+{
+	struct amdgpu_fence *a_fence;
+	struct amd_sched_fence *s_fence;
+
+	s_fence = to_amd_sched_fence(f);
+	if (s_fence)
+		return s_fence->scheduler->ring_id;
+	a_fence = to_amdgpu_fence(f);
+	if (a_fence)
+		return a_fence->ring->idx;
+	return 0;
+}
+
 static void amdgpu_sa_bo_remove_locked(struct amdgpu_sa_bo *sa_bo)
 {
 	struct amdgpu_sa_manager *sa_manager = sa_bo->manager;
@@ -147,7 +161,7 @@ static void amdgpu_sa_bo_remove_locked(struct amdgpu_sa_bo *sa_bo)
 	}
 	list_del_init(&sa_bo->olist);
 	list_del_init(&sa_bo->flist);
-	amdgpu_fence_unref(&sa_bo->fence);
+	fence_put(sa_bo->fence);
 	kfree(sa_bo);
 }
 
@@ -160,7 +174,8 @@ static void amdgpu_sa_bo_try_free(struct amdgpu_sa_manager *sa_manager)
 
 	sa_bo = list_entry(sa_manager->hole->next, struct amdgpu_sa_bo, olist);
 	list_for_each_entry_safe_from(sa_bo, tmp, &sa_manager->olist, olist) {
-		if (sa_bo->fence == NULL || !amdgpu_fence_signaled(sa_bo->fence)) {
+		if (sa_bo->fence == NULL ||
+		    !fence_is_signaled(sa_bo->fence)) {
 			return;
 		}
 		amdgpu_sa_bo_remove_locked(sa_bo);
@@ -245,7 +260,7 @@ static bool amdgpu_sa_event(struct amdgpu_sa_manager *sa_manager,
 }
 
 static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
-				   struct amdgpu_fence **fences,
+				   struct fence **fences,
 				   unsigned *tries)
 {
 	struct amdgpu_sa_bo *best_bo = NULL;
@@ -274,7 +289,7 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
 		sa_bo = list_first_entry(&sa_manager->flist[i],
 					 struct amdgpu_sa_bo, flist);
 
-		if (!amdgpu_fence_signaled(sa_bo->fence)) {
+		if (!fence_is_signaled(sa_bo->fence)) {
 			fences[i] = sa_bo->fence;
 			continue;
 		}
@@ -298,7 +313,8 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
 	}
 
 	if (best_bo) {
-		++tries[best_bo->fence->ring->idx];
+		uint32_t idx = amdgpu_sa_get_ring_from_fence(best_bo->fence);
+		++tries[idx];
 		sa_manager->hole = best_bo->olist.prev;
 
 		/* we knew that this one is signaled,
@@ -314,9 +330,10 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 		     struct amdgpu_sa_bo **sa_bo,
 		     unsigned size, unsigned align)
 {
-	struct amdgpu_fence *fences[AMDGPU_MAX_RINGS];
+	struct fence *fences[AMDGPU_MAX_RINGS];
 	unsigned tries[AMDGPU_MAX_RINGS];
 	int i, r;
+	signed long t;
 
 	BUG_ON(align > sa_manager->align);
 	BUG_ON(size > sa_manager->size);
@@ -350,7 +367,9 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 		} while (amdgpu_sa_bo_next_hole(sa_manager, fences, tries));
 
 		spin_unlock(&sa_manager->wq.lock);
-		r = amdgpu_fence_wait_any(adev, fences, false);
+		t = amdgpu_fence_wait_any(adev, fences, AMDGPU_MAX_RINGS,
+					  false, MAX_SCHEDULE_TIMEOUT);
+		r = (t > 0) ? 0 : t;
 		spin_lock(&sa_manager->wq.lock);
 		/* if we have nothing to wait for block */
 		if (r == -ENOENT) {
@@ -369,7 +388,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 }
 
 void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo,
-		       struct amdgpu_fence *fence)
+		       struct fence *fence)
 {
 	struct amdgpu_sa_manager *sa_manager;
 
@@ -379,10 +398,11 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo,
 
 	sa_manager = (*sa_bo)->manager;
 	spin_lock(&sa_manager->wq.lock);
-	if (fence && !amdgpu_fence_signaled(fence)) {
-		(*sa_bo)->fence = amdgpu_fence_ref(fence);
-		list_add_tail(&(*sa_bo)->flist,
-			      &sa_manager->flist[fence->ring->idx]);
+	if (fence && !fence_is_signaled(fence)) {
+		uint32_t idx;
+		(*sa_bo)->fence = fence_get(fence);
+		idx = amdgpu_sa_get_ring_from_fence(fence);
+		list_add_tail(&(*sa_bo)->flist, &sa_manager->flist[idx]);
 	} else {
 		amdgpu_sa_bo_remove_locked(*sa_bo);
 	}
@@ -409,8 +429,16 @@ void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager,
 		seq_printf(m, "[0x%010llx 0x%010llx] size %8lld",
 			   soffset, eoffset, eoffset - soffset);
 		if (i->fence) {
-			seq_printf(m, " protected by 0x%016llx on ring %d",
-				   i->fence->seq, i->fence->ring->idx);
+			struct amdgpu_fence *a_fence = to_amdgpu_fence(i->fence);
+			struct amd_sched_fence *s_fence = to_amd_sched_fence(i->fence);
+			if (a_fence)
+				seq_printf(m, " protected by 0x%016llx on ring %d",
+					   a_fence->seq, a_fence->ring->idx);
+			if (s_fence)
+				seq_printf(m, " protected by 0x%016x on ring %d",
+					   s_fence->base.seqno,
+					   s_fence->scheduler->ring_id);
+
 		}
 		seq_printf(m, "\n");
 	}

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

@@ -0,0 +1,128 @@
+/*
+ * Copyright 2015 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 <linux/kthread.h>
+#include <linux/wait.h>
+#include <linux/sched.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+
+static struct fence *amdgpu_sched_dependency(struct amd_sched_job *job)
+{
+	struct amdgpu_job *sched_job = (struct amdgpu_job *)job;
+	return amdgpu_sync_get_fence(&sched_job->ibs->sync);
+}
+
+static struct fence *amdgpu_sched_run_job(struct amd_sched_job *job)
+{
+	struct amdgpu_job *sched_job;
+	struct amdgpu_fence *fence;
+	int r;
+
+	if (!job) {
+		DRM_ERROR("job is null\n");
+		return NULL;
+	}
+	sched_job = (struct amdgpu_job *)job;
+	mutex_lock(&sched_job->job_lock);
+	r = amdgpu_ib_schedule(sched_job->adev,
+			       sched_job->num_ibs,
+			       sched_job->ibs,
+			       sched_job->base.owner);
+	if (r)
+		goto err;
+	fence = amdgpu_fence_ref(sched_job->ibs[sched_job->num_ibs - 1].fence);
+
+	if (sched_job->free_job)
+		sched_job->free_job(sched_job);
+
+	mutex_unlock(&sched_job->job_lock);
+	return &fence->base;
+
+err:
+	DRM_ERROR("Run job error\n");
+	mutex_unlock(&sched_job->job_lock);
+	job->sched->ops->process_job(job);
+	return NULL;
+}
+
+static void amdgpu_sched_process_job(struct amd_sched_job *job)
+{
+	struct amdgpu_job *sched_job;
+
+	if (!job) {
+		DRM_ERROR("job is null\n");
+		return;
+	}
+	sched_job = (struct amdgpu_job *)job;
+	/* after processing job, free memory */
+	fence_put(&sched_job->base.s_fence->base);
+	kfree(sched_job);
+}
+
+struct amd_sched_backend_ops amdgpu_sched_ops = {
+	.dependency = amdgpu_sched_dependency,
+	.run_job = amdgpu_sched_run_job,
+	.process_job = amdgpu_sched_process_job
+};
+
+int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
+					 struct amdgpu_ring *ring,
+					 struct amdgpu_ib *ibs,
+					 unsigned num_ibs,
+					 int (*free_job)(struct amdgpu_job *),
+					 void *owner,
+					 struct fence **f)
+{
+	int r = 0;
+	if (amdgpu_enable_scheduler) {
+		struct amdgpu_job *job =
+			kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+		if (!job)
+			return -ENOMEM;
+		job->base.sched = ring->scheduler;
+		job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+		job->adev = adev;
+		job->ibs = ibs;
+		job->num_ibs = num_ibs;
+		job->base.owner = owner;
+		mutex_init(&job->job_lock);
+		job->free_job = free_job;
+		mutex_lock(&job->job_lock);
+		r = amd_sched_entity_push_job((struct amd_sched_job *)job);
+		if (r) {
+			mutex_unlock(&job->job_lock);
+			kfree(job);
+			return r;
+		}
+		*f = fence_get(&job->base.s_fence->base);
+		mutex_unlock(&job->job_lock);
+	} else {
+		r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
+		if (r)
+			return r;
+		*f = fence_get(&ibs[num_ibs - 1].fence->base);
+	}
+
+	return 0;
+}

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

@@ -87,7 +87,7 @@ bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring,
 
 void amdgpu_semaphore_free(struct amdgpu_device *adev,
 			   struct amdgpu_semaphore **semaphore,
-			   struct amdgpu_fence *fence)
+			   struct fence *fence)
 {
 	if (semaphore == NULL || *semaphore == NULL) {
 		return;

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

@@ -32,6 +32,11 @@
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
+struct amdgpu_sync_entry {
+	struct hlist_node	node;
+	struct fence		*fence;
+};
+
 /**
  * amdgpu_sync_create - zero init sync object
  *
@@ -49,36 +54,104 @@ void amdgpu_sync_create(struct amdgpu_sync *sync)
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
 		sync->sync_to[i] = NULL;
 
+	hash_init(sync->fences);
 	sync->last_vm_update = NULL;
 }
 
+static bool amdgpu_sync_same_dev(struct amdgpu_device *adev, struct fence *f)
+{
+	struct amdgpu_fence *a_fence = to_amdgpu_fence(f);
+	struct amd_sched_fence *s_fence = to_amd_sched_fence(f);
+
+	if (a_fence)
+		return a_fence->ring->adev == adev;
+	if (s_fence)
+		return (struct amdgpu_device *)s_fence->scheduler->priv == adev;
+	return false;
+}
+
+static bool amdgpu_sync_test_owner(struct fence *f, void *owner)
+{
+	struct amdgpu_fence *a_fence = to_amdgpu_fence(f);
+	struct amd_sched_fence *s_fence = to_amd_sched_fence(f);
+	if (s_fence)
+		return s_fence->owner == owner;
+	if (a_fence)
+		return a_fence->owner == owner;
+	return false;
+}
+
 /**
- * amdgpu_sync_fence - use the semaphore to sync to a fence
+ * amdgpu_sync_fence - remember to sync to this fence
  *
  * @sync: sync object to add fence to
  * @fence: fence to sync to
  *
- * Sync to the fence using the semaphore objects
  */
-void amdgpu_sync_fence(struct amdgpu_sync *sync,
-		       struct amdgpu_fence *fence)
+int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
+		      struct fence *f)
 {
+	struct amdgpu_sync_entry *e;
+	struct amdgpu_fence *fence;
 	struct amdgpu_fence *other;
+	struct fence *tmp, *later;
 
-	if (!fence)
-		return;
+	if (!f)
+		return 0;
+
+	if (amdgpu_sync_same_dev(adev, f) &&
+	    amdgpu_sync_test_owner(f, AMDGPU_FENCE_OWNER_VM)) {
+		if (sync->last_vm_update) {
+			tmp = sync->last_vm_update;
+			BUG_ON(f->context != tmp->context);
+			later = (f->seqno - tmp->seqno <= INT_MAX) ? f : tmp;
+			sync->last_vm_update = fence_get(later);
+			fence_put(tmp);
+		} else
+			sync->last_vm_update = fence_get(f);
+	}
+
+	fence = to_amdgpu_fence(f);
+	if (!fence || fence->ring->adev != adev) {
+		hash_for_each_possible(sync->fences, e, node, f->context) {
+			struct fence *new;
+			if (unlikely(e->fence->context != f->context))
+				continue;
+			new = fence_get(fence_later(e->fence, f));
+			if (new) {
+				fence_put(e->fence);
+				e->fence = new;
+			}
+			return 0;
+		}
+
+		e = kmalloc(sizeof(struct amdgpu_sync_entry), GFP_KERNEL);
+		if (!e)
+			return -ENOMEM;
+
+		hash_add(sync->fences, &e->node, f->context);
+		e->fence = fence_get(f);
+		return 0;
+	}
 
 	other = sync->sync_to[fence->ring->idx];
 	sync->sync_to[fence->ring->idx] = amdgpu_fence_ref(
 		amdgpu_fence_later(fence, other));
 	amdgpu_fence_unref(&other);
 
-	if (fence->owner == AMDGPU_FENCE_OWNER_VM) {
-		other = sync->last_vm_update;
-		sync->last_vm_update = amdgpu_fence_ref(
-			amdgpu_fence_later(fence, other));
-		amdgpu_fence_unref(&other);
-	}
+	return 0;
+}
+
+static void *amdgpu_sync_get_owner(struct fence *f)
+{
+	struct amdgpu_fence *a_fence = to_amdgpu_fence(f);
+	struct amd_sched_fence *s_fence = to_amd_sched_fence(f);
+
+	if (s_fence)
+		return s_fence->owner;
+	else if (a_fence)
+		return a_fence->owner;
+	return AMDGPU_FENCE_OWNER_UNDEFINED;
 }
 
 /**
@@ -97,7 +170,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 {
 	struct reservation_object_list *flist;
 	struct fence *f;
-	struct amdgpu_fence *fence;
+	void *fence_owner;
 	unsigned i;
 	int r = 0;
 
@@ -106,11 +179,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 
 	/* always sync to the exclusive fence */
 	f = reservation_object_get_excl(resv);
-	fence = f ? to_amdgpu_fence(f) : NULL;
-	if (fence && fence->ring->adev == adev)
-		amdgpu_sync_fence(sync, fence);
-	else if (f)
-		r = fence_wait(f, true);
+	r = amdgpu_sync_fence(adev, sync, f);
 
 	flist = reservation_object_get_list(resv);
 	if (!flist || r)
@@ -119,20 +188,72 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 	for (i = 0; i < flist->shared_count; ++i) {
 		f = rcu_dereference_protected(flist->shared[i],
 					      reservation_object_held(resv));
-		fence = f ? to_amdgpu_fence(f) : NULL;
-		if (fence && fence->ring->adev == adev) {
-			if (fence->owner != owner ||
-			    fence->owner == AMDGPU_FENCE_OWNER_UNDEFINED)
-				amdgpu_sync_fence(sync, fence);
-		} else if (f) {
-			r = fence_wait(f, true);
-			if (r)
-				break;
+		if (amdgpu_sync_same_dev(adev, f)) {
+			/* VM updates are only interesting
+			 * for other VM updates and moves.
+			 */
+			fence_owner = amdgpu_sync_get_owner(f);
+			if ((owner != AMDGPU_FENCE_OWNER_MOVE) &&
+			    (fence_owner != AMDGPU_FENCE_OWNER_MOVE) &&
+			    ((owner == AMDGPU_FENCE_OWNER_VM) !=
+			     (fence_owner == AMDGPU_FENCE_OWNER_VM)))
+				continue;
+
+			/* Ignore fence from the same owner as
+			 * long as it isn't undefined.
+			 */
+			if (owner != AMDGPU_FENCE_OWNER_UNDEFINED &&
+			    fence_owner == owner)
+				continue;
 		}
+
+		r = amdgpu_sync_fence(adev, sync, f);
+		if (r)
+			break;
 	}
 	return r;
 }
 
+struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync)
+{
+	struct amdgpu_sync_entry *e;
+	struct hlist_node *tmp;
+	struct fence *f;
+	int i;
+
+	hash_for_each_safe(sync->fences, i, tmp, e, node) {
+
+		f = e->fence;
+
+		hash_del(&e->node);
+		kfree(e);
+
+		if (!fence_is_signaled(f))
+			return f;
+
+		fence_put(f);
+	}
+	return NULL;
+}
+
+int amdgpu_sync_wait(struct amdgpu_sync *sync)
+{
+	struct amdgpu_sync_entry *e;
+	struct hlist_node *tmp;
+	int i, r;
+
+	hash_for_each_safe(sync->fences, i, tmp, e, node) {
+		r = fence_wait(e->fence, false);
+		if (r)
+			return r;
+
+		hash_del(&e->node);
+		fence_put(e->fence);
+		kfree(e);
+	}
+	return 0;
+}
+
 /**
  * amdgpu_sync_rings - sync ring to all registered fences
  *
@@ -164,9 +285,9 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
 			return -EINVAL;
 		}
 
-		if (count >= AMDGPU_NUM_SYNCS) {
+		if (amdgpu_enable_scheduler || (count >= AMDGPU_NUM_SYNCS)) {
 			/* not enough room, wait manually */
-			r = amdgpu_fence_wait(fence, false);
+			r = fence_wait(&fence->base, false);
 			if (r)
 				return r;
 			continue;
@@ -186,7 +307,7 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
 		if (!amdgpu_semaphore_emit_signal(other, semaphore)) {
 			/* signaling wasn't successful wait manually */
 			amdgpu_ring_undo(other);
-			r = amdgpu_fence_wait(fence, false);
+			r = fence_wait(&fence->base, false);
 			if (r)
 				return r;
 			continue;
@@ -196,7 +317,7 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
 		if (!amdgpu_semaphore_emit_wait(ring, semaphore)) {
 			/* waiting wasn't successful wait manually */
 			amdgpu_ring_undo(other);
-			r = amdgpu_fence_wait(fence, false);
+			r = fence_wait(&fence->base, false);
 			if (r)
 				return r;
 			continue;
@@ -220,15 +341,23 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
  */
 void amdgpu_sync_free(struct amdgpu_device *adev,
 		      struct amdgpu_sync *sync,
-		      struct amdgpu_fence *fence)
+		      struct fence *fence)
 {
+	struct amdgpu_sync_entry *e;
+	struct hlist_node *tmp;
 	unsigned i;
 
+	hash_for_each_safe(sync->fences, i, tmp, e, node) {
+		hash_del(&e->node);
+		fence_put(e->fence);
+		kfree(e);
+	}
+
 	for (i = 0; i < AMDGPU_NUM_SYNCS; ++i)
 		amdgpu_semaphore_free(adev, &sync->semaphores[i], fence);
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
 		amdgpu_fence_unref(&sync->sync_to[i]);
 
-	amdgpu_fence_unref(&sync->last_vm_update);
+	fence_put(sync->last_vm_update);
 }

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

@@ -77,7 +77,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 		void *gtt_map, *vram_map;
 		void **gtt_start, **gtt_end;
 		void **vram_start, **vram_end;
-		struct amdgpu_fence *fence = NULL;
+		struct fence *fence = NULL;
 
 		r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
 				     AMDGPU_GEM_DOMAIN_GTT, 0, NULL, gtt_obj + i);
@@ -116,13 +116,13 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 			goto out_lclean_unpin;
 		}
 
-		r = amdgpu_fence_wait(fence, false);
+		r = fence_wait(fence, false);
 		if (r) {
 			DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
 			goto out_lclean_unpin;
 		}
 
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 
 		r = amdgpu_bo_kmap(vram_obj, &vram_map);
 		if (r) {
@@ -161,13 +161,13 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 			goto out_lclean_unpin;
 		}
 
-		r = amdgpu_fence_wait(fence, false);
+		r = fence_wait(fence, false);
 		if (r) {
 			DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
 			goto out_lclean_unpin;
 		}
 
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 
 		r = amdgpu_bo_kmap(gtt_obj[i], &gtt_map);
 		if (r) {
@@ -214,7 +214,7 @@ out_lclean:
 			amdgpu_bo_unref(&gtt_obj[i]);
 		}
 		if (fence)
-			amdgpu_fence_unref(&fence);
+			fence_put(fence);
 		break;
 	}
 
@@ -238,7 +238,7 @@ void amdgpu_test_moves(struct amdgpu_device *adev)
 
 static int amdgpu_test_create_and_emit_fence(struct amdgpu_device *adev,
 					     struct amdgpu_ring *ring,
-					     struct amdgpu_fence **fence)
+					     struct fence **fence)
 {
 	uint32_t handle = ring->idx ^ 0xdeafbeef;
 	int r;
@@ -269,15 +269,16 @@ static int amdgpu_test_create_and_emit_fence(struct amdgpu_device *adev,
 			DRM_ERROR("Failed to get dummy destroy msg\n");
 			return r;
 		}
-
 	} else {
+		struct amdgpu_fence *a_fence = NULL;
 		r = amdgpu_ring_lock(ring, 64);
 		if (r) {
 			DRM_ERROR("Failed to lock ring A %d\n", ring->idx);
 			return r;
 		}
-		amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_UNDEFINED, fence);
+		amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_UNDEFINED, &a_fence);
 		amdgpu_ring_unlock_commit(ring);
+		*fence = &a_fence->base;
 	}
 	return 0;
 }
@@ -286,7 +287,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
 			   struct amdgpu_ring *ringA,
 			   struct amdgpu_ring *ringB)
 {
-	struct amdgpu_fence *fence1 = NULL, *fence2 = NULL;
+	struct fence *fence1 = NULL, *fence2 = NULL;
 	struct amdgpu_semaphore *semaphore = NULL;
 	int r;
 
@@ -322,7 +323,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
 
 	mdelay(1000);
 
-	if (amdgpu_fence_signaled(fence1)) {
+	if (fence_is_signaled(fence1)) {
 		DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
 		goto out_cleanup;
 	}
@@ -335,7 +336,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
 	amdgpu_semaphore_emit_signal(ringB, semaphore);
 	amdgpu_ring_unlock_commit(ringB);
 
-	r = amdgpu_fence_wait(fence1, false);
+	r = fence_wait(fence1, false);
 	if (r) {
 		DRM_ERROR("Failed to wait for sync fence 1\n");
 		goto out_cleanup;
@@ -343,7 +344,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
 
 	mdelay(1000);
 
-	if (amdgpu_fence_signaled(fence2)) {
+	if (fence_is_signaled(fence2)) {
 		DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
 		goto out_cleanup;
 	}
@@ -356,7 +357,7 @@ void amdgpu_test_ring_sync(struct amdgpu_device *adev,
 	amdgpu_semaphore_emit_signal(ringB, semaphore);
 	amdgpu_ring_unlock_commit(ringB);
 
-	r = amdgpu_fence_wait(fence2, false);
+	r = fence_wait(fence2, false);
 	if (r) {
 		DRM_ERROR("Failed to wait for sync fence 1\n");
 		goto out_cleanup;
@@ -366,10 +367,10 @@ out_cleanup:
 	amdgpu_semaphore_free(adev, &semaphore, NULL);
 
 	if (fence1)
-		amdgpu_fence_unref(&fence1);
+		fence_put(fence1);
 
 	if (fence2)
-		amdgpu_fence_unref(&fence2);
+		fence_put(fence2);
 
 	if (r)
 		printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
@@ -380,7 +381,7 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
 			    struct amdgpu_ring *ringB,
 			    struct amdgpu_ring *ringC)
 {
-	struct amdgpu_fence *fenceA = NULL, *fenceB = NULL;
+	struct fence *fenceA = NULL, *fenceB = NULL;
 	struct amdgpu_semaphore *semaphore = NULL;
 	bool sigA, sigB;
 	int i, r;
@@ -416,11 +417,11 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
 
 	mdelay(1000);
 
-	if (amdgpu_fence_signaled(fenceA)) {
+	if (fence_is_signaled(fenceA)) {
 		DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
 		goto out_cleanup;
 	}
-	if (amdgpu_fence_signaled(fenceB)) {
+	if (fence_is_signaled(fenceB)) {
 		DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
 		goto out_cleanup;
 	}
@@ -435,8 +436,8 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
 
 	for (i = 0; i < 30; ++i) {
 		mdelay(100);
-		sigA = amdgpu_fence_signaled(fenceA);
-		sigB = amdgpu_fence_signaled(fenceB);
+		sigA = fence_is_signaled(fenceA);
+		sigB = fence_is_signaled(fenceB);
 		if (sigA || sigB)
 			break;
 	}
@@ -461,12 +462,12 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev,
 
 	mdelay(1000);
 
-	r = amdgpu_fence_wait(fenceA, false);
+	r = fence_wait(fenceA, false);
 	if (r) {
 		DRM_ERROR("Failed to wait for sync fence A\n");
 		goto out_cleanup;
 	}
-	r = amdgpu_fence_wait(fenceB, false);
+	r = fence_wait(fenceB, false);
 	if (r) {
 		DRM_ERROR("Failed to wait for sync fence B\n");
 		goto out_cleanup;
@@ -476,10 +477,10 @@ out_cleanup:
 	amdgpu_semaphore_free(adev, &semaphore, NULL);
 
 	if (fenceA)
-		amdgpu_fence_unref(&fenceA);
+		fence_put(fenceA);
 
 	if (fenceB)
-		amdgpu_fence_unref(&fenceB);
+		fence_put(fenceB);
 
 	if (r)
 		printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);

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

@@ -228,7 +228,7 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 	struct amdgpu_device *adev;
 	struct amdgpu_ring *ring;
 	uint64_t old_start, new_start;
-	struct amdgpu_fence *fence;
+	struct fence *fence;
 	int r;
 
 	adev = amdgpu_get_adev(bo->bdev);
@@ -269,9 +269,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 			       new_mem->num_pages * PAGE_SIZE, /* bytes */
 			       bo->resv, &fence);
 	/* FIXME: handle copy error */
-	r = ttm_bo_move_accel_cleanup(bo, &fence->base,
+	r = ttm_bo_move_accel_cleanup(bo, fence,
 				      evict, no_wait_gpu, new_mem);
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	return r;
 }
 
@@ -859,7 +859,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
 
 	r = amdgpu_bo_create(adev, 256 * 1024, PAGE_SIZE, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0,
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
 			     NULL, &adev->stollen_vga_memory);
 	if (r) {
 		return r;
@@ -987,46 +988,48 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 		       uint64_t dst_offset,
 		       uint32_t byte_count,
 		       struct reservation_object *resv,
-		       struct amdgpu_fence **fence)
+		       struct fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_sync sync;
 	uint32_t max_bytes;
 	unsigned num_loops, num_dw;
+	struct amdgpu_ib *ib;
 	unsigned i;
 	int r;
 
-	/* sync other rings */
-	amdgpu_sync_create(&sync);
-	if (resv) {
-		r = amdgpu_sync_resv(adev, &sync, resv, false);
-		if (r) {
-			DRM_ERROR("sync failed (%d).\n", r);
-			amdgpu_sync_free(adev, &sync, NULL);
-			return r;
-		}
-	}
-
 	max_bytes = adev->mman.buffer_funcs->copy_max_bytes;
 	num_loops = DIV_ROUND_UP(byte_count, max_bytes);
 	num_dw = num_loops * adev->mman.buffer_funcs->copy_num_dw;
 
-	/* for fence and sync */
-	num_dw += 64 + AMDGPU_NUM_SYNCS * 8;
+	/* for IB padding */
+	while (num_dw & 0x7)
+		num_dw++;
+
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
 
-	r = amdgpu_ring_lock(ring, num_dw);
+	r = amdgpu_ib_get(ring, NULL, num_dw * 4, ib);
 	if (r) {
-		DRM_ERROR("ring lock failed (%d).\n", r);
-		amdgpu_sync_free(adev, &sync, NULL);
+		kfree(ib);
 		return r;
 	}
 
-	amdgpu_sync_rings(&sync, ring);
+	ib->length_dw = 0;
+
+	if (resv) {
+		r = amdgpu_sync_resv(adev, &ib->sync, resv,
+				     AMDGPU_FENCE_OWNER_UNDEFINED);
+		if (r) {
+			DRM_ERROR("sync failed (%d).\n", r);
+			goto error_free;
+		}
+	}
 
 	for (i = 0; i < num_loops; i++) {
 		uint32_t cur_size_in_bytes = min(byte_count, max_bytes);
 
-		amdgpu_emit_copy_buffer(adev, ring, src_offset, dst_offset,
+		amdgpu_emit_copy_buffer(adev, ib, src_offset, dst_offset,
 					cur_size_in_bytes);
 
 		src_offset += cur_size_in_bytes;
@@ -1034,17 +1037,24 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 		byte_count -= cur_size_in_bytes;
 	}
 
-	r = amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_MOVE, fence);
-	if (r) {
-		amdgpu_ring_unlock_undo(ring);
-		amdgpu_sync_free(adev, &sync, NULL);
-		return r;
-	}
-
-	amdgpu_ring_unlock_commit(ring);
-	amdgpu_sync_free(adev, &sync, *fence);
+	amdgpu_vm_pad_ib(adev, ib);
+	WARN_ON(ib->length_dw > num_dw);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vm_free_job,
+						 AMDGPU_FENCE_OWNER_MOVE,
+						 fence);
+	if (r)
+		goto error_free;
 
+	if (!amdgpu_enable_scheduler) {
+		amdgpu_ib_free(adev, ib);
+		kfree(ib);
+	}
 	return 0;
+error_free:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
+	return r;
 }
 
 #if defined(CONFIG_DEBUG_FS)

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

@@ -52,6 +52,7 @@
 #endif
 #define FIRMWARE_TONGA		"amdgpu/tonga_uvd.bin"
 #define FIRMWARE_CARRIZO	"amdgpu/carrizo_uvd.bin"
+#define FIRMWARE_FIJI		"amdgpu/fiji_uvd.bin"
 
 /**
  * amdgpu_uvd_cs_ctx - Command submission parser context
@@ -81,6 +82,7 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS);
 #endif
 MODULE_FIRMWARE(FIRMWARE_TONGA);
 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
+MODULE_FIRMWARE(FIRMWARE_FIJI);
 
 static void amdgpu_uvd_note_usage(struct amdgpu_device *adev);
 static void amdgpu_uvd_idle_work_handler(struct work_struct *work);
@@ -116,6 +118,9 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 	case CHIP_TONGA:
 		fw_name = FIRMWARE_TONGA;
 		break;
+	case CHIP_FIJI:
+		fw_name = FIRMWARE_FIJI;
+		break;
 	case CHIP_CARRIZO:
 		fw_name = FIRMWARE_CARRIZO;
 		break;
@@ -149,7 +154,9 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 	bo_size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8)
 		 +  AMDGPU_UVD_STACK_SIZE + AMDGPU_UVD_HEAP_SIZE;
 	r = amdgpu_bo_create(adev, bo_size, PAGE_SIZE, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->uvd.vcpu_bo);
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     NULL, &adev->uvd.vcpu_bo);
 	if (r) {
 		dev_err(adev->dev, "(%d) failed to allocate UVD bo\n", r);
 		return r;
@@ -216,31 +223,32 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 
 int amdgpu_uvd_suspend(struct amdgpu_device *adev)
 {
-	unsigned size;
-	void *ptr;
-	const struct common_firmware_header *hdr;
-	int i;
+	struct amdgpu_ring *ring = &adev->uvd.ring;
+	int i, r;
 
 	if (adev->uvd.vcpu_bo == NULL)
 		return 0;
 
-	for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i)
-		if (atomic_read(&adev->uvd.handles[i]))
-			break;
-
-	if (i == AMDGPU_MAX_UVD_HANDLES)
-		return 0;
+	for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
+		uint32_t handle = atomic_read(&adev->uvd.handles[i]);
+		if (handle != 0) {
+			struct fence *fence;
 
-	hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
+			amdgpu_uvd_note_usage(adev);
 
-	size = amdgpu_bo_size(adev->uvd.vcpu_bo);
-	size -= le32_to_cpu(hdr->ucode_size_bytes);
+			r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence);
+			if (r) {
+				DRM_ERROR("Error destroying UVD (%d)!\n", r);
+				continue;
+			}
 
-	ptr = adev->uvd.cpu_addr;
-	ptr += le32_to_cpu(hdr->ucode_size_bytes);
+			fence_wait(fence, false);
+			fence_put(fence);
 
-	adev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
-	memcpy(adev->uvd.saved_bo, ptr, size);
+			adev->uvd.filp[i] = NULL;
+			atomic_set(&adev->uvd.handles[i], 0);
+		}
+	}
 
 	return 0;
 }
@@ -265,12 +273,7 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev)
 	ptr = adev->uvd.cpu_addr;
 	ptr += le32_to_cpu(hdr->ucode_size_bytes);
 
-	if (adev->uvd.saved_bo != NULL) {
-		memcpy(ptr, adev->uvd.saved_bo, size);
-		kfree(adev->uvd.saved_bo);
-		adev->uvd.saved_bo = NULL;
-	} else
-		memset(ptr, 0, size);
+	memset(ptr, 0, size);
 
 	return 0;
 }
@@ -283,7 +286,7 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 	for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
 		uint32_t handle = atomic_read(&adev->uvd.handles[i]);
 		if (handle != 0 && adev->uvd.filp[i] == filp) {
-			struct amdgpu_fence *fence;
+			struct fence *fence;
 
 			amdgpu_uvd_note_usage(adev);
 
@@ -293,8 +296,8 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 				continue;
 			}
 
-			amdgpu_fence_wait(fence, false);
-			amdgpu_fence_unref(&fence);
+			fence_wait(fence, false);
+			fence_put(fence);
 
 			adev->uvd.filp[i] = NULL;
 			atomic_set(&adev->uvd.handles[i], 0);
@@ -374,7 +377,8 @@ static int amdgpu_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
 	unsigned height_in_mb = ALIGN(height / 16, 2);
 	unsigned fs_in_mb = width_in_mb * height_in_mb;
 
-	unsigned image_size, tmp, min_dpb_size, num_dpb_buffer, min_ctx_size;
+	unsigned image_size, tmp, min_dpb_size, num_dpb_buffer;
+	unsigned min_ctx_size = 0;
 
 	image_size = width * height;
 	image_size += image_size / 2;
@@ -507,28 +511,25 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
 {
 	struct amdgpu_device *adev = ctx->parser->adev;
 	int32_t *msg, msg_type, handle;
-	struct fence *f;
 	void *ptr;
-
-	int i, r;
+	long r;
+	int i;
 
 	if (offset & 0x3F) {
 		DRM_ERROR("UVD messages must be 64 byte aligned!\n");
 		return -EINVAL;
 	}
 
-	f = reservation_object_get_excl(bo->tbo.resv);
-	if (f) {
-		r = amdgpu_fence_wait((struct amdgpu_fence *)f, false);
-		if (r) {
-			DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
-			return r;
-		}
+	r = reservation_object_wait_timeout_rcu(bo->tbo.resv, true, false,
+						MAX_SCHEDULE_TIMEOUT);
+	if (r < 0) {
+		DRM_ERROR("Failed waiting for UVD message (%ld)!\n", r);
+		return r;
 	}
 
 	r = amdgpu_bo_kmap(bo, &ptr);
 	if (r) {
-		DRM_ERROR("Failed mapping the UVD message (%d)!\n", r);
+		DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r);
 		return r;
 	}
 
@@ -803,14 +804,24 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
 	return 0;
 }
 
+static int amdgpu_uvd_free_job(
+	struct amdgpu_job *sched_job)
+{
+	amdgpu_ib_free(sched_job->adev, sched_job->ibs);
+	kfree(sched_job->ibs);
+	return 0;
+}
+
 static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 			       struct amdgpu_bo *bo,
-			       struct amdgpu_fence **fence)
+			       struct fence **fence)
 {
 	struct ttm_validate_buffer tv;
 	struct ww_acquire_ctx ticket;
 	struct list_head head;
-	struct amdgpu_ib ib;
+	struct amdgpu_ib *ib = NULL;
+	struct fence *f = NULL;
+	struct amdgpu_device *adev = ring->adev;
 	uint64_t addr;
 	int i, r;
 
@@ -832,34 +843,49 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 	r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
 	if (r)
 		goto err;
-
-	r = amdgpu_ib_get(ring, NULL, 64, &ib);
-	if (r)
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib) {
+		r = -ENOMEM;
 		goto err;
+	}
+	r = amdgpu_ib_get(ring, NULL, 64, ib);
+	if (r)
+		goto err1;
 
 	addr = amdgpu_bo_gpu_offset(bo);
-	ib.ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
-	ib.ptr[1] = addr;
-	ib.ptr[2] = PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0);
-	ib.ptr[3] = addr >> 32;
-	ib.ptr[4] = PACKET0(mmUVD_GPCOM_VCPU_CMD, 0);
-	ib.ptr[5] = 0;
+	ib->ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
+	ib->ptr[1] = addr;
+	ib->ptr[2] = PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0);
+	ib->ptr[3] = addr >> 32;
+	ib->ptr[4] = PACKET0(mmUVD_GPCOM_VCPU_CMD, 0);
+	ib->ptr[5] = 0;
 	for (i = 6; i < 16; ++i)
-		ib.ptr[i] = PACKET2(0);
-	ib.length_dw = 16;
+		ib->ptr[i] = PACKET2(0);
+	ib->length_dw = 16;
 
-	r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_uvd_free_job,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
 	if (r)
-		goto err;
-	ttm_eu_fence_buffer_objects(&ticket, &head, &ib.fence->base);
+		goto err2;
 
-	if (fence)
-		*fence = amdgpu_fence_ref(ib.fence);
+	ttm_eu_fence_buffer_objects(&ticket, &head, f);
 
-	amdgpu_ib_free(ring->adev, &ib);
+	if (fence)
+		*fence = fence_get(f);
 	amdgpu_bo_unref(&bo);
-	return 0;
+	fence_put(f);
+	if (amdgpu_enable_scheduler)
+		return 0;
 
+	amdgpu_ib_free(ring->adev, ib);
+	kfree(ib);
+	return 0;
+err2:
+	amdgpu_ib_free(ring->adev, ib);
+err1:
+	kfree(ib);
 err:
 	ttm_eu_backoff_reservation(&ticket, &head);
 	return r;
@@ -869,7 +895,7 @@ err:
    crash the vcpu so just try to emmit a dummy create/destroy msg to
    avoid this */
 int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
-			      struct amdgpu_fence **fence)
+			      struct fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_bo *bo;
@@ -877,7 +903,9 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	int r, i;
 
 	r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &bo);
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     NULL, &bo);
 	if (r)
 		return r;
 
@@ -916,7 +944,7 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 }
 
 int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
-			       struct amdgpu_fence **fence)
+			       struct fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_bo *bo;
@@ -924,7 +952,9 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 	int r, i;
 
 	r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &bo);
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     NULL, &bo);
 	if (r)
 		return r;
 

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

@@ -29,9 +29,9 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev);
 int amdgpu_uvd_suspend(struct amdgpu_device *adev);
 int amdgpu_uvd_resume(struct amdgpu_device *adev);
 int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
-			      struct amdgpu_fence **fence);
+			      struct fence **fence);
 int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
-			       struct amdgpu_fence **fence);
+			       struct fence **fence);
 void amdgpu_uvd_free_handles(struct amdgpu_device *adev,
 			     struct drm_file *filp);
 int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx);

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

@@ -48,6 +48,7 @@
 #endif
 #define FIRMWARE_TONGA		"amdgpu/tonga_vce.bin"
 #define FIRMWARE_CARRIZO	"amdgpu/carrizo_vce.bin"
+#define FIRMWARE_FIJI		"amdgpu/fiji_vce.bin"
 
 #ifdef CONFIG_DRM_AMDGPU_CIK
 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
@@ -58,6 +59,7 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS);
 #endif
 MODULE_FIRMWARE(FIRMWARE_TONGA);
 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
+MODULE_FIRMWARE(FIRMWARE_FIJI);
 
 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
 
@@ -101,6 +103,9 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
 	case CHIP_CARRIZO:
 		fw_name = FIRMWARE_CARRIZO;
 		break;
+	case CHIP_FIJI:
+		fw_name = FIRMWARE_FIJI;
+		break;
 
 	default:
 		return -EINVAL;
@@ -136,7 +141,9 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
 	/* allocate firmware, stack and heap BO */
 
 	r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->vce.vcpu_bo);
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     NULL, &adev->vce.vcpu_bo);
 	if (r) {
 		dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
 		return r;
@@ -334,6 +341,14 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 	}
 }
 
+static int amdgpu_vce_free_job(
+	struct amdgpu_job *sched_job)
+{
+	amdgpu_ib_free(sched_job->adev, sched_job->ibs);
+	kfree(sched_job->ibs);
+	return 0;
+}
+
 /**
  * amdgpu_vce_get_create_msg - generate a VCE create msg
  *
@@ -345,59 +360,69 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
  * Open up a stream for HW test
  */
 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
-			      struct amdgpu_fence **fence)
+			      struct fence **fence)
 {
 	const unsigned ib_size_dw = 1024;
-	struct amdgpu_ib ib;
+	struct amdgpu_ib *ib = NULL;
+	struct fence *f = NULL;
+	struct amdgpu_device *adev = ring->adev;
 	uint64_t dummy;
 	int i, r;
 
-	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
+	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
+		kfree(ib);
 		return r;
 	}
 
-	dummy = ib.gpu_addr + 1024;
+	dummy = ib->gpu_addr + 1024;
 
 	/* stitch together an VCE create msg */
-	ib.length_dw = 0;
-	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
-	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
-	ib.ptr[ib.length_dw++] = handle;
-
-	ib.ptr[ib.length_dw++] = 0x00000030; /* len */
-	ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */
-	ib.ptr[ib.length_dw++] = 0x00000000;
-	ib.ptr[ib.length_dw++] = 0x00000042;
-	ib.ptr[ib.length_dw++] = 0x0000000a;
-	ib.ptr[ib.length_dw++] = 0x00000001;
-	ib.ptr[ib.length_dw++] = 0x00000080;
-	ib.ptr[ib.length_dw++] = 0x00000060;
-	ib.ptr[ib.length_dw++] = 0x00000100;
-	ib.ptr[ib.length_dw++] = 0x00000100;
-	ib.ptr[ib.length_dw++] = 0x0000000c;
-	ib.ptr[ib.length_dw++] = 0x00000000;
-
-	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
-	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
-	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
-	ib.ptr[ib.length_dw++] = dummy;
-	ib.ptr[ib.length_dw++] = 0x00000001;
-
-	for (i = ib.length_dw; i < ib_size_dw; ++i)
-		ib.ptr[i] = 0x0;
-
-	r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-	}
-
+	ib->length_dw = 0;
+	ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
+	ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
+	ib->ptr[ib->length_dw++] = handle;
+
+	ib->ptr[ib->length_dw++] = 0x00000030; /* len */
+	ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
+	ib->ptr[ib->length_dw++] = 0x00000000;
+	ib->ptr[ib->length_dw++] = 0x00000042;
+	ib->ptr[ib->length_dw++] = 0x0000000a;
+	ib->ptr[ib->length_dw++] = 0x00000001;
+	ib->ptr[ib->length_dw++] = 0x00000080;
+	ib->ptr[ib->length_dw++] = 0x00000060;
+	ib->ptr[ib->length_dw++] = 0x00000100;
+	ib->ptr[ib->length_dw++] = 0x00000100;
+	ib->ptr[ib->length_dw++] = 0x0000000c;
+	ib->ptr[ib->length_dw++] = 0x00000000;
+
+	ib->ptr[ib->length_dw++] = 0x00000014; /* len */
+	ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
+	ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
+	ib->ptr[ib->length_dw++] = dummy;
+	ib->ptr[ib->length_dw++] = 0x00000001;
+
+	for (i = ib->length_dw; i < ib_size_dw; ++i)
+		ib->ptr[i] = 0x0;
+
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vce_free_job,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err;
 	if (fence)
-		*fence = amdgpu_fence_ref(ib.fence);
-
-	amdgpu_ib_free(ring->adev, &ib);
-
+		*fence = fence_get(f);
+	fence_put(f);
+	if (amdgpu_enable_scheduler)
+		return 0;
+err:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
 	return r;
 }
 
@@ -412,49 +437,59 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
  * Close up a stream for HW test or if userspace failed to do so
  */
 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
-			       struct amdgpu_fence **fence)
+			       struct fence **fence)
 {
 	const unsigned ib_size_dw = 1024;
-	struct amdgpu_ib ib;
+	struct amdgpu_ib *ib = NULL;
+	struct fence *f = NULL;
+	struct amdgpu_device *adev = ring->adev;
 	uint64_t dummy;
 	int i, r;
 
-	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
+
+	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
 	if (r) {
+		kfree(ib);
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		return r;
 	}
 
-	dummy = ib.gpu_addr + 1024;
+	dummy = ib->gpu_addr + 1024;
 
 	/* stitch together an VCE destroy msg */
-	ib.length_dw = 0;
-	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
-	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
-	ib.ptr[ib.length_dw++] = handle;
-
-	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
-	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
-	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
-	ib.ptr[ib.length_dw++] = dummy;
-	ib.ptr[ib.length_dw++] = 0x00000001;
-
-	ib.ptr[ib.length_dw++] = 0x00000008; /* len */
-	ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */
-
-	for (i = ib.length_dw; i < ib_size_dw; ++i)
-		ib.ptr[i] = 0x0;
-
-	r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-	}
-
+	ib->length_dw = 0;
+	ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
+	ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
+	ib->ptr[ib->length_dw++] = handle;
+
+	ib->ptr[ib->length_dw++] = 0x00000014; /* len */
+	ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
+	ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
+	ib->ptr[ib->length_dw++] = dummy;
+	ib->ptr[ib->length_dw++] = 0x00000001;
+
+	ib->ptr[ib->length_dw++] = 0x00000008; /* len */
+	ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */
+
+	for (i = ib->length_dw; i < ib_size_dw; ++i)
+		ib->ptr[i] = 0x0;
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vce_free_job,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err;
 	if (fence)
-		*fence = amdgpu_fence_ref(ib.fence);
-
-	amdgpu_ib_free(ring->adev, &ib);
-
+		*fence = fence_get(f);
+	fence_put(f);
+	if (amdgpu_enable_scheduler)
+		return 0;
+err:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
 	return r;
 }
 
@@ -800,9 +835,13 @@ int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
  */
 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
 {
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int r;
 
+	/* skip vce ring1 ib test for now, since it's not reliable */
+	if (ring == &ring->adev->vce.ring[1])
+		return 0;
+
 	r = amdgpu_vce_get_create_msg(ring, 1, NULL);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r);
@@ -815,13 +854,13 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
 		goto error;
 	}
 
-	r = amdgpu_fence_wait(fence, false);
+	r = fence_wait(fence, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
 	} else {
 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
 	}
 error:
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	return r;
 }

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

@@ -29,9 +29,9 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev);
 int amdgpu_vce_suspend(struct amdgpu_device *adev);
 int amdgpu_vce_resume(struct amdgpu_device *adev);
 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
-			      struct amdgpu_fence **fence);
+			      struct fence **fence);
 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
-			       struct amdgpu_fence **fence);
+			       struct fence **fence);
 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp);
 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx);
 bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring,

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

@@ -127,16 +127,16 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
 /**
  * amdgpu_vm_grab_id - allocate the next free VMID
  *
- * @ring: ring we want to submit job to
  * @vm: vm to allocate id for
+ * @ring: ring we want to submit job to
+ * @sync: sync object where we add dependencies
  *
- * Allocate an id for the vm (cayman+).
- * Returns the fence we need to sync to (if any).
+ * Allocate an id for the vm, adding fences to the sync obj as necessary.
  *
- * Global and local mutex must be locked!
+ * Global mutex must be locked!
  */
-struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
-				       struct amdgpu_vm *vm)
+int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
+		      struct amdgpu_sync *sync)
 {
 	struct amdgpu_fence *best[AMDGPU_MAX_RINGS] = {};
 	struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
@@ -148,7 +148,7 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
 	/* check if the id is still valid */
 	if (vm_id->id && vm_id->last_id_use &&
 	    vm_id->last_id_use == adev->vm_manager.active[vm_id->id])
-		return NULL;
+		return 0;
 
 	/* we definately need to flush */
 	vm_id->pd_gpu_addr = ~0ll;
@@ -161,7 +161,7 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
 			/* found a free one */
 			vm_id->id = i;
 			trace_amdgpu_vm_grab_id(i, ring->idx);
-			return NULL;
+			return 0;
 		}
 
 		if (amdgpu_fence_is_earlier(fence, best[fence->ring->idx])) {
@@ -172,15 +172,19 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
 
 	for (i = 0; i < 2; ++i) {
 		if (choices[i]) {
+			struct amdgpu_fence *fence;
+
+			fence  = adev->vm_manager.active[choices[i]];
 			vm_id->id = choices[i];
+
 			trace_amdgpu_vm_grab_id(choices[i], ring->idx);
-			return adev->vm_manager.active[choices[i]];
+			return amdgpu_sync_fence(ring->adev, sync, &fence->base);
 		}
 	}
 
 	/* should never happen */
 	BUG();
-	return NULL;
+	return -EINVAL;
 }
 
 /**
@@ -196,17 +200,29 @@ struct amdgpu_fence *amdgpu_vm_grab_id(struct amdgpu_ring *ring,
  */
 void amdgpu_vm_flush(struct amdgpu_ring *ring,
 		     struct amdgpu_vm *vm,
-		     struct amdgpu_fence *updates)
+		     struct fence *updates)
 {
 	uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory);
 	struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx];
+	struct fence *flushed_updates = vm_id->flushed_updates;
+	bool is_earlier = false;
 
-	if (pd_addr != vm_id->pd_gpu_addr || !vm_id->flushed_updates ||
-	    amdgpu_fence_is_earlier(vm_id->flushed_updates, updates)) {
+	if (flushed_updates && updates) {
+		BUG_ON(flushed_updates->context != updates->context);
+		is_earlier = (updates->seqno - flushed_updates->seqno <=
+			      INT_MAX) ? true : false;
+	}
+
+	if (pd_addr != vm_id->pd_gpu_addr || !flushed_updates ||
+	    is_earlier) {
 
 		trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id);
-		amdgpu_fence_unref(&vm_id->flushed_updates);
-		vm_id->flushed_updates = amdgpu_fence_ref(updates);
+		if (is_earlier) {
+			vm_id->flushed_updates = fence_get(updates);
+			fence_put(flushed_updates);
+		}
+		if (!flushed_updates)
+			vm_id->flushed_updates = fence_get(updates);
 		vm_id->pd_gpu_addr = pd_addr;
 		amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr);
 	}
@@ -300,6 +316,15 @@ static void amdgpu_vm_update_pages(struct amdgpu_device *adev,
 	}
 }
 
+int amdgpu_vm_free_job(struct amdgpu_job *sched_job)
+{
+	int i;
+	for (i = 0; i < sched_job->num_ibs; i++)
+		amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
+	kfree(sched_job->ibs);
+	return 0;
+}
+
 /**
  * amdgpu_vm_clear_bo - initially clear the page dir/table
  *
@@ -310,7 +335,8 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 			      struct amdgpu_bo *bo)
 {
 	struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
-	struct amdgpu_ib ib;
+	struct fence *fence = NULL;
+	struct amdgpu_ib *ib;
 	unsigned entries;
 	uint64_t addr;
 	int r;
@@ -330,24 +356,33 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	addr = amdgpu_bo_gpu_offset(bo);
 	entries = amdgpu_bo_size(bo) / 8;
 
-	r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, &ib);
-	if (r)
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
 		goto error_unreserve;
 
-	ib.length_dw = 0;
-
-	amdgpu_vm_update_pages(adev, &ib, addr, 0, entries, 0, 0, 0);
-	amdgpu_vm_pad_ib(adev, &ib);
-	WARN_ON(ib.length_dw > 64);
-
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
+	r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, ib);
 	if (r)
 		goto error_free;
 
-	amdgpu_bo_fence(bo, ib.fence, true);
-
+	ib->length_dw = 0;
+
+	amdgpu_vm_update_pages(adev, ib, addr, 0, entries, 0, 0, 0);
+	amdgpu_vm_pad_ib(adev, ib);
+	WARN_ON(ib->length_dw > 64);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vm_free_job,
+						 AMDGPU_FENCE_OWNER_VM,
+						 &fence);
+	if (!r)
+		amdgpu_bo_fence(bo, fence, true);
+	fence_put(fence);
+	if (amdgpu_enable_scheduler) {
+		amdgpu_bo_unreserve(bo);
+		return 0;
+	}
 error_free:
-	amdgpu_ib_free(adev, &ib);
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
 
 error_unreserve:
 	amdgpu_bo_unreserve(bo);
@@ -400,7 +435,9 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	uint32_t incr = AMDGPU_VM_PTE_COUNT * 8;
 	uint64_t last_pde = ~0, last_pt = ~0;
 	unsigned count = 0, pt_idx, ndw;
-	struct amdgpu_ib ib;
+	struct amdgpu_ib *ib;
+	struct fence *fence = NULL;
+
 	int r;
 
 	/* padding, etc. */
@@ -413,10 +450,14 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	if (ndw > 0xfffff)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, ndw * 4, &ib);
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
+
+	r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
 	if (r)
 		return r;
-	ib.length_dw = 0;
+	ib->length_dw = 0;
 
 	/* walk over the address space and update the page directory */
 	for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) {
@@ -436,7 +477,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 		    ((last_pt + incr * count) != pt)) {
 
 			if (count) {
-				amdgpu_vm_update_pages(adev, &ib, last_pde,
+				amdgpu_vm_update_pages(adev, ib, last_pde,
 						       last_pt, count, incr,
 						       AMDGPU_PTE_VALID, 0);
 			}
@@ -450,23 +491,37 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	}
 
 	if (count)
-		amdgpu_vm_update_pages(adev, &ib, last_pde, last_pt, count,
+		amdgpu_vm_update_pages(adev, ib, last_pde, last_pt, count,
 				       incr, AMDGPU_PTE_VALID, 0);
 
-	if (ib.length_dw != 0) {
-		amdgpu_vm_pad_ib(adev, &ib);
-		amdgpu_sync_resv(adev, &ib.sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
-		WARN_ON(ib.length_dw > ndw);
-		r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
-		if (r) {
-			amdgpu_ib_free(adev, &ib);
-			return r;
-		}
-		amdgpu_bo_fence(pd, ib.fence, true);
+	if (ib->length_dw != 0) {
+		amdgpu_vm_pad_ib(adev, ib);
+		amdgpu_sync_resv(adev, &ib->sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
+		WARN_ON(ib->length_dw > ndw);
+		r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+							 &amdgpu_vm_free_job,
+							 AMDGPU_FENCE_OWNER_VM,
+							 &fence);
+		if (r)
+			goto error_free;
+
+		amdgpu_bo_fence(pd, fence, true);
+		fence_put(vm->page_directory_fence);
+		vm->page_directory_fence = fence_get(fence);
+		fence_put(fence);
+	}
+
+	if (!amdgpu_enable_scheduler || ib->length_dw == 0) {
+		amdgpu_ib_free(adev, ib);
+		kfree(ib);
 	}
-	amdgpu_ib_free(adev, &ib);
 
 	return 0;
+
+error_free:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
+	return r;
 }
 
 /**
@@ -572,9 +627,14 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
 {
 	uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
 	uint64_t last_pte = ~0, last_dst = ~0;
+	void *owner = AMDGPU_FENCE_OWNER_VM;
 	unsigned count = 0;
 	uint64_t addr;
 
+	/* sync to everything on unmapping */
+	if (!(flags & AMDGPU_PTE_VALID))
+		owner = AMDGPU_FENCE_OWNER_UNDEFINED;
+
 	/* walk over the address space and update the page tables */
 	for (addr = start; addr < end; ) {
 		uint64_t pt_idx = addr >> amdgpu_vm_block_size;
@@ -583,8 +643,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
 		uint64_t pte;
 		int r;
 
-		amdgpu_sync_resv(adev, &ib->sync, pt->tbo.resv,
-				 AMDGPU_FENCE_OWNER_VM);
+		amdgpu_sync_resv(adev, &ib->sync, pt->tbo.resv, owner);
 		r = reservation_object_reserve_shared(pt->tbo.resv);
 		if (r)
 			return r;
@@ -640,7 +699,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_device *adev,
  */
 static void amdgpu_vm_fence_pts(struct amdgpu_vm *vm,
 				uint64_t start, uint64_t end,
-				struct amdgpu_fence *fence)
+				struct fence *fence)
 {
 	unsigned i;
 
@@ -670,12 +729,13 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 				       struct amdgpu_vm *vm,
 				       struct amdgpu_bo_va_mapping *mapping,
 				       uint64_t addr, uint32_t gtt_flags,
-				       struct amdgpu_fence **fence)
+				       struct fence **fence)
 {
 	struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
 	unsigned nptes, ncmds, ndw;
 	uint32_t flags = gtt_flags;
-	struct amdgpu_ib ib;
+	struct amdgpu_ib *ib;
+	struct fence *f = NULL;
 	int r;
 
 	/* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here
@@ -722,46 +782,54 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (ndw > 0xfffff)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, ndw * 4, &ib);
-	if (r)
-		return r;
-	ib.length_dw = 0;
-
-	if (!(flags & AMDGPU_PTE_VALID)) {
-		unsigned i;
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
 
-		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			struct amdgpu_fence *f = vm->ids[i].last_id_use;
-			amdgpu_sync_fence(&ib.sync, f);
-		}
+	r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
+	if (r) {
+		kfree(ib);
+		return r;
 	}
 
-	r = amdgpu_vm_update_ptes(adev, vm, &ib, mapping->it.start,
+	ib->length_dw = 0;
+
+	r = amdgpu_vm_update_ptes(adev, vm, ib, mapping->it.start,
 				  mapping->it.last + 1, addr + mapping->offset,
 				  flags, gtt_flags);
 
 	if (r) {
-		amdgpu_ib_free(adev, &ib);
+		amdgpu_ib_free(adev, ib);
+		kfree(ib);
 		return r;
 	}
 
-	amdgpu_vm_pad_ib(adev, &ib);
-	WARN_ON(ib.length_dw > ndw);
+	amdgpu_vm_pad_ib(adev, ib);
+	WARN_ON(ib->length_dw > ndw);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vm_free_job,
+						 AMDGPU_FENCE_OWNER_VM,
+						 &f);
+	if (r)
+		goto error_free;
 
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_VM);
-	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		return r;
-	}
 	amdgpu_vm_fence_pts(vm, mapping->it.start,
-			    mapping->it.last + 1, ib.fence);
+			    mapping->it.last + 1, f);
 	if (fence) {
-		amdgpu_fence_unref(fence);
-		*fence = amdgpu_fence_ref(ib.fence);
+		fence_put(*fence);
+		*fence = fence_get(f);
+	}
+	fence_put(f);
+	if (!amdgpu_enable_scheduler) {
+		amdgpu_ib_free(adev, ib);
+		kfree(ib);
 	}
-	amdgpu_ib_free(adev, &ib);
-
 	return 0;
+
+error_free:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
+	return r;
 }
 
 /**
@@ -794,21 +862,25 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 		addr = 0;
 	}
 
-	if (addr == bo_va->addr)
-		return 0;
-
 	flags = amdgpu_ttm_tt_pte_flags(adev, bo_va->bo->tbo.ttm, mem);
 
-	list_for_each_entry(mapping, &bo_va->mappings, list) {
+	spin_lock(&vm->status_lock);
+	if (!list_empty(&bo_va->vm_status))
+		list_splice_init(&bo_va->valids, &bo_va->invalids);
+	spin_unlock(&vm->status_lock);
+
+	list_for_each_entry(mapping, &bo_va->invalids, list) {
 		r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, addr,
 						flags, &bo_va->last_pt_update);
 		if (r)
 			return r;
 	}
 
-	bo_va->addr = addr;
 	spin_lock(&vm->status_lock);
+	list_splice_init(&bo_va->invalids, &bo_va->valids);
 	list_del_init(&bo_va->vm_status);
+	if (!mem)
+		list_add(&bo_va->vm_status, &vm->cleared);
 	spin_unlock(&vm->status_lock);
 
 	return 0;
@@ -861,7 +933,7 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
 			     struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 {
 	struct amdgpu_bo_va *bo_va = NULL;
-	int r;
+	int r = 0;
 
 	spin_lock(&vm->status_lock);
 	while (!list_empty(&vm->invalidated)) {
@@ -878,8 +950,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
 	spin_unlock(&vm->status_lock);
 
 	if (bo_va)
-		amdgpu_sync_fence(sync, bo_va->last_pt_update);
-	return 0;
+		r = amdgpu_sync_fence(adev, sync, bo_va->last_pt_update);
+
+	return r;
 }
 
 /**
@@ -907,10 +980,10 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 	}
 	bo_va->vm = vm;
 	bo_va->bo = bo;
-	bo_va->addr = 0;
 	bo_va->ref_count = 1;
 	INIT_LIST_HEAD(&bo_va->bo_list);
-	INIT_LIST_HEAD(&bo_va->mappings);
+	INIT_LIST_HEAD(&bo_va->valids);
+	INIT_LIST_HEAD(&bo_va->invalids);
 	INIT_LIST_HEAD(&bo_va->vm_status);
 
 	mutex_lock(&vm->mutex);
@@ -999,12 +1072,10 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 	mapping->offset = offset;
 	mapping->flags = flags;
 
-	list_add(&mapping->list, &bo_va->mappings);
+	list_add(&mapping->list, &bo_va->invalids);
 	interval_tree_insert(&mapping->it, &vm->va);
 	trace_amdgpu_vm_bo_map(bo_va, mapping);
 
-	bo_va->addr = 0;
-
 	/* Make sure the page tables are allocated */
 	saddr >>= amdgpu_vm_block_size;
 	eaddr >>= amdgpu_vm_block_size;
@@ -1028,7 +1099,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 
 		r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8,
 				     AMDGPU_GPU_PAGE_SIZE, true,
-				     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &pt);
+				     AMDGPU_GEM_DOMAIN_VRAM,
+				     AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
+				     NULL, &pt);
 		if (r)
 			goto error_free;
 
@@ -1085,17 +1158,27 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 {
 	struct amdgpu_bo_va_mapping *mapping;
 	struct amdgpu_vm *vm = bo_va->vm;
+	bool valid = true;
 
 	saddr /= AMDGPU_GPU_PAGE_SIZE;
 
-	list_for_each_entry(mapping, &bo_va->mappings, list) {
+	list_for_each_entry(mapping, &bo_va->valids, list) {
 		if (mapping->it.start == saddr)
 			break;
 	}
 
-	if (&mapping->list == &bo_va->mappings) {
-		amdgpu_bo_unreserve(bo_va->bo);
-		return -ENOENT;
+	if (&mapping->list == &bo_va->valids) {
+		valid = false;
+
+		list_for_each_entry(mapping, &bo_va->invalids, list) {
+			if (mapping->it.start == saddr)
+				break;
+		}
+
+		if (&mapping->list == &bo_va->invalids) {
+			amdgpu_bo_unreserve(bo_va->bo);
+			return -ENOENT;
+		}
 	}
 
 	mutex_lock(&vm->mutex);
@@ -1103,12 +1186,10 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 	interval_tree_remove(&mapping->it, &vm->va);
 	trace_amdgpu_vm_bo_unmap(bo_va, mapping);
 
-	if (bo_va->addr) {
-		/* clear the old address */
+	if (valid)
 		list_add(&mapping->list, &vm->freed);
-	} else {
+	else
 		kfree(mapping);
-	}
 	mutex_unlock(&vm->mutex);
 	amdgpu_bo_unreserve(bo_va->bo);
 
@@ -1139,16 +1220,19 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 	list_del(&bo_va->vm_status);
 	spin_unlock(&vm->status_lock);
 
-	list_for_each_entry_safe(mapping, next, &bo_va->mappings, list) {
+	list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
 		list_del(&mapping->list);
 		interval_tree_remove(&mapping->it, &vm->va);
 		trace_amdgpu_vm_bo_unmap(bo_va, mapping);
-		if (bo_va->addr)
-			list_add(&mapping->list, &vm->freed);
-		else
-			kfree(mapping);
+		list_add(&mapping->list, &vm->freed);
+	}
+	list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) {
+		list_del(&mapping->list);
+		interval_tree_remove(&mapping->it, &vm->va);
+		kfree(mapping);
 	}
-	amdgpu_fence_unref(&bo_va->last_pt_update);
+
+	fence_put(bo_va->last_pt_update);
 	kfree(bo_va);
 
 	mutex_unlock(&vm->mutex);
@@ -1169,12 +1253,10 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 	struct amdgpu_bo_va *bo_va;
 
 	list_for_each_entry(bo_va, &bo->va, bo_list) {
-		if (bo_va->addr) {
-			spin_lock(&bo_va->vm->status_lock);
-			list_del(&bo_va->vm_status);
+		spin_lock(&bo_va->vm->status_lock);
+		if (list_empty(&bo_va->vm_status))
 			list_add(&bo_va->vm_status, &bo_va->vm->invalidated);
-			spin_unlock(&bo_va->vm->status_lock);
-		}
+		spin_unlock(&bo_va->vm->status_lock);
 	}
 }
 
@@ -1202,6 +1284,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	vm->va = RB_ROOT;
 	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->invalidated);
+	INIT_LIST_HEAD(&vm->cleared);
 	INIT_LIST_HEAD(&vm->freed);
 
 	pd_size = amdgpu_vm_directory_size(adev);
@@ -1215,8 +1298,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 		return -ENOMEM;
 	}
 
+	vm->page_directory_fence = NULL;
+
 	r = amdgpu_bo_create(adev, pd_size, align, true,
-			     AMDGPU_GEM_DOMAIN_VRAM, 0,
+			     AMDGPU_GEM_DOMAIN_VRAM,
+			     AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
 			     NULL, &vm->page_directory);
 	if (r)
 		return r;
@@ -1263,9 +1349,10 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	kfree(vm->page_tables);
 
 	amdgpu_bo_unref(&vm->page_directory);
+	fence_put(vm->page_directory_fence);
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		amdgpu_fence_unref(&vm->ids[i].flushed_updates);
+		fence_put(vm->ids[i].flushed_updates);
 		amdgpu_fence_unref(&vm->ids[i].last_id_use);
 	}
 

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

@@ -139,7 +139,8 @@ amdgpu_atombios_dp_aux_transfer(struct drm_dp_aux *aux, struct drm_dp_aux_msg *m
 
 	tx_buf[0] = msg->address & 0xff;
 	tx_buf[1] = msg->address >> 8;
-	tx_buf[2] = msg->request << 4;
+	tx_buf[2] = (msg->request << 4) |
+		((msg->address >> 16) & 0xf);
 	tx_buf[3] = msg->size ? (msg->size - 1) : 0;
 
 	switch (msg->request & ~DP_AUX_I2C_MOT) {

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

@@ -812,7 +812,7 @@ amdgpu_atombios_encoder_setup_dig_transmitter(struct drm_encoder *encoder, int a
 			else
 				args.v1.ucConfig |= ATOM_TRANSMITTER_CONFIG_DIG1_ENCODER;
 
-			if ((adev->flags & AMDGPU_IS_APU) &&
+			if ((adev->flags & AMD_IS_APU) &&
 			    (amdgpu_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_UNIPHY)) {
 				if (is_dp ||
 				    !amdgpu_dig_monitor_is_duallink(encoder, amdgpu_encoder->pixel_clock)) {

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

@@ -64,6 +64,8 @@
 #include "oss/oss_2_0_d.h"
 #include "oss/oss_2_0_sh_mask.h"
 
+#include "amdgpu_amdkfd.h"
+
 /*
  * Indirect registers accessor
  */
@@ -836,7 +838,7 @@ static u32 cik_get_xclk(struct amdgpu_device *adev)
 {
 	u32 reference_clock = adev->clock.spll.reference_freq;
 
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		if (RREG32_SMC(ixGENERAL_PWRMGT) & GENERAL_PWRMGT__GPU_COUNTER_CLK_MASK)
 			return reference_clock / 2;
 	} else {
@@ -1233,7 +1235,7 @@ static void cik_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask)
 	if (reset_mask & AMDGPU_RESET_VMC)
 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_VMC_MASK;
 
-	if (!(adev->flags & AMDGPU_IS_APU)) {
+	if (!(adev->flags & AMD_IS_APU)) {
 		if (reset_mask & AMDGPU_RESET_MC)
 			srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_MC_MASK;
 	}
@@ -1409,7 +1411,7 @@ static void cik_gpu_pci_config_reset(struct amdgpu_device *adev)
 		dev_warn(adev->dev, "Wait for MC idle timed out !\n");
 	}
 
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		kv_save_regs_for_reset(adev, &kv_save);
 
 	/* disable BM */
@@ -1427,7 +1429,7 @@ static void cik_gpu_pci_config_reset(struct amdgpu_device *adev)
 	}
 
 	/* does asic init need to be run first??? */
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		kv_restore_regs_for_reset(adev, &kv_save);
 }
 
@@ -1568,7 +1570,7 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev)
 	if (amdgpu_pcie_gen2 == 0)
 		return;
 
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return;
 
 	ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask);
@@ -1728,7 +1730,7 @@ static void cik_program_aspm(struct amdgpu_device *adev)
 		return;
 
 	/* XXX double check APUs */
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return;
 
 	orig = data = RREG32_PCIE(ixPCIE_LC_N_FTS_CNTL);
@@ -2448,14 +2450,21 @@ static int cik_common_suspend(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_amdkfd_suspend(adev);
+
 	return cik_common_hw_fini(adev);
 }
 
 static int cik_common_resume(void *handle)
 {
+	int r;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	return cik_common_hw_init(adev);
+	r = cik_common_hw_init(adev);
+	if (r)
+		return r;
+
+	return amdgpu_amdkfd_resume(adev);
 }
 
 static bool cik_common_is_idle(void *handle)

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

@@ -188,6 +188,19 @@ static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring)
 	WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc);
 }
 
+static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+					  SDMA_NOP_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * cik_sdma_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -213,8 +226,8 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 4)
-		amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
+	cik_sdma_ring_insert_nop(ring, (12 - (ring->wptr & 7)) % 8);
+
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
 	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
@@ -501,6 +514,8 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev)
 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 		fw_data = (const __le32 *)
 			(adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 		WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
@@ -614,6 +629,7 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib ib;
+	struct fence *f = NULL;
 	unsigned i;
 	unsigned index;
 	int r;
@@ -629,12 +645,11 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 	gpu_addr = adev->wb.gpu_addr + (index * 4);
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
-
+	memset(&ib, 0, sizeof(ib));
 	r = amdgpu_ib_get(ring, NULL, 256, &ib);
 	if (r) {
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
-		return r;
+		goto err0;
 	}
 
 	ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
@@ -643,20 +658,16 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[3] = 1;
 	ib.ptr[4] = 0xDEADBEEF;
 	ib.length_dw = 5;
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err1;
 
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
+	r = fence_wait(f, false);
 	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = amdgpu_fence_wait(ib.fence, false);
-	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
-		return r;
+		goto err1;
 	}
 	for (i = 0; i < adev->usec_timeout; i++) {
 		tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -666,12 +677,17 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	if (i < adev->usec_timeout) {
 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
-			 ib.fence->ring->idx, i);
+			 ring->idx, i);
+		goto err1;
 	} else {
 		DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
 		r = -EINVAL;
 	}
+
+err1:
+	fence_put(f);
 	amdgpu_ib_free(adev, &ib);
+err0:
 	amdgpu_wb_free(adev, index);
 	return r;
 }
@@ -814,8 +830,19 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+					SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0) |
+					SDMA_NOP_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+					SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
 }
 
 /**
@@ -1302,6 +1329,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
 	.test_ring = cik_sdma_ring_test_ring,
 	.test_ib = cik_sdma_ring_test_ib,
 	.is_lockup = cik_sdma_ring_is_lockup,
+	.insert_nop = cik_sdma_ring_insert_nop,
 };
 
 static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev)
@@ -1338,18 +1366,18 @@ static void cik_sdma_set_irq_funcs(struct amdgpu_device *adev)
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void cik_sdma_emit_copy_buffer(struct amdgpu_ring *ring,
+static void cik_sdma_emit_copy_buffer(struct amdgpu_ib *ib,
 				      uint64_t src_offset,
 				      uint64_t dst_offset,
 				      uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**
@@ -1362,16 +1390,16 @@ static void cik_sdma_emit_copy_buffer(struct amdgpu_ring *ring,
  *
  * Fill GPU buffers using the DMA engine (CIK).
  */
-static void cik_sdma_emit_fill_buffer(struct amdgpu_ring *ring,
+static void cik_sdma_emit_fill_buffer(struct amdgpu_ib *ib,
 				      uint32_t src_data,
 				      uint64_t dst_offset,
 				      uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {
@@ -1404,5 +1432,6 @@ static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev)
 	if (adev->vm_manager.vm_pte_funcs == NULL) {
 		adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs;
 		adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+		adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
 	}
 }

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

@@ -487,6 +487,7 @@
 					 (((op) & 0xFF) << 0))
 /* sDMA opcodes */
 #define	SDMA_OPCODE_NOP					  0
+#	define SDMA_NOP_COUNT(x)			  (((x) & 0x3FFF) << 16)
 #define	SDMA_OPCODE_COPY				  1
 #       define SDMA_COPY_SUB_OPCODE_LINEAR                0
 #       define SDMA_COPY_SUB_OPCODE_TILED                 1
@@ -552,6 +553,12 @@
 #define VCE_CMD_IB_AUTO		0x00000005
 #define VCE_CMD_SEMAPHORE	0x00000006
 
+/* if PTR32, these are the bases for scratch and lds */
+#define	PRIVATE_BASE(x)	((x) << 0) /* scratch */
+#define	SHARED_BASE(x)	((x) << 16) /* LDS */
+
+#define KFD_CIK_SDMA_QUEUE_OFFSET	0x200
+
 /* valid for both DEFAULT_MTYPE and APE1_MTYPE */
 enum {
 	MTYPE_CACHED = 0,

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

@@ -1596,9 +1596,9 @@ static int cz_dpm_update_low_memory_pstate(struct amdgpu_device *adev)
 
 	if (pi->sys_info.nb_dpm_enable) {
 		if (ps->force_high)
-			cz_dpm_nbdpm_lm_pstate_enable(adev, true);
-		else
 			cz_dpm_nbdpm_lm_pstate_enable(adev, false);
+		else
+			cz_dpm_nbdpm_lm_pstate_enable(adev, true);
 	}
 
 	return ret;

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

@@ -126,9 +126,31 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmXDMA_MEM_POWER_CNTL, 0x00000101, 0x00000000,
 };
 
+static const u32 golden_settings_fiji_a10[] =
+{
+	mmDCI_CLK_CNTL, 0x00000080, 0x00000000,
+	mmFBC_DEBUG_COMP, 0x000000f0, 0x00000070,
+	mmFBC_MISC, 0x1f311fff, 0x12300000,
+	mmHDMI_CONTROL, 0x31000111, 0x00000011,
+};
+
+static const u32 fiji_mgcg_cgcg_init[] =
+{
+	mmXDMA_CLOCK_GATING_CNTL, 0xffffffff, 0x00000100,
+	mmXDMA_MEM_POWER_CNTL, 0x00000101, 0x00000000,
+};
+
 static void dce_v10_0_init_golden_registers(struct amdgpu_device *adev)
 {
 	switch (adev->asic_type) {
+	case CHIP_FIJI:
+		amdgpu_program_register_sequence(adev,
+						 fiji_mgcg_cgcg_init,
+						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
+		amdgpu_program_register_sequence(adev,
+						 golden_settings_fiji_a10,
+						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
+		break;
 	case CHIP_TONGA:
 		amdgpu_program_register_sequence(adev,
 						 tonga_mgcg_cgcg_init,
@@ -803,11 +825,11 @@ static u32 dce_v10_0_line_buffer_adjust(struct amdgpu_device *adev,
 			buffer_alloc = 2;
 		} else if (mode->crtc_hdisplay < 4096) {
 			mem_cfg = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		} else {
 			DRM_DEBUG_KMS("Mode too big for LB!\n");
 			mem_cfg = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		}
 	} else {
 		mem_cfg = 1;
@@ -1331,7 +1353,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev,
 	tmp = REG_SET_FIELD(wm_mask, DPG_WATERMARK_MASK_CONTROL, URGENCY_WATERMARK_MASK, 2);
 	WREG32(mmDPG_WATERMARK_MASK_CONTROL + amdgpu_crtc->crtc_offset, tmp);
 	tmp = RREG32(mmDPG_PIPE_URGENCY_CONTROL + amdgpu_crtc->crtc_offset);
-	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_LOW_WATERMARK, latency_watermark_a);
+	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_LOW_WATERMARK, latency_watermark_b);
 	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_HIGH_WATERMARK, line_time);
 	WREG32(mmDPG_PIPE_URGENCY_CONTROL + amdgpu_crtc->crtc_offset, tmp);
 	/* restore original selection */
@@ -2888,6 +2910,7 @@ static int dce_v10_0_early_init(void *handle)
 	dce_v10_0_set_irq_funcs(adev);
 
 	switch (adev->asic_type) {
+	case CHIP_FIJI:
 	case CHIP_TONGA:
 		adev->mode_info.num_crtc = 6; /* XXX 7??? */
 		adev->mode_info.num_hpd = 6;

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

@@ -801,11 +801,11 @@ static u32 dce_v11_0_line_buffer_adjust(struct amdgpu_device *adev,
 			buffer_alloc = 2;
 		} else if (mode->crtc_hdisplay < 4096) {
 			mem_cfg = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		} else {
 			DRM_DEBUG_KMS("Mode too big for LB!\n");
 			mem_cfg = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		}
 	} else {
 		mem_cfg = 1;
@@ -1329,7 +1329,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev,
 	tmp = REG_SET_FIELD(wm_mask, DPG_WATERMARK_MASK_CONTROL, URGENCY_WATERMARK_MASK, 2);
 	WREG32(mmDPG_WATERMARK_MASK_CONTROL + amdgpu_crtc->crtc_offset, tmp);
 	tmp = RREG32(mmDPG_PIPE_URGENCY_CONTROL + amdgpu_crtc->crtc_offset);
-	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_LOW_WATERMARK, latency_watermark_a);
+	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_LOW_WATERMARK, latency_watermark_b);
 	tmp = REG_SET_FIELD(tmp, DPG_PIPE_URGENCY_CONTROL, URGENCY_HIGH_WATERMARK, line_time);
 	WREG32(mmDPG_PIPE_URGENCY_CONTROL + amdgpu_crtc->crtc_offset, tmp);
 	/* restore original selection */

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

@@ -770,11 +770,11 @@ static u32 dce_v8_0_line_buffer_adjust(struct amdgpu_device *adev,
 			buffer_alloc = 2;
 		} else if (mode->crtc_hdisplay < 4096) {
 			tmp = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		} else {
 			DRM_DEBUG_KMS("Mode too big for LB!\n");
 			tmp = 0;
-			buffer_alloc = (adev->flags & AMDGPU_IS_APU) ? 2 : 4;
+			buffer_alloc = (adev->flags & AMD_IS_APU) ? 2 : 4;
 		}
 	} else {
 		tmp = 1;

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

@@ -0,0 +1,181 @@
+/*
+ * Copyright 2014 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 <linux/firmware.h>
+#include "drmP.h"
+#include "amdgpu.h"
+#include "fiji_smumgr.h"
+
+MODULE_FIRMWARE("amdgpu/fiji_smc.bin");
+
+static void fiji_dpm_set_funcs(struct amdgpu_device *adev);
+
+static int fiji_dpm_early_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	fiji_dpm_set_funcs(adev);
+
+	return 0;
+}
+
+static int fiji_dpm_init_microcode(struct amdgpu_device *adev)
+{
+	char fw_name[30] = "amdgpu/fiji_smc.bin";
+	int err;
+
+	err = request_firmware(&adev->pm.fw, fw_name, adev->dev);
+	if (err)
+		goto out;
+	err = amdgpu_ucode_validate(adev->pm.fw);
+
+out:
+	if (err) {
+		DRM_ERROR("Failed to load firmware \"%s\"", fw_name);
+		release_firmware(adev->pm.fw);
+		adev->pm.fw = NULL;
+	}
+	return err;
+}
+
+static int fiji_dpm_sw_init(void *handle)
+{
+	int ret;
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	ret = fiji_dpm_init_microcode(adev);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int fiji_dpm_sw_fini(void *handle)
+{
+	return 0;
+}
+
+static int fiji_dpm_hw_init(void *handle)
+{
+	int ret;
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	mutex_lock(&adev->pm.mutex);
+
+	ret = fiji_smu_init(adev);
+	if (ret) {
+		DRM_ERROR("SMU initialization failed\n");
+		goto fail;
+	}
+
+	ret = fiji_smu_start(adev);
+	if (ret) {
+		DRM_ERROR("SMU start failed\n");
+		goto fail;
+	}
+
+	mutex_unlock(&adev->pm.mutex);
+	return 0;
+
+fail:
+	adev->firmware.smu_load = false;
+	mutex_unlock(&adev->pm.mutex);
+	return -EINVAL;
+}
+
+static int fiji_dpm_hw_fini(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	mutex_lock(&adev->pm.mutex);
+	fiji_smu_fini(adev);
+	mutex_unlock(&adev->pm.mutex);
+	return 0;
+}
+
+static int fiji_dpm_suspend(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	fiji_dpm_hw_fini(adev);
+
+	return 0;
+}
+
+static int fiji_dpm_resume(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	fiji_dpm_hw_init(adev);
+
+	return 0;
+}
+
+static int fiji_dpm_set_clockgating_state(void *handle,
+			enum amd_clockgating_state state)
+{
+	return 0;
+}
+
+static int fiji_dpm_set_powergating_state(void *handle,
+			enum amd_powergating_state state)
+{
+	return 0;
+}
+
+const struct amd_ip_funcs fiji_dpm_ip_funcs = {
+	.early_init = fiji_dpm_early_init,
+	.late_init = NULL,
+	.sw_init = fiji_dpm_sw_init,
+	.sw_fini = fiji_dpm_sw_fini,
+	.hw_init = fiji_dpm_hw_init,
+	.hw_fini = fiji_dpm_hw_fini,
+	.suspend = fiji_dpm_suspend,
+	.resume = fiji_dpm_resume,
+	.is_idle = NULL,
+	.wait_for_idle = NULL,
+	.soft_reset = NULL,
+	.print_status = NULL,
+	.set_clockgating_state = fiji_dpm_set_clockgating_state,
+	.set_powergating_state = fiji_dpm_set_powergating_state,
+};
+
+static const struct amdgpu_dpm_funcs fiji_dpm_funcs = {
+	.get_temperature = NULL,
+	.pre_set_power_state = NULL,
+	.set_power_state = NULL,
+	.post_set_power_state = NULL,
+	.display_configuration_changed = NULL,
+	.get_sclk = NULL,
+	.get_mclk = NULL,
+	.print_power_state = NULL,
+	.debugfs_print_current_performance_level = NULL,
+	.force_performance_level = NULL,
+	.vblank_too_short = NULL,
+	.powergate_uvd = NULL,
+};
+
+static void fiji_dpm_set_funcs(struct amdgpu_device *adev)
+{
+	if (NULL == adev->pm.funcs)
+		adev->pm.funcs = &fiji_dpm_funcs;
+}

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

@@ -0,0 +1,182 @@
+/*
+ * Copyright 2014 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.
+ *
+ */
+
+#ifndef FIJI_PP_SMC_H
+#define FIJI_PP_SMC_H
+
+#pragma pack(push, 1)
+
+#define PPSMC_SWSTATE_FLAG_DC                           0x01
+#define PPSMC_SWSTATE_FLAG_UVD                          0x02
+#define PPSMC_SWSTATE_FLAG_VCE                          0x04
+
+#define PPSMC_THERMAL_PROTECT_TYPE_INTERNAL             0x00
+#define PPSMC_THERMAL_PROTECT_TYPE_EXTERNAL             0x01
+#define PPSMC_THERMAL_PROTECT_TYPE_NONE                 0xff
+
+#define PPSMC_SYSTEMFLAG_GPIO_DC                        0x01
+#define PPSMC_SYSTEMFLAG_STEPVDDC                       0x02
+#define PPSMC_SYSTEMFLAG_GDDR5                          0x04
+
+#define PPSMC_SYSTEMFLAG_DISABLE_BABYSTEP               0x08
+
+#define PPSMC_SYSTEMFLAG_REGULATOR_HOT                  0x10
+#define PPSMC_SYSTEMFLAG_REGULATOR_HOT_ANALOG           0x20
+
+#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_MASK              0x07
+#define PPSMC_EXTRAFLAGS_AC2DC_DONT_WAIT_FOR_VBLANK     0x08
+
+#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTODPMLOWSTATE   0x00
+#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTOINITIALSTATE  0x01
+
+#define PPSMC_DPM2FLAGS_TDPCLMP                         0x01
+#define PPSMC_DPM2FLAGS_PWRSHFT                         0x02
+#define PPSMC_DPM2FLAGS_OCP                             0x04
+
+#define PPSMC_DISPLAY_WATERMARK_LOW                     0
+#define PPSMC_DISPLAY_WATERMARK_HIGH                    1
+
+#define PPSMC_STATEFLAG_AUTO_PULSE_SKIP    		0x01
+#define PPSMC_STATEFLAG_POWERBOOST         		0x02
+#define PPSMC_STATEFLAG_PSKIP_ON_TDP_FAULT 		0x04
+#define PPSMC_STATEFLAG_POWERSHIFT         		0x08
+#define PPSMC_STATEFLAG_SLOW_READ_MARGIN   		0x10
+#define PPSMC_STATEFLAG_DEEPSLEEP_THROTTLE 		0x20
+#define PPSMC_STATEFLAG_DEEPSLEEP_BYPASS   		0x40
+
+#define FDO_MODE_HARDWARE 0
+#define FDO_MODE_PIECE_WISE_LINEAR 1
+
+enum FAN_CONTROL {
+	FAN_CONTROL_FUZZY,
+	FAN_CONTROL_TABLE
+};
+
+//Gemini Modes
+#define PPSMC_GeminiModeNone   0  //Single GPU board
+#define PPSMC_GeminiModeMaster 1  //Master GPU on a Gemini board
+#define PPSMC_GeminiModeSlave  2  //Slave GPU on a Gemini board
+
+#define PPSMC_Result_OK             			((uint16_t)0x01)
+#define PPSMC_Result_NoMore         			((uint16_t)0x02)
+#define PPSMC_Result_NotNow         			((uint16_t)0x03)
+#define PPSMC_Result_Failed         			((uint16_t)0xFF)
+#define PPSMC_Result_UnknownCmd     			((uint16_t)0xFE)
+#define PPSMC_Result_UnknownVT      			((uint16_t)0xFD)
+
+typedef uint16_t PPSMC_Result;
+
+#define PPSMC_isERROR(x) ((uint16_t)0x80 & (x))
+
+#define PPSMC_MSG_Halt                      		((uint16_t)0x10)
+#define PPSMC_MSG_Resume                    		((uint16_t)0x11)
+#define PPSMC_MSG_EnableDPMLevel            		((uint16_t)0x12)
+#define PPSMC_MSG_ZeroLevelsDisabled        		((uint16_t)0x13)
+#define PPSMC_MSG_OneLevelsDisabled         		((uint16_t)0x14)
+#define PPSMC_MSG_TwoLevelsDisabled         		((uint16_t)0x15)
+#define PPSMC_MSG_EnableThermalInterrupt    		((uint16_t)0x16)
+#define PPSMC_MSG_RunningOnAC               		((uint16_t)0x17)
+#define PPSMC_MSG_LevelUp                   		((uint16_t)0x18)
+#define PPSMC_MSG_LevelDown                 		((uint16_t)0x19)
+#define PPSMC_MSG_ResetDPMCounters          		((uint16_t)0x1a)
+#define PPSMC_MSG_SwitchToSwState           		((uint16_t)0x20)
+#define PPSMC_MSG_SwitchToSwStateLast       		((uint16_t)0x3f)
+#define PPSMC_MSG_SwitchToInitialState      		((uint16_t)0x40)
+#define PPSMC_MSG_NoForcedLevel             		((uint16_t)0x41)
+#define PPSMC_MSG_ForceHigh                 		((uint16_t)0x42)
+#define PPSMC_MSG_ForceMediumOrHigh         		((uint16_t)0x43)
+#define PPSMC_MSG_SwitchToMinimumPower      		((uint16_t)0x51)
+#define PPSMC_MSG_ResumeFromMinimumPower    		((uint16_t)0x52)
+#define PPSMC_MSG_EnableCac                 		((uint16_t)0x53)
+#define PPSMC_MSG_DisableCac                		((uint16_t)0x54)
+#define PPSMC_DPMStateHistoryStart          		((uint16_t)0x55)
+#define PPSMC_DPMStateHistoryStop           		((uint16_t)0x56)
+#define PPSMC_CACHistoryStart               		((uint16_t)0x57)
+#define PPSMC_CACHistoryStop                		((uint16_t)0x58)
+#define PPSMC_TDPClampingActive             		((uint16_t)0x59)
+#define PPSMC_TDPClampingInactive           		((uint16_t)0x5A)
+#define PPSMC_StartFanControl               		((uint16_t)0x5B)
+#define PPSMC_StopFanControl                		((uint16_t)0x5C)
+#define PPSMC_NoDisplay                     		((uint16_t)0x5D)
+#define PPSMC_HasDisplay                    		((uint16_t)0x5E)
+#define PPSMC_MSG_UVDPowerOFF               		((uint16_t)0x60)
+#define PPSMC_MSG_UVDPowerON                		((uint16_t)0x61)
+#define PPSMC_MSG_EnableULV                 		((uint16_t)0x62)
+#define PPSMC_MSG_DisableULV                		((uint16_t)0x63)
+#define PPSMC_MSG_EnterULV                  		((uint16_t)0x64)
+#define PPSMC_MSG_ExitULV                   		((uint16_t)0x65)
+#define PPSMC_PowerShiftActive              		((uint16_t)0x6A)
+#define PPSMC_PowerShiftInactive            		((uint16_t)0x6B)
+#define PPSMC_OCPActive                     		((uint16_t)0x6C)
+#define PPSMC_OCPInactive                   		((uint16_t)0x6D)
+#define PPSMC_CACLongTermAvgEnable          		((uint16_t)0x6E)
+#define PPSMC_CACLongTermAvgDisable         		((uint16_t)0x6F)
+#define PPSMC_MSG_InferredStateSweep_Start  		((uint16_t)0x70)
+#define PPSMC_MSG_InferredStateSweep_Stop   		((uint16_t)0x71)
+#define PPSMC_MSG_SwitchToLowestInfState    		((uint16_t)0x72)
+#define PPSMC_MSG_SwitchToNonInfState       		((uint16_t)0x73)
+#define PPSMC_MSG_AllStateSweep_Start       		((uint16_t)0x74)
+#define PPSMC_MSG_AllStateSweep_Stop        		((uint16_t)0x75)
+#define PPSMC_MSG_SwitchNextLowerInfState   		((uint16_t)0x76)
+#define PPSMC_MSG_SwitchNextHigherInfState  		((uint16_t)0x77)
+#define PPSMC_MSG_MclkRetrainingTest        		((uint16_t)0x78)
+#define PPSMC_MSG_ForceTDPClamping          		((uint16_t)0x79)
+#define PPSMC_MSG_CollectCAC_PowerCorreln   		((uint16_t)0x7A)
+#define PPSMC_MSG_CollectCAC_WeightCalib    		((uint16_t)0x7B)
+#define PPSMC_MSG_CollectCAC_SQonly         		((uint16_t)0x7C)
+#define PPSMC_MSG_CollectCAC_TemperaturePwr 		((uint16_t)0x7D)
+#define PPSMC_MSG_ExtremitiesTest_Start     		((uint16_t)0x7E)
+#define PPSMC_MSG_ExtremitiesTest_Stop      		((uint16_t)0x7F)
+#define PPSMC_FlushDataCache                		((uint16_t)0x80)
+#define PPSMC_FlushInstrCache               		((uint16_t)0x81)
+#define PPSMC_MSG_SetEnabledLevels          		((uint16_t)0x82)
+#define PPSMC_MSG_SetForcedLevels           		((uint16_t)0x83)
+#define PPSMC_MSG_ResetToDefaults           		((uint16_t)0x84)
+#define PPSMC_MSG_SetForcedLevelsAndJump    		((uint16_t)0x85)
+#define PPSMC_MSG_SetCACHistoryMode         		((uint16_t)0x86)
+#define PPSMC_MSG_EnableDTE                 		((uint16_t)0x87)
+#define PPSMC_MSG_DisableDTE                		((uint16_t)0x88)
+#define PPSMC_MSG_SmcSpaceSetAddress        		((uint16_t)0x89)
+#define PPSMC_MSG_SmcSpaceWriteDWordInc     		((uint16_t)0x8A)
+#define PPSMC_MSG_SmcSpaceWriteWordInc      		((uint16_t)0x8B)
+#define PPSMC_MSG_SmcSpaceWriteByteInc      		((uint16_t)0x8C)
+
+#define PPSMC_MSG_BREAK                     		((uint16_t)0xF8)
+
+#define PPSMC_MSG_Test                      		((uint16_t)0x100)
+#define PPSMC_MSG_DRV_DRAM_ADDR_HI            		((uint16_t)0x250)
+#define PPSMC_MSG_DRV_DRAM_ADDR_LO            		((uint16_t)0x251)
+#define PPSMC_MSG_SMU_DRAM_ADDR_HI            		((uint16_t)0x252)
+#define PPSMC_MSG_SMU_DRAM_ADDR_LO            		((uint16_t)0x253)
+#define PPSMC_MSG_LoadUcodes                  		((uint16_t)0x254)
+
+typedef uint16_t PPSMC_Msg;
+
+#define PPSMC_EVENT_STATUS_THERMAL          		0x00000001
+#define PPSMC_EVENT_STATUS_REGULATORHOT     		0x00000002
+#define PPSMC_EVENT_STATUS_DC               		0x00000004
+#define PPSMC_EVENT_STATUS_GPIO17           		0x00000008
+
+#pragma pack(pop)
+
+#endif

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

@@ -0,0 +1,857 @@
+/*
+ * Copyright 2014 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 <linux/firmware.h>
+#include "drmP.h"
+#include "amdgpu.h"
+#include "fiji_ppsmc.h"
+#include "fiji_smumgr.h"
+#include "smu_ucode_xfer_vi.h"
+#include "amdgpu_ucode.h"
+
+#include "smu/smu_7_1_3_d.h"
+#include "smu/smu_7_1_3_sh_mask.h"
+
+#define FIJI_SMC_SIZE 0x20000
+
+static int fiji_set_smc_sram_address(struct amdgpu_device *adev, uint32_t smc_address, uint32_t limit)
+{
+	uint32_t val;
+
+	if (smc_address & 3)
+		return -EINVAL;
+
+	if ((smc_address + 3) > limit)
+		return -EINVAL;
+
+	WREG32(mmSMC_IND_INDEX_0, smc_address);
+
+	val = RREG32(mmSMC_IND_ACCESS_CNTL);
+	val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 0);
+	WREG32(mmSMC_IND_ACCESS_CNTL, val);
+
+	return 0;
+}
+
+static int fiji_copy_bytes_to_smc(struct amdgpu_device *adev, uint32_t smc_start_address, const uint8_t *src, uint32_t byte_count, uint32_t limit)
+{
+	uint32_t addr;
+	uint32_t data, orig_data;
+	int result = 0;
+	uint32_t extra_shift;
+	unsigned long flags;
+
+	if (smc_start_address & 3)
+		return -EINVAL;
+
+	if ((smc_start_address + byte_count) > limit)
+		return -EINVAL;
+
+	addr = smc_start_address;
+
+	spin_lock_irqsave(&adev->smc_idx_lock, flags);
+	while (byte_count >= 4) {
+		/* Bytes are written into the SMC addres space with the MSB first */
+		data = (src[0] << 24) + (src[1] << 16) + (src[2] << 8) + src[3];
+
+		result = fiji_set_smc_sram_address(adev, addr, limit);
+
+		if (result)
+			goto out;
+
+		WREG32(mmSMC_IND_DATA_0, data);
+
+		src += 4;
+		byte_count -= 4;
+		addr += 4;
+	}
+
+	if (0 != byte_count) {
+		/* Now write odd bytes left, do a read modify write cycle */
+		data = 0;
+
+		result = fiji_set_smc_sram_address(adev, addr, limit);
+		if (result)
+			goto out;
+
+		orig_data = RREG32(mmSMC_IND_DATA_0);
+		extra_shift = 8 * (4 - byte_count);
+
+		while (byte_count > 0) {
+			data = (data << 8) + *src++;
+			byte_count--;
+		}
+
+		data <<= extra_shift;
+		data |= (orig_data & ~((~0UL) << extra_shift));
+
+		result = fiji_set_smc_sram_address(adev, addr, limit);
+		if (result)
+			goto out;
+
+		WREG32(mmSMC_IND_DATA_0, data);
+	}
+
+out:
+	spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
+	return result;
+}
+
+static int fiji_program_jump_on_start(struct amdgpu_device *adev)
+{
+	static unsigned char data[] = {0xE0, 0x00, 0x80, 0x40};
+	fiji_copy_bytes_to_smc(adev, 0x0, data, 4, sizeof(data)+1);
+
+	return 0;
+}
+
+static bool fiji_is_smc_ram_running(struct amdgpu_device *adev)
+{
+	uint32_t val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
+	val = REG_GET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable);
+
+	return ((0 == val) && (0x20100 <= RREG32_SMC(ixSMC_PC_C)));
+}
+
+static int wait_smu_response(struct amdgpu_device *adev)
+{
+	int i;
+	uint32_t val;
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32(mmSMC_RESP_0);
+		if (REG_GET_FIELD(val, SMC_RESP_0, SMC_RESP))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout)
+		return -EINVAL;
+
+	return 0;
+}
+
+static int fiji_send_msg_to_smc_offset(struct amdgpu_device *adev)
+{
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send previous message\n");
+		return -EINVAL;
+	}
+
+	WREG32(mmSMC_MSG_ARG_0, 0x20000);
+	WREG32(mmSMC_MESSAGE_0, PPSMC_MSG_Test);
+
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send message\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int fiji_send_msg_to_smc(struct amdgpu_device *adev, PPSMC_Msg msg)
+{
+	if (!fiji_is_smc_ram_running(adev))
+	{
+		return -EINVAL;;
+	}
+
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send previous message\n");
+		return -EINVAL;
+	}
+
+	WREG32(mmSMC_MESSAGE_0, msg);
+
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send message\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int fiji_send_msg_to_smc_without_waiting(struct amdgpu_device *adev,
+						PPSMC_Msg msg)
+{
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send previous message\n");
+		return -EINVAL;
+	}
+
+	WREG32(mmSMC_MESSAGE_0, msg);
+
+	return 0;
+}
+
+static int fiji_send_msg_to_smc_with_parameter(struct amdgpu_device *adev,
+						PPSMC_Msg msg,
+						uint32_t parameter)
+{
+	if (!fiji_is_smc_ram_running(adev))
+		return -EINVAL;
+
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send previous message\n");
+		return -EINVAL;
+	}
+
+	WREG32(mmSMC_MSG_ARG_0, parameter);
+
+	return fiji_send_msg_to_smc(adev, msg);
+}
+
+static int fiji_send_msg_to_smc_with_parameter_without_waiting(
+					struct amdgpu_device *adev,
+					PPSMC_Msg msg, uint32_t parameter)
+{
+	if (wait_smu_response(adev)) {
+		DRM_ERROR("Failed to send previous message\n");
+		return -EINVAL;
+	}
+
+	WREG32(mmSMC_MSG_ARG_0, parameter);
+
+	return fiji_send_msg_to_smc_without_waiting(adev, msg);
+}
+
+#if 0 /* not used yet */
+static int fiji_wait_for_smc_inactive(struct amdgpu_device *adev)
+{
+	int i;
+	uint32_t val;
+
+	if (!fiji_is_smc_ram_running(adev))
+		return -EINVAL;
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
+		if (REG_GET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, cken) == 0)
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout)
+		return -EINVAL;
+
+	return 0;
+}
+#endif
+
+static int fiji_smu_upload_firmware_image(struct amdgpu_device *adev)
+{
+	const struct smc_firmware_header_v1_0 *hdr;
+	uint32_t ucode_size;
+	uint32_t ucode_start_address;
+	const uint8_t *src;
+	uint32_t val;
+	uint32_t byte_count;
+	uint32_t *data;
+	unsigned long flags;
+
+	if (!adev->pm.fw)
+		return -EINVAL;
+
+	hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data;
+	amdgpu_ucode_print_smc_hdr(&hdr->header);
+
+	adev->pm.fw_version = le32_to_cpu(hdr->header.ucode_version);
+	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes);
+	ucode_start_address = le32_to_cpu(hdr->ucode_start_addr);
+	src = (const uint8_t *)
+		(adev->pm.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+
+	if (ucode_size & 3) {
+		DRM_ERROR("SMC ucode is not 4 bytes aligned\n");
+		return -EINVAL;
+	}
+
+	if (ucode_size > FIJI_SMC_SIZE) {
+		DRM_ERROR("SMC address is beyond the SMC RAM area\n");
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&adev->smc_idx_lock, flags);
+	WREG32(mmSMC_IND_INDEX_0, ucode_start_address);
+
+	val = RREG32(mmSMC_IND_ACCESS_CNTL);
+	val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 1);
+	WREG32(mmSMC_IND_ACCESS_CNTL, val);
+
+	byte_count = ucode_size;
+	data = (uint32_t *)src;
+	for (; byte_count >= 4; data++, byte_count -= 4)
+		WREG32(mmSMC_IND_DATA_0, data[0]);
+
+	val = RREG32(mmSMC_IND_ACCESS_CNTL);
+	val = REG_SET_FIELD(val, SMC_IND_ACCESS_CNTL, AUTO_INCREMENT_IND_0, 0);
+	WREG32(mmSMC_IND_ACCESS_CNTL, val);
+	spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
+
+	return 0;
+}
+
+#if 0 /* not used yet */
+static int fiji_read_smc_sram_dword(struct amdgpu_device *adev,
+				uint32_t smc_address,
+				uint32_t *value,
+				uint32_t limit)
+{
+	int result;
+	unsigned long flags;
+
+	spin_lock_irqsave(&adev->smc_idx_lock, flags);
+	result = fiji_set_smc_sram_address(adev, smc_address, limit);
+	if (result == 0)
+		*value = RREG32(mmSMC_IND_DATA_0);
+	spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
+	return result;
+}
+
+static int fiji_write_smc_sram_dword(struct amdgpu_device *adev,
+				uint32_t smc_address,
+				uint32_t value,
+				uint32_t limit)
+{
+	int result;
+	unsigned long flags;
+
+	spin_lock_irqsave(&adev->smc_idx_lock, flags);
+	result = fiji_set_smc_sram_address(adev, smc_address, limit);
+	if (result == 0)
+		WREG32(mmSMC_IND_DATA_0, value);
+	spin_unlock_irqrestore(&adev->smc_idx_lock, flags);
+	return result;
+}
+
+static int fiji_smu_stop_smc(struct amdgpu_device *adev)
+{
+	uint32_t val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
+	val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
+	WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
+
+	val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
+	val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 1);
+	WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
+
+	return 0;
+}
+#endif
+
+static enum AMDGPU_UCODE_ID fiji_convert_fw_type(uint32_t fw_type)
+{
+	switch (fw_type) {
+		case UCODE_ID_SDMA0:
+			return AMDGPU_UCODE_ID_SDMA0;
+		case UCODE_ID_SDMA1:
+			return AMDGPU_UCODE_ID_SDMA1;
+		case UCODE_ID_CP_CE:
+			return AMDGPU_UCODE_ID_CP_CE;
+		case UCODE_ID_CP_PFP:
+			return AMDGPU_UCODE_ID_CP_PFP;
+		case UCODE_ID_CP_ME:
+			return AMDGPU_UCODE_ID_CP_ME;
+		case UCODE_ID_CP_MEC:
+		case UCODE_ID_CP_MEC_JT1:
+		case UCODE_ID_CP_MEC_JT2:
+			return AMDGPU_UCODE_ID_CP_MEC1;
+		case UCODE_ID_RLC_G:
+			return AMDGPU_UCODE_ID_RLC_G;
+		default:
+			DRM_ERROR("ucode type is out of range!\n");
+			return AMDGPU_UCODE_ID_MAXIMUM;
+	}
+}
+
+static int fiji_smu_populate_single_firmware_entry(struct amdgpu_device *adev,
+						uint32_t fw_type,
+						struct SMU_Entry *entry)
+{
+	enum AMDGPU_UCODE_ID id = fiji_convert_fw_type(fw_type);
+	struct amdgpu_firmware_info *ucode = &adev->firmware.ucode[id];
+	const struct gfx_firmware_header_v1_0 *header = NULL;
+	uint64_t gpu_addr;
+	uint32_t data_size;
+
+	if (ucode->fw == NULL)
+		return -EINVAL;
+	gpu_addr  = ucode->mc_addr;
+	header = (const struct gfx_firmware_header_v1_0 *)ucode->fw->data;
+	data_size = le32_to_cpu(header->header.ucode_size_bytes);
+
+	if ((fw_type == UCODE_ID_CP_MEC_JT1) ||
+		(fw_type == UCODE_ID_CP_MEC_JT2)) {
+		gpu_addr += le32_to_cpu(header->jt_offset) << 2;
+		data_size = le32_to_cpu(header->jt_size) << 2;
+	}
+
+	entry->version = (uint16_t)le32_to_cpu(header->header.ucode_version);
+	entry->id = (uint16_t)fw_type;
+	entry->image_addr_high = upper_32_bits(gpu_addr);
+	entry->image_addr_low = lower_32_bits(gpu_addr);
+	entry->meta_data_addr_high = 0;
+	entry->meta_data_addr_low = 0;
+	entry->data_size_byte = data_size;
+	entry->num_register_entries = 0;
+
+	if (fw_type == UCODE_ID_RLC_G)
+		entry->flags = 1;
+	else
+		entry->flags = 0;
+
+	return 0;
+}
+
+static int fiji_smu_request_load_fw(struct amdgpu_device *adev)
+{
+	struct fiji_smu_private_data *private = (struct fiji_smu_private_data *)adev->smu.priv;
+	struct SMU_DRAMData_TOC *toc;
+	uint32_t fw_to_load;
+
+	WREG32_SMC(ixSOFT_REGISTERS_TABLE_28, 0);
+
+	fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_SMU_DRAM_ADDR_HI, private->smu_buffer_addr_high);
+	fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_SMU_DRAM_ADDR_LO, private->smu_buffer_addr_low);
+
+	toc = (struct SMU_DRAMData_TOC *)private->header;
+	toc->num_entries = 0;
+	toc->structure_version = 1;
+
+	if (!adev->firmware.smu_load)
+		return 0;
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_RLC_G,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for RLC\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_CE,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for CE\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_PFP,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for PFP\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_ME,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for ME\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for MEC\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC_JT1,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for MEC_JT1\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC_JT2,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for MEC_JT2\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_SDMA0,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for SDMA0\n");
+		return -EINVAL;
+	}
+
+	if (fiji_smu_populate_single_firmware_entry(adev, UCODE_ID_SDMA1,
+			&toc->entry[toc->num_entries++])) {
+		DRM_ERROR("Failed to get firmware entry for SDMA1\n");
+		return -EINVAL;
+	}
+
+	fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_DRV_DRAM_ADDR_HI, private->header_addr_high);
+	fiji_send_msg_to_smc_with_parameter(adev, PPSMC_MSG_DRV_DRAM_ADDR_LO, private->header_addr_low);
+
+	fw_to_load = UCODE_ID_RLC_G_MASK |
+			UCODE_ID_SDMA0_MASK |
+			UCODE_ID_SDMA1_MASK |
+			UCODE_ID_CP_CE_MASK |
+			UCODE_ID_CP_ME_MASK |
+			UCODE_ID_CP_PFP_MASK |
+			UCODE_ID_CP_MEC_MASK;
+
+	if (fiji_send_msg_to_smc_with_parameter_without_waiting(adev, PPSMC_MSG_LoadUcodes, fw_to_load)) {
+		DRM_ERROR("Fail to request SMU load ucode\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static uint32_t fiji_smu_get_mask_for_fw_type(uint32_t fw_type)
+{
+	switch (fw_type) {
+		case AMDGPU_UCODE_ID_SDMA0:
+			return UCODE_ID_SDMA0_MASK;
+		case AMDGPU_UCODE_ID_SDMA1:
+			return UCODE_ID_SDMA1_MASK;
+		case AMDGPU_UCODE_ID_CP_CE:
+			return UCODE_ID_CP_CE_MASK;
+		case AMDGPU_UCODE_ID_CP_PFP:
+			return UCODE_ID_CP_PFP_MASK;
+		case AMDGPU_UCODE_ID_CP_ME:
+			return UCODE_ID_CP_ME_MASK;
+		case AMDGPU_UCODE_ID_CP_MEC1:
+			return UCODE_ID_CP_MEC_MASK;
+		case AMDGPU_UCODE_ID_CP_MEC2:
+			return UCODE_ID_CP_MEC_MASK;
+		case AMDGPU_UCODE_ID_RLC_G:
+			return UCODE_ID_RLC_G_MASK;
+		default:
+			DRM_ERROR("ucode type is out of range!\n");
+			return 0;
+	}
+}
+
+static int fiji_smu_check_fw_load_finish(struct amdgpu_device *adev,
+					uint32_t fw_type)
+{
+	uint32_t fw_mask = fiji_smu_get_mask_for_fw_type(fw_type);
+	int i;
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		if (fw_mask == (RREG32_SMC(ixSOFT_REGISTERS_TABLE_28) & fw_mask))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("check firmware loading failed\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int fiji_smu_start_in_protection_mode(struct amdgpu_device *adev)
+{
+	int result;
+	uint32_t val;
+	int i;
+
+	/* Assert reset */
+	val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
+	val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
+	WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
+
+	result = fiji_smu_upload_firmware_image(adev);
+	if (result)
+		return result;
+
+	/* Clear status */
+	WREG32_SMC(ixSMU_STATUS, 0);
+
+	/* Enable clock */
+	val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
+	val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 0);
+	WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
+
+	/* De-assert reset */
+	val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
+	val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 0);
+	WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
+
+	/* Set SMU Auto Start */
+	val = RREG32_SMC(ixSMU_INPUT_DATA);
+	val = REG_SET_FIELD(val, SMU_INPUT_DATA, AUTO_START, 1);
+	WREG32_SMC(ixSMU_INPUT_DATA, val);
+
+	/* Clear firmware interrupt enable flag */
+	WREG32_SMC(ixFIRMWARE_FLAGS, 0);
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixRCU_UC_EVENTS);
+		if (REG_GET_FIELD(val, RCU_UC_EVENTS, INTERRUPTS_ENABLED))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("Interrupt is not enabled by firmware\n");
+		return -EINVAL;
+	}
+
+	/* Call Test SMU message with 0x20000 offset
+	 * to trigger SMU start
+	 */
+	fiji_send_msg_to_smc_offset(adev);
+	DRM_INFO("[FM]try triger smu start\n");
+	/* Wait for done bit to be set */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixSMU_STATUS);
+		if (REG_GET_FIELD(val, SMU_STATUS, SMU_DONE))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("Timeout for SMU start\n");
+		return -EINVAL;
+	}
+
+	/* Check pass/failed indicator */
+	val = RREG32_SMC(ixSMU_STATUS);
+	if (!REG_GET_FIELD(val, SMU_STATUS, SMU_PASS)) {
+		DRM_ERROR("SMU Firmware start failed\n");
+		return -EINVAL;
+	}
+	DRM_INFO("[FM]smu started\n");
+	/* Wait for firmware to initialize */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixFIRMWARE_FLAGS);
+		if(REG_GET_FIELD(val, FIRMWARE_FLAGS, INTERRUPTS_ENABLED))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("SMU firmware initialization failed\n");
+		return -EINVAL;
+	}
+	DRM_INFO("[FM]smu initialized\n");
+
+	return 0;
+}
+
+static int fiji_smu_start_in_non_protection_mode(struct amdgpu_device *adev)
+{
+	int i, result;
+	uint32_t val;
+
+	/* wait for smc boot up */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixRCU_UC_EVENTS);
+		val = REG_GET_FIELD(val, RCU_UC_EVENTS, boot_seq_done);
+		if (val)
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("SMC boot sequence is not completed\n");
+		return -EINVAL;
+	}
+
+	/* Clear firmware interrupt enable flag */
+	WREG32_SMC(ixFIRMWARE_FLAGS, 0);
+
+	/* Assert reset */
+	val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
+	val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 1);
+	WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
+
+	result = fiji_smu_upload_firmware_image(adev);
+	if (result)
+		return result;
+
+	/* Set smc instruct start point at 0x0 */
+	fiji_program_jump_on_start(adev);
+
+	/* Enable clock */
+	val = RREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0);
+	val = REG_SET_FIELD(val, SMC_SYSCON_CLOCK_CNTL_0, ck_disable, 0);
+	WREG32_SMC(ixSMC_SYSCON_CLOCK_CNTL_0, val);
+
+	/* De-assert reset */
+	val = RREG32_SMC(ixSMC_SYSCON_RESET_CNTL);
+	val = REG_SET_FIELD(val, SMC_SYSCON_RESET_CNTL, rst_reg, 0);
+	WREG32_SMC(ixSMC_SYSCON_RESET_CNTL, val);
+
+	/* Wait for firmware to initialize */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32_SMC(ixFIRMWARE_FLAGS);
+		if (REG_GET_FIELD(val, FIRMWARE_FLAGS, INTERRUPTS_ENABLED))
+			break;
+		udelay(1);
+	}
+
+	if (i == adev->usec_timeout) {
+		DRM_ERROR("Timeout for SMC firmware initialization\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int fiji_smu_start(struct amdgpu_device *adev)
+{
+	int result;
+	uint32_t val;
+
+	if (!fiji_is_smc_ram_running(adev)) {
+		val = RREG32_SMC(ixSMU_FIRMWARE);
+		if (!REG_GET_FIELD(val, SMU_FIRMWARE, SMU_MODE)) {
+			DRM_INFO("[FM]start smu in nonprotection mode\n");
+			result = fiji_smu_start_in_non_protection_mode(adev);
+			if (result)
+				return result;
+		} else {
+			DRM_INFO("[FM]start smu in protection mode\n");
+			result = fiji_smu_start_in_protection_mode(adev);
+			if (result)
+				return result;
+		}
+	}
+
+	return fiji_smu_request_load_fw(adev);
+}
+
+static const struct amdgpu_smumgr_funcs fiji_smumgr_funcs = {
+	.check_fw_load_finish = fiji_smu_check_fw_load_finish,
+	.request_smu_load_fw = NULL,
+	.request_smu_specific_fw = NULL,
+};
+
+int fiji_smu_init(struct amdgpu_device *adev)
+{
+	struct fiji_smu_private_data *private;
+	uint32_t image_size = ((sizeof(struct SMU_DRAMData_TOC) / 4096) + 1) * 4096;
+	uint32_t smu_internal_buffer_size = 200*4096;
+	struct amdgpu_bo **toc_buf = &adev->smu.toc_buf;
+	struct amdgpu_bo **smu_buf = &adev->smu.smu_buf;
+	uint64_t mc_addr;
+	void *toc_buf_ptr;
+	void *smu_buf_ptr;
+	int ret;
+
+	private = kzalloc(sizeof(struct fiji_smu_private_data), GFP_KERNEL);
+	if (NULL == private)
+		return -ENOMEM;
+
+	/* allocate firmware buffers */
+	if (adev->firmware.smu_load)
+		amdgpu_ucode_init_bo(adev);
+
+	adev->smu.priv = private;
+	adev->smu.fw_flags = 0;
+
+	/* Allocate FW image data structure and header buffer */
+	ret = amdgpu_bo_create(adev, image_size, PAGE_SIZE,
+			       true, AMDGPU_GEM_DOMAIN_VRAM,
+			       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			       NULL, toc_buf);
+	if (ret) {
+		DRM_ERROR("Failed to allocate memory for TOC buffer\n");
+		return -ENOMEM;
+	}
+
+	/* Allocate buffer for SMU internal buffer */
+	ret = amdgpu_bo_create(adev, smu_internal_buffer_size, PAGE_SIZE,
+			       true, AMDGPU_GEM_DOMAIN_VRAM,
+			       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			       NULL, smu_buf);
+	if (ret) {
+		DRM_ERROR("Failed to allocate memory for SMU internal buffer\n");
+		return -ENOMEM;
+	}
+
+	/* Retrieve GPU address for header buffer and internal buffer */
+	ret = amdgpu_bo_reserve(adev->smu.toc_buf, false);
+	if (ret) {
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to reserve the TOC buffer\n");
+		return -EINVAL;
+	}
+
+	ret = amdgpu_bo_pin(adev->smu.toc_buf, AMDGPU_GEM_DOMAIN_VRAM, &mc_addr);
+	if (ret) {
+		amdgpu_bo_unreserve(adev->smu.toc_buf);
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to pin the TOC buffer\n");
+		return -EINVAL;
+	}
+
+	ret = amdgpu_bo_kmap(*toc_buf, &toc_buf_ptr);
+	if (ret) {
+		amdgpu_bo_unreserve(adev->smu.toc_buf);
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to map the TOC buffer\n");
+		return -EINVAL;
+	}
+
+	amdgpu_bo_unreserve(adev->smu.toc_buf);
+	private->header_addr_low = lower_32_bits(mc_addr);
+	private->header_addr_high = upper_32_bits(mc_addr);
+	private->header = toc_buf_ptr;
+
+	ret = amdgpu_bo_reserve(adev->smu.smu_buf, false);
+	if (ret) {
+		amdgpu_bo_unref(&adev->smu.smu_buf);
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to reserve the SMU internal buffer\n");
+		return -EINVAL;
+	}
+
+	ret = amdgpu_bo_pin(adev->smu.smu_buf, AMDGPU_GEM_DOMAIN_VRAM, &mc_addr);
+	if (ret) {
+		amdgpu_bo_unreserve(adev->smu.smu_buf);
+		amdgpu_bo_unref(&adev->smu.smu_buf);
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to pin the SMU internal buffer\n");
+		return -EINVAL;
+	}
+
+	ret = amdgpu_bo_kmap(*smu_buf, &smu_buf_ptr);
+	if (ret) {
+		amdgpu_bo_unreserve(adev->smu.smu_buf);
+		amdgpu_bo_unref(&adev->smu.smu_buf);
+		amdgpu_bo_unref(&adev->smu.toc_buf);
+		DRM_ERROR("Failed to map the SMU internal buffer\n");
+		return -EINVAL;
+	}
+
+	amdgpu_bo_unreserve(adev->smu.smu_buf);
+	private->smu_buffer_addr_low = lower_32_bits(mc_addr);
+	private->smu_buffer_addr_high = upper_32_bits(mc_addr);
+
+	adev->smu.smumgr_funcs = &fiji_smumgr_funcs;
+
+	return 0;
+}
+
+int fiji_smu_fini(struct amdgpu_device *adev)
+{
+	amdgpu_bo_unref(&adev->smu.toc_buf);
+	amdgpu_bo_unref(&adev->smu.smu_buf);
+	kfree(adev->smu.priv);
+	adev->smu.priv = NULL;
+	if (adev->firmware.fw_buf)
+		amdgpu_ucode_fini_bo(adev);
+
+	return 0;
+}

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

@@ -0,0 +1,42 @@
+/*
+ * Copyright 2014 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.
+ *
+ */
+
+#ifndef FIJI_SMUMGR_H
+#define FIJI_SMUMGR_H
+
+#include "fiji_ppsmc.h"
+
+int fiji_smu_init(struct amdgpu_device *adev);
+int fiji_smu_fini(struct amdgpu_device *adev);
+int fiji_smu_start(struct amdgpu_device *adev);
+
+struct fiji_smu_private_data
+{
+	uint8_t *header;
+	uint32_t smu_buffer_addr_high;
+	uint32_t smu_buffer_addr_low;
+	uint32_t header_addr_high;
+	uint32_t header_addr_low;
+};
+
+#endif

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -2173,7 +2173,7 @@ static void gfx_v7_0_gpu_init(struct amdgpu_device *adev)
 
 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
 	adev->gfx.config.mem_max_burst_length_bytes = 256;
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		/* Get memory bank mapping mode. */
 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
@@ -2648,6 +2648,7 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib ib;
+	struct fence *f = NULL;
 	uint32_t scratch;
 	uint32_t tmp = 0;
 	unsigned i;
@@ -2659,29 +2660,27 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
 		return r;
 	}
 	WREG32(scratch, 0xCAFEDEAD);
+	memset(&ib, 0, sizeof(ib));
 	r = amdgpu_ib_get(ring, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
-		amdgpu_gfx_scratch_free(adev, scratch);
-		return r;
+		goto err1;
 	}
 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
 	ib.ptr[2] = 0xDEADBEEF;
 	ib.length_dw = 3;
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		amdgpu_gfx_scratch_free(adev, scratch);
-		amdgpu_ib_free(adev, &ib);
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = amdgpu_fence_wait(ib.fence, false);
+
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err2;
+
+	r = fence_wait(f, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
-		amdgpu_gfx_scratch_free(adev, scratch);
-		amdgpu_ib_free(adev, &ib);
-		return r;
+		goto err2;
 	}
 	for (i = 0; i < adev->usec_timeout; i++) {
 		tmp = RREG32(scratch);
@@ -2691,14 +2690,19 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	if (i < adev->usec_timeout) {
 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
-			 ib.fence->ring->idx, i);
+			 ring->idx, i);
+		goto err2;
 	} else {
 		DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
 			  scratch, tmp);
 		r = -EINVAL;
 	}
-	amdgpu_gfx_scratch_free(adev, scratch);
+
+err2:
+	fence_put(f);
 	amdgpu_ib_free(adev, &ib);
+err1:
+	amdgpu_gfx_scratch_free(adev, scratch);
 	return r;
 }
 
@@ -3758,7 +3762,7 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 	int r;
 
 	/* allocate rlc buffers */
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		if (adev->asic_type == CHIP_KAVERI) {
 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
 			adev->gfx.rlc.reg_list_size =
@@ -3782,7 +3786,9 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 		/* save restore block */
 		if (adev->gfx.rlc.save_restore_obj == NULL) {
 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
-					     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->gfx.rlc.save_restore_obj);
+					     AMDGPU_GEM_DOMAIN_VRAM,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     NULL, &adev->gfx.rlc.save_restore_obj);
 			if (r) {
 				dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", r);
 				return r;
@@ -3823,7 +3829,9 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 
 		if (adev->gfx.rlc.clear_state_obj == NULL) {
 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
-					     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->gfx.rlc.clear_state_obj);
+					     AMDGPU_GEM_DOMAIN_VRAM,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     NULL, &adev->gfx.rlc.clear_state_obj);
 			if (r) {
 				dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r);
 				gfx_v7_0_rlc_fini(adev);
@@ -3860,7 +3868,9 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 	if (adev->gfx.rlc.cp_table_size) {
 		if (adev->gfx.rlc.cp_table_obj == NULL) {
 			r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
-					     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->gfx.rlc.cp_table_obj);
+					     AMDGPU_GEM_DOMAIN_VRAM,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     NULL, &adev->gfx.rlc.cp_table_obj);
 			if (r) {
 				dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r);
 				gfx_v7_0_rlc_fini(adev);
@@ -5594,6 +5604,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.is_lockup = gfx_v7_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
@@ -5610,6 +5621,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.test_ring = gfx_v7_0_ring_test_ring,
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.is_lockup = gfx_v7_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)

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

@@ -87,6 +87,13 @@ MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
 MODULE_FIRMWARE("amdgpu/topaz_mec2.bin");
 MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
 
+MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
+MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
+MODULE_FIRMWARE("amdgpu/fiji_me.bin");
+MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
+MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
+MODULE_FIRMWARE("amdgpu/fiji_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},
@@ -217,6 +224,71 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 };
 
+static const u32 fiji_golden_common_all[] =
+{
+	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
+	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
+	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
+	mmGB_ADDR_CONFIG, 0xffffffff, 0x12011003,
+	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
+	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
+	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00007FBF,
+	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00007FAF
+};
+
+static const u32 golden_settings_fiji_a10[] =
+{
+	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
+	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
+	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
+	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x00000100,
+	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
+	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
+	mmTCC_CTRL, 0x00100000, 0xf30fff7f,
+	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
+	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x7d6cf5e4,
+	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x3928b1a0,
+};
+
+static const u32 fiji_mgcg_cgcg_init[] =
+{
+	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffc0,
+	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
+	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
+	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
+	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
+	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
+	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
+	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
+	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
+	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
+	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
+	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
+	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
+	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
+	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
+	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
+	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
+	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
+	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
+	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
+	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
+	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
+};
+
 static const u32 golden_settings_iceland_a11[] =
 {
 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
@@ -439,6 +511,18 @@ static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
 						 iceland_golden_common_all,
 						 (const u32)ARRAY_SIZE(iceland_golden_common_all));
 		break;
+	case CHIP_FIJI:
+		amdgpu_program_register_sequence(adev,
+						 fiji_mgcg_cgcg_init,
+						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
+		amdgpu_program_register_sequence(adev,
+						 golden_settings_fiji_a10,
+						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
+		amdgpu_program_register_sequence(adev,
+						 fiji_golden_common_all,
+						 (const u32)ARRAY_SIZE(fiji_golden_common_all));
+		break;
+
 	case CHIP_TONGA:
 		amdgpu_program_register_sequence(adev,
 						 tonga_mgcg_cgcg_init,
@@ -526,6 +610,7 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib ib;
+	struct fence *f = NULL;
 	uint32_t scratch;
 	uint32_t tmp = 0;
 	unsigned i;
@@ -537,29 +622,27 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
 		return r;
 	}
 	WREG32(scratch, 0xCAFEDEAD);
+	memset(&ib, 0, sizeof(ib));
 	r = amdgpu_ib_get(ring, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
-		amdgpu_gfx_scratch_free(adev, scratch);
-		return r;
+		goto err1;
 	}
 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
 	ib.ptr[2] = 0xDEADBEEF;
 	ib.length_dw = 3;
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		amdgpu_gfx_scratch_free(adev, scratch);
-		amdgpu_ib_free(adev, &ib);
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = amdgpu_fence_wait(ib.fence, false);
+
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err2;
+
+	r = fence_wait(f, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
-		amdgpu_gfx_scratch_free(adev, scratch);
-		amdgpu_ib_free(adev, &ib);
-		return r;
+		goto err2;
 	}
 	for (i = 0; i < adev->usec_timeout; i++) {
 		tmp = RREG32(scratch);
@@ -569,14 +652,18 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	if (i < adev->usec_timeout) {
 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
-			 ib.fence->ring->idx, i);
+			 ring->idx, i);
+		goto err2;
 	} else {
 		DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
 			  scratch, tmp);
 		r = -EINVAL;
 	}
-	amdgpu_gfx_scratch_free(adev, scratch);
+err2:
+	fence_put(f);
 	amdgpu_ib_free(adev, &ib);
+err1:
+	amdgpu_gfx_scratch_free(adev, scratch);
 	return r;
 }
 
@@ -601,6 +688,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_CARRIZO:
 		chip_name = "carrizo";
 		break;
+	case CHIP_FIJI:
+		chip_name = "fiji";
+		break;
 	default:
 		BUG();
 	}
@@ -1236,6 +1326,7 @@ static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
 			adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden;
 			WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden);
 		}
+	case CHIP_FIJI:
 	case CHIP_TONGA:
 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
 			switch (reg_offset) {
@@ -1914,7 +2005,7 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev,
 }
 
 /**
- * gmc_v8_0_init_compute_vmid - gart enable
+ * gfx_v8_0_init_compute_vmid - gart enable
  *
  * @rdev: amdgpu_device pointer
  *
@@ -1924,7 +2015,7 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev,
 #define DEFAULT_SH_MEM_BASES	(0x6000)
 #define FIRST_COMPUTE_VMID	(8)
 #define LAST_COMPUTE_VMID	(16)
-static void gmc_v8_0_init_compute_vmid(struct amdgpu_device *adev)
+static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
 {
 	int i;
 	uint32_t sh_mem_config;
@@ -1984,6 +2075,23 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
 		break;
+	case CHIP_FIJI:
+		adev->gfx.config.max_shader_engines = 4;
+		adev->gfx.config.max_tile_pipes = 16;
+		adev->gfx.config.max_cu_per_sh = 16;
+		adev->gfx.config.max_sh_per_se = 1;
+		adev->gfx.config.max_backends_per_se = 4;
+		adev->gfx.config.max_texture_channel_caches = 8;
+		adev->gfx.config.max_gprs = 256;
+		adev->gfx.config.max_gs_threads = 32;
+		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 = 0x130;
+		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
+		break;
 	case CHIP_TONGA:
 		adev->gfx.config.max_shader_engines = 4;
 		adev->gfx.config.max_tile_pipes = 8;
@@ -2078,7 +2186,7 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
 
 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
 	adev->gfx.config.mem_max_burst_length_bytes = 256;
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		/* Get memory bank mapping mode. */
 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
@@ -2174,7 +2282,7 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
 	vi_srbm_select(adev, 0, 0, 0, 0);
 	mutex_unlock(&adev->srbm_mutex);
 
-	gmc_v8_0_init_compute_vmid(adev);
+	gfx_v8_0_init_compute_vmid(adev);
 
 	mutex_lock(&adev->grbm_idx_mutex);
 	/*
@@ -2490,6 +2598,7 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
 	switch (adev->asic_type) {
 	case CHIP_TONGA:
+	case CHIP_FIJI:
 		amdgpu_ring_write(ring, 0x16000012);
 		amdgpu_ring_write(ring, 0x0000002A);
 		break;
@@ -3131,7 +3240,8 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev)
 
 		/* enable the doorbell if requested */
 		if (use_doorbell) {
-			if (adev->asic_type == CHIP_CARRIZO) {
+			if ((adev->asic_type == CHIP_CARRIZO) ||
+			    (adev->asic_type == CHIP_FIJI)) {
 				WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER,
 				       AMDGPU_DOORBELL_KIQ << 2);
 				WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER,
@@ -3875,7 +3985,8 @@ static bool gfx_v8_0_ring_emit_semaphore(struct amdgpu_ring *ring,
 	unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
 
 	if (ring->adev->asic_type == CHIP_TOPAZ ||
-	    ring->adev->asic_type == CHIP_TONGA)
+	    ring->adev->asic_type == CHIP_TONGA ||
+	    ring->adev->asic_type == CHIP_FIJI)
 		/* we got a hw semaphore bug in VI TONGA, return false to switch back to sw fence wait */
 		return false;
 	else {
@@ -4268,6 +4379,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.is_lockup = gfx_v8_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
@@ -4284,6 +4396,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.test_ring = gfx_v8_0_ring_test_ring,
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.is_lockup = gfx_v8_0_ring_is_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)

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

@@ -523,17 +523,11 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev)
 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 			    amdgpu_vm_block_size - 9);
@@ -636,7 +630,7 @@ static int gmc_v7_0_vm_init(struct amdgpu_device *adev)
 	adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
 
 	/* base offset of vram pages */
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
 		tmp <<= 22;
 		adev->vm_manager.vram_base_offset = tmp;
@@ -841,7 +835,7 @@ static int gmc_v7_0_early_init(void *handle)
 	gmc_v7_0_set_gart_funcs(adev);
 	gmc_v7_0_set_irq_funcs(adev);
 
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
 	} else {
 		u32 tmp = RREG32(mmMC_SEQ_MISC0);
@@ -852,6 +846,13 @@ static int gmc_v7_0_early_init(void *handle)
 	return 0;
 }
 
+static int gmc_v7_0_late_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
+}
+
 static int gmc_v7_0_sw_init(void *handle)
 {
 	int r;
@@ -957,7 +958,7 @@ static int gmc_v7_0_hw_init(void *handle)
 
 	gmc_v7_0_mc_program(adev);
 
-	if (!(adev->flags & AMDGPU_IS_APU)) {
+	if (!(adev->flags & AMD_IS_APU)) {
 		r = gmc_v7_0_mc_load_microcode(adev);
 		if (r) {
 			DRM_ERROR("Failed to load MC firmware!\n");
@@ -976,6 +977,7 @@ static int gmc_v7_0_hw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
 	gmc_v7_0_gart_disable(adev);
 
 	return 0;
@@ -1172,7 +1174,7 @@ static int gmc_v7_0_soft_reset(void *handle)
 
 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
-		if (!(adev->flags & AMDGPU_IS_APU))
+		if (!(adev->flags & AMD_IS_APU))
 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
 	}
@@ -1282,7 +1284,7 @@ static int gmc_v7_0_set_clockgating_state(void *handle,
 	if (state == AMD_CG_STATE_GATE)
 		gate = true;
 
-	if (!(adev->flags & AMDGPU_IS_APU)) {
+	if (!(adev->flags & AMD_IS_APU)) {
 		gmc_v7_0_enable_mc_mgcg(adev, gate);
 		gmc_v7_0_enable_mc_ls(adev, gate);
 	}
@@ -1301,7 +1303,7 @@ static int gmc_v7_0_set_powergating_state(void *handle,
 
 const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
 	.early_init = gmc_v7_0_early_init,
-	.late_init = NULL,
+	.late_init = gmc_v7_0_late_init,
 	.sw_init = gmc_v7_0_sw_init,
 	.sw_fini = gmc_v7_0_sw_fini,
 	.hw_init = gmc_v7_0_hw_init,

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

@@ -44,6 +44,7 @@ static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
 
 MODULE_FIRMWARE("amdgpu/topaz_mc.bin");
 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
+MODULE_FIRMWARE("amdgpu/fiji_mc.bin");
 
 static const u32 golden_settings_tonga_a11[] =
 {
@@ -61,6 +62,19 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 };
 
+static const u32 golden_settings_fiji_a10[] =
+{
+	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
+	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
+	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
+	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
+};
+
+static const u32 fiji_mgcg_cgcg_init[] =
+{
+	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
+};
+
 static const u32 golden_settings_iceland_a11[] =
 {
 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
@@ -90,6 +104,14 @@ static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
 						 golden_settings_iceland_a11,
 						 (const u32)ARRAY_SIZE(golden_settings_iceland_a11));
 		break;
+	case CHIP_FIJI:
+		amdgpu_program_register_sequence(adev,
+						 fiji_mgcg_cgcg_init,
+						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
+		amdgpu_program_register_sequence(adev,
+						 golden_settings_fiji_a10,
+						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
+		break;
 	case CHIP_TONGA:
 		amdgpu_program_register_sequence(adev,
 						 tonga_mgcg_cgcg_init,
@@ -202,6 +224,9 @@ static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_TONGA:
 		chip_name = "tonga";
 		break;
+	case CHIP_FIJI:
+		chip_name = "fiji";
+		break;
 	case CHIP_CARRIZO:
 		return 0;
 	default: BUG();
@@ -628,19 +653,12 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
-	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 			    amdgpu_vm_block_size - 9);
@@ -737,7 +755,7 @@ static int gmc_v8_0_vm_init(struct amdgpu_device *adev)
 	adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
 
 	/* base offset of vram pages */
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
 		tmp <<= 22;
 		adev->vm_manager.vram_base_offset = tmp;
@@ -816,7 +834,7 @@ static int gmc_v8_0_early_init(void *handle)
 	gmc_v8_0_set_gart_funcs(adev);
 	gmc_v8_0_set_irq_funcs(adev);
 
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
 	} else {
 		u32 tmp = RREG32(mmMC_SEQ_MISC0);
@@ -827,6 +845,13 @@ static int gmc_v8_0_early_init(void *handle)
 	return 0;
 }
 
+static int gmc_v8_0_late_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
+}
+
 static int gmc_v8_0_sw_init(void *handle)
 {
 	int r;
@@ -934,7 +959,7 @@ static int gmc_v8_0_hw_init(void *handle)
 
 	gmc_v8_0_mc_program(adev);
 
-	if (!(adev->flags & AMDGPU_IS_APU)) {
+	if (!(adev->flags & AMD_IS_APU)) {
 		r = gmc_v8_0_mc_load_microcode(adev);
 		if (r) {
 			DRM_ERROR("Failed to load MC firmware!\n");
@@ -953,6 +978,7 @@ static int gmc_v8_0_hw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
 	gmc_v8_0_gart_disable(adev);
 
 	return 0;
@@ -1147,7 +1173,7 @@ static int gmc_v8_0_soft_reset(void *handle)
 
 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
-		if (!(adev->flags & AMDGPU_IS_APU))
+		if (!(adev->flags & AMD_IS_APU))
 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
 	}
@@ -1263,7 +1289,7 @@ static int gmc_v8_0_set_powergating_state(void *handle,
 
 const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
 	.early_init = gmc_v8_0_early_init,
-	.late_init = NULL,
+	.late_init = gmc_v8_0_late_init,
 	.sw_init = gmc_v8_0_sw_init,
 	.sw_fini = gmc_v8_0_sw_fini,
 	.hw_init = gmc_v8_0_hw_init,

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

@@ -2163,5 +2163,10 @@
 #define SDMA_PKT_NOP_HEADER_sub_op_shift  8
 #define SDMA_PKT_NOP_HEADER_SUB_OP(x) (((x) & SDMA_PKT_NOP_HEADER_sub_op_mask) << SDMA_PKT_NOP_HEADER_sub_op_shift)
 
+/*define for count field*/
+#define SDMA_PKT_NOP_HEADER_count_offset 0
+#define SDMA_PKT_NOP_HEADER_count_mask   0x00003FFF
+#define SDMA_PKT_NOP_HEADER_count_shift  16
+#define SDMA_PKT_NOP_HEADER_COUNT(x) (((x) & SDMA_PKT_NOP_HEADER_count_mask) << SDMA_PKT_NOP_HEADER_count_shift)
 
 #endif /* __ICELAND_SDMA_PKT_OPEN_H_ */

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

@@ -623,7 +623,9 @@ int iceland_smu_init(struct amdgpu_device *adev)
 
 	/* Allocate FW image data structure and header buffer */
 	ret = amdgpu_bo_create(adev, image_size, PAGE_SIZE,
-			       true, AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, toc_buf);
+			       true, AMDGPU_GEM_DOMAIN_VRAM,
+			       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			       NULL, toc_buf);
 	if (ret) {
 		DRM_ERROR("Failed to allocate memory for TOC buffer\n");
 		return -ENOMEM;

drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c

@@ -146,6 +146,8 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 
 		if (adev->firmware.smu_load) {
 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
@@ -218,6 +220,19 @@ static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring)
 	WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
 }
 
+static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * sdma_v2_4_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -245,8 +260,8 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 2)
-		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
+	sdma_v2_4_ring_insert_nop(ring, (10 - (ring->wptr & 7)) % 8);
+
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid));
 	/* base must be 32 byte aligned */
@@ -673,6 +688,7 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib ib;
+	struct fence *f = NULL;
 	unsigned i;
 	unsigned index;
 	int r;
@@ -688,12 +704,11 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 	gpu_addr = adev->wb.gpu_addr + (index * 4);
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
-
+	memset(&ib, 0, sizeof(ib));
 	r = amdgpu_ib_get(ring, NULL, 256, &ib);
 	if (r) {
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
-		return r;
+		goto err0;
 	}
 
 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
@@ -707,19 +722,16 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[7] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 	ib.length_dw = 8;
 
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = amdgpu_fence_wait(ib.fence, false);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err1;
+
+	r = fence_wait(f, false);
 	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
-		return r;
+		goto err1;
 	}
 	for (i = 0; i < adev->usec_timeout; i++) {
 		tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -729,12 +741,17 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	if (i < adev->usec_timeout) {
 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
-			 ib.fence->ring->idx, i);
+			 ring->idx, i);
+		goto err1;
 	} else {
 		DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
 		r = -EINVAL;
 	}
+
+err1:
+	fence_put(f);
 	amdgpu_ib_free(adev, &ib);
+err0:
 	amdgpu_wb_free(adev, index);
 	return r;
 }
@@ -877,8 +894,19 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
+				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 }
 
 /**
@@ -1312,6 +1340,7 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
 	.test_ring = sdma_v2_4_ring_test_ring,
 	.test_ib = sdma_v2_4_ring_test_ib,
 	.is_lockup = sdma_v2_4_ring_is_lockup,
+	.insert_nop = sdma_v2_4_ring_insert_nop,
 };
 
 static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev)
@@ -1348,19 +1377,19 @@ static void sdma_v2_4_set_irq_funcs(struct amdgpu_device *adev)
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ring *ring,
+static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ib *ib,
 				       uint64_t src_offset,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
-			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**
@@ -1373,16 +1402,16 @@ static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ring *ring,
  *
  * Fill GPU buffers using the DMA engine (VI).
  */
-static void sdma_v2_4_emit_fill_buffer(struct amdgpu_ring *ring,
+static void sdma_v2_4_emit_fill_buffer(struct amdgpu_ib *ib,
 				       uint32_t src_data,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {
@@ -1415,5 +1444,6 @@ static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev)
 	if (adev->vm_manager.vm_pte_funcs == NULL) {
 		adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs;
 		adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+		adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
 	}
 }

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

@@ -53,6 +53,8 @@ MODULE_FIRMWARE("amdgpu/tonga_sdma.bin");
 MODULE_FIRMWARE("amdgpu/tonga_sdma1.bin");
 MODULE_FIRMWARE("amdgpu/carrizo_sdma.bin");
 MODULE_FIRMWARE("amdgpu/carrizo_sdma1.bin");
+MODULE_FIRMWARE("amdgpu/fiji_sdma.bin");
+MODULE_FIRMWARE("amdgpu/fiji_sdma1.bin");
 
 static const u32 sdma_offsets[SDMA_MAX_INSTANCE] =
 {
@@ -80,6 +82,24 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100
 };
 
+static const u32 golden_settings_fiji_a10[] =
+{
+	mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007,
+	mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100,
+	mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100,
+	mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100,
+	mmSDMA1_CHICKEN_BITS, 0xfc910007, 0x00810007,
+	mmSDMA1_GFX_IB_CNTL, 0x800f0111, 0x00000100,
+	mmSDMA1_RLC0_IB_CNTL, 0x800f0111, 0x00000100,
+	mmSDMA1_RLC1_IB_CNTL, 0x800f0111, 0x00000100,
+};
+
+static const u32 fiji_mgcg_cgcg_init[] =
+{
+	mmSDMA0_CLK_CTRL, 0xff000ff0, 0x00000100,
+	mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100
+};
+
 static const u32 cz_golden_settings_a11[] =
 {
 	mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007,
@@ -122,6 +142,14 @@ static const u32 cz_mgcg_cgcg_init[] =
 static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev)
 {
 	switch (adev->asic_type) {
+	case CHIP_FIJI:
+		amdgpu_program_register_sequence(adev,
+						 fiji_mgcg_cgcg_init,
+						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
+		amdgpu_program_register_sequence(adev,
+						 golden_settings_fiji_a10,
+						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
+		break;
 	case CHIP_TONGA:
 		amdgpu_program_register_sequence(adev,
 						 tonga_mgcg_cgcg_init,
@@ -167,6 +195,9 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
 	case CHIP_TONGA:
 		chip_name = "tonga";
 		break;
+	case CHIP_FIJI:
+		chip_name = "fiji";
+		break;
 	case CHIP_CARRIZO:
 		chip_name = "carrizo";
 		break;
@@ -187,6 +218,8 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
 		adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
 		adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+		if (adev->sdma[i].feature_version >= 20)
+			adev->sdma[i].burst_nop = true;
 
 		if (adev->firmware.smu_load) {
 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
@@ -273,6 +306,19 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
 	}
 }
 
+static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
+{
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+	int i;
+
+	for (i = 0; i < count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			amdgpu_ring_write(ring, ring->nop |
+				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
+		else
+			amdgpu_ring_write(ring, ring->nop);
+}
+
 /**
  * sdma_v3_0_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -299,8 +345,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* IB packet must end on a 8 DW boundary */
-	while ((ring->wptr & 7) != 2)
-		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
+	sdma_v3_0_ring_insert_nop(ring, (10 - (ring->wptr & 7)) % 8);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid));
@@ -763,6 +808,7 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib ib;
+	struct fence *f = NULL;
 	unsigned i;
 	unsigned index;
 	int r;
@@ -778,12 +824,11 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 	gpu_addr = adev->wb.gpu_addr + (index * 4);
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
-
+	memset(&ib, 0, sizeof(ib));
 	r = amdgpu_ib_get(ring, NULL, 256, &ib);
 	if (r) {
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
-		return r;
+		goto err0;
 	}
 
 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
@@ -797,19 +842,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
 	ib.length_dw = 8;
 
-	r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
-	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
-		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
-		return r;
-	}
-	r = amdgpu_fence_wait(ib.fence, false);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL,
+						 AMDGPU_FENCE_OWNER_UNDEFINED,
+						 &f);
+	if (r)
+		goto err1;
+
+	r = fence_wait(f, false);
 	if (r) {
-		amdgpu_ib_free(adev, &ib);
-		amdgpu_wb_free(adev, index);
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
-		return r;
+		goto err1;
 	}
 	for (i = 0; i < adev->usec_timeout; i++) {
 		tmp = le32_to_cpu(adev->wb.wb[index]);
@@ -819,12 +861,16 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	if (i < adev->usec_timeout) {
 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
-			 ib.fence->ring->idx, i);
+			 ring->idx, i);
+		goto err1;
 	} else {
 		DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
 		r = -EINVAL;
 	}
+err1:
+	fence_put(f);
 	amdgpu_ib_free(adev, &ib);
+err0:
 	amdgpu_wb_free(adev, index);
 	return r;
 }
@@ -967,8 +1013,19 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib,
  */
 static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib)
 {
-	while (ib->length_dw & 0x7)
-		ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+	struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+	u32 pad_count;
+	int i;
+
+	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	for (i = 0; i < pad_count; i++)
+		if (sdma && sdma->burst_nop && (i == 0))
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
+				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
+		else
+			ib->ptr[ib->length_dw++] =
+				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 }
 
 /**
@@ -1406,6 +1463,7 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
 	.test_ring = sdma_v3_0_ring_test_ring,
 	.test_ib = sdma_v3_0_ring_test_ib,
 	.is_lockup = sdma_v3_0_ring_is_lockup,
+	.insert_nop = sdma_v3_0_ring_insert_nop,
 };
 
 static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev)
@@ -1442,19 +1500,19 @@ static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev)
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ring *ring,
+static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ib *ib,
 				       uint64_t src_offset,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
-			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**
@@ -1467,16 +1525,16 @@ static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ring *ring,
  *
  * Fill GPU buffers using the DMA engine (VI).
  */
-static void sdma_v3_0_emit_fill_buffer(struct amdgpu_ring *ring,
+static void sdma_v3_0_emit_fill_buffer(struct amdgpu_ib *ib,
 				       uint32_t src_data,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
-	amdgpu_ring_write(ring, src_data);
-	amdgpu_ring_write(ring, byte_count);
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = src_data;
+	ib->ptr[ib->length_dw++] = byte_count;
 }
 
 static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {
@@ -1509,5 +1567,6 @@ static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 	if (adev->vm_manager.vm_pte_funcs == NULL) {
 		adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs;
 		adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+		adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
 	}
 }

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

@@ -2236,5 +2236,10 @@
 #define SDMA_PKT_NOP_HEADER_sub_op_shift  8
 #define SDMA_PKT_NOP_HEADER_SUB_OP(x) (((x) & SDMA_PKT_NOP_HEADER_sub_op_mask) << SDMA_PKT_NOP_HEADER_sub_op_shift)
 
+/*define for count field*/
+#define SDMA_PKT_NOP_HEADER_count_offset 0
+#define SDMA_PKT_NOP_HEADER_count_mask   0x00003FFF
+#define SDMA_PKT_NOP_HEADER_count_shift  16
+#define SDMA_PKT_NOP_HEADER_COUNT(x) (((x) & SDMA_PKT_NOP_HEADER_count_mask) << SDMA_PKT_NOP_HEADER_count_shift)
 
 #endif /* __TONGA_SDMA_PKT_OPEN_H_ */

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

@@ -761,7 +761,9 @@ int tonga_smu_init(struct amdgpu_device *adev)
 
 	/* Allocate FW image data structure and header buffer */
 	ret = amdgpu_bo_create(adev, image_size, PAGE_SIZE,
-				true, AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, toc_buf);
+			       true, AMDGPU_GEM_DOMAIN_VRAM,
+			       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			       NULL, toc_buf);
 	if (ret) {
 		DRM_ERROR("Failed to allocate memory for TOC buffer\n");
 		return -ENOMEM;
@@ -769,7 +771,9 @@ int tonga_smu_init(struct amdgpu_device *adev)
 
 	/* Allocate buffer for SMU internal buffer */
 	ret = amdgpu_bo_create(adev, smu_internal_buffer_size, PAGE_SIZE,
-				true, AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, smu_buf);
+			       true, AMDGPU_GEM_DOMAIN_VRAM,
+			       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			       NULL, smu_buf);
 	if (ret) {
 		DRM_ERROR("Failed to allocate memory for SMU internal buffer\n");
 		return -ENOMEM;

drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c

@@ -534,7 +534,7 @@ static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
 static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int r;
 
 	r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
@@ -555,14 +555,14 @@ static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring)
 		goto error;
 	}
 
-	r = amdgpu_fence_wait(fence, false);
+	r = fence_wait(fence, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
 		goto error;
 	}
 	DRM_INFO("ib test on ring %d succeeded\n",  ring->idx);
 error:
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	amdgpu_asic_set_uvd_clocks(adev, 0, 0);
 	return r;
 }
@@ -886,6 +886,7 @@ static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
 	.test_ring = uvd_v4_2_ring_test_ring,
 	.test_ib = uvd_v4_2_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c

@@ -580,7 +580,7 @@ static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
 static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int r;
 
 	r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
@@ -601,14 +601,14 @@ static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring)
 		goto error;
 	}
 
-	r = amdgpu_fence_wait(fence, false);
+	r = fence_wait(fence, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
 		goto error;
 	}
 	DRM_INFO("ib test on ring %d succeeded\n",  ring->idx);
 error:
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	amdgpu_asic_set_uvd_clocks(adev, 0, 0);
 	return r;
 }
@@ -825,6 +825,7 @@ static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
 	.test_ring = uvd_v5_0_ring_test_ring,
 	.test_ib = uvd_v5_0_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c

@@ -575,7 +575,7 @@ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
  */
 static int uvd_v6_0_ring_test_ib(struct amdgpu_ring *ring)
 {
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int r;
 
 	r = amdgpu_uvd_get_create_msg(ring, 1, NULL);
@@ -590,14 +590,14 @@ static int uvd_v6_0_ring_test_ib(struct amdgpu_ring *ring)
 		goto error;
 	}
 
-	r = amdgpu_fence_wait(fence, false);
+	r = fence_wait(fence, false);
 	if (r) {
 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
 		goto error;
 	}
 	DRM_INFO("ib test on ring %d succeeded\n",  ring->idx);
 error:
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	return r;
 }
 
@@ -805,6 +805,7 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = {
 	.test_ring = uvd_v6_0_ring_test_ring,
 	.test_ib = uvd_v6_0_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void uvd_v6_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/vce_v2_0.c

@@ -643,6 +643,7 @@ static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs = {
 	.test_ring = amdgpu_vce_ring_test_ring,
 	.test_ib = amdgpu_vce_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void vce_v2_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/vce_v3_0.c

@@ -32,8 +32,8 @@
 #include "vid.h"
 #include "vce/vce_3_0_d.h"
 #include "vce/vce_3_0_sh_mask.h"
-#include "oss/oss_2_0_d.h"
-#include "oss/oss_2_0_sh_mask.h"
+#include "oss/oss_3_0_d.h"
+#include "oss/oss_3_0_sh_mask.h"
 #include "gca/gfx_8_0_d.h"
 #include "smu/smu_7_1_2_d.h"
 #include "smu/smu_7_1_2_sh_mask.h"
@@ -205,7 +205,14 @@ static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
 	u32 tmp;
 	unsigned ret;
 
-	if (adev->flags & AMDGPU_IS_APU)
+	/* Fiji is single pipe */
+	if (adev->asic_type == CHIP_FIJI) {
+		ret = AMDGPU_VCE_HARVEST_VCE1;
+		return ret;
+	}
+
+	/* Tonga and CZ are dual or single pipe */
+	if (adev->flags & AMD_IS_APU)
 		tmp = (RREG32_SMC(ixVCE_HARVEST_FUSE_MACRO__ADDRESS) &
 		       VCE_HARVEST_FUSE_MACRO__MASK) >>
 			VCE_HARVEST_FUSE_MACRO__SHIFT;
@@ -419,17 +426,41 @@ static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
 static bool vce_v3_0_is_idle(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	u32 mask = 0;
+	int idx;
+
+	for (idx = 0; idx < 2; ++idx) {
+		if (adev->vce.harvest_config & (1 << idx))
+			continue;
+
+		if (idx == 0)
+			mask |= SRBM_STATUS2__VCE0_BUSY_MASK;
+		else
+			mask |= SRBM_STATUS2__VCE1_BUSY_MASK;
+	}
 
-	return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
+	return !(RREG32(mmSRBM_STATUS2) & mask);
 }
 
 static int vce_v3_0_wait_for_idle(void *handle)
 {
 	unsigned i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	u32 mask = 0;
+	int idx;
+
+	for (idx = 0; idx < 2; ++idx) {
+		if (adev->vce.harvest_config & (1 << idx))
+			continue;
+
+		if (idx == 0)
+			mask |= SRBM_STATUS2__VCE0_BUSY_MASK;
+		else
+			mask |= SRBM_STATUS2__VCE1_BUSY_MASK;
+	}
 
 	for (i = 0; i < adev->usec_timeout; i++) {
-		if (!(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK))
+		if (!(RREG32(mmSRBM_STATUS2) & mask))
 			return 0;
 	}
 	return -ETIMEDOUT;
@@ -438,9 +469,21 @@ static int vce_v3_0_wait_for_idle(void *handle)
 static int vce_v3_0_soft_reset(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	u32 mask = 0;
+	int idx;
 
-	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_VCE_MASK,
-			~SRBM_SOFT_RESET__SOFT_RESET_VCE_MASK);
+	for (idx = 0; idx < 2; ++idx) {
+		if (adev->vce.harvest_config & (1 << idx))
+			continue;
+
+		if (idx == 0)
+			mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK;
+		else
+			mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK;
+	}
+	WREG32_P(mmSRBM_SOFT_RESET, mask,
+		 ~(SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK |
+		   SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK));
 	mdelay(5);
 
 	return vce_v3_0_start(adev);
@@ -601,6 +644,7 @@ static const struct amdgpu_ring_funcs vce_v3_0_ring_funcs = {
 	.test_ring = amdgpu_vce_ring_test_ring,
 	.test_ib = amdgpu_vce_ring_test_ib,
 	.is_lockup = amdgpu_ring_test_lockup,
+	.insert_nop = amdgpu_ring_insert_nop,
 };
 
 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)

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

@@ -203,6 +203,17 @@ static const u32 tonga_mgcg_cgcg_init[] =
 	mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104,
 };
 
+static const u32 fiji_mgcg_cgcg_init[] =
+{
+	mmCGTT_DRM_CLK_CTRL0, 0xffffffff, 0x00600100,
+	mmPCIE_INDEX, 0xffffffff, 0x0140001c,
+	mmPCIE_DATA, 0x000f0000, 0x00000000,
+	mmSMC_IND_INDEX_4, 0xffffffff, 0xC060000C,
+	mmSMC_IND_DATA_4, 0xc0000fff, 0x00000100,
+	mmCGTT_DRM_CLK_CTRL0, 0xff000fff, 0x00000100,
+	mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104,
+};
+
 static const u32 iceland_mgcg_cgcg_init[] =
 {
 	mmPCIE_INDEX, 0xffffffff, ixPCIE_CNTL2,
@@ -232,6 +243,11 @@ static void vi_init_golden_registers(struct amdgpu_device *adev)
 						 iceland_mgcg_cgcg_init,
 						 (const u32)ARRAY_SIZE(iceland_mgcg_cgcg_init));
 		break;
+	case CHIP_FIJI:
+		amdgpu_program_register_sequence(adev,
+						 fiji_mgcg_cgcg_init,
+						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
+		break;
 	case CHIP_TONGA:
 		amdgpu_program_register_sequence(adev,
 						 tonga_mgcg_cgcg_init,
@@ -261,7 +277,7 @@ static u32 vi_get_xclk(struct amdgpu_device *adev)
 	u32 reference_clock = adev->clock.spll.reference_freq;
 	u32 tmp;
 
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return reference_clock;
 
 	tmp = RREG32_SMC(ixCG_CLKPIN_CNTL_2);
@@ -362,6 +378,26 @@ static struct amdgpu_allowed_register_entry cz_allowed_read_registers[] = {
 
 static struct amdgpu_allowed_register_entry vi_allowed_read_registers[] = {
 	{mmGRBM_STATUS, false},
+	{mmGRBM_STATUS2, false},
+	{mmGRBM_STATUS_SE0, false},
+	{mmGRBM_STATUS_SE1, false},
+	{mmGRBM_STATUS_SE2, false},
+	{mmGRBM_STATUS_SE3, false},
+	{mmSRBM_STATUS, false},
+	{mmSRBM_STATUS2, false},
+	{mmSRBM_STATUS3, false},
+	{mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET, false},
+	{mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET, false},
+	{mmCP_STAT, false},
+	{mmCP_STALLED_STAT1, false},
+	{mmCP_STALLED_STAT2, false},
+	{mmCP_STALLED_STAT3, false},
+	{mmCP_CPF_BUSY_STAT, false},
+	{mmCP_CPF_STALLED_STAT1, false},
+	{mmCP_CPF_STATUS, false},
+	{mmCP_CPC_BUSY_STAT, false},
+	{mmCP_CPC_STALLED_STAT1, false},
+	{mmCP_CPC_STATUS, false},
 	{mmGB_ADDR_CONFIG, false},
 	{mmMC_ARB_RAMCFG, false},
 	{mmGB_TILE_MODE0, false},
@@ -449,6 +485,7 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num,
 		asic_register_table = tonga_allowed_read_registers;
 		size = ARRAY_SIZE(tonga_allowed_read_registers);
 		break;
+	case CHIP_FIJI:
 	case CHIP_TONGA:
 	case CHIP_CARRIZO:
 		asic_register_table = cz_allowed_read_registers;
@@ -751,7 +788,7 @@ static void vi_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask)
 		srbm_soft_reset =
 			REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
 
-	if (!(adev->flags & AMDGPU_IS_APU)) {
+	if (!(adev->flags & AMD_IS_APU)) {
 		if (reset_mask & AMDGPU_RESET_MC)
 		srbm_soft_reset =
 			REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
@@ -971,7 +1008,7 @@ static void vi_pcie_gen3_enable(struct amdgpu_device *adev)
 	if (amdgpu_pcie_gen2 == 0)
 		return;
 
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return;
 
 	ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask);
@@ -999,7 +1036,7 @@ static void vi_enable_doorbell_aperture(struct amdgpu_device *adev,
 	u32 tmp;
 
 	/* not necessary on CZ */
-	if (adev->flags & AMDGPU_IS_APU)
+	if (adev->flags & AMD_IS_APU)
 		return;
 
 	tmp = RREG32(mmBIF_DOORBELL_APER_EN);
@@ -1127,6 +1164,74 @@ static const struct amdgpu_ip_block_version tonga_ip_blocks[] =
 	},
 };
 
+static const struct amdgpu_ip_block_version fiji_ip_blocks[] =
+{
+	/* ORDER MATTERS! */
+	{
+		.type = AMD_IP_BLOCK_TYPE_COMMON,
+		.major = 2,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &vi_common_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_GMC,
+		.major = 8,
+		.minor = 5,
+		.rev = 0,
+		.funcs = &gmc_v8_0_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_IH,
+		.major = 3,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &tonga_ih_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_SMC,
+		.major = 7,
+		.minor = 1,
+		.rev = 0,
+		.funcs = &fiji_dpm_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_DCE,
+		.major = 10,
+		.minor = 1,
+		.rev = 0,
+		.funcs = &dce_v10_0_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_GFX,
+		.major = 8,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &gfx_v8_0_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_SDMA,
+		.major = 3,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &sdma_v3_0_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_UVD,
+		.major = 6,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &uvd_v6_0_ip_funcs,
+	},
+	{
+		.type = AMD_IP_BLOCK_TYPE_VCE,
+		.major = 3,
+		.minor = 0,
+		.rev = 0,
+		.funcs = &vce_v3_0_ip_funcs,
+	},
+};
+
 static const struct amdgpu_ip_block_version cz_ip_blocks[] =
 {
 	/* ORDER MATTERS! */
@@ -1202,6 +1307,10 @@ int vi_set_ip_blocks(struct amdgpu_device *adev)
 		adev->ip_blocks = topaz_ip_blocks;
 		adev->num_ip_blocks = ARRAY_SIZE(topaz_ip_blocks);
 		break;
+	case CHIP_FIJI:
+		adev->ip_blocks = fiji_ip_blocks;
+		adev->num_ip_blocks = ARRAY_SIZE(fiji_ip_blocks);
+		break;
 	case CHIP_TONGA:
 		adev->ip_blocks = tonga_ip_blocks;
 		adev->num_ip_blocks = ARRAY_SIZE(tonga_ip_blocks);
@@ -1248,7 +1357,7 @@ static int vi_common_early_init(void *handle)
 	bool smc_enabled = false;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	if (adev->flags & AMDGPU_IS_APU) {
+	if (adev->flags & AMD_IS_APU) {
 		adev->smc_rreg = &cz_smc_rreg;
 		adev->smc_wreg = &cz_smc_wreg;
 	} else {
@@ -1279,6 +1388,7 @@ static int vi_common_early_init(void *handle)
 		if (amdgpu_smc_load_fw && smc_enabled)
 			adev->firmware.smu_load = true;
 		break;
+	case CHIP_FIJI:
 	case CHIP_TONGA:
 		adev->has_uvd = true;
 		adev->cg_flags = 0;

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

@@ -30,7 +30,7 @@ int cz_smu_start(struct amdgpu_device *adev);
 int cz_smu_fini(struct amdgpu_device *adev);
 
 extern const struct amd_ip_funcs tonga_dpm_ip_funcs;
-
+extern const struct amd_ip_funcs fiji_dpm_ip_funcs;
 extern const struct amd_ip_funcs iceland_dpm_ip_funcs;
 
 #endif

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

@@ -66,6 +66,11 @@
 
 #define AMDGPU_NUM_OF_VMIDS			8
 
+#define		PIPEID(x)					((x) << 0)
+#define		MEID(x)						((x) << 2)
+#define		VMID(x)						((x) << 4)
+#define		QUEUEID(x)					((x) << 8)
+
 #define RB_BITMAP_WIDTH_PER_SH     2
 
 #define MC_SEQ_MISC0__MT__MASK	0xf0000000

drivers/gpu/drm/amd/amdkfd/Kconfig

@@ -4,6 +4,6 @@
 
 config HSA_AMD
 	tristate "HSA kernel driver for AMD GPU devices"
-	depends on DRM_RADEON && AMD_IOMMU_V2 && X86_64
+	depends on (DRM_RADEON || DRM_AMDGPU) && AMD_IOMMU_V2 && X86_64
 	help
 	  Enable this if you want to use HSA features on AMD GPU devices.

drivers/gpu/drm/amd/amdkfd/Makefile

@@ -2,7 +2,8 @@
 # Makefile for Heterogenous System Architecture support for AMD GPU devices
 #
 
-ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/
+ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/  \
+		-Idrivers/gpu/drm/amd/include/asic_reg
 
 amdkfd-y	:= kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
 		kfd_pasid.o kfd_doorbell.o kfd_flat_memory.o \

drivers/gpu/drm/amd/amdkfd/cik_regs.h

@@ -65,17 +65,6 @@
 
 #define	AQL_ENABLE					1
 
-#define	SDMA_RB_VMID(x)					(x << 24)
-#define	SDMA_RB_ENABLE					(1 << 0)
-#define	SDMA_RB_SIZE(x)					((x) << 1) /* log2 */
-#define	SDMA_RPTR_WRITEBACK_ENABLE			(1 << 12)
-#define	SDMA_RPTR_WRITEBACK_TIMER(x)			((x) << 16) /* log2 */
-#define	SDMA_OFFSET(x)					(x << 0)
-#define	SDMA_DB_ENABLE					(1 << 28)
-#define	SDMA_ATC					(1 << 0)
-#define	SDMA_VA_PTR32					(1 << 4)
-#define	SDMA_VA_SHARED_BASE(x)				(x << 8)
-
 #define GRBM_GFX_INDEX					0x30800
 
 #define	ATC_VMID_PASID_MAPPING_VALID			(1U << 31)

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c

@@ -31,7 +31,7 @@
 #include <uapi/linux/kfd_ioctl.h>
 #include <linux/time.h>
 #include <linux/mm.h>
-#include <uapi/asm-generic/mman-common.h>
+#include <linux/mman.h>
 #include <asm/processor.h>
 #include "kfd_priv.h"
 #include "kfd_device_queue_manager.h"

drivers/gpu/drm/amd/amdkfd/kfd_device.c

@@ -80,7 +80,12 @@ static const struct kfd_deviceid supported_devices[] = {
 	{ 0x1318, &kaveri_device_info },	/* Kaveri */
 	{ 0x131B, &kaveri_device_info },	/* Kaveri */
 	{ 0x131C, &kaveri_device_info },	/* Kaveri */
-	{ 0x131D, &kaveri_device_info }		/* Kaveri */
+	{ 0x131D, &kaveri_device_info },	/* Kaveri */
+	{ 0x9870, &carrizo_device_info },	/* Carrizo */
+	{ 0x9874, &carrizo_device_info },	/* Carrizo */
+	{ 0x9875, &carrizo_device_info },	/* Carrizo */
+	{ 0x9876, &carrizo_device_info },	/* Carrizo */
+	{ 0x9877, &carrizo_device_info }	/* Carrizo */
 };
 
 static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c

@@ -23,6 +23,7 @@
 
 #include "kfd_device_queue_manager.h"
 #include "cik_regs.h"
+#include "oss/oss_2_4_sh_mask.h"
 
 static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
@@ -135,13 +136,16 @@ static int register_process_cik(struct device_queue_manager *dqm,
 static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 				struct qcm_process_device *qpd)
 {
-	uint32_t value = SDMA_ATC;
+	uint32_t value = (1 << SDMA0_RLC0_VIRTUAL_ADDR__ATC__SHIFT);
 
 	if (q->process->is_32bit_user_mode)
-		value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd));
+		value |= (1 << SDMA0_RLC0_VIRTUAL_ADDR__PTR32__SHIFT) |
+				get_sh_mem_bases_32(qpd_to_pdd(qpd));
 	else
-		value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64(
-							qpd_to_pdd(qpd)));
+		value |= ((get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd))) <<
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE__SHIFT) &
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE_MASK;
+
 	q->properties.sdma_vm_addr = value;
 }
 

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c

@@ -22,6 +22,10 @@
  */
 
 #include "kfd_device_queue_manager.h"
+#include "gca/gfx_8_0_enum.h"
+#include "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_enum.h"
+#include "oss/oss_3_0_sh_mask.h"
 
 static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
@@ -37,14 +41,40 @@ static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 
 void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops)
 {
-	pr_warn("amdkfd: VI DQM is not currently supported\n");
-
 	ops->set_cache_memory_policy = set_cache_memory_policy_vi;
 	ops->register_process = register_process_vi;
 	ops->initialize = initialize_cpsch_vi;
 	ops->init_sdma_vm = init_sdma_vm;
 }
 
+static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble)
+{
+	/* In 64-bit mode, we can only control the top 3 bits of the LDS,
+	 * scratch and GPUVM apertures.
+	 * The hardware fills in the remaining 59 bits according to the
+	 * following pattern:
+	 * LDS:		X0000000'00000000 - X0000001'00000000 (4GB)
+	 * Scratch:	X0000001'00000000 - X0000002'00000000 (4GB)
+	 * GPUVM:	Y0010000'00000000 - Y0020000'00000000 (1TB)
+	 *
+	 * (where X/Y is the configurable nybble with the low-bit 0)
+	 *
+	 * LDS and scratch will have the same top nybble programmed in the
+	 * top 3 bits of SH_MEM_BASES.PRIVATE_BASE.
+	 * GPUVM can have a different top nybble programmed in the
+	 * top 3 bits of SH_MEM_BASES.SHARED_BASE.
+	 * We don't bother to support different top nybbles
+	 * for LDS/Scratch and GPUVM.
+	 */
+
+	BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE ||
+		top_address_nybble == 0);
+
+	return top_address_nybble << 12 |
+			(top_address_nybble << 12) <<
+			SH_MEM_BASES__SHARED_BASE__SHIFT;
+}
+
 static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
 				   enum cache_policy default_policy,
@@ -52,18 +82,83 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				   void __user *alternate_aperture_base,
 				   uint64_t alternate_aperture_size)
 {
-	return false;
+	uint32_t default_mtype;
+	uint32_t ape1_mtype;
+
+	default_mtype = (default_policy == cache_policy_coherent) ?
+			MTYPE_CC :
+			MTYPE_NC;
+
+	ape1_mtype = (alternate_policy == cache_policy_coherent) ?
+			MTYPE_CC :
+			MTYPE_NC;
+
+	qpd->sh_mem_config = (qpd->sh_mem_config &
+			SH_MEM_CONFIG__ADDRESS_MODE_MASK) |
+		SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
+		default_mtype << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
+		ape1_mtype << SH_MEM_CONFIG__APE1_MTYPE__SHIFT |
+		SH_MEM_CONFIG__PRIVATE_ATC_MASK;
+
+	return true;
 }
 
 static int register_process_vi(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd)
 {
-	return -1;
+	struct kfd_process_device *pdd;
+	unsigned int temp;
+
+	BUG_ON(!dqm || !qpd);
+
+	pdd = qpd_to_pdd(qpd);
+
+	/* check if sh_mem_config register already configured */
+	if (qpd->sh_mem_config == 0) {
+		qpd->sh_mem_config =
+			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
+			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
+			MTYPE_CC << SH_MEM_CONFIG__APE1_MTYPE__SHIFT |
+			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
+
+		qpd->sh_mem_ape1_limit = 0;
+		qpd->sh_mem_ape1_base = 0;
+	}
+
+	if (qpd->pqm->process->is_32bit_user_mode) {
+		temp = get_sh_mem_bases_32(pdd);
+		qpd->sh_mem_bases = temp << SH_MEM_BASES__SHARED_BASE__SHIFT;
+		qpd->sh_mem_config |= SH_MEM_ADDRESS_MODE_HSA32 <<
+					SH_MEM_CONFIG__ADDRESS_MODE__SHIFT;
+	} else {
+		temp = get_sh_mem_bases_nybble_64(pdd);
+		qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
+		qpd->sh_mem_config |= SH_MEM_ADDRESS_MODE_HSA64 <<
+			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT;
+	}
+
+	pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n",
+		qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases);
+
+	return 0;
 }
 
 static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 				struct qcm_process_device *qpd)
 {
+	uint32_t value = (1 << SDMA0_RLC0_VIRTUAL_ADDR__ATC__SHIFT);
+
+	if (q->process->is_32bit_user_mode)
+		value |= (1 << SDMA0_RLC0_VIRTUAL_ADDR__PTR32__SHIFT) |
+				get_sh_mem_bases_32(qpd_to_pdd(qpd));
+	else
+		value |= ((get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd))) <<
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE__SHIFT) &
+				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE_MASK;
+
+	q->properties.sdma_vm_addr = value;
 }
 
 static int initialize_cpsch_vi(struct device_queue_manager *dqm)

drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c

@@ -33,7 +33,7 @@
 #include <linux/time.h>
 #include "kfd_priv.h"
 #include <linux/mm.h>
-#include <uapi/asm-generic/mman-common.h>
+#include <linux/mman.h>
 #include <asm/processor.h>
 
 /*