drm/radeon: add support for SET_APPEND_CNT packet3 (v2)

This adds support to the command parser for the set append counter
packet3, this is required to support atomic counters on
evergreen/cayman GPUs.

v2: fixup some of the hardcoded numbers with real register names
(Christian)

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/radeon/evergreen_cs.c

@@ -2608,6 +2608,51 @@ static int evergreen_packet3_check(struct radeon_cs_parser *p,
 			}
 		}
 		break;
+	case PACKET3_SET_APPEND_CNT:
+	{
+		uint32_t areg;
+		uint32_t allowed_reg_base;
+		uint32_t source_sel;
+		if (pkt->count != 2) {
+			DRM_ERROR("bad SET_APPEND_CNT (invalid count)\n");
+			return -EINVAL;
+		}
+
+		allowed_reg_base = GDS_APPEND_COUNT_0;
+		allowed_reg_base -= PACKET3_SET_CONTEXT_REG_START;
+		allowed_reg_base >>= 2;
+
+		areg = idx_value >> 16;
+		if (areg < allowed_reg_base || areg > (allowed_reg_base + 11)) {
+			dev_warn(p->dev, "forbidden register for append cnt 0x%08x at %d\n",
+				 areg, idx);
+			return -EINVAL;
+		}
+
+		source_sel = G_PACKET3_SET_APPEND_CNT_SRC_SELECT(idx_value);
+		if (source_sel == PACKET3_SAC_SRC_SEL_MEM) {
+			uint64_t offset;
+			uint32_t swap;
+			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
+			if (r) {
+				DRM_ERROR("bad SET_APPEND_CNT (missing reloc)\n");
+				return -EINVAL;
+			}
+			offset = radeon_get_ib_value(p, idx + 1);
+			swap = offset & 0x3;
+			offset &= ~0x3;
+
+			offset += ((u64)(radeon_get_ib_value(p, idx + 2) & 0xff)) << 32;
+
+			offset += reloc->gpu_offset;
+			ib[idx+1] = (offset & 0xfffffffc) | swap;
+			ib[idx+2] = upper_32_bits(offset) & 0xff;
+		} else {
+			DRM_ERROR("bad SET_APPEND_CNT (unsupported operation)\n");
+			return -EINVAL;
+		}
+		break;
+	}
 	case PACKET3_NOP:
 		break;
 	default:

drivers/gpu/drm/radeon/evergreend.h

@@ -1689,6 +1689,36 @@
 #define	PACKET3_SET_CONTEXT_REG_INDIRECT		0x73
 #define	PACKET3_SET_RESOURCE_INDIRECT			0x74
 #define	PACKET3_SET_APPEND_CNT			        0x75
+/* SET_APPEND_CNT - documentation
+ * 1. header
+ * 2. COMMAND
+ *  1:0 - SOURCE SEL
+ *  15:2 - Reserved
+ *  31:16 - WR_REG_OFFSET - context register to write source data to.
+ *          (one of R_02872C_GDS_APPEND_COUNT_0-11)
+ * 3. CONTROL
+ *  (for source == mem)
+ *  31:2 SRC_ADDRESS_LO
+ *  0:1 SWAP
+ *  (for source == GDS)
+ *  31:0 GDS offset
+ *  (for source == DATA)
+ *  31:0 DATA
+ *  (for source == REG)
+ *  31:0 REG
+ * 4. SRC_ADDRESS_HI[7:0]
+ * kernel driver 2.44 only supports SRC == MEM.
+ */
+#define PACKET3_SET_APPEND_CNT_SRC_SELECT(x) ((x) << 0)
+#define G_PACKET3_SET_APPEND_CNT_SRC_SELECT(x) ((x & 0x3) >> 0)
+/* source is from the data in CONTROL */
+#define PACKET3_SAC_SRC_SEL_DATA 0x0
+/* source is from register */
+#define PACKET3_SAC_SRC_SEL_REG  0x1
+/* source is from GDS offset in CONTROL */
+#define PACKET3_SAC_SRC_SEL_GDS  0x2
+/* source is from memory address */
+#define PACKET3_SAC_SRC_SEL_MEM  0x3
 
 #define	SQ_RESOURCE_CONSTANT_WORD7_0				0x3001c
 #define		S__SQ_CONSTANT_TYPE(x)			(((x) & 3) << 30)
@@ -2005,6 +2035,19 @@
 
 #define GDS_ADDR_BASE					0x28720
 
+#define GDS_APPEND_COUNT_0				0x2872C
+#define GDS_APPEND_COUNT_1				0x28730
+#define GDS_APPEND_COUNT_2				0x28734
+#define GDS_APPEND_COUNT_3				0x28738
+#define GDS_APPEND_COUNT_4				0x2873C
+#define GDS_APPEND_COUNT_5				0x28740
+#define GDS_APPEND_COUNT_6				0x28744
+#define GDS_APPEND_COUNT_7				0x28748
+#define GDS_APPEND_COUNT_8				0x2874c
+#define GDS_APPEND_COUNT_9				0x28750
+#define GDS_APPEND_COUNT_10				0x28754
+#define GDS_APPEND_COUNT_11				0x28758
+
 #define	CB_IMMED0_BASE					0x28b9c
 #define	CB_IMMED1_BASE					0x28ba0
 #define	CB_IMMED2_BASE					0x28ba4

drivers/gpu/drm/radeon/radeon_drv.c

@@ -93,9 +93,10 @@
  *   2.41.0 - evergreen/cayman: Add SET_BASE/DRAW_INDIRECT command parsing support
  *   2.42.0 - Add VCE/VUI (Video Usability Information) support
  *   2.43.0 - RADEON_INFO_GPU_RESET_COUNTER
+ *   2.44.0 - SET_APPEND_CNT packet3 support
  */
 #define KMS_DRIVER_MAJOR	2
-#define KMS_DRIVER_MINOR	43
+#define KMS_DRIVER_MINOR	44
 #define KMS_DRIVER_PATCHLEVEL	0
 int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags);
 int radeon_driver_unload_kms(struct drm_device *dev);