drm/radeon/si: properly set up the clearstate buffer for pg (v2)

The format of the clearstate buffer used for pg (powergating)
changed between NI and SI.  This formats it properly for what
the hardware expects on SI.

v2: fix addresses
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/radeon/evergreen.c

@@ -142,6 +142,9 @@ extern void cayman_vm_decode_fault(struct radeon_device *rdev,
 				   u32 status, u32 addr);
 void cik_init_cp_pg_table(struct radeon_device *rdev);
 
+extern u32 si_get_csb_size(struct radeon_device *rdev);
+extern void si_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer);
+
 static const u32 evergreen_golden_registers[] =
 {
 	0x3f90, 0xffff0000, 0xff000000,
@@ -3893,7 +3896,7 @@ int sumo_rlc_init(struct radeon_device *rdev)
 	const u32 *src_ptr;
 	volatile u32 *dst_ptr;
 	u32 dws, data, i, j, k, reg_num;
-	u32 reg_list_num, reg_list_hdr_blk_index, reg_list_blk_index;
+	u32 reg_list_num, reg_list_hdr_blk_index, reg_list_blk_index = 0;
 	u64 reg_list_mc_addr;
 	const struct cs_section_def *cs_data;
 	int r;
@@ -3937,7 +3940,7 @@ int sumo_rlc_init(struct radeon_device *rdev)
 		dst_ptr = rdev->rlc.sr_ptr;
 		if (rdev->family >= CHIP_TAHITI) {
 			/* SI */
-			for (i = 0; i < dws; i++)
+			for (i = 0; i < rdev->rlc.reg_list_size; i++)
 				dst_ptr[i] = src_ptr[i];
 		} else {
 			/* ON/LN/TN */
@@ -3963,6 +3966,10 @@ int sumo_rlc_init(struct radeon_device *rdev)
 
 	if (cs_data) {
 		/* clear state block */
+		if (rdev->family >= CHIP_TAHITI) {
+			rdev->rlc.clear_state_size = si_get_csb_size(rdev);
+			dws = rdev->rlc.clear_state_size + (256 / 4);
+		} else {
 			reg_list_num = 0;
 			dws = 0;
 			for (i = 0; cs_data[i].section != NULL; i++) {
@@ -3974,9 +3981,10 @@ int sumo_rlc_init(struct radeon_device *rdev)
 			reg_list_blk_index = (3 * reg_list_num + 2);
 			dws += reg_list_blk_index;
 			rdev->rlc.clear_state_size = dws;
+		}
 
 		if (rdev->rlc.clear_state_obj == NULL) {
-			r = radeon_bo_create(rdev, rdev->rlc.clear_state_size * 4, PAGE_SIZE, true,
+			r = radeon_bo_create(rdev, dws * 4, PAGE_SIZE, true,
 					     RADEON_GEM_DOMAIN_VRAM, NULL, &rdev->rlc.clear_state_obj);
 			if (r) {
 				dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
@@ -4006,6 +4014,13 @@ int sumo_rlc_init(struct radeon_device *rdev)
 		}
 		/* set up the cs buffer */
 		dst_ptr = rdev->rlc.cs_ptr;
+		if (rdev->family >= CHIP_TAHITI) {
+			reg_list_mc_addr = rdev->rlc.clear_state_gpu_addr + 256;
+			dst_ptr[0] = upper_32_bits(reg_list_mc_addr);
+			dst_ptr[1] = lower_32_bits(reg_list_mc_addr);
+			dst_ptr[2] = rdev->rlc.clear_state_size;
+			si_get_csb_buffer(rdev, &dst_ptr[(256/4)]);
+		} else {
 			reg_list_hdr_blk_index = 0;
 			reg_list_mc_addr = rdev->rlc.clear_state_gpu_addr + (reg_list_blk_index * 4);
 			data = upper_32_bits(reg_list_mc_addr);
@@ -4035,7 +4050,7 @@ int sumo_rlc_init(struct radeon_device *rdev)
 				}
 			}
 			dst_ptr[reg_list_hdr_blk_index] = RLC_CLEAR_STATE_END_MARKER;
-
+		}
 		radeon_bo_kunmap(rdev->rlc.clear_state_obj);
 		radeon_bo_unreserve(rdev->rlc.clear_state_obj);
 	}

drivers/gpu/drm/radeon/si.c

@@ -5286,6 +5286,97 @@ void si_update_pg(struct radeon_device *rdev,
 	si_enable_gfx_cgpg(rdev, enable);
 }
 
+u32 si_get_csb_size(struct radeon_device *rdev)
+{
+	u32 count = 0;
+	const struct cs_section_def *sect = NULL;
+	const struct cs_extent_def *ext = NULL;
+
+	if (rdev->rlc.cs_data == NULL)
+		return 0;
+
+	/* begin clear state */
+	count += 2;
+	/* context control state */
+	count += 3;
+
+	for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
+		for (ext = sect->section; ext->extent != NULL; ++ext) {
+			if (sect->id == SECT_CONTEXT)
+				count += 2 + ext->reg_count;
+			else
+				return 0;
+		}
+	}
+	/* pa_sc_raster_config */
+	count += 3;
+	/* end clear state */
+	count += 2;
+	/* clear state */
+	count += 2;
+
+	return count;
+}
+
+void si_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer)
+{
+	u32 count = 0, i;
+	const struct cs_section_def *sect = NULL;
+	const struct cs_extent_def *ext = NULL;
+
+	if (rdev->rlc.cs_data == NULL)
+		return;
+	if (buffer == NULL)
+		return;
+
+	buffer[count++] = PACKET3(PACKET3_PREAMBLE_CNTL, 0);
+	buffer[count++] = PACKET3_PREAMBLE_BEGIN_CLEAR_STATE;
+
+	buffer[count++] = PACKET3(PACKET3_CONTEXT_CONTROL, 1);
+	buffer[count++] = 0x80000000;
+	buffer[count++] = 0x80000000;
+
+	for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) {
+		for (ext = sect->section; ext->extent != NULL; ++ext) {
+			if (sect->id == SECT_CONTEXT) {
+				buffer[count++] = PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count);
+				buffer[count++] = ext->reg_index - 0xa000;
+				for (i = 0; i < ext->reg_count; i++)
+					buffer[count++] = ext->extent[i];
+			} else {
+				return;
+			}
+		}
+	}
+
+	buffer[count++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1);
+	buffer[count++] = PA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START;
+	switch (rdev->family) {
+	case CHIP_TAHITI:
+	case CHIP_PITCAIRN:
+		buffer[count++] = 0x2a00126a;
+		break;
+	case CHIP_VERDE:
+		buffer[count++] = 0x0000124a;
+		break;
+	case CHIP_OLAND:
+		buffer[count++] = 0x00000082;
+		break;
+	case CHIP_HAINAN:
+		buffer[count++] = 0x00000000;
+		break;
+	default:
+		buffer[count++] = 0x00000000;
+		break;
+	}
+
+	buffer[count++] = PACKET3(PACKET3_PREAMBLE_CNTL, 0);
+	buffer[count++] = PACKET3_PREAMBLE_END_CLEAR_STATE;
+
+	buffer[count++] = PACKET3(PACKET3_CLEAR_STATE, 0);
+	buffer[count++] = 0;
+}
+
 static void si_init_pg(struct radeon_device *rdev)
 {
 	if (rdev->pg_flags) {