drm/radeon/cik: 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/cik.c

@@ -5622,7 +5622,7 @@ static void cik_init_gfx_cgpg(struct radeon_device *rdev)
 	if (rdev->rlc.cs_data) {
 		WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
 		WREG32(RLC_GPM_SCRATCH_DATA, upper_32_bits(rdev->rlc.clear_state_gpu_addr));
-		WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_gpu_addr);
+		WREG32(RLC_GPM_SCRATCH_DATA, lower_32_bits(rdev->rlc.clear_state_gpu_addr));
 		WREG32(RLC_GPM_SCRATCH_DATA, rdev->rlc.clear_state_size);
 	} else {
 		WREG32(RLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
@@ -5670,6 +5670,97 @@ static void cik_update_gfx_pg(struct radeon_device *rdev, bool enable)
 	cik_enable_gfx_dynamic_mgpg(rdev, enable);
 }
 
+u32 cik_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/pa_sc_raster_config1 */
+	count += 4;
+	/* end clear state */
+	count += 2;
+	/* clear state */
+	count += 2;
+
+	return count;
+}
+
+void cik_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, 2);
+	buffer[count++] = PA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START;
+	switch (rdev->family) {
+	case CHIP_BONAIRE:
+		buffer[count++] = 0x16000012;
+		buffer[count++] = 0x00000000;
+		break;
+	case CHIP_KAVERI:
+		buffer[count++] = 0x00000000; /* XXX */
+		buffer[count++] = 0x00000000;
+		break;
+	case CHIP_KABINI:
+		buffer[count++] = 0x00000000; /* XXX */
+		buffer[count++] = 0x00000000;
+		break;
+	default:
+		buffer[count++] = 0x00000000;
+		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 cik_init_pg(struct radeon_device *rdev)
 {
 	if (rdev->pg_flags) {

drivers/gpu/drm/radeon/evergreen.c

@@ -144,6 +144,8 @@ 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);
+extern u32 cik_get_csb_size(struct radeon_device *rdev);
+extern void cik_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer);
 
 static const u32 evergreen_golden_registers[] =
 {
@@ -3903,6 +3905,9 @@ int sumo_rlc_init(struct radeon_device *rdev)
 
 	src_ptr = rdev->rlc.reg_list;
 	dws = rdev->rlc.reg_list_size;
+	if (rdev->family >= CHIP_BONAIRE) {
+		dws += (5 * 16) + 48 + 48 + 64;
+	}
 	cs_data = rdev->rlc.cs_data;
 
 	if (src_ptr) {
@@ -3966,7 +3971,9 @@ int sumo_rlc_init(struct radeon_device *rdev)
 
 	if (cs_data) {
 		/* clear state block */
-		if (rdev->family >= CHIP_TAHITI) {
+		if (rdev->family >= CHIP_BONAIRE) {
+			rdev->rlc.clear_state_size = dws = cik_get_csb_size(rdev);
+		} else if (rdev->family >= CHIP_TAHITI) {
 			rdev->rlc.clear_state_size = si_get_csb_size(rdev);
 			dws = rdev->rlc.clear_state_size + (256 / 4);
 		} else {
@@ -4014,7 +4021,9 @@ int sumo_rlc_init(struct radeon_device *rdev)
 		}
 		/* set up the cs buffer */
 		dst_ptr = rdev->rlc.cs_ptr;
-		if (rdev->family >= CHIP_TAHITI) {
+		if (rdev->family >= CHIP_BONAIRE) {
+			cik_get_csb_buffer(rdev, dst_ptr);
+		} else 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);