struct radeon_bo *db_s_read_bo;
struct radeon_bo *db_s_write_bo;
bool sx_misc_kill_all_prims;
+ bool cb_dirty;
+ bool db_dirty;
+ bool streamout_dirty;
};
static u32 evergreen_cs_get_aray_mode(u32 tiling_flags)
}
track->cb_target_mask = 0xFFFFFFFF;
track->cb_shader_mask = 0xFFFFFFFF;
+ track->cb_dirty = true;
track->db_depth_view = 0xFFFFC000;
track->db_depth_size = 0xFFFFFFFF;
track->db_s_write_offset = 0xFFFFFFFF;
track->db_s_read_bo = NULL;
track->db_s_write_bo = NULL;
+ track->db_dirty = true;
for (i = 0; i < 4; i++) {
track->vgt_strmout_size[i] = 0;
track->vgt_strmout_bo_offset[i] = 0xFFFFFFFF;
track->vgt_strmout_bo_mc[i] = 0xFFFFFFFF;
}
+ track->streamout_dirty = true;
track->sx_misc_kill_all_prims = false;
}
static int evergreen_cs_track_check(struct radeon_cs_parser *p)
{
struct evergreen_cs_track *track = p->track;
- unsigned tmp, i, j;
+ unsigned tmp, i;
int r;
+ unsigned buffer_mask = 0;
/* check streamout */
- for (i = 0; i < 4; i++) {
- if (track->vgt_strmout_config & (1 << i)) {
- for (j = 0; j < 4; j++) {
- if ((track->vgt_strmout_buffer_config >> (i * 4)) & (1 << j)) {
- if (track->vgt_strmout_bo[j]) {
- u64 offset = (u64)track->vgt_strmout_bo_offset[j] +
- (u64)track->vgt_strmout_size[j];
- if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
- DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
- j, offset,
- radeon_bo_size(track->vgt_strmout_bo[j]));
- return -EINVAL;
- }
- } else {
- dev_warn(p->dev, "No buffer for streamout %d\n", j);
+ if (track->streamout_dirty && track->vgt_strmout_config) {
+ for (i = 0; i < 4; i++) {
+ if (track->vgt_strmout_config & (1 << i)) {
+ buffer_mask |= (track->vgt_strmout_buffer_config >> (i * 4)) & 0xf;
+ }
+ }
+
+ for (i = 0; i < 4; i++) {
+ if (buffer_mask & (1 << i)) {
+ if (track->vgt_strmout_bo[i]) {
+ u64 offset = (u64)track->vgt_strmout_bo_offset[i] +
+ (u64)track->vgt_strmout_size[i];
+ if (offset > radeon_bo_size(track->vgt_strmout_bo[i])) {
+ DRM_ERROR("streamout %d bo too small: 0x%llx, 0x%lx\n",
+ i, offset,
+ radeon_bo_size(track->vgt_strmout_bo[i]));
return -EINVAL;
}
+ } else {
+ dev_warn(p->dev, "No buffer for streamout %d\n", i);
+ return -EINVAL;
}
}
}
+ track->streamout_dirty = false;
}
if (track->sx_misc_kill_all_prims)
/* check that we have a cb for each enabled target
*/
- tmp = track->cb_target_mask;
- for (i = 0; i < 8; i++) {
- if ((tmp >> (i * 4)) & 0xF) {
- /* at least one component is enabled */
- if (track->cb_color_bo[i] == NULL) {
- dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
- __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
- return -EINVAL;
- }
- /* check cb */
- r = evergreen_cs_track_validate_cb(p, i);
- if (r) {
- return r;
+ if (track->cb_dirty) {
+ tmp = track->cb_target_mask;
+ for (i = 0; i < 8; i++) {
+ if ((tmp >> (i * 4)) & 0xF) {
+ /* at least one component is enabled */
+ if (track->cb_color_bo[i] == NULL) {
+ dev_warn(p->dev, "%s:%d mask 0x%08X | 0x%08X no cb for %d\n",
+ __func__, __LINE__, track->cb_target_mask, track->cb_shader_mask, i);
+ return -EINVAL;
+ }
+ /* check cb */
+ r = evergreen_cs_track_validate_cb(p, i);
+ if (r) {
+ return r;
+ }
}
}
+ track->cb_dirty = false;
}
- /* Check stencil buffer */
- if (G_028800_STENCIL_ENABLE(track->db_depth_control)) {
- r = evergreen_cs_track_validate_stencil(p);
- if (r)
- return r;
- }
- /* Check depth buffer */
- if (G_028800_Z_WRITE_ENABLE(track->db_depth_control)) {
- r = evergreen_cs_track_validate_depth(p);
- if (r)
- return r;
+ if (track->db_dirty) {
+ /* Check stencil buffer */
+ if (G_028800_STENCIL_ENABLE(track->db_depth_control)) {
+ r = evergreen_cs_track_validate_stencil(p);
+ if (r)
+ return r;
+ }
+ /* Check depth buffer */
+ if (G_028800_Z_WRITE_ENABLE(track->db_depth_control)) {
+ r = evergreen_cs_track_validate_depth(p);
+ if (r)
+ return r;
+ }
+ track->db_dirty = false;
}
return 0;
break;
case DB_DEPTH_CONTROL:
track->db_depth_control = radeon_get_ib_value(p, idx);
+ track->db_dirty = true;
break;
case CAYMAN_DB_EQAA:
if (p->rdev->family < CHIP_CAYMAN) {
DB_MACRO_TILE_ASPECT(mtaspect);
}
}
+ track->db_dirty = true;
break;
case DB_STENCIL_INFO:
track->db_s_info = radeon_get_ib_value(p, idx);
+ track->db_dirty = true;
break;
case DB_DEPTH_VIEW:
track->db_depth_view = radeon_get_ib_value(p, idx);
+ track->db_dirty = true;
break;
case DB_DEPTH_SIZE:
track->db_depth_size = radeon_get_ib_value(p, idx);
track->db_depth_size_idx = idx;
+ track->db_dirty = true;
break;
case R_02805C_DB_DEPTH_SLICE:
track->db_depth_slice = radeon_get_ib_value(p, idx);
+ track->db_dirty = true;
break;
case DB_Z_READ_BASE:
r = evergreen_cs_packet_next_reloc(p, &reloc);
track->db_z_read_offset = radeon_get_ib_value(p, idx);
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->db_z_read_bo = reloc->robj;
+ track->db_dirty = true;
break;
case DB_Z_WRITE_BASE:
r = evergreen_cs_packet_next_reloc(p, &reloc);
track->db_z_write_offset = radeon_get_ib_value(p, idx);
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->db_z_write_bo = reloc->robj;
+ track->db_dirty = true;
break;
case DB_STENCIL_READ_BASE:
r = evergreen_cs_packet_next_reloc(p, &reloc);
track->db_s_read_offset = radeon_get_ib_value(p, idx);
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->db_s_read_bo = reloc->robj;
+ track->db_dirty = true;
break;
case DB_STENCIL_WRITE_BASE:
r = evergreen_cs_packet_next_reloc(p, &reloc);
track->db_s_write_offset = radeon_get_ib_value(p, idx);
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->db_s_write_bo = reloc->robj;
+ track->db_dirty = true;
break;
case VGT_STRMOUT_CONFIG:
track->vgt_strmout_config = radeon_get_ib_value(p, idx);
+ track->streamout_dirty = true;
break;
case VGT_STRMOUT_BUFFER_CONFIG:
track->vgt_strmout_buffer_config = radeon_get_ib_value(p, idx);
+ track->streamout_dirty = true;
break;
case VGT_STRMOUT_BUFFER_BASE_0:
case VGT_STRMOUT_BUFFER_BASE_1:
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->vgt_strmout_bo[tmp] = reloc->robj;
track->vgt_strmout_bo_mc[tmp] = reloc->lobj.gpu_offset;
+ track->streamout_dirty = true;
break;
case VGT_STRMOUT_BUFFER_SIZE_0:
case VGT_STRMOUT_BUFFER_SIZE_1:
tmp = (reg - VGT_STRMOUT_BUFFER_SIZE_0) / 16;
/* size in register is DWs, convert to bytes */
track->vgt_strmout_size[tmp] = radeon_get_ib_value(p, idx) * 4;
+ track->streamout_dirty = true;
break;
case CP_COHER_BASE:
r = evergreen_cs_packet_next_reloc(p, &reloc);
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
case CB_TARGET_MASK:
track->cb_target_mask = radeon_get_ib_value(p, idx);
+ track->cb_dirty = true;
break;
case CB_SHADER_MASK:
track->cb_shader_mask = radeon_get_ib_value(p, idx);
+ track->cb_dirty = true;
break;
case PA_SC_AA_CONFIG:
if (p->rdev->family >= CHIP_CAYMAN) {
case CB_COLOR7_VIEW:
tmp = (reg - CB_COLOR0_VIEW) / 0x3c;
track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
+ track->cb_dirty = true;
break;
case CB_COLOR8_VIEW:
case CB_COLOR9_VIEW:
case CB_COLOR11_VIEW:
tmp = ((reg - CB_COLOR8_VIEW) / 0x1c) + 8;
track->cb_color_view[tmp] = radeon_get_ib_value(p, idx);
+ track->cb_dirty = true;
break;
case CB_COLOR0_INFO:
case CB_COLOR1_INFO:
ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
}
+ track->cb_dirty = true;
break;
case CB_COLOR8_INFO:
case CB_COLOR9_INFO:
ib[idx] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
track->cb_color_info[tmp] |= CB_ARRAY_MODE(evergreen_cs_get_aray_mode(reloc->lobj.tiling_flags));
}
+ track->cb_dirty = true;
break;
case CB_COLOR0_PITCH:
case CB_COLOR1_PITCH:
tmp = (reg - CB_COLOR0_PITCH) / 0x3c;
track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
track->cb_color_pitch_idx[tmp] = idx;
+ track->cb_dirty = true;
break;
case CB_COLOR8_PITCH:
case CB_COLOR9_PITCH:
tmp = ((reg - CB_COLOR8_PITCH) / 0x1c) + 8;
track->cb_color_pitch[tmp] = radeon_get_ib_value(p, idx);
track->cb_color_pitch_idx[tmp] = idx;
+ track->cb_dirty = true;
break;
case CB_COLOR0_SLICE:
case CB_COLOR1_SLICE:
tmp = (reg - CB_COLOR0_SLICE) / 0x3c;
track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
track->cb_color_slice_idx[tmp] = idx;
+ track->cb_dirty = true;
break;
case CB_COLOR8_SLICE:
case CB_COLOR9_SLICE:
tmp = ((reg - CB_COLOR8_SLICE) / 0x1c) + 8;
track->cb_color_slice[tmp] = radeon_get_ib_value(p, idx);
track->cb_color_slice_idx[tmp] = idx;
+ track->cb_dirty = true;
break;
case CB_COLOR0_ATTRIB:
case CB_COLOR1_ATTRIB:
}
tmp = ((reg - CB_COLOR0_ATTRIB) / 0x3c);
track->cb_color_attrib[tmp] = ib[idx];
+ track->cb_dirty = true;
break;
case CB_COLOR8_ATTRIB:
case CB_COLOR9_ATTRIB:
}
tmp = ((reg - CB_COLOR8_ATTRIB) / 0x1c) + 8;
track->cb_color_attrib[tmp] = ib[idx];
+ track->cb_dirty = true;
break;
case CB_COLOR0_DIM:
case CB_COLOR1_DIM:
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->cb_color_base_last[tmp] = ib[idx];
track->cb_color_bo[tmp] = reloc->robj;
+ track->cb_dirty = true;
break;
case CB_COLOR8_BASE:
case CB_COLOR9_BASE:
ib[idx] += (u32)((reloc->lobj.gpu_offset >> 8) & 0xffffffff);
track->cb_color_base_last[tmp] = ib[idx];
track->cb_color_bo[tmp] = reloc->robj;
+ track->cb_dirty = true;
break;
case CB_IMMED0_BASE:
case CB_IMMED1_BASE:
{
int pred_op;
int tmp;
+ uint64_t offset;
+
if (pkt->count != 1) {
DRM_ERROR("bad SET PREDICATION\n");
return -EINVAL;
return -EINVAL;
}
- ib[idx + 0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx + 1] = tmp + (upper_32_bits(reloc->lobj.gpu_offset) & 0xff);
+ offset = reloc->lobj.gpu_offset +
+ (idx_value & 0xfffffff0) +
+ ((u64)(tmp & 0xff) << 32);
+
+ ib[idx + 0] = offset;
+ ib[idx + 1] = (tmp & 0xffffff00) | (upper_32_bits(offset) & 0xff);
}
break;
case PACKET3_CONTEXT_CONTROL:
}
break;
case PACKET3_INDEX_BASE:
+ {
+ uint64_t offset;
+
if (pkt->count != 1) {
DRM_ERROR("bad INDEX_BASE\n");
return -EINVAL;
DRM_ERROR("bad INDEX_BASE\n");
return -EINVAL;
}
- ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ idx_value +
+ ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
+
+ ib[idx+0] = offset;
+ ib[idx+1] = upper_32_bits(offset) & 0xff;
+
r = evergreen_cs_track_check(p);
if (r) {
dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
return r;
}
break;
+ }
case PACKET3_DRAW_INDEX:
+ {
+ uint64_t offset;
if (pkt->count != 3) {
DRM_ERROR("bad DRAW_INDEX\n");
return -EINVAL;
DRM_ERROR("bad DRAW_INDEX\n");
return -EINVAL;
}
- ib[idx+0] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+1] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ idx_value +
+ ((u64)(radeon_get_ib_value(p, idx+1) & 0xff) << 32);
+
+ ib[idx+0] = offset;
+ ib[idx+1] = upper_32_bits(offset) & 0xff;
+
r = evergreen_cs_track_check(p);
if (r) {
dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
return r;
}
break;
+ }
case PACKET3_DRAW_INDEX_2:
+ {
+ uint64_t offset;
+
if (pkt->count != 4) {
DRM_ERROR("bad DRAW_INDEX_2\n");
return -EINVAL;
DRM_ERROR("bad DRAW_INDEX_2\n");
return -EINVAL;
}
- ib[idx+1] = idx_value + (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ radeon_get_ib_value(p, idx+1) +
+ ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
+
+ ib[idx+1] = offset;
+ ib[idx+2] = upper_32_bits(offset) & 0xff;
+
r = evergreen_cs_track_check(p);
if (r) {
dev_warn(p->dev, "%s:%d invalid cmd stream\n", __func__, __LINE__);
return r;
}
break;
+ }
case PACKET3_DRAW_INDEX_AUTO:
if (pkt->count != 1) {
DRM_ERROR("bad DRAW_INDEX_AUTO\n");
}
/* bit 4 is reg (0) or mem (1) */
if (idx_value & 0x10) {
+ uint64_t offset;
+
r = evergreen_cs_packet_next_reloc(p, &reloc);
if (r) {
DRM_ERROR("bad WAIT_REG_MEM\n");
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
+ ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
+
+ ib[idx+1] = (ib[idx+1] & 0x3) | (offset & 0xfffffffc);
+ ib[idx+2] = upper_32_bits(offset) & 0xff;
}
break;
case PACKET3_SURFACE_SYNC:
return -EINVAL;
}
if (pkt->count) {
+ uint64_t offset;
+
r = evergreen_cs_packet_next_reloc(p, &reloc);
if (r) {
DRM_ERROR("bad EVENT_WRITE\n");
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+ offset = reloc->lobj.gpu_offset +
+ (radeon_get_ib_value(p, idx+1) & 0xfffffff8) +
+ ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
+
+ ib[idx+1] = offset & 0xfffffff8;
+ ib[idx+2] = upper_32_bits(offset) & 0xff;
}
break;
case PACKET3_EVENT_WRITE_EOP:
+ {
+ uint64_t offset;
+
if (pkt->count != 4) {
DRM_ERROR("bad EVENT_WRITE_EOP\n");
return -EINVAL;
DRM_ERROR("bad EVENT_WRITE_EOP\n");
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
+ ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
+
+ ib[idx+1] = offset & 0xfffffffc;
+ ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
break;
+ }
case PACKET3_EVENT_WRITE_EOS:
+ {
+ uint64_t offset;
+
if (pkt->count != 3) {
DRM_ERROR("bad EVENT_WRITE_EOS\n");
return -EINVAL;
DRM_ERROR("bad EVENT_WRITE_EOS\n");
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset = reloc->lobj.gpu_offset +
+ (radeon_get_ib_value(p, idx+1) & 0xfffffffc) +
+ ((u64)(radeon_get_ib_value(p, idx+2) & 0xff) << 32);
+
+ ib[idx+1] = offset & 0xfffffffc;
+ ib[idx+2] = (ib[idx+2] & 0xffffff00) | (upper_32_bits(offset) & 0xff);
break;
+ }
case PACKET3_SET_CONFIG_REG:
start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
end_reg = 4 * pkt->count + start_reg - 4;
ib[idx+1+(i*8)+3] += moffset;
break;
case SQ_TEX_VTX_VALID_BUFFER:
+ {
+ uint64_t offset64;
/* vtx base */
r = evergreen_cs_packet_next_reloc(p, &reloc);
if (r) {
if (p->rdev && (size + offset) > radeon_bo_size(reloc->robj)) {
/* force size to size of the buffer */
dev_warn(p->dev, "vbo resource seems too big for the bo\n");
- ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj);
+ ib[idx+1+(i*8)+1] = radeon_bo_size(reloc->robj) - offset;
}
- ib[idx+1+(i*8)+0] += (u32)((reloc->lobj.gpu_offset) & 0xffffffff);
- ib[idx+1+(i*8)+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+
+ offset64 = reloc->lobj.gpu_offset + offset;
+ ib[idx+1+(i*8)+0] = offset64;
+ ib[idx+1+(i*8)+2] = (ib[idx+1+(i*8)+2] & 0xffffff00) |
+ (upper_32_bits(offset64) & 0xff);
break;
+ }
case SQ_TEX_VTX_INVALID_TEXTURE:
case SQ_TEX_VTX_INVALID_BUFFER:
default:
offset + 4, radeon_bo_size(reloc->robj));
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+ offset += reloc->lobj.gpu_offset;
+ ib[idx+1] = offset;
+ ib[idx+2] = upper_32_bits(offset) & 0xff;
}
/* Reading data from SRC_ADDRESS. */
if (((idx_value >> 1) & 0x3) == 2) {
offset + 4, radeon_bo_size(reloc->robj));
return -EINVAL;
}
- ib[idx+3] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+4] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+ offset += reloc->lobj.gpu_offset;
+ ib[idx+3] = offset;
+ ib[idx+4] = upper_32_bits(offset) & 0xff;
}
break;
case PACKET3_COPY_DW:
offset + 4, radeon_bo_size(reloc->robj));
return -EINVAL;
}
- ib[idx+1] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+2] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+ offset += reloc->lobj.gpu_offset;
+ ib[idx+1] = offset;
+ ib[idx+2] = upper_32_bits(offset) & 0xff;
} else {
/* SRC is a reg. */
reg = radeon_get_ib_value(p, idx+1) << 2;
offset + 4, radeon_bo_size(reloc->robj));
return -EINVAL;
}
- ib[idx+3] += (u32)(reloc->lobj.gpu_offset & 0xffffffff);
- ib[idx+4] += upper_32_bits(reloc->lobj.gpu_offset) & 0xff;
+ offset += reloc->lobj.gpu_offset;
+ ib[idx+3] = offset;
+ ib[idx+4] = upper_32_bits(offset) & 0xff;
} else {
/* DST is a reg. */
reg = radeon_get_ib_value(p, idx+3) << 2;