aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/radeon/r100.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/radeon/r100.c')
-rw-r--r--drivers/gpu/drm/radeon/r100.c773
1 files changed, 759 insertions, 14 deletions
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index c550932a108f..f1ba8ff41130 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -110,7 +110,7 @@ int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
110 if (i < 0 || i > rdev->gart.num_gpu_pages) { 110 if (i < 0 || i > rdev->gart.num_gpu_pages) {
111 return -EINVAL; 111 return -EINVAL;
112 } 112 }
113 rdev->gart.table.ram.ptr[i] = cpu_to_le32((uint32_t)addr); 113 rdev->gart.table.ram.ptr[i] = cpu_to_le32(lower_32_bits(addr));
114 return 0; 114 return 0;
115} 115}
116 116
@@ -173,8 +173,12 @@ void r100_mc_setup(struct radeon_device *rdev)
173 DRM_ERROR("Failed to register debugfs file for R100 MC !\n"); 173 DRM_ERROR("Failed to register debugfs file for R100 MC !\n");
174 } 174 }
175 /* Write VRAM size in case we are limiting it */ 175 /* Write VRAM size in case we are limiting it */
176 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size); 176 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
177 tmp = rdev->mc.vram_location + rdev->mc.vram_size - 1; 177 /* Novell bug 204882 for RN50/M6/M7 with 8/16/32MB VRAM,
178 * if the aperture is 64MB but we have 32MB VRAM
179 * we report only 32MB VRAM but we have to set MC_FB_LOCATION
180 * to 64MB, otherwise the gpu accidentially dies */
181 tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1;
178 tmp = REG_SET(RADEON_MC_FB_TOP, tmp >> 16); 182 tmp = REG_SET(RADEON_MC_FB_TOP, tmp >> 16);
179 tmp |= REG_SET(RADEON_MC_FB_START, rdev->mc.vram_location >> 16); 183 tmp |= REG_SET(RADEON_MC_FB_START, rdev->mc.vram_location >> 16);
180 WREG32(RADEON_MC_FB_LOCATION, tmp); 184 WREG32(RADEON_MC_FB_LOCATION, tmp);
@@ -215,7 +219,6 @@ int r100_mc_init(struct radeon_device *rdev)
215 r100_pci_gart_disable(rdev); 219 r100_pci_gart_disable(rdev);
216 220
217 /* Setup GPU memory space */ 221 /* Setup GPU memory space */
218 rdev->mc.vram_location = 0xFFFFFFFFUL;
219 rdev->mc.gtt_location = 0xFFFFFFFFUL; 222 rdev->mc.gtt_location = 0xFFFFFFFFUL;
220 if (rdev->flags & RADEON_IS_AGP) { 223 if (rdev->flags & RADEON_IS_AGP) {
221 r = radeon_agp_init(rdev); 224 r = radeon_agp_init(rdev);
@@ -719,13 +722,14 @@ int r100_cs_packet_parse(struct radeon_cs_parser *p,
719 unsigned idx) 722 unsigned idx)
720{ 723{
721 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx]; 724 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
722 uint32_t header = ib_chunk->kdata[idx]; 725 uint32_t header;
723 726
724 if (idx >= ib_chunk->length_dw) { 727 if (idx >= ib_chunk->length_dw) {
725 DRM_ERROR("Can not parse packet at %d after CS end %d !\n", 728 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
726 idx, ib_chunk->length_dw); 729 idx, ib_chunk->length_dw);
727 return -EINVAL; 730 return -EINVAL;
728 } 731 }
732 header = ib_chunk->kdata[idx];
729 pkt->idx = idx; 733 pkt->idx = idx;
730 pkt->type = CP_PACKET_GET_TYPE(header); 734 pkt->type = CP_PACKET_GET_TYPE(header);
731 pkt->count = CP_PACKET_GET_COUNT(header); 735 pkt->count = CP_PACKET_GET_COUNT(header);
@@ -753,6 +757,102 @@ int r100_cs_packet_parse(struct radeon_cs_parser *p,
753} 757}
754 758
755/** 759/**
760 * r100_cs_packet_next_vline() - parse userspace VLINE packet
761 * @parser: parser structure holding parsing context.
762 *
763 * Userspace sends a special sequence for VLINE waits.
764 * PACKET0 - VLINE_START_END + value
765 * PACKET0 - WAIT_UNTIL +_value
766 * RELOC (P3) - crtc_id in reloc.
767 *
768 * This function parses this and relocates the VLINE START END
769 * and WAIT UNTIL packets to the correct crtc.
770 * It also detects a switched off crtc and nulls out the
771 * wait in that case.
772 */
773int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
774{
775 struct radeon_cs_chunk *ib_chunk;
776 struct drm_mode_object *obj;
777 struct drm_crtc *crtc;
778 struct radeon_crtc *radeon_crtc;
779 struct radeon_cs_packet p3reloc, waitreloc;
780 int crtc_id;
781 int r;
782 uint32_t header, h_idx, reg;
783
784 ib_chunk = &p->chunks[p->chunk_ib_idx];
785
786 /* parse the wait until */
787 r = r100_cs_packet_parse(p, &waitreloc, p->idx);
788 if (r)
789 return r;
790
791 /* check its a wait until and only 1 count */
792 if (waitreloc.reg != RADEON_WAIT_UNTIL ||
793 waitreloc.count != 0) {
794 DRM_ERROR("vline wait had illegal wait until segment\n");
795 r = -EINVAL;
796 return r;
797 }
798
799 if (ib_chunk->kdata[waitreloc.idx + 1] != RADEON_WAIT_CRTC_VLINE) {
800 DRM_ERROR("vline wait had illegal wait until\n");
801 r = -EINVAL;
802 return r;
803 }
804
805 /* jump over the NOP */
806 r = r100_cs_packet_parse(p, &p3reloc, p->idx);
807 if (r)
808 return r;
809
810 h_idx = p->idx - 2;
811 p->idx += waitreloc.count;
812 p->idx += p3reloc.count;
813
814 header = ib_chunk->kdata[h_idx];
815 crtc_id = ib_chunk->kdata[h_idx + 5];
816 reg = ib_chunk->kdata[h_idx] >> 2;
817 mutex_lock(&p->rdev->ddev->mode_config.mutex);
818 obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
819 if (!obj) {
820 DRM_ERROR("cannot find crtc %d\n", crtc_id);
821 r = -EINVAL;
822 goto out;
823 }
824 crtc = obj_to_crtc(obj);
825 radeon_crtc = to_radeon_crtc(crtc);
826 crtc_id = radeon_crtc->crtc_id;
827
828 if (!crtc->enabled) {
829 /* if the CRTC isn't enabled - we need to nop out the wait until */
830 ib_chunk->kdata[h_idx + 2] = PACKET2(0);
831 ib_chunk->kdata[h_idx + 3] = PACKET2(0);
832 } else if (crtc_id == 1) {
833 switch (reg) {
834 case AVIVO_D1MODE_VLINE_START_END:
835 header &= R300_CP_PACKET0_REG_MASK;
836 header |= AVIVO_D2MODE_VLINE_START_END >> 2;
837 break;
838 case RADEON_CRTC_GUI_TRIG_VLINE:
839 header &= R300_CP_PACKET0_REG_MASK;
840 header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
841 break;
842 default:
843 DRM_ERROR("unknown crtc reloc\n");
844 r = -EINVAL;
845 goto out;
846 }
847 ib_chunk->kdata[h_idx] = header;
848 ib_chunk->kdata[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
849 }
850out:
851 mutex_unlock(&p->rdev->ddev->mode_config.mutex);
852 return r;
853}
854
855/**
756 * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3 856 * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
757 * @parser: parser structure holding parsing context. 857 * @parser: parser structure holding parsing context.
758 * @data: pointer to relocation data 858 * @data: pointer to relocation data
@@ -814,6 +914,7 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
814 unsigned idx; 914 unsigned idx;
815 bool onereg; 915 bool onereg;
816 int r; 916 int r;
917 u32 tile_flags = 0;
817 918
818 ib = p->ib->ptr; 919 ib = p->ib->ptr;
819 ib_chunk = &p->chunks[p->chunk_ib_idx]; 920 ib_chunk = &p->chunks[p->chunk_ib_idx];
@@ -825,6 +926,15 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
825 } 926 }
826 for (i = 0; i <= pkt->count; i++, idx++, reg += 4) { 927 for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
827 switch (reg) { 928 switch (reg) {
929 case RADEON_CRTC_GUI_TRIG_VLINE:
930 r = r100_cs_packet_parse_vline(p);
931 if (r) {
932 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
933 idx, reg);
934 r100_cs_dump_packet(p, pkt);
935 return r;
936 }
937 break;
828 /* FIXME: only allow PACKET3 blit? easier to check for out of 938 /* FIXME: only allow PACKET3 blit? easier to check for out of
829 * range access */ 939 * range access */
830 case RADEON_DST_PITCH_OFFSET: 940 case RADEON_DST_PITCH_OFFSET:
@@ -838,7 +948,20 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
838 } 948 }
839 tmp = ib_chunk->kdata[idx] & 0x003fffff; 949 tmp = ib_chunk->kdata[idx] & 0x003fffff;
840 tmp += (((u32)reloc->lobj.gpu_offset) >> 10); 950 tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
841 ib[idx] = (ib_chunk->kdata[idx] & 0xffc00000) | tmp; 951
952 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
953 tile_flags |= RADEON_DST_TILE_MACRO;
954 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
955 if (reg == RADEON_SRC_PITCH_OFFSET) {
956 DRM_ERROR("Cannot src blit from microtiled surface\n");
957 r100_cs_dump_packet(p, pkt);
958 return -EINVAL;
959 }
960 tile_flags |= RADEON_DST_TILE_MICRO;
961 }
962
963 tmp |= tile_flags;
964 ib[idx] = (ib_chunk->kdata[idx] & 0x3fc00000) | tmp;
842 break; 965 break;
843 case RADEON_RB3D_DEPTHOFFSET: 966 case RADEON_RB3D_DEPTHOFFSET:
844 case RADEON_RB3D_COLOROFFSET: 967 case RADEON_RB3D_COLOROFFSET:
@@ -869,6 +992,11 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
869 case R300_TX_OFFSET_0+52: 992 case R300_TX_OFFSET_0+52:
870 case R300_TX_OFFSET_0+56: 993 case R300_TX_OFFSET_0+56:
871 case R300_TX_OFFSET_0+60: 994 case R300_TX_OFFSET_0+60:
995 /* rn50 has no 3D engine so fail on any 3d setup */
996 if (ASIC_IS_RN50(p->rdev)) {
997 DRM_ERROR("attempt to use RN50 3D engine failed\n");
998 return -EINVAL;
999 }
872 r = r100_cs_packet_next_reloc(p, &reloc); 1000 r = r100_cs_packet_next_reloc(p, &reloc);
873 if (r) { 1001 if (r) {
874 DRM_ERROR("No reloc for ib[%d]=0x%04X\n", 1002 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
@@ -878,6 +1006,25 @@ static int r100_packet0_check(struct radeon_cs_parser *p,
878 } 1006 }
879 ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset); 1007 ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
880 break; 1008 break;
1009 case R300_RB3D_COLORPITCH0:
1010 case RADEON_RB3D_COLORPITCH:
1011 r = r100_cs_packet_next_reloc(p, &reloc);
1012 if (r) {
1013 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1014 idx, reg);
1015 r100_cs_dump_packet(p, pkt);
1016 return r;
1017 }
1018
1019 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1020 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1021 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1022 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1023
1024 tmp = ib_chunk->kdata[idx] & ~(0x7 << 16);
1025 tmp |= tile_flags;
1026 ib[idx] = tmp;
1027 break;
881 default: 1028 default:
882 /* FIXME: we don't want to allow anyothers packet */ 1029 /* FIXME: we don't want to allow anyothers packet */
883 break; 1030 break;
@@ -1256,29 +1403,100 @@ static void r100_vram_get_type(struct radeon_device *rdev)
1256 } 1403 }
1257} 1404}
1258 1405
1259void r100_vram_info(struct radeon_device *rdev) 1406static u32 r100_get_accessible_vram(struct radeon_device *rdev)
1260{ 1407{
1261 r100_vram_get_type(rdev); 1408 u32 aper_size;
1409 u8 byte;
1410
1411 aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1412
1413 /* Set HDP_APER_CNTL only on cards that are known not to be broken,
1414 * that is has the 2nd generation multifunction PCI interface
1415 */
1416 if (rdev->family == CHIP_RV280 ||
1417 rdev->family >= CHIP_RV350) {
1418 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
1419 ~RADEON_HDP_APER_CNTL);
1420 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
1421 return aper_size * 2;
1422 }
1423
1424 /* Older cards have all sorts of funny issues to deal with. First
1425 * check if it's a multifunction card by reading the PCI config
1426 * header type... Limit those to one aperture size
1427 */
1428 pci_read_config_byte(rdev->pdev, 0xe, &byte);
1429 if (byte & 0x80) {
1430 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
1431 DRM_INFO("Limiting VRAM to one aperture\n");
1432 return aper_size;
1433 }
1434
1435 /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
1436 * have set it up. We don't write this as it's broken on some ASICs but
1437 * we expect the BIOS to have done the right thing (might be too optimistic...)
1438 */
1439 if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
1440 return aper_size * 2;
1441 return aper_size;
1442}
1443
1444void r100_vram_init_sizes(struct radeon_device *rdev)
1445{
1446 u64 config_aper_size;
1447 u32 accessible;
1448
1449 config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1262 1450
1263 if (rdev->flags & RADEON_IS_IGP) { 1451 if (rdev->flags & RADEON_IS_IGP) {
1264 uint32_t tom; 1452 uint32_t tom;
1265 /* read NB_TOM to get the amount of ram stolen for the GPU */ 1453 /* read NB_TOM to get the amount of ram stolen for the GPU */
1266 tom = RREG32(RADEON_NB_TOM); 1454 tom = RREG32(RADEON_NB_TOM);
1267 rdev->mc.vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16); 1455 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
1268 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size); 1456 /* for IGPs we need to keep VRAM where it was put by the BIOS */
1457 rdev->mc.vram_location = (tom & 0xffff) << 16;
1458 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1459 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1269 } else { 1460 } else {
1270 rdev->mc.vram_size = RREG32(RADEON_CONFIG_MEMSIZE); 1461 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
1271 /* Some production boards of m6 will report 0 1462 /* Some production boards of m6 will report 0
1272 * if it's 8 MB 1463 * if it's 8 MB
1273 */ 1464 */
1274 if (rdev->mc.vram_size == 0) { 1465 if (rdev->mc.real_vram_size == 0) {
1275 rdev->mc.vram_size = 8192 * 1024; 1466 rdev->mc.real_vram_size = 8192 * 1024;
1276 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size); 1467 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1277 } 1468 }
1469 /* let driver place VRAM */
1470 rdev->mc.vram_location = 0xFFFFFFFFUL;
1471 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
1472 * Novell bug 204882 + along with lots of ubuntu ones */
1473 if (config_aper_size > rdev->mc.real_vram_size)
1474 rdev->mc.mc_vram_size = config_aper_size;
1475 else
1476 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1278 } 1477 }
1279 1478
1479 /* work out accessible VRAM */
1480 accessible = r100_get_accessible_vram(rdev);
1481
1280 rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0); 1482 rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
1281 rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0); 1483 rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
1484
1485 if (accessible > rdev->mc.aper_size)
1486 accessible = rdev->mc.aper_size;
1487
1488 if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
1489 rdev->mc.mc_vram_size = rdev->mc.aper_size;
1490
1491 if (rdev->mc.real_vram_size > rdev->mc.aper_size)
1492 rdev->mc.real_vram_size = rdev->mc.aper_size;
1493}
1494
1495void r100_vram_info(struct radeon_device *rdev)
1496{
1497 r100_vram_get_type(rdev);
1498
1499 r100_vram_init_sizes(rdev);
1282} 1500}
1283 1501
1284 1502
@@ -1533,3 +1751,530 @@ int r100_debugfs_mc_info_init(struct radeon_device *rdev)
1533 return 0; 1751 return 0;
1534#endif 1752#endif
1535} 1753}
1754
1755int r100_set_surface_reg(struct radeon_device *rdev, int reg,
1756 uint32_t tiling_flags, uint32_t pitch,
1757 uint32_t offset, uint32_t obj_size)
1758{
1759 int surf_index = reg * 16;
1760 int flags = 0;
1761
1762 /* r100/r200 divide by 16 */
1763 if (rdev->family < CHIP_R300)
1764 flags = pitch / 16;
1765 else
1766 flags = pitch / 8;
1767
1768 if (rdev->family <= CHIP_RS200) {
1769 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1770 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1771 flags |= RADEON_SURF_TILE_COLOR_BOTH;
1772 if (tiling_flags & RADEON_TILING_MACRO)
1773 flags |= RADEON_SURF_TILE_COLOR_MACRO;
1774 } else if (rdev->family <= CHIP_RV280) {
1775 if (tiling_flags & (RADEON_TILING_MACRO))
1776 flags |= R200_SURF_TILE_COLOR_MACRO;
1777 if (tiling_flags & RADEON_TILING_MICRO)
1778 flags |= R200_SURF_TILE_COLOR_MICRO;
1779 } else {
1780 if (tiling_flags & RADEON_TILING_MACRO)
1781 flags |= R300_SURF_TILE_MACRO;
1782 if (tiling_flags & RADEON_TILING_MICRO)
1783 flags |= R300_SURF_TILE_MICRO;
1784 }
1785
1786 DRM_DEBUG("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
1787 WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
1788 WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
1789 WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
1790 return 0;
1791}
1792
1793void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
1794{
1795 int surf_index = reg * 16;
1796 WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
1797}
1798
1799void r100_bandwidth_update(struct radeon_device *rdev)
1800{
1801 fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
1802 fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
1803 fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
1804 uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
1805 fixed20_12 memtcas_ff[8] = {
1806 fixed_init(1),
1807 fixed_init(2),
1808 fixed_init(3),
1809 fixed_init(0),
1810 fixed_init_half(1),
1811 fixed_init_half(2),
1812 fixed_init(0),
1813 };
1814 fixed20_12 memtcas_rs480_ff[8] = {
1815 fixed_init(0),
1816 fixed_init(1),
1817 fixed_init(2),
1818 fixed_init(3),
1819 fixed_init(0),
1820 fixed_init_half(1),
1821 fixed_init_half(2),
1822 fixed_init_half(3),
1823 };
1824 fixed20_12 memtcas2_ff[8] = {
1825 fixed_init(0),
1826 fixed_init(1),
1827 fixed_init(2),
1828 fixed_init(3),
1829 fixed_init(4),
1830 fixed_init(5),
1831 fixed_init(6),
1832 fixed_init(7),
1833 };
1834 fixed20_12 memtrbs[8] = {
1835 fixed_init(1),
1836 fixed_init_half(1),
1837 fixed_init(2),
1838 fixed_init_half(2),
1839 fixed_init(3),
1840 fixed_init_half(3),
1841 fixed_init(4),
1842 fixed_init_half(4)
1843 };
1844 fixed20_12 memtrbs_r4xx[8] = {
1845 fixed_init(4),
1846 fixed_init(5),
1847 fixed_init(6),
1848 fixed_init(7),
1849 fixed_init(8),
1850 fixed_init(9),
1851 fixed_init(10),
1852 fixed_init(11)
1853 };
1854 fixed20_12 min_mem_eff;
1855 fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
1856 fixed20_12 cur_latency_mclk, cur_latency_sclk;
1857 fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
1858 disp_drain_rate2, read_return_rate;
1859 fixed20_12 time_disp1_drop_priority;
1860 int c;
1861 int cur_size = 16; /* in octawords */
1862 int critical_point = 0, critical_point2;
1863/* uint32_t read_return_rate, time_disp1_drop_priority; */
1864 int stop_req, max_stop_req;
1865 struct drm_display_mode *mode1 = NULL;
1866 struct drm_display_mode *mode2 = NULL;
1867 uint32_t pixel_bytes1 = 0;
1868 uint32_t pixel_bytes2 = 0;
1869
1870 if (rdev->mode_info.crtcs[0]->base.enabled) {
1871 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
1872 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
1873 }
1874 if (rdev->mode_info.crtcs[1]->base.enabled) {
1875 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
1876 pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
1877 }
1878
1879 min_mem_eff.full = rfixed_const_8(0);
1880 /* get modes */
1881 if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
1882 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
1883 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
1884 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
1885 /* check crtc enables */
1886 if (mode2)
1887 mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
1888 if (mode1)
1889 mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
1890 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
1891 }
1892
1893 /*
1894 * determine is there is enough bw for current mode
1895 */
1896 mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
1897 temp_ff.full = rfixed_const(100);
1898 mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
1899 sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
1900 sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
1901
1902 temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
1903 temp_ff.full = rfixed_const(temp);
1904 mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
1905
1906 pix_clk.full = 0;
1907 pix_clk2.full = 0;
1908 peak_disp_bw.full = 0;
1909 if (mode1) {
1910 temp_ff.full = rfixed_const(1000);
1911 pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
1912 pix_clk.full = rfixed_div(pix_clk, temp_ff);
1913 temp_ff.full = rfixed_const(pixel_bytes1);
1914 peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
1915 }
1916 if (mode2) {
1917 temp_ff.full = rfixed_const(1000);
1918 pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
1919 pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
1920 temp_ff.full = rfixed_const(pixel_bytes2);
1921 peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
1922 }
1923
1924 mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
1925 if (peak_disp_bw.full >= mem_bw.full) {
1926 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
1927 "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
1928 }
1929
1930 /* Get values from the EXT_MEM_CNTL register...converting its contents. */
1931 temp = RREG32(RADEON_MEM_TIMING_CNTL);
1932 if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
1933 mem_trcd = ((temp >> 2) & 0x3) + 1;
1934 mem_trp = ((temp & 0x3)) + 1;
1935 mem_tras = ((temp & 0x70) >> 4) + 1;
1936 } else if (rdev->family == CHIP_R300 ||
1937 rdev->family == CHIP_R350) { /* r300, r350 */
1938 mem_trcd = (temp & 0x7) + 1;
1939 mem_trp = ((temp >> 8) & 0x7) + 1;
1940 mem_tras = ((temp >> 11) & 0xf) + 4;
1941 } else if (rdev->family == CHIP_RV350 ||
1942 rdev->family <= CHIP_RV380) {
1943 /* rv3x0 */
1944 mem_trcd = (temp & 0x7) + 3;
1945 mem_trp = ((temp >> 8) & 0x7) + 3;
1946 mem_tras = ((temp >> 11) & 0xf) + 6;
1947 } else if (rdev->family == CHIP_R420 ||
1948 rdev->family == CHIP_R423 ||
1949 rdev->family == CHIP_RV410) {
1950 /* r4xx */
1951 mem_trcd = (temp & 0xf) + 3;
1952 if (mem_trcd > 15)
1953 mem_trcd = 15;
1954 mem_trp = ((temp >> 8) & 0xf) + 3;
1955 if (mem_trp > 15)
1956 mem_trp = 15;
1957 mem_tras = ((temp >> 12) & 0x1f) + 6;
1958 if (mem_tras > 31)
1959 mem_tras = 31;
1960 } else { /* RV200, R200 */
1961 mem_trcd = (temp & 0x7) + 1;
1962 mem_trp = ((temp >> 8) & 0x7) + 1;
1963 mem_tras = ((temp >> 12) & 0xf) + 4;
1964 }
1965 /* convert to FF */
1966 trcd_ff.full = rfixed_const(mem_trcd);
1967 trp_ff.full = rfixed_const(mem_trp);
1968 tras_ff.full = rfixed_const(mem_tras);
1969
1970 /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
1971 temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
1972 data = (temp & (7 << 20)) >> 20;
1973 if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
1974 if (rdev->family == CHIP_RS480) /* don't think rs400 */
1975 tcas_ff = memtcas_rs480_ff[data];
1976 else
1977 tcas_ff = memtcas_ff[data];
1978 } else
1979 tcas_ff = memtcas2_ff[data];
1980
1981 if (rdev->family == CHIP_RS400 ||
1982 rdev->family == CHIP_RS480) {
1983 /* extra cas latency stored in bits 23-25 0-4 clocks */
1984 data = (temp >> 23) & 0x7;
1985 if (data < 5)
1986 tcas_ff.full += rfixed_const(data);
1987 }
1988
1989 if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
1990 /* on the R300, Tcas is included in Trbs.
1991 */
1992 temp = RREG32(RADEON_MEM_CNTL);
1993 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
1994 if (data == 1) {
1995 if (R300_MEM_USE_CD_CH_ONLY & temp) {
1996 temp = RREG32(R300_MC_IND_INDEX);
1997 temp &= ~R300_MC_IND_ADDR_MASK;
1998 temp |= R300_MC_READ_CNTL_CD_mcind;
1999 WREG32(R300_MC_IND_INDEX, temp);
2000 temp = RREG32(R300_MC_IND_DATA);
2001 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2002 } else {
2003 temp = RREG32(R300_MC_READ_CNTL_AB);
2004 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2005 }
2006 } else {
2007 temp = RREG32(R300_MC_READ_CNTL_AB);
2008 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2009 }
2010 if (rdev->family == CHIP_RV410 ||
2011 rdev->family == CHIP_R420 ||
2012 rdev->family == CHIP_R423)
2013 trbs_ff = memtrbs_r4xx[data];
2014 else
2015 trbs_ff = memtrbs[data];
2016 tcas_ff.full += trbs_ff.full;
2017 }
2018
2019 sclk_eff_ff.full = sclk_ff.full;
2020
2021 if (rdev->flags & RADEON_IS_AGP) {
2022 fixed20_12 agpmode_ff;
2023 agpmode_ff.full = rfixed_const(radeon_agpmode);
2024 temp_ff.full = rfixed_const_666(16);
2025 sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
2026 }
2027 /* TODO PCIE lanes may affect this - agpmode == 16?? */
2028
2029 if (ASIC_IS_R300(rdev)) {
2030 sclk_delay_ff.full = rfixed_const(250);
2031 } else {
2032 if ((rdev->family == CHIP_RV100) ||
2033 rdev->flags & RADEON_IS_IGP) {
2034 if (rdev->mc.vram_is_ddr)
2035 sclk_delay_ff.full = rfixed_const(41);
2036 else
2037 sclk_delay_ff.full = rfixed_const(33);
2038 } else {
2039 if (rdev->mc.vram_width == 128)
2040 sclk_delay_ff.full = rfixed_const(57);
2041 else
2042 sclk_delay_ff.full = rfixed_const(41);
2043 }
2044 }
2045
2046 mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
2047
2048 if (rdev->mc.vram_is_ddr) {
2049 if (rdev->mc.vram_width == 32) {
2050 k1.full = rfixed_const(40);
2051 c = 3;
2052 } else {
2053 k1.full = rfixed_const(20);
2054 c = 1;
2055 }
2056 } else {
2057 k1.full = rfixed_const(40);
2058 c = 3;
2059 }
2060
2061 temp_ff.full = rfixed_const(2);
2062 mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
2063 temp_ff.full = rfixed_const(c);
2064 mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
2065 temp_ff.full = rfixed_const(4);
2066 mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
2067 mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
2068 mc_latency_mclk.full += k1.full;
2069
2070 mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
2071 mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
2072
2073 /*
2074 HW cursor time assuming worst case of full size colour cursor.
2075 */
2076 temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2077 temp_ff.full += trcd_ff.full;
2078 if (temp_ff.full < tras_ff.full)
2079 temp_ff.full = tras_ff.full;
2080 cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
2081
2082 temp_ff.full = rfixed_const(cur_size);
2083 cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
2084 /*
2085 Find the total latency for the display data.
2086 */
2087 disp_latency_overhead.full = rfixed_const(80);
2088 disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
2089 mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2090 mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2091
2092 if (mc_latency_mclk.full > mc_latency_sclk.full)
2093 disp_latency.full = mc_latency_mclk.full;
2094 else
2095 disp_latency.full = mc_latency_sclk.full;
2096
2097 /* setup Max GRPH_STOP_REQ default value */
2098 if (ASIC_IS_RV100(rdev))
2099 max_stop_req = 0x5c;
2100 else
2101 max_stop_req = 0x7c;
2102
2103 if (mode1) {
2104 /* CRTC1
2105 Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2106 GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2107 */
2108 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2109
2110 if (stop_req > max_stop_req)
2111 stop_req = max_stop_req;
2112
2113 /*
2114 Find the drain rate of the display buffer.
2115 */
2116 temp_ff.full = rfixed_const((16/pixel_bytes1));
2117 disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
2118
2119 /*
2120 Find the critical point of the display buffer.
2121 */
2122 crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
2123 crit_point_ff.full += rfixed_const_half(0);
2124
2125 critical_point = rfixed_trunc(crit_point_ff);
2126
2127 if (rdev->disp_priority == 2) {
2128 critical_point = 0;
2129 }
2130
2131 /*
2132 The critical point should never be above max_stop_req-4. Setting
2133 GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
2134 */
2135 if (max_stop_req - critical_point < 4)
2136 critical_point = 0;
2137
2138 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
2139 /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
2140 critical_point = 0x10;
2141 }
2142
2143 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
2144 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
2145 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2146 temp &= ~(RADEON_GRPH_START_REQ_MASK);
2147 if ((rdev->family == CHIP_R350) &&
2148 (stop_req > 0x15)) {
2149 stop_req -= 0x10;
2150 }
2151 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2152 temp |= RADEON_GRPH_BUFFER_SIZE;
2153 temp &= ~(RADEON_GRPH_CRITICAL_CNTL |
2154 RADEON_GRPH_CRITICAL_AT_SOF |
2155 RADEON_GRPH_STOP_CNTL);
2156 /*
2157 Write the result into the register.
2158 */
2159 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2160 (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2161
2162#if 0
2163 if ((rdev->family == CHIP_RS400) ||
2164 (rdev->family == CHIP_RS480)) {
2165 /* attempt to program RS400 disp regs correctly ??? */
2166 temp = RREG32(RS400_DISP1_REG_CNTL);
2167 temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
2168 RS400_DISP1_STOP_REQ_LEVEL_MASK);
2169 WREG32(RS400_DISP1_REQ_CNTL1, (temp |
2170 (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2171 (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2172 temp = RREG32(RS400_DMIF_MEM_CNTL1);
2173 temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
2174 RS400_DISP1_CRITICAL_POINT_STOP_MASK);
2175 WREG32(RS400_DMIF_MEM_CNTL1, (temp |
2176 (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
2177 (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
2178 }
2179#endif
2180
2181 DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
2182 /* (unsigned int)info->SavedReg->grph_buffer_cntl, */
2183 (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
2184 }
2185
2186 if (mode2) {
2187 u32 grph2_cntl;
2188 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
2189
2190 if (stop_req > max_stop_req)
2191 stop_req = max_stop_req;
2192
2193 /*
2194 Find the drain rate of the display buffer.
2195 */
2196 temp_ff.full = rfixed_const((16/pixel_bytes2));
2197 disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
2198
2199 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
2200 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
2201 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2202 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
2203 if ((rdev->family == CHIP_R350) &&
2204 (stop_req > 0x15)) {
2205 stop_req -= 0x10;
2206 }
2207 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2208 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
2209 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL |
2210 RADEON_GRPH_CRITICAL_AT_SOF |
2211 RADEON_GRPH_STOP_CNTL);
2212
2213 if ((rdev->family == CHIP_RS100) ||
2214 (rdev->family == CHIP_RS200))
2215 critical_point2 = 0;
2216 else {
2217 temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
2218 temp_ff.full = rfixed_const(temp);
2219 temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
2220 if (sclk_ff.full < temp_ff.full)
2221 temp_ff.full = sclk_ff.full;
2222
2223 read_return_rate.full = temp_ff.full;
2224
2225 if (mode1) {
2226 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
2227 time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
2228 } else {
2229 time_disp1_drop_priority.full = 0;
2230 }
2231 crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
2232 crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
2233 crit_point_ff.full += rfixed_const_half(0);
2234
2235 critical_point2 = rfixed_trunc(crit_point_ff);
2236
2237 if (rdev->disp_priority == 2) {
2238 critical_point2 = 0;
2239 }
2240
2241 if (max_stop_req - critical_point2 < 4)
2242 critical_point2 = 0;
2243
2244 }
2245
2246 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
2247 /* some R300 cards have problem with this set to 0 */
2248 critical_point2 = 0x10;
2249 }
2250
2251 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2252 (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2253
2254 if ((rdev->family == CHIP_RS400) ||
2255 (rdev->family == CHIP_RS480)) {
2256#if 0
2257 /* attempt to program RS400 disp2 regs correctly ??? */
2258 temp = RREG32(RS400_DISP2_REQ_CNTL1);
2259 temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
2260 RS400_DISP2_STOP_REQ_LEVEL_MASK);
2261 WREG32(RS400_DISP2_REQ_CNTL1, (temp |
2262 (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2263 (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2264 temp = RREG32(RS400_DISP2_REQ_CNTL2);
2265 temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
2266 RS400_DISP2_CRITICAL_POINT_STOP_MASK);
2267 WREG32(RS400_DISP2_REQ_CNTL2, (temp |
2268 (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
2269 (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
2270#endif
2271 WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
2272 WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
2273 WREG32(RS400_DMIF_MEM_CNTL1, 0x29CA71DC);
2274 WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
2275 }
2276
2277 DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
2278 (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
2279 }
2280}
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-17 11:49:59 -0500