From: Roy Spliet Date: Sat, 23 May 2015 08:37:46 +0000 (+0200) Subject: drm/nouveau/fb/ramnv50: Ressurect timing code, use proper timing/rammap handlers X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=35fe024acffc2c29bade5a68a09962bf7ea3c8ed;p=GitHub%2Fexynos8895%2Fandroid_kernel_samsung_universal8895.git drm/nouveau/fb/ramnv50: Ressurect timing code, use proper timing/rammap handlers Might need some generalisation to < GT200. For those: use at your own risk! Signed-off-by: Roy Spliet Signed-off-by: Ben Skeggs --- diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/ramcfg.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/ramcfg.h index c6fb6aa2c143..f09b6bf69098 100644 --- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/ramcfg.h +++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/ramcfg.h @@ -34,6 +34,22 @@ struct nvbios_ramcfg { unsigned ramcfg_timing; unsigned ramcfg_DLLoff; union { + struct { + unsigned ramcfg_00_03_01:1; + unsigned ramcfg_00_03_02:1; + unsigned ramcfg_00_03_08:1; + unsigned ramcfg_00_03_10:1; + unsigned ramcfg_00_04_02:1; + unsigned ramcfg_00_04_04:1; + unsigned ramcfg_00_04_20:1; + unsigned ramcfg_00_05:8; + unsigned ramcfg_00_06:8; + unsigned ramcfg_00_07:8; + unsigned ramcfg_00_08:8; + unsigned ramcfg_00_09:8; + unsigned ramcfg_00_0a_0f:4; + unsigned ramcfg_00_0a_f0:4; + }; struct { unsigned ramcfg_10_02_01:1; unsigned ramcfg_10_02_02:1; diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/rammap.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/rammap.h index 609a905ec780..2044fc911ceb 100644 --- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/rammap.h +++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/rammap.h @@ -15,6 +15,8 @@ u32 nvbios_rammapEm(struct nvkm_bios *, u16 mhz, u32 nvbios_rammapSe(struct nvkm_bios *, u32 data, u8 ever, u8 ehdr, u8 ecnt, u8 elen, int idx, u8 *ver, u8 *hdr); +u32 nvbios_rammapSp_from_perf(struct nvkm_bios *bios, u32 data, u8 size, int idx, + struct nvbios_ramcfg *p); u32 nvbios_rammapSp(struct nvkm_bios *, u32 data, u8 ever, u8 ehdr, u8 ecnt, u8 elen, int idx, u8 *ver, u8 *hdr, struct nvbios_ramcfg *); diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c index a688d3b4ef37..d6e4af7606ba 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c @@ -140,6 +140,35 @@ nvbios_rammapSe(struct nvkm_bios *bios, u32 data, return 0; } +u32 +nvbios_rammapSp_from_perf(struct nvkm_bios *bios, u32 data, u8 size, int idx, + struct nvbios_ramcfg *p) +{ + data += (idx * size); + + if (size < 11) + return 0x00000000; + + p->ramcfg_timing = nv_ro08(bios, data + 0x01); + p->ramcfg_00_03_01 = (nv_ro08(bios, data + 0x03) & 0x01) >> 0; + p->ramcfg_00_03_02 = (nv_ro08(bios, data + 0x03) & 0x02) >> 1; + p->ramcfg_DLLoff = (nv_ro08(bios, data + 0x03) & 0x04) >> 2; + p->ramcfg_00_03_08 = (nv_ro08(bios, data + 0x03) & 0x08) >> 3; + p->ramcfg_00_03_10 = (nv_ro08(bios, data + 0x03) & 0x10) >> 4; + p->ramcfg_00_04_02 = (nv_ro08(bios, data + 0x04) & 0x02) >> 1; + p->ramcfg_00_04_04 = (nv_ro08(bios, data + 0x04) & 0x04) >> 2; + p->ramcfg_00_04_20 = (nv_ro08(bios, data + 0x04) & 0x20) >> 5; + p->ramcfg_00_05 = (nv_ro08(bios, data + 0x05) & 0xff) >> 0; + p->ramcfg_00_06 = (nv_ro08(bios, data + 0x06) & 0xff) >> 0; + p->ramcfg_00_07 = (nv_ro08(bios, data + 0x07) & 0xff) >> 0; + p->ramcfg_00_08 = (nv_ro08(bios, data + 0x08) & 0xff) >> 0; + p->ramcfg_00_09 = (nv_ro08(bios, data + 0x09) & 0xff) >> 0; + p->ramcfg_00_0a_0f = (nv_ro08(bios, data + 0x0a) & 0x0f) >> 0; + p->ramcfg_00_0a_f0 = (nv_ro08(bios, data + 0x0a) & 0xf0) >> 4; + + return data; +} + u32 nvbios_rammapSp(struct nvkm_bios *bios, u32 data, u8 ever, u8 ehdr, u8 ecnt, u8 elen, int idx, diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c index 67715c69e158..37ccc4dbeaf5 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramnv50.c @@ -29,6 +29,7 @@ #include #include #include +#include #include #include @@ -55,6 +56,85 @@ struct nv50_ram { struct nv50_ramseq hwsq; }; +#define T(t) cfg->timing_10_##t +static int +nv50_ram_timing_calc(struct nvkm_fb *pfb, u32 *timing) +{ + struct nv50_ram *ram = (void *)pfb->ram; + struct nvbios_ramcfg *cfg = &ram->base.target.bios; + u32 cur2, cur3, cur4, cur7, cur8; + u8 unkt3b; + + cur2 = nv_rd32(pfb, 0x100228); + cur3 = nv_rd32(pfb, 0x10022c); + cur4 = nv_rd32(pfb, 0x100230); + cur7 = nv_rd32(pfb, 0x10023c); + cur8 = nv_rd32(pfb, 0x100240); + + switch ((!T(CWL)) * ram->base.type) { + case NV_MEM_TYPE_DDR2: + T(CWL) = T(CL) - 1; + break; + case NV_MEM_TYPE_GDDR3: + T(CWL) = ((cur2 & 0xff000000) >> 24) + 1; + break; + } + + /* XXX: N=1 is not proper statistics */ + if (nv_device(pfb)->chipset == 0xa0) { + unkt3b = 0x19 + ram->base.next->bios.rammap_00_16_40; + timing[6] = (0x2d + T(CL) - T(CWL) + + ram->base.next->bios.rammap_00_16_40) << 16 | + T(CWL) << 8 | + (0x2f + T(CL) - T(CWL)); + } else { + unkt3b = 0x16; + timing[6] = (0x2b + T(CL) - T(CWL)) << 16 | + max_t(s8, T(CWL) - 2, 1) << 8 | + (0x2e + T(CL) - T(CWL)); + } + + timing[0] = (T(RP) << 24 | T(RAS) << 16 | T(RFC) << 8 | T(RC)); + timing[1] = (T(WR) + 1 + T(CWL)) << 24 | + max_t(u8, T(18), 1) << 16 | + (T(WTR) + 1 + T(CWL)) << 8 | + (3 + T(CL) - T(CWL)); + timing[2] = (T(CWL) - 1) << 24 | + (T(RRD) << 16) | + (T(RCDWR) << 8) | + T(RCDRD); + timing[3] = (unkt3b - 2 + T(CL)) << 24 | + unkt3b << 16 | + (T(CL) - 1) << 8 | + (T(CL) - 1); + timing[4] = (cur4 & 0xffff0000) | + T(13) << 8 | + T(13); + timing[5] = T(RFC) << 24 | + max_t(u8, T(RCDRD), T(RCDWR)) << 16 | + T(RP); + /* Timing 6 is already done above */ + timing[7] = (cur7 & 0xff00ffff) | (T(CL) - 1) << 16; + timing[8] = (cur8 & 0xffffff00); + + /* XXX: P.version == 1 only has DDR2 and GDDR3? */ + if (pfb->ram->type == NV_MEM_TYPE_DDR2) { + timing[5] |= (T(CL) + 3) << 8; + timing[8] |= (T(CL) - 4); + } else if (pfb->ram->type == NV_MEM_TYPE_GDDR3) { + timing[5] |= (T(CL) + 2) << 8; + timing[8] |= (T(CL) - 2); + } + + nv_debug(pfb, " 220: %08x %08x %08x %08x\n", + timing[0], timing[1], timing[2], timing[3]); + nv_debug(pfb, " 230: %08x %08x %08x %08x\n", + timing[4], timing[5], timing[6], timing[7]); + nv_debug(pfb, " 240: %08x\n", timing[8]); + return 0; +} +#undef T + #define QFX5800NVA0 1 static int @@ -65,22 +145,25 @@ nv50_ram_calc(struct nvkm_fb *pfb, u32 freq) struct nv50_ramseq *hwsq = &ram->hwsq; struct nvbios_perfE perfE; struct nvbios_pll mpll; - struct { - u32 data; - u8 size; - } ramcfg, timing; - u8 ver, hdr, cnt, len, strap; + struct nvkm_ram_data *next; + u8 ver, hdr, cnt, len, strap, size; + u32 data; u32 r100da0; int N1, M1, N2, M2, P; int ret, i; + u32 timing[9]; + + next = &ram->base.target; + next->freq = freq; + ram->base.next = next; /* lookup closest matching performance table entry for frequency */ i = 0; do { - ramcfg.data = nvbios_perfEp(bios, i++, &ver, &hdr, &cnt, - &ramcfg.size, &perfE); - if (!ramcfg.data || (ver < 0x25 || ver >= 0x40) || - (ramcfg.size < 2)) { + data = nvbios_perfEp(bios, i++, &ver, &hdr, &cnt, + &size, &perfE); + if (!data || (ver < 0x25 || ver >= 0x40) || + (size < 2)) { nv_error(pfb, "invalid/missing perftab entry\n"); return -EINVAL; } @@ -93,23 +176,48 @@ nv50_ram_calc(struct nvkm_fb *pfb, u32 freq) return -EINVAL; } - ramcfg.data += hdr + (strap * ramcfg.size); + data = nvbios_rammapSp_from_perf(bios, data + hdr, size, strap, + &next->bios); + if (!data) { + nv_error(pfb, "invalid/missing rammap entry "); + return -EINVAL; + } /* lookup memory timings, if bios says they're present */ - strap = nv_ro08(bios, ramcfg.data + 0x01); - if (strap != 0xff) { - timing.data = nvbios_timingEe(bios, strap, &ver, &hdr, - &cnt, &len); - if (!timing.data || ver != 0x10 || hdr < 0x12) { + if (next->bios.ramcfg_timing != 0xff) { + data = nvbios_timingEp(bios, next->bios.ramcfg_timing, + &ver, &hdr, &cnt, &len, &next->bios); + if (!data || ver != 0x10 || hdr < 0x12) { nv_error(pfb, "invalid/missing timing entry " "%02x %04x %02x %02x\n", - strap, timing.data, ver, hdr); + strap, data, ver, hdr); return -EINVAL; } - } else { - timing.data = 0; } + nv50_ram_timing_calc(pfb, timing); + + ret = ram_init(hwsq, nv_subdev(pfb)); + if (ret) + return ret; + + /* Determine ram-specific MR values */ + ram->base.mr[0] = ram_rd32(hwsq, mr[0]); + ram->base.mr[1] = ram_rd32(hwsq, mr[1]); + ram->base.mr[2] = ram_rd32(hwsq, mr[2]); + + switch (ram->base.type) { + case NV_MEM_TYPE_GDDR3: + ret = nvkm_gddr3_calc(&ram->base); + break; + default: + ret = -ENOSYS; + break; + } + + if (ret) + return ret; + /* XXX: 750MHz seems rather arbitrary */ if (freq <= 750000) { r100da0 = 0x00000010; @@ -117,10 +225,6 @@ nv50_ram_calc(struct nvkm_fb *pfb, u32 freq) r100da0 = 0x00000000; } - ret = ram_init(hwsq, nv_subdev(pfb)); - if (ret) - return ret; - ram_wait(hwsq, 0x01, 0x00); /* wait for !vblank */ ram_wait(hwsq, 0x01, 0x01); /* wait for vblank */ ram_wr32(hwsq, 0x611200, 0x00003300); @@ -177,17 +281,15 @@ nv50_ram_calc(struct nvkm_fb *pfb, u32 freq) break; } - ram_mask(hwsq, timing[3], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[1], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[6], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[7], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[8], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[0], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[2], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[4], 0x00000000, 0x00000000); /*XXX*/ - ram_mask(hwsq, timing[5], 0x00000000, 0x00000000); /*XXX*/ - - ram_mask(hwsq, timing[0], 0x00000000, 0x00000000); /*XXX*/ + ram_mask(hwsq, timing[3], 0xffffffff, timing[3]); + ram_mask(hwsq, timing[1], 0xffffffff, timing[1]); + ram_mask(hwsq, timing[6], 0xffffffff, timing[6]); + ram_mask(hwsq, timing[7], 0xffffffff, timing[7]); + ram_mask(hwsq, timing[8], 0xffffffff, timing[8]); + ram_mask(hwsq, timing[0], 0xffffffff, timing[0]); + ram_mask(hwsq, timing[2], 0xffffffff, timing[2]); + ram_mask(hwsq, timing[4], 0xffffffff, timing[4]); + ram_mask(hwsq, timing[5], 0xffffffff, timing[5]); #if QFX5800NVA0 ram_nuke(hwsq, 0x100e24);