From c7c039fd31be82ecb8d48477955e76badd38141a Mon Sep 17 00:00:00 2001 From: Roy Spliet Date: Mon, 9 Jan 2012 15:23:07 +1000 Subject: drm/nouveau/pm: implement DDR2/DDR3/GDDR3/GDDR5 MR generation and validation Roy Spliet: - Implement according to specs - Simplify - Make array for mc latency registers Martin Peres: - squash and split all the commits from Roy - rework following Ben Skeggs comments - add a form of timings validation - store the initial timings for later use Ben Skeggs - merge slightly modified tidy-up patch with this one - remove perflvl-dropping logic for the moment Signed-off-by: Roy Spliet Signed-off-by: Martin Peres Signed-off-by: Ben Skeggs --- drivers/gpu/drm/nouveau/nouveau_drv.h | 38 +-- drivers/gpu/drm/nouveau/nouveau_mem.c | 588 ++++++++++++++++++++++++++-------- drivers/gpu/drm/nouveau/nouveau_pm.c | 1 + drivers/gpu/drm/nouveau/nv50_vram.c | 1 + 4 files changed, 476 insertions(+), 152 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h index c5105326db7f..f5700418da3f 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drv.h +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h @@ -432,25 +432,26 @@ struct nouveau_pm_voltage { int nr_level; }; +/* Exclusive upper limits */ +#define NV_MEM_CL_DDR2_MAX 8 +#define NV_MEM_WR_DDR2_MAX 9 +#define NV_MEM_CL_DDR3_MAX 17 +#define NV_MEM_WR_DDR3_MAX 17 +#define NV_MEM_CL_GDDR3_MAX 16 +#define NV_MEM_WR_GDDR3_MAX 18 +#define NV_MEM_CL_GDDR5_MAX 21 +#define NV_MEM_WR_GDDR5_MAX 20 + struct nouveau_pm_memtiming { int id; - u32 reg_0; /* 0x10f290 on Fermi, 0x100220 for older */ - u32 reg_1; - u32 reg_2; - u32 reg_3; - u32 reg_4; - u32 reg_5; - u32 reg_6; - u32 reg_7; - u32 reg_8; - /* To be written to 0x1002c0 */ - u8 CL; - u8 WR; + + u32 reg[9]; + u32 mr[4]; + u8 tCWL; - bool odt; - bool dll_disable; - bool ron_pull; + u8 odt; + u8 drive_strength; }; struct nouveau_pm_tbl_header { @@ -527,8 +528,10 @@ struct nouveau_pm_threshold_temp { struct nouveau_pm_memtimings { bool supported; + struct nouveau_pm_memtiming boot; struct nouveau_pm_memtiming *timing; int nr_timing; + int nr_timing_valid; }; struct nouveau_pm_fan { @@ -796,6 +799,7 @@ struct drm_nouveau_private { } vram_type; uint64_t vram_size; uint64_t vram_sys_base; + bool vram_rank_B; uint64_t fb_available_size; uint64_t fb_mappable_pages; @@ -927,10 +931,6 @@ extern void nv10_mem_put_tile_region(struct drm_device *dev, struct nouveau_fence *fence); extern const struct ttm_mem_type_manager_func nouveau_vram_manager; extern const struct ttm_mem_type_manager_func nouveau_gart_manager; -void nv30_mem_timing_entry(struct drm_device *dev, - struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, uint8_t magic_number, - struct nouveau_pm_memtiming *timing); /* nouveau_notifier.c */ extern int nouveau_notifier_init_channel(struct nouveau_channel *); diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c index 3337f2cfb5f6..1cd29c02a7e9 100644 --- a/drivers/gpu/drm/nouveau/nouveau_mem.c +++ b/drivers/gpu/drm/nouveau/nouveau_mem.c @@ -471,50 +471,41 @@ nouveau_mem_gart_init(struct drm_device *dev) return 0; } -/* XXX: For now a dummy. More samples required, possibly even a card - * Called from nouveau_perf.c */ -void nv30_mem_timing_entry(struct drm_device *dev, - struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, uint8_t magic_number, - struct nouveau_pm_memtiming *timing) -{ - - NV_DEBUG(dev, "Timing entry format unknown, " - "please contact nouveau developers"); -} - -void nv40_mem_timing_entry(struct drm_device *dev, - struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, - struct nouveau_pm_memtiming *timing) +static void +nv40_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) { - timing->reg_0 = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC); + t->reg[0] = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC); /* XXX: I don't trust the -1's and +1's... they must come * from somewhere! */ - timing->reg_1 = (e->tWR + 2 + (timing->tCWL - 1)) << 24 | - 1 << 16 | - (e->tWTR + 2 + (timing->tCWL - 1)) << 8 | - (e->tCL + 2 - (timing->tCWL - 1)); - - timing->reg_2 = 0x20200000 | ((timing->tCWL - 1) << 24 | - e->tRRD << 16 | e->tRCDWR << 8 | e->tRCDRD); - - NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x\n", timing->id, - timing->reg_0, timing->reg_1, timing->reg_2); + t->reg[1] = (e->tWR + 2 + (t->tCWL - 1)) << 24 | + 1 << 16 | + (e->tWTR + 2 + (t->tCWL - 1)) << 8 | + (e->tCL + 2 - (t->tCWL - 1)); + + t->reg[2] = 0x20200000 | + ((t->tCWL - 1) << 24 | + e->tRRD << 16 | + e->tRCDWR << 8 | + e->tRCDRD); + + NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x\n", t->id, + t->reg[0], t->reg[1], t->reg[2]); } -void nv50_mem_timing_entry(struct drm_device *dev, struct bit_entry *P, - struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, - struct nouveau_pm_memtiming *timing) +static void +nv50_mem_timing_entry(struct drm_device *dev, struct bit_entry *P, + struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) { struct drm_nouveau_private *dev_priv = dev->dev_private; - uint8_t unk18 = 1, - unk20 = 0, - unk21 = 0, - tmp7_3; + uint8_t unk18 = 1, unk20 = 0, unk21 = 0, tmp7_3; switch (min(hdr->entry_len, (u8) 22)) { case 22: @@ -523,132 +514,414 @@ void nv50_mem_timing_entry(struct drm_device *dev, struct bit_entry *P, unk20 = e->tUNK_20; case 20: if (e->tCWL > 0) - timing->tCWL = e->tCWL; + t->tCWL = e->tCWL; case 19: unk18 = e->tUNK_18; break; } - timing->reg_0 = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC); + t->reg[0] = (e->tRP << 24 | e->tRAS << 16 | e->tRFC << 8 | e->tRC); - timing->reg_1 = (e->tWR + 2 + (timing->tCWL - 1)) << 24 | + t->reg[1] = (e->tWR + 2 + (t->tCWL - 1)) << 24 | max(unk18, (u8) 1) << 16 | - (e->tWTR + 2 + (timing->tCWL - 1)) << 8; + (e->tWTR + 2 + (t->tCWL - 1)) << 8; - timing->reg_2 = ((timing->tCWL - 1) << 24 | e->tRRD << 16 | - e->tRCDWR << 8 | e->tRCDRD); + t->reg[2] = ((t->tCWL - 1) << 24 | + e->tRRD << 16 | + e->tRCDWR << 8 | + e->tRCDRD); - timing->reg_4 = e->tUNK_13 << 8 | e->tUNK_13; + t->reg[4] = e->tUNK_13 << 8 | e->tUNK_13; - timing->reg_5 = (e->tRFC << 24 | max(e->tRCDRD, e->tRCDWR) << 16 | - e->tRP); + t->reg[5] = (e->tRFC << 24 | max(e->tRCDRD, e->tRCDWR) << 16 | e->tRP); - timing->reg_8 = (nv_rd32(dev, 0x100240) & 0xffffff00); + t->reg[8] = boot->reg[8] & 0xffffff00; if (P->version == 1) { - timing->reg_1 |= (e->tCL + 2 - (timing->tCWL - 1)); + t->reg[1] |= (e->tCL + 2 - (t->tCWL - 1)); - timing->reg_3 = (0x14 + e->tCL) << 24 | - 0x16 << 16 | - (e->tCL - 1) << 8 | - (e->tCL - 1); + t->reg[3] = (0x14 + e->tCL) << 24 | + 0x16 << 16 | + (e->tCL - 1) << 8 | + (e->tCL - 1); - timing->reg_4 |= (nv_rd32(dev, 0x100230) & 0xffff0000); + t->reg[4] |= boot->reg[4] & 0xffff0000; - timing->reg_6 = (0x33 - timing->tCWL) << 16 | - timing->tCWL << 8 | - (0x2E + e->tCL - timing->tCWL); + t->reg[6] = (0x33 - t->tCWL) << 16 | + t->tCWL << 8 | + (0x2e + e->tCL - t->tCWL); - timing->reg_7 = 0x4000202 | (e->tCL - 1) << 16; + t->reg[7] = 0x4000202 | (e->tCL - 1) << 16; /* XXX: P.version == 1 only has DDR2 and GDDR3? */ if (dev_priv->vram_type == NV_MEM_TYPE_DDR2) { - timing->reg_5 |= (e->tCL + 3) << 8; - timing->reg_6 |= (timing->tCWL - 2) << 8; - timing->reg_8 |= (e->tCL - 4); + t->reg[5] |= (e->tCL + 3) << 8; + t->reg[6] |= (t->tCWL - 2) << 8; + t->reg[8] |= (e->tCL - 4); } else { - timing->reg_5 |= (e->tCL + 2) << 8; - timing->reg_6 |= timing->tCWL << 8; - timing->reg_8 |= (e->tCL - 2); + t->reg[5] |= (e->tCL + 2) << 8; + t->reg[6] |= t->tCWL << 8; + t->reg[8] |= (e->tCL - 2); } } else { - timing->reg_1 |= (5 + e->tCL - (timing->tCWL)); + t->reg[1] |= (5 + e->tCL - (t->tCWL)); /* XXX: 0xb? 0x30? */ - timing->reg_3 = (0x30 + e->tCL) << 24 | - (nv_rd32(dev, 0x10022c) & 0x00ff0000) | - (0xB + e->tCL) << 8 | - (e->tCL - 1); + t->reg[3] = (0x30 + e->tCL) << 24 | + (boot->reg[3] & 0x00ff0000)| + (0xb + e->tCL) << 8 | + (e->tCL - 1); - timing->reg_4 |= (unk20 << 24 | unk21 << 16); + t->reg[4] |= (unk20 << 24 | unk21 << 16); /* XXX: +6? */ - timing->reg_5 |= (timing->tCWL + 6) << 8; + t->reg[5] |= (t->tCWL + 6) << 8; - timing->reg_6 = (0x5A + e->tCL) << 16 | - (6 - e->tCL + timing->tCWL) << 8 | - (0x50 + e->tCL - timing->tCWL); + t->reg[6] = (0x5a + e->tCL) << 16 | + (6 - e->tCL + t->tCWL) << 8 | + (0x50 + e->tCL - t->tCWL); - tmp7_3 = (nv_rd32(dev, 0x10023c) & 0xff000000) >> 24; - timing->reg_7 = (tmp7_3 << 24) | - ((tmp7_3 - 6 + e->tCL) << 16) | - 0x202; + tmp7_3 = (boot->reg[7] & 0xff000000) >> 24; + t->reg[7] = (tmp7_3 << 24) | + ((tmp7_3 - 6 + e->tCL) << 16) | + 0x202; } - NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x %08x\n", timing->id, - timing->reg_0, timing->reg_1, - timing->reg_2, timing->reg_3); + NV_DEBUG(dev, "Entry %d: 220: %08x %08x %08x %08x\n", t->id, + t->reg[0], t->reg[1], t->reg[2], t->reg[3]); NV_DEBUG(dev, " 230: %08x %08x %08x %08x\n", - timing->reg_4, timing->reg_5, - timing->reg_6, timing->reg_7); - NV_DEBUG(dev, " 240: %08x\n", timing->reg_8); + t->reg[4], t->reg[5], t->reg[6], t->reg[7]); + NV_DEBUG(dev, " 240: %08x\n", t->reg[8]); } -void nvc0_mem_timing_entry(struct drm_device *dev, - struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, - struct nouveau_pm_memtiming *timing) +static void +nvc0_mem_timing_entry(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) { - timing->tCWL = e->tCWL; + if (e->tCWL > 0) + t->tCWL = e->tCWL; - timing->reg_0 = (e->tRP << 24 | (e->tRAS & 0x7f) << 17 | - e->tRFC << 8 | e->tRC); + t->reg[0] = (e->tRP << 24 | (e->tRAS & 0x7f) << 17 | + e->tRFC << 8 | e->tRC); - timing->reg_1 = (nv_rd32(dev, 0x10f294) & 0xff000000) | - (e->tRCDWR & 0x0f) << 20 | - (e->tRCDRD & 0x0f) << 14 | - (e->tCWL << 7) | - (e->tCL & 0x0f); + t->reg[1] = (boot->reg[1] & 0xff000000) | + (e->tRCDWR & 0x0f) << 20 | + (e->tRCDRD & 0x0f) << 14 | + (e->tCWL << 7) | + (e->tCL & 0x0f); - timing->reg_2 = (nv_rd32(dev, 0x10f298) & 0xff0000ff) | - e->tWR << 16 | e->tWTR << 8; + t->reg[2] = (boot->reg[2] & 0xff0000ff) | + e->tWR << 16 | e->tWTR << 8; - timing->reg_3 = (e->tUNK_20&0xf) << 9 | - (e->tUNK_21 & 0xf) << 5 | - (e->tUNK_13 & 0x1f); + t->reg[3] = (e->tUNK_20 & 0xf) << 9 | + (e->tUNK_21 & 0xf) << 5 | + (e->tUNK_13 & 0x1f); - timing->reg_4 = (nv_rd32(dev, 0x10f2a0) & 0xfff00fff) | - (e->tRRD&0x1f) << 15; + t->reg[4] = (boot->reg[4] & 0xfff00fff) | + (e->tRRD&0x1f) << 15; - NV_DEBUG(dev, "Entry %d: 290: %08x %08x %08x %08x\n", timing->id, - timing->reg_0, timing->reg_1, - timing->reg_2, timing->reg_3); - NV_DEBUG(dev, " 2a0: %08x\n", - timing->reg_4); + NV_DEBUG(dev, "Entry %d: 290: %08x %08x %08x %08x\n", t->id, + t->reg[0], t->reg[1], t->reg[2], t->reg[3]); + NV_DEBUG(dev, " 2a0: %08x\n", t->reg[4]); } -void -nouveau_mem_features_entry(uint8_t p_version, struct nouveau_pm_tbl_header *hdr, - struct nouveau_pm_tbl_entry *e, - struct nouveau_pm_memtiming *timing) +/** + * MR generation methods + */ + +static bool +nouveau_mem_ddr2_mr(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) +{ + t->drive_strength = 0; + if (hdr->entry_len < 15) { + t->odt = boot->odt; + } else { + t->odt = e->RAM_FT1 & 0x07; + } + + if (e->tCL >= NV_MEM_CL_DDR2_MAX) { + NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL); + return false; + } + + if (e->tWR >= NV_MEM_WR_DDR2_MAX) { + NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR); + return false; + } + + if (t->odt > 3) { + NV_WARN(dev, "(%u) Invalid odt value, assuming disabled: %x", + t->id, t->odt); + t->odt = 0; + } + + t->mr[0] = (boot->mr[0] & 0x100f) | + (e->tCL) << 4 | + (e->tWR - 1) << 9; + t->mr[1] = (boot->mr[1] & 0x101fbb) | + (t->odt & 0x1) << 2 | + (t->odt & 0x2) << 5; + + NV_DEBUG(dev, "(%u) MR: %08x", t->id, t->mr[0]); + return true; +} + +uint8_t nv_mem_wr_lut_ddr3[NV_MEM_WR_DDR3_MAX] = { + 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 5, 6, 6, 7, 7, 0, 0}; + +static bool +nouveau_mem_ddr3_mr(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) +{ + u8 cl = e->tCL - 4; + + t->drive_strength = 0; + if (hdr->entry_len < 15) { + t->odt = boot->odt; + } else { + t->odt = e->RAM_FT1 & 0x07; + } + + if (e->tCL >= NV_MEM_CL_DDR3_MAX || e->tCL < 4) { + NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL); + return false; + } + + if (e->tWR >= NV_MEM_WR_DDR3_MAX || e->tWR < 4) { + NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR); + return false; + } + + if (e->tCWL < 5) { + NV_WARN(dev, "(%u) Invalid tCWL: %u", t->id, e->tCWL); + return false; + } + + t->mr[0] = (boot->mr[0] & 0x180b) | + /* CAS */ + (cl & 0x7) << 4 | + (cl & 0x8) >> 1 | + (nv_mem_wr_lut_ddr3[e->tWR]) << 9; + t->mr[1] = (boot->mr[1] & 0x101dbb) | + (t->odt & 0x1) << 2 | + (t->odt & 0x2) << 5 | + (t->odt & 0x4) << 7; + t->mr[2] = (boot->mr[2] & 0x20ffb7) | (e->tCWL - 5) << 3; + + NV_DEBUG(dev, "(%u) MR: %08x %08x", t->id, t->mr[0], t->mr[2]); + return true; +} + +uint8_t nv_mem_cl_lut_gddr3[NV_MEM_CL_GDDR3_MAX] = { + 0, 0, 0, 0, 4, 5, 6, 7, 0, 1, 2, 3, 8, 9, 10, 11}; +uint8_t nv_mem_wr_lut_gddr3[NV_MEM_WR_GDDR3_MAX] = { + 0, 0, 0, 0, 0, 2, 3, 8, 9, 10, 11, 0, 0, 1, 1, 0, 3}; + +static bool +nouveau_mem_gddr3_mr(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) { - if (p_version == 1) { - /* XXX: Todo */ - } else if (p_version == 2) { - timing->odt = e->RAM_FT1 & 0x1; - timing->dll_disable = (e->RAM_FT1 & 0x2) >> 1; - timing->ron_pull = (e->RAM_FT1 & 0x4) >> 2; + if (hdr->entry_len < 15) { + t->drive_strength = boot->drive_strength; + t->odt = boot->odt; + } else { + t->drive_strength = (e->RAM_FT1 & 0x30) >> 4; + t->odt = e->RAM_FT1 & 0x07; } + + if (e->tCL >= NV_MEM_CL_GDDR3_MAX) { + NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL); + return false; + } + + if (e->tWR >= NV_MEM_WR_GDDR3_MAX) { + NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR); + return false; + } + + if (t->odt > 3) { + NV_WARN(dev, "(%u) Invalid odt value, assuming autocal: %x", + t->id, t->odt); + t->odt = 0; + } + + t->mr[0] = (boot->mr[0] & 0xe0b) | + /* CAS */ + ((nv_mem_cl_lut_gddr3[e->tCL] & 0x7) << 4) | + ((nv_mem_cl_lut_gddr3[e->tCL] & 0x8) >> 2); + t->mr[1] = (boot->mr[1] & 0x100f40) | t->drive_strength | + (t->odt << 2) | + (nv_mem_wr_lut_gddr3[e->tWR] & 0xf) << 4; + + NV_DEBUG(dev, "(%u) MR: %08x %08x", t->id, t->mr[0], t->mr[1]); + return true; +} + +static bool +nouveau_mem_gddr5_mr(struct drm_device *dev, struct nouveau_pm_tbl_header *hdr, + struct nouveau_pm_tbl_entry *e, + struct nouveau_pm_memtiming *t, + struct nouveau_pm_memtiming *boot) +{ + if (hdr->entry_len < 15) { + t->drive_strength = boot->drive_strength; + t->odt = boot->odt; + } else { + t->drive_strength = (e->RAM_FT1 & 0x30) >> 4; + t->odt = e->RAM_FT1 & 0x03; + } + + if (e->tCL >= NV_MEM_CL_GDDR5_MAX) { + NV_WARN(dev, "(%u) Invalid tCL: %u", t->id, e->tCL); + return false; + } + + if (e->tWR >= NV_MEM_WR_GDDR5_MAX) { + NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR); + return false; + } + + if (t->odt > 3) { + NV_WARN(dev, "(%u) Invalid odt value, assuming autocal: %x", + t->id, t->odt); + t->odt = 0; + } + + t->mr[0] = (boot->mr[0] & 0x007) | + ((e->tCL - 5) << 3) | + ((e->tWR - 4) << 8); + t->mr[1] = (boot->mr[1] & 0x1007f0) | + t->drive_strength | + (t->odt << 2); + + NV_DEBUG(dev, "(%u) MR: %08x %08x", t->id, t->mr[0], t->mr[1]); + return true; +} + +static void +nouveau_mem_copy_current_timings(struct drm_device *dev, + struct nouveau_pm_memtiming *t) +{ + struct drm_nouveau_private *dev_priv = dev->dev_private; + u32 timing_base, timing_regs, mr_base; + int i; + + if (dev_priv->card_type >= 0xC0) { + timing_base = 0x10f290; + mr_base = 0x10f300; + } else { + timing_base = 0x100220; + mr_base = 0x1002c0; + } + + t->id = -1; + + switch (dev_priv->card_type) { + case NV_50: + timing_regs = 9; + break; + case NV_C0: + case NV_D0: + timing_regs = 5; + break; + case NV_30: + case NV_40: + timing_regs = 3; + break; + default: + timing_regs = 0; + return; + } + for(i = 0; i < timing_regs; i++) + t->reg[i] = nv_rd32(dev, timing_base + (0x04 * i)); + + t->tCWL = 0; + if (dev_priv->card_type < NV_C0) { + t->tCWL = ((nv_rd32(dev, 0x100228) & 0x0f000000) >> 24) + 1; + } + + t->mr[0] = nv_rd32(dev, mr_base); + t->mr[1] = nv_rd32(dev, mr_base + 0x04); + t->mr[2] = nv_rd32(dev, mr_base + 0x20); + t->mr[3] = nv_rd32(dev, mr_base + 0x24); + + t->odt = 0; + t->drive_strength = 0; + + switch (dev_priv->vram_type) { + case NV_MEM_TYPE_DDR3: + t->odt |= (t->mr[1] & 0x200) >> 7; + case NV_MEM_TYPE_DDR2: + t->odt |= (t->mr[1] & 0x04) >> 2 | + (t->mr[1] & 0x40) >> 5; + break; + case NV_MEM_TYPE_GDDR3: + case NV_MEM_TYPE_GDDR5: + t->drive_strength = t->mr[1] & 0x03; + t->odt = (t->mr[1] & 0x0c) >> 2; + break; + default: + break; + } +} + +static bool +nouveau_mem_compare_timings(struct drm_device *dev, + struct nouveau_pm_memtiming *t1, + struct nouveau_pm_memtiming *t2) +{ + struct drm_nouveau_private *dev_priv = dev->dev_private; + + switch (dev_priv->card_type) { + case 0x50: + if (t1->reg[8] != t2->reg[8] || + t1->reg[7] != t2->reg[7] || + t1->reg[6] != t2->reg[6] || + t1->reg[5] != t2->reg[5]) + return false; + case 0xC0: + if (t1->reg[4] != t2->reg[4] || + t1->reg[3] != t2->reg[3]) + return false; + case 0x40: + if (t1->reg[2] != t2->reg[2] || + t1->reg[1] != t2->reg[1] || + t1->reg[0] != t2->reg[0]) + return false; + break; + default: + return false; + } + + /* RSpliet: may generate many false negatives */ + switch (dev_priv->vram_type) { + case NV_MEM_TYPE_GDDR3: + case NV_MEM_TYPE_GDDR5: + if (t1->mr[0] == t2->mr[0] || + t1->mr[1] != t2->mr[1]) + return true; + break; + case NV_MEM_TYPE_DDR3: + if (t1->mr[2] == t2->mr[2]) + return true; + case NV_MEM_TYPE_DDR2: + if (t1->mr[0] == t2->mr[0]) + return true; + break; + default: + return false; + } + + return false; } /** @@ -665,10 +938,24 @@ nouveau_mem_timing_init(struct drm_device *dev) struct nvbios *bios = &dev_priv->vbios; struct bit_entry P; struct nouveau_pm_tbl_header *hdr = NULL; - uint8_t tCWL; + bool valid_generation = false; u8 *entry; int i; + memtimings->nr_timing = 0; + memtimings->nr_timing_valid = 0; + memtimings->supported = 0; + + if (dev_priv->card_type < NV_40) { + NV_ERROR(dev, "Timing entry format unknown for card_type %x. " + "please contact nouveau developers", + dev_priv->card_type); + return; + } + + /* Copy the current timings */ + nouveau_mem_copy_current_timings(dev, &memtimings->boot); + if (bios->type == NVBIOS_BIT) { if (bit_table(dev, 'P', &P)) return; @@ -710,12 +997,6 @@ nouveau_mem_timing_init(struct drm_device *dev) if (!memtimings->timing) return; - /* Get tCWL from the timing reg for NV_40 and NV_50 - * Used in calculations later... source unknown */ - tCWL = 0; - if (dev_priv->card_type < NV_C0) - tCWL = ((nv_rd32(dev, 0x100228) & 0x0f000000) >> 24) + 1; - entry = (u8 *) hdr + hdr->header_len; for (i = 0; i < hdr->entry_cnt; i++, entry += hdr->entry_len) { struct nouveau_pm_memtiming *timing = &pm->memtimings.timing[i]; @@ -723,29 +1004,70 @@ nouveau_mem_timing_init(struct drm_device *dev) (struct nouveau_pm_tbl_entry *) entry; if (entry[0] == 0) continue; + memtimings->nr_timing_valid++; timing->id = i; - timing->WR = entry[0]; - timing->CL = entry[2]; - timing->tCWL = tCWL; - - nouveau_mem_features_entry(P.version, hdr, entry_struct, - &pm->memtimings.timing[i]); + timing->tCWL = memtimings->boot.tCWL; - if (dev_priv->card_type <= NV_40) { + /* generate the timngs */ + if (dev_priv->card_type == NV_40) { nv40_mem_timing_entry(dev, hdr, entry_struct, - &pm->memtimings.timing[i]); + &pm->memtimings.timing[i], + &memtimings->boot); } else if (dev_priv->card_type == NV_50) { nv50_mem_timing_entry(dev, &P, hdr, entry_struct, - &pm->memtimings.timing[i]); + &pm->memtimings.timing[i], + &memtimings->boot); } else if (dev_priv->card_type == NV_C0) { nvc0_mem_timing_entry(dev, hdr, entry_struct, - &pm->memtimings.timing[i]); + &pm->memtimings.timing[i], + &memtimings->boot); + } + + /* generate the MR/EMR/... */ + switch (dev_priv->vram_type) { + case NV_MEM_TYPE_GDDR3: + nouveau_mem_gddr3_mr(dev, hdr, entry_struct, timing, + &memtimings->boot); + break; + case NV_MEM_TYPE_GDDR5: + nouveau_mem_gddr5_mr(dev, hdr, entry_struct, timing, + &memtimings->boot); + break; + case NV_MEM_TYPE_DDR2: + nouveau_mem_ddr2_mr(dev, hdr, entry_struct, timing, + &memtimings->boot); + break; + case NV_MEM_TYPE_DDR3: + nouveau_mem_ddr3_mr(dev, hdr, entry_struct, timing, + &memtimings->boot); + break; + default: + valid_generation = false; + break; + } + + /* some kind of validation */ + if (nouveau_mem_compare_timings(dev, timing, + &memtimings->boot)) { + NV_DEBUG(dev, "Copy boot timings from entry %d\n", + timing->id); + memtimings->boot = *timing; + valid_generation = true; } } memtimings->nr_timing = hdr->entry_cnt; - memtimings->supported = (P.version == 1); + memtimings->supported = (P.version == 1) && valid_generation; + + /* if there are no timing entries that cannot + * re-generate the current timings + */ + if (memtimings->nr_timing_valid > 0 && !valid_generation) { + NV_INFO(dev, + "Memory timings management may not be working." + " please report to nouveau devs\n"); + } } void diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.c b/drivers/gpu/drm/nouveau/nouveau_pm.c index 6e3851e1a1d7..8a742596d095 100644 --- a/drivers/gpu/drm/nouveau/nouveau_pm.c +++ b/drivers/gpu/drm/nouveau/nouveau_pm.c @@ -808,6 +808,7 @@ nouveau_pm_init(struct drm_device *dev) ret = nouveau_pm_perflvl_get(dev, &pm->boot); if (ret == 0) { strncpy(pm->boot.name, "boot", 4); + pm->boot.timing = &pm->memtimings.boot; pm->cur = &pm->boot; nouveau_pm_perflvl_info(&pm->boot, info, sizeof(info)); diff --git a/drivers/gpu/drm/nouveau/nv50_vram.c b/drivers/gpu/drm/nouveau/nv50_vram.c index 1467efc6d61d..a52b9a51e010 100644 --- a/drivers/gpu/drm/nouveau/nv50_vram.c +++ b/drivers/gpu/drm/nouveau/nv50_vram.c @@ -207,6 +207,7 @@ nv50_vram_init(struct drm_device *dev) break; } + dev_priv->vram_rank_B = (nv_rd32(dev, NV04_PFB_CFG0) & 0x100) >> 8; dev_priv->vram_size = nv_rd32(dev, 0x10020c); dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32; dev_priv->vram_size &= 0xffffffff00ULL; -- cgit v1.2.1