Merge branch 'perf/core' into perf/uprobes

Merge in latest upstream (and the latest perf development tree),
to prepare for tooling changes, and also to pick up v3.4 MM
changes that the uprobes code needs to take care of.

Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 587 insertions, 222 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index c3a5745e9c79..b08065f981df 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -26,7 +26,8 @@
  * DEALINGS IN THE SOFTWARE.
  *
  * Authors:
- *    Keith Whitwell <keith@tungstengraphics.com>
+ *    Ben Skeggs <bskeggs@redhat.com>
+ *    Roy Spliet <r.spliet@student.tudelft.nl>
  */
 
 
@@ -192,75 +193,6 @@ nouveau_mem_gart_fini(struct drm_device *dev)
 	}
 }
 
-static uint32_t
-nouveau_mem_detect_nv04(struct drm_device *dev)
-{
-	uint32_t boot0 = nv_rd32(dev, NV04_PFB_BOOT_0);
-
-	if (boot0 & 0x00000100)
-		return (((boot0 >> 12) & 0xf) * 2 + 2) * 1024 * 1024;
-
-	switch (boot0 & NV04_PFB_BOOT_0_RAM_AMOUNT) {
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_32MB:
-		return 32 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_16MB:
-		return 16 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_8MB:
-		return 8 * 1024 * 1024;
-	case NV04_PFB_BOOT_0_RAM_AMOUNT_4MB:
-		return 4 * 1024 * 1024;
-	}
-
-	return 0;
-}
-
-static uint32_t
-nouveau_mem_detect_nforce(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct pci_dev *bridge;
-	uint32_t mem;
-
-	bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 1));
-	if (!bridge) {
-		NV_ERROR(dev, "no bridge device\n");
-		return 0;
-	}
-
-	if (dev_priv->flags & NV_NFORCE) {
-		pci_read_config_dword(bridge, 0x7C, &mem);
-		return (uint64_t)(((mem >> 6) & 31) + 1)*1024*1024;
-	} else
-	if (dev_priv->flags & NV_NFORCE2) {
-		pci_read_config_dword(bridge, 0x84, &mem);
-		return (uint64_t)(((mem >> 4) & 127) + 1)*1024*1024;
-	}
-
-	NV_ERROR(dev, "impossible!\n");
-	return 0;
-}
-
-int
-nouveau_mem_detect(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->card_type == NV_04) {
-		dev_priv->vram_size = nouveau_mem_detect_nv04(dev);
-	} else
-	if (dev_priv->flags & (NV_NFORCE | NV_NFORCE2)) {
-		dev_priv->vram_size = nouveau_mem_detect_nforce(dev);
-	} else
-	if (dev_priv->card_type < NV_50) {
-		dev_priv->vram_size  = nv_rd32(dev, NV04_PFB_FIFO_DATA);
-		dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
-	}
-
-	if (dev_priv->vram_size)
-		return 0;
-	return -ENOMEM;
-}
-
 bool
 nouveau_mem_flags_valid(struct drm_device *dev, u32 tile_flags)
 {
@@ -385,11 +317,29 @@ nouveau_mem_init_agp(struct drm_device *dev)
 	return 0;
 }
 
+static const struct vram_types {
+	int value;
+	const char *name;
+} vram_type_map[] = {
+	{ NV_MEM_TYPE_STOLEN , "stolen system memory" },
+	{ NV_MEM_TYPE_SGRAM  , "SGRAM" },
+	{ NV_MEM_TYPE_SDRAM  , "SDRAM" },
+	{ NV_MEM_TYPE_DDR1   , "DDR1" },
+	{ NV_MEM_TYPE_DDR2   , "DDR2" },
+	{ NV_MEM_TYPE_DDR3   , "DDR3" },
+	{ NV_MEM_TYPE_GDDR2  , "GDDR2" },
+	{ NV_MEM_TYPE_GDDR3  , "GDDR3" },
+	{ NV_MEM_TYPE_GDDR4  , "GDDR4" },
+	{ NV_MEM_TYPE_GDDR5  , "GDDR5" },
+	{ NV_MEM_TYPE_UNKNOWN, "unknown type" }
+};
+
 int
 nouveau_mem_vram_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
+	const struct vram_types *vram_type;
 	int ret, dma_bits;
 
 	dma_bits = 32;
@@ -427,7 +377,21 @@ nouveau_mem_vram_init(struct drm_device *dev)
 		return ret;
 	}
 
-	NV_INFO(dev, "Detected %dMiB VRAM\n", (int)(dev_priv->vram_size >> 20));
+	vram_type = vram_type_map;
+	while (vram_type->value != NV_MEM_TYPE_UNKNOWN) {
+		if (nouveau_vram_type) {
+			if (!strcasecmp(nouveau_vram_type, vram_type->name))
+				break;
+			dev_priv->vram_type = vram_type->value;
+		} else {
+			if (vram_type->value == dev_priv->vram_type)
+				break;
+		}
+		vram_type++;
+	}
+
+	NV_INFO(dev, "Detected %dMiB VRAM (%s)\n",
+		(int)(dev_priv->vram_size >> 20), vram_type->name);
 	if (dev_priv->vram_sys_base) {
 		NV_INFO(dev, "Stolen system memory at: 0x%010llx\n",
 			dev_priv->vram_sys_base);
@@ -508,216 +472,617 @@ 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, uint8_t magic_number,
-							struct nouveau_pm_memtiming *timing) {
-
-	timing->reg_0 = (e->tRC << 24 | e->tRFC << 16 | e->tRAS << 8 | e->tRP);
+static int
+nv40_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	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 + magic_number) << 24 |
-				  1 << 16 |
-				  (e->tUNK_1 + 2 + magic_number) << 8 |
-				  (e->tCL + 2 - magic_number);
-	timing->reg_2 = (magic_number << 24 | e->tUNK_12 << 16 | e->tUNK_11 << 8 | e->tUNK_10);
-	timing->reg_2 |= 0x20200000;
-
-	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]);
+	return 0;
 }
 
-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, uint8_t magic_number,struct nouveau_pm_memtiming *timing) {
+static int
+nv50_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct bit_entry P;
+	uint8_t unk18 = 1, unk20 = 0, unk21 = 0, tmp7_3;
 
-	uint8_t unk18 = 1,
-		unk19 = 1,
-		unk20 = 0,
-		unk21 = 0;
+	if (bit_table(dev, 'P', &P))
+		return -EINVAL;
 
-	switch (min(hdr->entry_len, (u8) 22)) {
+	switch (min(len, (u8) 22)) {
 	case 22:
 		unk21 = e->tUNK_21;
 	case 21:
 		unk20 = e->tUNK_20;
 	case 20:
-		unk19 = e->tUNK_19;
+		if (e->tCWL > 0)
+			t->tCWL = e->tCWL;
 	case 19:
 		unk18 = e->tUNK_18;
 		break;
 	}
 
-	timing->reg_0 = (e->tRC << 24 | e->tRFC << 16 | e->tRAS << 8 | e->tRP);
+	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 + unk19 + 1 + magic_number) << 24 |
-				  max(unk18, (u8) 1) << 16 |
-				  (e->tUNK_1 + unk19 + 1 + magic_number) << 8;
-	if (dev_priv->chipset == 0xa8) {
-		timing->reg_1 |= (e->tCL - 1);
-	} else {
-		timing->reg_1 |= (e->tCL + 2 - magic_number);
-	}
-	timing->reg_2 = (e->tUNK_12 << 16 | e->tUNK_11 << 8 | e->tUNK_10);
-
-	timing->reg_5 = (e->tRAS << 24 | e->tRC);
-	timing->reg_5 += max(e->tUNK_10, e->tUNK_11) << 16;
-
-	if (P->version == 1) {
-		timing->reg_2 |= magic_number << 24;
-		timing->reg_3 = (0x14 + e->tCL) << 24 |
-						0x16 << 16 |
-						(e->tCL - 1) << 8 |
-						(e->tCL - 1);
-		timing->reg_4 = (nv_rd32(dev,0x10022c) & 0xffff0000) | e->tUNK_13 << 8  | e->tUNK_13;
-		timing->reg_5 |= (e->tCL + 2) << 8;
-		timing->reg_7 = 0x4000202 | (e->tCL - 1) << 16;
+	t->reg[1] = (e->tWR + 2 + (t->tCWL - 1)) << 24 |
+				max(unk18, (u8) 1) << 16 |
+				(e->tWTR + 2 + (t->tCWL - 1)) << 8;
+
+	t->reg[2] = ((t->tCWL - 1) << 24 |
+		    e->tRRD << 16 |
+		    e->tRCDWR << 8 |
+		    e->tRCDRD);
+
+	t->reg[4] = e->tUNK_13 << 8  | e->tUNK_13;
+
+	t->reg[5] = (e->tRFC << 24 | max(e->tRCDRD, e->tRCDWR) << 16 | e->tRP);
+
+	t->reg[8] = boot->reg[8] & 0xffffff00;
+
+	if (P.version == 1) {
+		t->reg[1] |= (e->tCL + 2 - (t->tCWL - 1));
+
+		t->reg[3] = (0x14 + e->tCL) << 24 |
+			    0x16 << 16 |
+			    (e->tCL - 1) << 8 |
+			    (e->tCL - 1);
+
+		t->reg[4] |= boot->reg[4] & 0xffff0000;
+
+		t->reg[6] = (0x33 - t->tCWL) << 16 |
+			    t->tCWL << 8 |
+			    (0x2e + e->tCL - t->tCWL);
+
+		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) {
+			t->reg[5] |= (e->tCL + 3) << 8;
+			t->reg[6] |= (t->tCWL - 2) << 8;
+			t->reg[8] |= (e->tCL - 4);
+		} else {
+			t->reg[5] |= (e->tCL + 2) << 8;
+			t->reg[6] |= t->tCWL << 8;
+			t->reg[8] |= (e->tCL - 2);
+		}
 	} else {
-		timing->reg_2 |= (unk19 - 1) << 24;
-		/* XXX: reg_10022c for recentish cards pretty much unknown*/
-		timing->reg_3 = e->tCL - 1;
-		timing->reg_4 = (unk20 << 24 | unk21 << 16 |
-							e->tUNK_13 << 8  | e->tUNK_13);
+		t->reg[1] |= (5 + e->tCL - (t->tCWL));
+
+		/* XXX: 0xb? 0x30? */
+		t->reg[3] = (0x30 + e->tCL) << 24 |
+			    (boot->reg[3] & 0x00ff0000)|
+			    (0xb + e->tCL) << 8 |
+			    (e->tCL - 1);
+
+		t->reg[4] |= (unk20 << 24 | unk21 << 16);
+
 		/* XXX: +6? */
-		timing->reg_5 |= (unk19 + 6) << 8;
+		t->reg[5] |= (t->tCWL + 6) << 8;
 
-		/* XXX: reg_10023c currently unknown
-		 * 10023c seen as 06xxxxxx, 0bxxxxxx or 0fxxxxxx */
-		timing->reg_7 = 0x202;
+		t->reg[6] = (0x5a + e->tCL) << 16 |
+			    (6 - e->tCL + t->tCWL) << 8 |
+			    (0x50 + e->tCL - t->tCWL);
+
+		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);
-}
-
-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) {
-	timing->reg_0 = (e->tRC << 24 | (e->tRFC & 0x7f) << 17 | e->tRAS << 8 | e->tRP);
-	timing->reg_1 = (nv_rd32(dev,0x10f294) & 0xff000000) | (e->tUNK_11&0x0f) << 20 | (e->tUNK_19 << 7) | (e->tCL & 0x0f);
-	timing->reg_2 = (nv_rd32(dev,0x10f298) & 0xff0000ff) | e->tWR << 16 | e->tUNK_1 << 8;
-	timing->reg_3 = e->tUNK_20 << 9 | e->tUNK_13;
-	timing->reg_4 = (nv_rd32(dev,0x10f2a0) & 0xfff000ff) | e->tUNK_12 << 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 %08x %08x %08x\n",
-		 timing->reg_4, timing->reg_5,
-		 timing->reg_6, timing->reg_7);
+		 t->reg[4], t->reg[5], t->reg[6], t->reg[7]);
+	NV_DEBUG(dev, "         240: %08x\n", t->reg[8]);
+	return 0;
+}
+
+static int
+nvc0_mem_timing_calc(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (e->tCWL > 0)
+		t->tCWL = e->tCWL;
+
+	t->reg[0] = (e->tRP << 24 | (e->tRAS & 0x7f) << 17 |
+		     e->tRFC << 8 | e->tRC);
+
+	t->reg[1] = (boot->reg[1] & 0xff000000) |
+		    (e->tRCDWR & 0x0f) << 20 |
+		    (e->tRCDRD & 0x0f) << 14 |
+		    (t->tCWL << 7) |
+		    (e->tCL & 0x0f);
+
+	t->reg[2] = (boot->reg[2] & 0xff0000ff) |
+		    e->tWR << 16 | e->tWTR << 8;
+
+	t->reg[3] = (e->tUNK_20 & 0x1f) << 9 |
+		    (e->tUNK_21 & 0xf) << 5 |
+		    (e->tUNK_13 & 0x1f);
+
+	t->reg[4] = (boot->reg[4] & 0xfff00fff) |
+		    (e->tRRD&0x1f) << 15;
+
+	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]);
+	return 0;
 }
 
 /**
- * Processes the Memory Timing BIOS table, stores generated
- * register values
- * @pre init scripts were run, memtiming regs are initialized
+ * MR generation methods
  */
-void
-nouveau_mem_timing_init(struct drm_device *dev)
+
+static int
+nouveau_mem_ddr2_mr(struct drm_device *dev, u32 freq,
+		    struct nouveau_pm_tbl_entry *e, u8 len,
+		    struct nouveau_pm_memtiming *boot,
+		    struct nouveau_pm_memtiming *t)
+{
+	t->drive_strength = 0;
+	if (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 -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_DDR2_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	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 0;
+}
+
+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 int
+nouveau_mem_ddr3_mr(struct drm_device *dev, u32 freq,
+		    struct nouveau_pm_tbl_entry *e, u8 len,
+		    struct nouveau_pm_memtiming *boot,
+		    struct nouveau_pm_memtiming *t)
+{
+	u8 cl = e->tCL - 4;
+
+	t->drive_strength = 0;
+	if (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 -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_DDR3_MAX || e->tWR < 4) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	if (e->tCWL < 5) {
+		NV_WARN(dev, "(%u) Invalid tCWL: %u", t->id, e->tCWL);
+		return -ERANGE;
+	}
+
+	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 0;
+}
+
+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 int
+nouveau_mem_gddr3_mr(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (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 -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_GDDR3_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	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;
+	t->mr[2] = boot->mr[2];
+
+	NV_DEBUG(dev, "(%u) MR: %08x %08x %08x", t->id,
+		      t->mr[0], t->mr[1], t->mr[2]);
+	return 0;
+}
+
+static int
+nouveau_mem_gddr5_mr(struct drm_device *dev, u32 freq,
+		     struct nouveau_pm_tbl_entry *e, u8 len,
+		     struct nouveau_pm_memtiming *boot,
+		     struct nouveau_pm_memtiming *t)
+{
+	if (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 -ERANGE;
+	}
+
+	if (e->tWR >= NV_MEM_WR_GDDR5_MAX) {
+		NV_WARN(dev, "(%u) Invalid tWR: %u", t->id, e->tWR);
+		return -ERANGE;
+	}
+
+	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 0;
+}
+
+int
+nouveau_mem_timing_calc(struct drm_device *dev, u32 freq,
+			struct nouveau_pm_memtiming *t)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pm_engine *pm = &dev_priv->engine.pm;
-	struct nouveau_pm_memtimings *memtimings = &pm->memtimings;
-	struct nvbios *bios = &dev_priv->vbios;
-	struct bit_entry P;
-	struct nouveau_pm_tbl_header *hdr = NULL;
-	uint8_t magic_number;
-	u8 *entry;
-	int i;
+	struct nouveau_pm_memtiming *boot = &pm->boot.timing;
+	struct nouveau_pm_tbl_entry *e;
+	u8 ver, len, *ptr, *ramcfg;
+	int ret;
+
+	ptr = nouveau_perf_timing(dev, freq, &ver, &len);
+	if (!ptr || ptr[0] == 0x00) {
+		*t = *boot;
+		return 0;
+	}
+	e = (struct nouveau_pm_tbl_entry *)ptr;
+
+	t->tCWL = boot->tCWL;
+
+	switch (dev_priv->card_type) {
+	case NV_40:
+		ret = nv40_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	case NV_50:
+		ret = nv50_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	case NV_C0:
+		ret = nvc0_mem_timing_calc(dev, freq, e, len, boot, t);
+		break;
+	default:
+		ret = -ENODEV;
+		break;
+	}
 
-	if (bios->type == NVBIOS_BIT) {
-		if (bit_table(dev, 'P', &P))
-			return;
+	switch (dev_priv->vram_type * !ret) {
+	case NV_MEM_TYPE_GDDR3:
+		ret = nouveau_mem_gddr3_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_GDDR5:
+		ret = nouveau_mem_gddr5_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_DDR2:
+		ret = nouveau_mem_ddr2_mr(dev, freq, e, len, boot, t);
+		break;
+	case NV_MEM_TYPE_DDR3:
+		ret = nouveau_mem_ddr3_mr(dev, freq, e, len, boot, t);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	ramcfg = nouveau_perf_ramcfg(dev, freq, &ver, &len);
+	if (ramcfg) {
+		int dll_off;
 
-		if (P.version == 1)
-			hdr = (struct nouveau_pm_tbl_header *) ROMPTR(dev, P.data[4]);
+		if (ver == 0x00)
+			dll_off = !!(ramcfg[3] & 0x04);
 		else
-		if (P.version == 2)
-			hdr = (struct nouveau_pm_tbl_header *) ROMPTR(dev, P.data[8]);
-		else {
-			NV_WARN(dev, "unknown mem for BIT P %d\n", P.version);
+			dll_off = !!(ramcfg[2] & 0x40);
+
+		switch (dev_priv->vram_type) {
+		case NV_MEM_TYPE_GDDR3:
+			t->mr[1] &= ~0x00000040;
+			t->mr[1] |=  0x00000040 * dll_off;
+			break;
+		default:
+			t->mr[1] &= ~0x00000001;
+			t->mr[1] |=  0x00000001 * dll_off;
+			break;
 		}
+	}
+
+	return ret;
+}
+
+void
+nouveau_mem_timing_read(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 {
-		NV_DEBUG(dev, "BMP version too old for memory\n");
-		return;
+		timing_base = 0x100220;
+		mr_base = 0x1002c0;
 	}
 
-	if (!hdr) {
-		NV_DEBUG(dev, "memory timing table pointer invalid\n");
+	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;
+	} else if (dev_priv->card_type <= NV_D0) {
+		t->tCWL = ((nv_rd32(dev, 0x10f294) & 0x00000f80) >> 7);
+	}
 
-	if (hdr->version != 0x10) {
-		NV_WARN(dev, "memory timing table 0x%02x unknown\n", hdr->version);
-		return;
+	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;
 	}
+}
 
-	/* validate record length */
-	if (hdr->entry_len < 15) {
-		NV_ERROR(dev, "mem timing table length unknown: %d\n", hdr->entry_len);
-		return;
+int
+nouveau_mem_exec(struct nouveau_mem_exec_func *exec,
+		 struct nouveau_pm_level *perflvl)
+{
+	struct drm_nouveau_private *dev_priv = exec->dev->dev_private;
+	struct nouveau_pm_memtiming *info = &perflvl->timing;
+	u32 tMRD = 1000, tCKSRE = 0, tCKSRX = 0, tXS = 0, tDLLK = 0;
+	u32 mr[3] = { info->mr[0], info->mr[1], info->mr[2] };
+	u32 mr1_dlloff;
+
+	switch (dev_priv->vram_type) {
+	case NV_MEM_TYPE_DDR2:
+		tDLLK = 2000;
+		mr1_dlloff = 0x00000001;
+		break;
+	case NV_MEM_TYPE_DDR3:
+		tDLLK = 12000;
+		mr1_dlloff = 0x00000001;
+		break;
+	case NV_MEM_TYPE_GDDR3:
+		tDLLK = 40000;
+		mr1_dlloff = 0x00000040;
+		break;
+	default:
+		NV_ERROR(exec->dev, "cannot reclock unsupported memtype\n");
+		return -ENODEV;
 	}
 
-	/* parse vbios entries into common format */
-	memtimings->timing =
-		kcalloc(hdr->entry_cnt, sizeof(*memtimings->timing), GFP_KERNEL);
-	if (!memtimings->timing)
-		return;
+	/* fetch current MRs */
+	switch (dev_priv->vram_type) {
+	case NV_MEM_TYPE_GDDR3:
+	case NV_MEM_TYPE_DDR3:
+		mr[2] = exec->mrg(exec, 2);
+	default:
+		mr[1] = exec->mrg(exec, 1);
+		mr[0] = exec->mrg(exec, 0);
+		break;
+	}
 
-	/* Get "some number" from the timing reg for NV_40 and NV_50
-	 * Used in calculations later... source unknown */
-	magic_number = 0;
-	if (P.version == 1) {
-		magic_number = (nv_rd32(dev, 0x100228) & 0x0f000000) >> 24;
+	/* DLL 'on' -> DLL 'off' mode, disable before entering self-refresh  */
+	if (!(mr[1] & mr1_dlloff) && (info->mr[1] & mr1_dlloff)) {
+		exec->precharge(exec);
+		exec->mrs (exec, 1, mr[1] | mr1_dlloff);
+		exec->wait(exec, tMRD);
 	}
 
-	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];
-		if (entry[0] == 0)
-			continue;
+	/* enter self-refresh mode */
+	exec->precharge(exec);
+	exec->refresh(exec);
+	exec->refresh(exec);
+	exec->refresh_auto(exec, false);
+	exec->refresh_self(exec, true);
+	exec->wait(exec, tCKSRE);
+
+	/* modify input clock frequency */
+	exec->clock_set(exec);
+
+	/* exit self-refresh mode */
+	exec->wait(exec, tCKSRX);
+	exec->precharge(exec);
+	exec->refresh_self(exec, false);
+	exec->refresh_auto(exec, true);
+	exec->wait(exec, tXS);
+
+	/* update MRs */
+	if (mr[2] != info->mr[2]) {
+		exec->mrs (exec, 2, info->mr[2]);
+		exec->wait(exec, tMRD);
+	}
+
+	if (mr[1] != info->mr[1]) {
+		/* need to keep DLL off until later, at least on GDDR3 */
+		exec->mrs (exec, 1, info->mr[1] | (mr[1] & mr1_dlloff));
+		exec->wait(exec, tMRD);
+	}
+
+	if (mr[0] != info->mr[0]) {
+		exec->mrs (exec, 0, info->mr[0]);
+		exec->wait(exec, tMRD);
+	}
 
-		timing->id = i;
-		timing->WR = entry[0];
-		timing->CL = entry[2];
+	/* update PFB timing registers */
+	exec->timing_set(exec);
 
-		if(dev_priv->card_type <= NV_40) {
-			nv40_mem_timing_entry(dev,hdr,(struct nouveau_pm_tbl_entry*) entry,magic_number,&pm->memtimings.timing[i]);
-		} else if(dev_priv->card_type == NV_50){
-			nv50_mem_timing_entry(dev,&P,hdr,(struct nouveau_pm_tbl_entry*) entry,magic_number,&pm->memtimings.timing[i]);
-		} else if(dev_priv->card_type == NV_C0) {
-			nvc0_mem_timing_entry(dev,hdr,(struct nouveau_pm_tbl_entry*) entry,&pm->memtimings.timing[i]);
+	/* DLL (enable + ) reset */
+	if (!(info->mr[1] & mr1_dlloff)) {
+		if (mr[1] & mr1_dlloff) {
+			exec->mrs (exec, 1, info->mr[1]);
+			exec->wait(exec, tMRD);
 		}
+		exec->mrs (exec, 0, info->mr[0] | 0x00000100);
+		exec->wait(exec, tMRD);
+		exec->mrs (exec, 0, info->mr[0] | 0x00000000);
+		exec->wait(exec, tMRD);
+		exec->wait(exec, tDLLK);
+		if (dev_priv->vram_type == NV_MEM_TYPE_GDDR3)
+			exec->precharge(exec);
 	}
 
-	memtimings->nr_timing = hdr->entry_cnt;
-	memtimings->supported = P.version == 1;
+	return 0;
 }
 
-void
-nouveau_mem_timing_fini(struct drm_device *dev)
+int
+nouveau_mem_vbios_type(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pm_memtimings *mem = &dev_priv->engine.pm.memtimings;
+	struct bit_entry M;
+	u8 ramcfg = (nv_rd32(dev, 0x101000) & 0x0000003c) >> 2;
+	if (!bit_table(dev, 'M', &M) || M.version != 2 || M.length < 5) {
+		u8 *table = ROMPTR(dev, M.data[3]);
+		if (table && table[0] == 0x10 && ramcfg < table[3]) {
+			u8 *entry = table + table[1] + (ramcfg * table[2]);
+			switch (entry[0] & 0x0f) {
+			case 0: return NV_MEM_TYPE_DDR2;
+			case 1: return NV_MEM_TYPE_DDR3;
+			case 2: return NV_MEM_TYPE_GDDR3;
+			case 3: return NV_MEM_TYPE_GDDR5;
+			default:
+				break;
+			}
 
-	if(mem->timing) {
-		kfree(mem->timing);
-		mem->timing = NULL;
+		}
 	}
+	return NV_MEM_TYPE_UNKNOWN;
 }
 
 static int