Merge remote branch 'nouveau/drm-nouveau-next' of ../drm-nouveau-next into drm-core-next

* 'nouveau/drm-nouveau-next' of ../drm-nouveau-next:
  drm/nouveau: fix hwmon device binding
  drm/nouveau: create grctx on the fly on all chipsets
  drm/nvc0: fix init without firmware present
  drm/nvc0/pgraph: fix 0x406028/0x405870 init
  drm/nvc0/pgraph: more unit names
  drm/nvc0/pfifo: support for chipsets with only one PSUBFIFO (0xc1)
  drm/nvc0: reserve only subc 0 for kernel use
  drm/nv50: sync up gr data error names with rnn, use for nvc0 also
  drm/nvc0: parse a couple more PGRAPH_INTR
  drm/nvc0: nuke left-over debug messages
  drm/nvc0: kill off a couple more magics
  drm/nouveau: Validate channel indices passed from userspace.
  drm/nouveau: Only select ACPI_VIDEO if its dependencies are met

14 files changed, 143 insertions, 79 deletions

diff --git a/drivers/gpu/drm/nouveau/Kconfig b/drivers/gpu/drm/nouveau/Kconfig
index 72730e9..21d6c29 100644
--- a/drivers/gpu/drm/nouveau/Kconfig
+++ b/drivers/gpu/drm/nouveau/Kconfig
@@ -10,7 +10,7 @@ config DRM_NOUVEAU
 	select FB
 	select FRAMEBUFFER_CONSOLE if !EMBEDDED
 	select FB_BACKLIGHT if DRM_NOUVEAU_BACKLIGHT
-	select ACPI_VIDEO if ACPI
+	select ACPI_VIDEO if ACPI && X86 && BACKLIGHT_CLASS_DEVICE && VIDEO_OUTPUT_CONTROL && INPUT
 	help
 	  Choose this option for open-source nVidia support.
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index e37977d..3960d66 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -121,7 +121,6 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 		      uint32_t vram_handle, uint32_t gart_handle)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_channel *chan;
 	unsigned long flags;
@@ -202,15 +201,6 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	/* disable the fifo caches */
 	pfifo->reassign(dev, false);
 
-	/* Create a graphics context for new channel */
-	if (dev_priv->card_type < NV_50) {
-		ret = pgraph->create_context(chan);
-		if (ret) {
-			nouveau_channel_put(&chan);
-			return ret;
-		}
-	}
-
 	/* Construct inital RAMFC for new channel */
 	ret = pfifo->create_context(chan);
 	if (ret) {
@@ -253,6 +243,9 @@ nouveau_channel_get(struct drm_device *dev, struct drm_file *file_priv, int id)
 	struct nouveau_channel *chan;
 	unsigned long flags;
 
+	if (unlikely(id < 0 || id >= NOUVEAU_MAX_CHANNEL_NR))
+		return ERR_PTR(-EINVAL);
+
 	spin_lock_irqsave(&dev_priv->channels.lock, flags);
 	chan = nouveau_channel_get_unlocked(dev_priv->channels.ptr[id]);
 	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
@@ -443,14 +436,20 @@ nouveau_ioctl_fifo_alloc(struct drm_device *dev, void *data,
 	else
 		init->pushbuf_domains = NOUVEAU_GEM_DOMAIN_GART;
 
-	init->subchan[0].handle = NvM2MF;
-	if (dev_priv->card_type < NV_50)
-		init->subchan[0].grclass = 0x0039;
-	else
-		init->subchan[0].grclass = 0x5039;
-	init->subchan[1].handle = NvSw;
-	init->subchan[1].grclass = NV_SW;
-	init->nr_subchan = 2;
+	if (dev_priv->card_type < NV_C0) {
+		init->subchan[0].handle = NvM2MF;
+		if (dev_priv->card_type < NV_50)
+			init->subchan[0].grclass = 0x0039;
+		else
+			init->subchan[0].grclass = 0x5039;
+		init->subchan[1].handle = NvSw;
+		init->subchan[1].grclass = NV_SW;
+		init->nr_subchan = 2;
+	} else {
+		init->subchan[0].handle  = 0x9039;
+		init->subchan[0].grclass = 0x9039;
+		init->nr_subchan = 1;
+	}
 
 	/* Named memory object area */
 	ret = drm_gem_handle_create(file_priv, chan->notifier_bo->gem,
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index e59f5bc..46e3257 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -1191,6 +1191,7 @@ extern int  nv50_graph_unload_context(struct drm_device *);
 extern int  nv50_grctx_init(struct nouveau_grctx *);
 extern void nv50_graph_tlb_flush(struct drm_device *dev);
 extern void nv86_graph_tlb_flush(struct drm_device *dev);
+extern struct nouveau_enum nv50_data_error_names[];
 
 /* nvc0_graph.c */
 extern int  nvc0_graph_init(struct drm_device *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index 88b2f29..221b846 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -165,7 +165,7 @@ nouveau_fence_emit(struct nouveau_fence *fence)
 		if (dev_priv->card_type < NV_C0)
 			BEGIN_RING(chan, NvSubSw, 0x0050, 1);
 		else
-			BEGIN_NVC0(chan, 2, NvSubSw, 0x0050, 1);
+			BEGIN_NVC0(chan, 2, NvSubM2MF, 0x0050, 1);
 	} else {
 		BEGIN_RING(chan, NvSubSw, 0x0150, 1);
 	}
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index d77b1fc..30b6544 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -651,7 +651,8 @@ found:
 		}
 		break;
 	case NVOBJ_ENGINE_GR:
-		if (dev_priv->card_type >= NV_50 && !chan->ramin_grctx) {
+		if ((dev_priv->card_type >= NV_20 && !chan->ramin_grctx) ||
+		    (dev_priv->card_type  < NV_20 && !chan->pgraph_ctx)) {
 			struct nouveau_pgraph_engine *pgraph =
 				&dev_priv->engine.graph;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.c b/drivers/gpu/drm/nouveau/nouveau_pm.c
index d938141..fb846a3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_pm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_pm.c
@@ -422,8 +422,7 @@ nouveau_hwmon_init(struct drm_device *dev)
 		return ret;
 	}
 	dev_set_drvdata(hwmon_dev, dev);
-	ret = sysfs_create_group(&hwmon_dev->kobj,
-					&hwmon_attrgroup);
+	ret = sysfs_create_group(&dev->pdev->dev.kobj, &hwmon_attrgroup);
 	if (ret) {
 		NV_ERROR(dev,
 			"Unable to create hwmon sysfs file: %d\n", ret);
diff --git a/drivers/gpu/drm/nouveau/nv40_fifo.c b/drivers/gpu/drm/nouveau/nv40_fifo.c
index c86e4d4..49b9a35 100644
--- a/drivers/gpu/drm/nouveau/nv40_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv40_fifo.c
@@ -64,7 +64,6 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 			      NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			      0x30000000 /* no idea.. */);
-	nv_wi32(dev, fc + 56, chan->ramin_grctx->pinst >> 4);
 	nv_wi32(dev, fc + 60, 0x0001FFFF);
 
 	/* enable the fifo dma operation */
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 0618846..19ef92a 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -62,6 +62,7 @@ nv40_graph_create_context(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_grctx ctx = {};
+	unsigned long flags;
 	int ret;
 
 	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
@@ -76,6 +77,17 @@ nv40_graph_create_context(struct nouveau_channel *chan)
 	nv40_grctx_init(&ctx);
 
 	nv_wo32(chan->ramin_grctx, 0, chan->ramin_grctx->pinst);
+
+	/* init grctx pointer in ramfc, and on PFIFO if channel is
+	 * already active there
+	 */
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	nv_wo32(chan->ramfc, 0x38, chan->ramin_grctx->pinst >> 4);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000000);
+	if ((nv_rd32(dev, 0x003204) & 0x0000001f) == chan->id)
+		nv_wr32(dev, 0x0032e0, chan->ramin_grctx->pinst >> 4);
+	nv_mask(dev, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index c510e74..2d7ea75 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -554,13 +554,48 @@ static struct nouveau_bitfield nv50_graph_trap_ccache[] = {
 };
 
 /* There must be a *lot* of these. Will take some time to gather them up. */
-static struct nouveau_enum nv50_data_error_names[] = {
-	{ 4,	"INVALID_VALUE" },
-	{ 5,	"INVALID_ENUM" },
-	{ 8,	"INVALID_OBJECT" },
-	{ 0xc,	"INVALID_BITFIELD" },
-	{ 0x28,	"MP_NO_REG_SPACE" },
-	{ 0x2b,	"MP_BLOCK_SIZE_MISMATCH" },
+struct nouveau_enum nv50_data_error_names[] = {
+	{ 0x00000003, "INVALID_QUERY_OR_TEXTURE" },
+	{ 0x00000004, "INVALID_VALUE" },
+	{ 0x00000005, "INVALID_ENUM" },
+	{ 0x00000008, "INVALID_OBJECT" },
+	{ 0x00000009, "READ_ONLY_OBJECT" },
+	{ 0x0000000a, "SUPERVISOR_OBJECT" },
+	{ 0x0000000b, "INVALID_ADDRESS_ALIGNMENT" },
+	{ 0x0000000c, "INVALID_BITFIELD" },
+	{ 0x0000000d, "BEGIN_END_ACTIVE" },
+	{ 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT" },
+	{ 0x0000000f, "VIEWPORT_ID_NEEDS_GP" },
+	{ 0x00000010, "RT_DOUBLE_BIND" },
+	{ 0x00000011, "RT_TYPES_MISMATCH" },
+	{ 0x00000012, "RT_LINEAR_WITH_ZETA" },
+	{ 0x00000015, "FP_TOO_FEW_REGS" },
+	{ 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH" },
+	{ 0x00000017, "RT_LINEAR_WITH_MSAA" },
+	{ 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT" },
+	{ 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT" },
+	{ 0x0000001a, "RT_INVALID_ALIGNMENT" },
+	{ 0x0000001b, "SAMPLER_OVER_LIMIT" },
+	{ 0x0000001c, "TEXTURE_OVER_LIMIT" },
+	{ 0x0000001e, "GP_TOO_MANY_OUTPUTS" },
+	{ 0x0000001f, "RT_BPP128_WITH_MS8" },
+	{ 0x00000021, "Z_OUT_OF_BOUNDS" },
+	{ 0x00000023, "XY_OUT_OF_BOUNDS" },
+	{ 0x00000027, "CP_MORE_PARAMS_THAN_SHARED" },
+	{ 0x00000028, "CP_NO_REG_SPACE_STRIPED" },
+	{ 0x00000029, "CP_NO_REG_SPACE_PACKED" },
+	{ 0x0000002a, "CP_NOT_ENOUGH_WARPS" },
+	{ 0x0000002b, "CP_BLOCK_SIZE_MISMATCH" },
+	{ 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS" },
+	{ 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS" },
+	{ 0x0000002e, "CP_NO_BLOCKDIM_LATCH" },
+	{ 0x00000031, "ENG2D_FORMAT_MISMATCH" },
+	{ 0x0000003f, "PRIMITIVE_ID_NEEDS_GP" },
+	{ 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT" },
+	{ 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT" },
+	{ 0x00000046, "LAYER_ID_NEEDS_GP" },
+	{ 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT" },
+	{ 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT" },
 	{}
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_fbcon.c b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
index cbb4a1a..fa5d4c2 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
@@ -200,8 +200,6 @@ nvc0_fbcon_accel_init(struct fb_info *info)
 		return ret;
 	}
 
-	printk(KERN_ERR "fb vma 0x%010llx\n", nvbo->vma.offset);
-
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0000, 1);
 	OUT_RING  (chan, 0x0000902d);
 	BEGIN_NVC0(chan, 2, NvSub2D, 0x0104, 2);
diff --git a/drivers/gpu/drm/nouveau/nvc0_fifo.c b/drivers/gpu/drm/nouveau/nvc0_fifo.c
index 82a4ded..e6f92c54 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fifo.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fifo.c
@@ -33,6 +33,7 @@ struct nvc0_fifo_priv {
 	struct nouveau_gpuobj *playlist[2];
 	int cur_playlist;
 	struct nouveau_vma user_vma;
+	int spoon_nr;
 };
 
 struct nvc0_fifo_chan {
@@ -324,13 +325,18 @@ nvc0_fifo_init(struct drm_device *dev)
 	nv_wr32(dev, 0x000204, 0xffffffff);
 	nv_wr32(dev, 0x002204, 0xffffffff);
 
+	priv->spoon_nr = hweight32(nv_rd32(dev, 0x002204));
+	NV_DEBUG(dev, "PFIFO: %d subfifo(s)\n", priv->spoon_nr);
+
 	/* assign engines to subfifos */
-	nv_wr32(dev, 0x002208, ~(1 << 0)); /* PGRAPH */
-	nv_wr32(dev, 0x00220c, ~(1 << 1)); /* PVP */
-	nv_wr32(dev, 0x002210, ~(1 << 1)); /* PPP */
-	nv_wr32(dev, 0x002214, ~(1 << 1)); /* PBSP */
-	nv_wr32(dev, 0x002218, ~(1 << 2)); /* PCE0 */
-	nv_wr32(dev, 0x00221c, ~(1 << 1)); /* PCE1 */
+	if (priv->spoon_nr >= 3) {
+		nv_wr32(dev, 0x002208, ~(1 << 0)); /* PGRAPH */
+		nv_wr32(dev, 0x00220c, ~(1 << 1)); /* PVP */
+		nv_wr32(dev, 0x002210, ~(1 << 1)); /* PPP */
+		nv_wr32(dev, 0x002214, ~(1 << 1)); /* PBSP */
+		nv_wr32(dev, 0x002218, ~(1 << 2)); /* PCE0 */
+		nv_wr32(dev, 0x00221c, ~(1 << 1)); /* PCE1 */
+	}
 
 	/* PSUBFIFO[n] */
 	for (i = 0; i < 3; i++) {
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.c b/drivers/gpu/drm/nouveau/nvc0_graph.c
index cf2f6aa..5feacd5 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.c
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.c
@@ -244,7 +244,6 @@ nvc0_graph_load_context(struct nouveau_channel *chan)
 	if (!nv_wait(dev, 0x409800, 0x00000010, 0x00000010))
 		NV_ERROR(dev, "PGRAPH: load_ctx timeout\n");
 
-	printk(KERN_ERR "load_ctx 0x%08x\n", nv_rd32(dev, 0x409b00));
 	return 0;
 }
 
@@ -334,8 +333,6 @@ nvc0_graph_create(struct drm_device *dev)
 	case 0xc0:
 		if (priv->tp_total == 11) { /* 465, 3/4/4/0, 4 */
 			priv->magic_not_rop_nr = 0x07;
-			priv->magic419bd0      = 0x0a360000;
-			priv->magic419be4      = 0x04c33a54;
 			/* filled values up to tp_total, the rest 0 */
 			priv->magicgpc980[0]   = 0x22111000;
 			priv->magicgpc980[1]   = 0x00000233;
@@ -345,8 +342,6 @@ nvc0_graph_create(struct drm_device *dev)
 		} else
 		if (priv->tp_total == 14) { /* 470, 3/3/4/4, 5 */
 			priv->magic_not_rop_nr = 0x05;
-			priv->magic419bd0      = 0x043c0000;
-			priv->magic419be4      = 0x09041208;
 			priv->magicgpc980[0]   = 0x11110000;
 			priv->magicgpc980[1]   = 0x00233222;
 			priv->magicgpc980[2]   = 0x00000000;
@@ -355,8 +350,6 @@ nvc0_graph_create(struct drm_device *dev)
 		} else
 		if (priv->tp_total == 15) { /* 480, 3/4/4/4, 6 */
 			priv->magic_not_rop_nr = 0x06;
-			priv->magic419bd0      = 0x023e0000;
-			priv->magic419be4      = 0x10414104;
 			priv->magicgpc980[0]   = 0x11110000;
 			priv->magicgpc980[1]   = 0x03332222;
 			priv->magicgpc980[2]   = 0x00000000;
@@ -366,8 +359,6 @@ nvc0_graph_create(struct drm_device *dev)
 		break;
 	case 0xc3: /* 450, 4/0/0/0, 2 */
 		priv->magic_not_rop_nr = 0x03;
-		priv->magic419bd0      = 0x00500000;
-		priv->magic419be4      = 0x00000000;
 		priv->magicgpc980[0]   = 0x00003210;
 		priv->magicgpc980[1]   = 0x00000000;
 		priv->magicgpc980[2]   = 0x00000000;
@@ -376,8 +367,6 @@ nvc0_graph_create(struct drm_device *dev)
 		break;
 	case 0xc4: /* 460, 3/4/0/0, 4 */
 		priv->magic_not_rop_nr = 0x01;
-		priv->magic419bd0      = 0x045c0000;
-		priv->magic419be4      = 0x09041208;
 		priv->magicgpc980[0]   = 0x02321100;
 		priv->magicgpc980[1]   = 0x00000000;
 		priv->magicgpc980[2]   = 0x00000000;
@@ -386,14 +375,12 @@ nvc0_graph_create(struct drm_device *dev)
 		break;
 	}
 
-	if (!priv->magic419bd0) {
+	if (!priv->magic_not_rop_nr) {
 		NV_ERROR(dev, "PGRAPH: unknown config: %d/%d/%d/%d, %d\n",
 			 priv->tp_nr[0], priv->tp_nr[1], priv->tp_nr[2],
 			 priv->tp_nr[3], priv->rop_nr);
 		/* use 0xc3's values... */
 		priv->magic_not_rop_nr = 0x03;
-		priv->magic419bd0      = 0x00500000;
-		priv->magic419be4      = 0x00000000;
 		priv->magicgpc980[0]   = 0x00003210;
 		priv->magicgpc980[1]   = 0x00000000;
 		priv->magicgpc980[2]   = 0x00000000;
@@ -597,7 +584,7 @@ nvc0_graph_init_ctxctl(struct drm_device *dev)
 	r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
 	ret = nvc0_fuc_load_fw(dev, 0x409000, "fuc409c", "fuc409d");
 	if (ret == 0)
-		nvc0_fuc_load_fw(dev, 0x41a000, "fuc41ac", "fuc41ad");
+		ret = nvc0_fuc_load_fw(dev, 0x41a000, "fuc41ac", "fuc41ad");
 	nv_wr32(dev, 0x000260, r000260);
 
 	if (ret)
@@ -699,18 +686,11 @@ nvc0_graph_init(struct drm_device *dev)
 	nv_wr32(dev, 0x400054, 0x34ce3464);
 
 	ret = nvc0_graph_init_ctxctl(dev);
-	if (ret)
-		return ret;
-
-	dev_priv->engine.graph.accel_blocked = false;
+	if (ret == 0)
+		dev_priv->engine.graph.accel_blocked = false;
 	return 0;
 }
 
-static struct nouveau_enum nvc0_graph_data_error[] = {
-	{ 5, "INVALID_ENUM" },
-	{}
-};
-
 static int
 nvc0_graph_isr_chid(struct drm_device *dev, u64 inst)
 {
@@ -753,9 +733,17 @@ nvc0_graph_isr(struct drm_device *dev)
 		stat &= ~0x00000010;
 	}
 
+	if (stat & 0x00000020) {
+		NV_INFO(dev, "PGRAPH: ILLEGAL_CLASS ch %d [0x%010llx] subc %d "
+			     "class 0x%04x mthd 0x%04x data 0x%08x\n",
+			chid, inst, subc, class, mthd, data);
+		nv_wr32(dev, 0x400100, 0x00000020);
+		stat &= ~0x00000020;
+	}
+
 	if (stat & 0x00100000) {
 		NV_INFO(dev, "PGRAPH: DATA_ERROR [");
-		nouveau_enum_print(nvc0_graph_data_error, code);
+		nouveau_enum_print(nv50_data_error_names, code);
 		printk("] ch %d [0x%010llx] subc %d class 0x%04x "
 		       "mthd 0x%04x data 0x%08x\n",
 		       chid, inst, subc, class, mthd, data);
@@ -763,6 +751,14 @@ nvc0_graph_isr(struct drm_device *dev)
 		stat &= ~0x00100000;
 	}
 
+	if (stat & 0x00200000) {
+		u32 trap = nv_rd32(dev, 0x400108);
+		NV_INFO(dev, "PGRAPH: TRAP ch %d status 0x%08x\n", chid, trap);
+		nv_wr32(dev, 0x400108, trap);
+		nv_wr32(dev, 0x400100, 0x00200000);
+		stat &= ~0x00200000;
+	}
+
 	if (stat & 0x00080000) {
 		u32 ustat = nv_rd32(dev, 0x409c18);
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.h b/drivers/gpu/drm/nouveau/nvc0_graph.h
index 1e1f24f..40e26f9 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.h
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.h
@@ -46,8 +46,6 @@ struct nvc0_graph_priv {
 	struct nouveau_gpuobj *unk4188b8;
 
 	u8  magic_not_rop_nr;
-	u32 magic419bd0;
-	u32 magic419be4;
 	u32 magicgpc980[4];
 	u32 magicgpc918;
 };
diff --git a/drivers/gpu/drm/nouveau/nvc0_grctx.c b/drivers/gpu/drm/nouveau/nvc0_grctx.c
index 88fa621..b9e68b2 100644
--- a/drivers/gpu/drm/nouveau/nvc0_grctx.c
+++ b/drivers/gpu/drm/nouveau/nvc0_grctx.c
@@ -1557,7 +1557,7 @@ nvc0_grctx_generate_unk47xx(struct drm_device *dev)
 }
 
 static void
-nvc0_grctx_generate_unk58xx(struct drm_device *dev)
+nvc0_grctx_generate_shaders(struct drm_device *dev)
 {
 	nv_wr32(dev, 0x405800, 0x078000bf);
 	nv_wr32(dev, 0x405830, 0x02180000);
@@ -1593,7 +1593,7 @@ nvc0_grctx_generate_unk64xx(struct drm_device *dev)
 }
 
 static void
-nvc0_grctx_generate_unk78xx(struct drm_device *dev)
+nvc0_grctx_generate_tpbus(struct drm_device *dev)
 {
 	nv_wr32(dev, 0x407804, 0x00000023);
 	nv_wr32(dev, 0x40780c, 0x0a418820);
@@ -1606,7 +1606,7 @@ nvc0_grctx_generate_unk78xx(struct drm_device *dev)
 }
 
 static void
-nvc0_grctx_generate_unk80xx(struct drm_device *dev)
+nvc0_grctx_generate_ccache(struct drm_device *dev)
 {
 	nv_wr32(dev, 0x408000, 0x00000000);
 	nv_wr32(dev, 0x408004, 0x00000000);
@@ -1801,7 +1801,7 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	struct nvc0_graph_chan *grch = chan->pgraph_ctx;
 	struct drm_device *dev = chan->dev;
 	int i, gpc, tp, id;
-	u32 r000260;
+	u32 r000260, tmp;
 
 	r000260 = nv_rd32(dev, 0x000260);
 	nv_wr32(dev, 0x000260, r000260 & ~1);
@@ -1811,11 +1811,11 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	nvc0_grctx_generate_macro(dev);
 	nvc0_grctx_generate_m2mf(dev);
 	nvc0_grctx_generate_unk47xx(dev);
-	nvc0_grctx_generate_unk58xx(dev);
+	nvc0_grctx_generate_shaders(dev);
 	nvc0_grctx_generate_unk60xx(dev);
 	nvc0_grctx_generate_unk64xx(dev);
-	nvc0_grctx_generate_unk78xx(dev);
-	nvc0_grctx_generate_unk80xx(dev);
+	nvc0_grctx_generate_tpbus(dev);
+	nvc0_grctx_generate_ccache(dev);
 	nvc0_grctx_generate_rop(dev);
 	nvc0_grctx_generate_gpc(dev);
 	nvc0_grctx_generate_tp(dev);
@@ -1843,8 +1843,12 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 		}
 	}
 
-	nv_wr32(dev, 0x406028, 0x00000443);
-	nv_wr32(dev, 0x405870, 0x00000443);
+	tmp = 0;
+	for (i = 0; i < priv->gpc_nr; i++)
+		tmp |= priv->tp_nr[i] << (i * 4);
+	nv_wr32(dev, 0x406028, tmp);
+	nv_wr32(dev, 0x405870, tmp);
+
 	nv_wr32(dev, 0x40602c, 0x00000000);
 	nv_wr32(dev, 0x405874, 0x00000000);
 	nv_wr32(dev, 0x406030, 0x00000000);
@@ -1875,9 +1879,11 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 	}
 
 	if (1) {
-		u32 data[6] = {};
+		u32 data[6] = {}, data2[2] = {};
 		u8 tpnr[GPC_MAX];
+		u8 shift, ntpcv;
 
+		/* calculate first set of magics */
 		memcpy(tpnr, priv->tp_nr, sizeof(priv->tp_nr));
 
 		for (tp = 0; tp < priv->tp_total; tp++) {
@@ -1892,6 +1898,20 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 		for (; tp < 32; tp++)
 			data[tp / 6] |= 7 << ((tp % 6) * 5);
 
+		/* and the second... */
+		shift = 0;
+		ntpcv = priv->tp_total;
+		while (!(ntpcv & (1 << 4))) {
+			ntpcv <<= 1;
+			shift++;
+		}
+
+		data2[0]  = (ntpcv << 16);
+		data2[0] |= (shift << 21);
+		data2[0] |= (((1 << (0 + 5)) % ntpcv) << 24);
+		for (i = 1; i < 7; i++)
+			data2[1] |= ((1 << (i + 5)) % ntpcv) << ((i - 1) * 5);
+
 		// GPC_BROADCAST
 		nv_wr32(dev, 0x418bb8, (priv->tp_total << 8) |
 					priv->magic_not_rop_nr);
@@ -1900,9 +1920,9 @@ nvc0_grctx_generate(struct nouveau_channel *chan)
 
 		// GPC_BROADCAST.TP_BROADCAST
 		nv_wr32(dev, 0x419bd0, (priv->tp_total << 8) |
-					priv->magic_not_rop_nr |
-					priv->magic419bd0);
-		nv_wr32(dev, 0x419be4, priv->magic419be4);
+				       priv->magic_not_rop_nr |
+				       data2[0]);
+		nv_wr32(dev, 0x419be4, data2[1]);
 		for (i = 0; i < 6; i++)
 			nv_wr32(dev, 0x419b00 + (i * 4), data[i]);