Nouveau - Mar 2013 - [PATCH 1/4] drm/nvc0: implement VRAM compression

---
 drivers/gpu/drm/nouveau/core/include/subdev/ltcg.h |    7 +
 drivers/gpu/drm/nouveau/core/subdev/fb/nvc0.c      |   55 +++++----
 drivers/gpu/drm/nouveau/core/subdev/ltcg/nvc0.c    |  129 +++++++++++++++++++-
 drivers/gpu/drm/nouveau/core/subdev/vm/nvc0.c      |   58 +++++++++-
 4 files changed, 220 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/core/include/subdev/ltcg.h
b/drivers/gpu/drm/nouveau/core/include/subdev/ltcg.h
index f351f63..a1985ed 100644
--- a/drivers/gpu/drm/nouveau/core/include/subdev/ltcg.h
+++ b/drivers/gpu/drm/nouveau/core/include/subdev/ltcg.h
@@ -4,8 +4,15 @@
 #include <core/subdev.h>
 #include <core/device.h>
 
+struct nouveau_mm_node;
+
 struct nouveau_ltcg {
 	struct nouveau_subdev base;
+
+	int  (*tags_alloc)(struct nouveau_ltcg *, u32 count,
+	                   struct nouveau_mm_node **);
+	void (*tags_free)(struct nouveau_ltcg *, struct nouveau_mm_node **);
+	void (*tags_clear)(struct nouveau_ltcg *, u32 first, u32 count);
 };
 
 static inline struct nouveau_ltcg *
diff --git a/drivers/gpu/drm/nouveau/core/subdev/fb/nvc0.c
b/drivers/gpu/drm/nouveau/core/subdev/fb/nvc0.c
index 3b2ddc6..86ad592 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/fb/nvc0.c
+++ b/drivers/gpu/drm/nouveau/core/subdev/fb/nvc0.c
@@ -23,6 +23,7 @@
  */
 
 #include <subdev/fb.h>
+#include <subdev/ltcg.h>
 #include <subdev/bios.h>
 
 struct nvc0_fb_priv {
@@ -31,34 +32,14 @@ struct nvc0_fb_priv {
 	dma_addr_t r100c10;
 };
 
-/* 0 = unsupported
- * 1 = non-compressed
- * 3 = compressed
- */
-static const u8 types[256] = {
-	1, 1, 3, 3, 3, 3, 0, 3, 3, 3, 3, 0, 0, 0, 0, 0,
-	0, 1, 0, 0, 0, 0, 0, 3, 3, 3, 3, 0, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3,
-	3, 3, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-	0, 3, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 0, 1, 1, 1, 1, 0,
-	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-	0, 0, 0, 3, 3, 3, 3, 1, 1, 1, 1, 0, 0, 0, 0, 0,
-	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3,
-	3, 3, 3, 1, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 3, 3,
-	3, 3, 0, 0, 0, 0, 0, 0, 3, 0, 0, 3, 0, 3, 0, 3,
-	3, 0, 3, 3, 3, 3, 3, 0, 0, 3, 0, 3, 0, 3, 3, 0,
-	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 1, 1, 0
-};
+extern const u8 nvc0_pte_storage_type_map[256];
+
 
 static bool
 nvc0_fb_memtype_valid(struct nouveau_fb *pfb, u32 tile_flags)
 {
 	u8 memtype = (tile_flags & 0x0000ff00) >> 8;
-	return likely((types[memtype] == 1));
+	return likely((nvc0_pte_storage_type_map[memtype] != 0xff));
 }
 
 static int
@@ -130,6 +111,7 @@ nvc0_fb_vram_new(struct nouveau_fb *pfb, u64 size, u32
align, u32 ncmin,
 	int type = (memtype & 0x0ff);
 	int back = (memtype & 0x800);
 	int ret;
+	const bool comp = nvc0_pte_storage_type_map[type] != type;
 
 	size  >>= 12;
 	align >>= 12;
@@ -142,10 +124,22 @@ nvc0_fb_vram_new(struct nouveau_fb *pfb, u64 size, u32
align, u32 ncmin,
 		return -ENOMEM;
 
 	INIT_LIST_HEAD(&mem->regions);
-	mem->memtype = type;
 	mem->size = size;
 
 	mutex_lock(&pfb->base.mutex);
+	if (comp) {
+		struct nouveau_ltcg *ltcg = nouveau_ltcg(pfb->base.base.parent);
+
+		/* compression only works with lpages */
+		if (align == (1 << (17 - 12))) {
+			int n = size >> 5;
+			ltcg->tags_alloc(ltcg, n, &mem->tag);
+		}
+		if (unlikely(!mem->tag))
+			type = nvc0_pte_storage_type_map[type];
+	}
+	mem->memtype = type;
+
 	do {
 		if (back)
 			ret = nouveau_mm_tail(mm, 1, size, ncmin, align, &r);
@@ -168,6 +162,17 @@ nvc0_fb_vram_new(struct nouveau_fb *pfb, u64 size, u32
align, u32 ncmin,
 	return 0;
 }
 
+static void
+nvc0_fb_vram_del(struct nouveau_fb *pfb, struct nouveau_mem **pmem)
+{
+	struct nouveau_ltcg *ltcg = nouveau_ltcg(pfb->base.base.parent);
+
+	if ((*pmem)->tag)
+		ltcg->tags_free(ltcg, &(*pmem)->tag);
+
+	nv50_fb_vram_del(pfb, pmem);
+}
+
 static int
 nvc0_fb_init(struct nouveau_object *object)
 {
@@ -215,7 +220,7 @@ nvc0_fb_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	priv->base.memtype_valid = nvc0_fb_memtype_valid;
 	priv->base.ram.init = nvc0_fb_vram_init;
 	priv->base.ram.get = nvc0_fb_vram_new;
-	priv->base.ram.put = nv50_fb_vram_del;
+	priv->base.ram.put = nvc0_fb_vram_del;
 
 	priv->r100c10_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 	if (priv->r100c10_page) {
diff --git a/drivers/gpu/drm/nouveau/core/subdev/ltcg/nvc0.c
b/drivers/gpu/drm/nouveau/core/subdev/ltcg/nvc0.c
index 078a2b9..a529563 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/ltcg/nvc0.c
+++ b/drivers/gpu/drm/nouveau/core/subdev/ltcg/nvc0.c
@@ -23,10 +23,17 @@
  */
 
 #include <subdev/ltcg.h>
+#include <subdev/fb.h>
+#include <subdev/timer.h>
 
 struct nvc0_ltcg_priv {
 	struct nouveau_ltcg base;
+	u32 part_nr;
+	u32 part_mask;
 	u32 subp_nr;
+	struct nouveau_mm tags;
+	u32 num_tags;
+	struct nouveau_mm_node *tag_ram;
 };
 
 static void
@@ -62,11 +69,104 @@ nvc0_ltcg_intr(struct nouveau_subdev *subdev)
 }
 
 static int
+nvc0_ltcg_tags_alloc(struct nouveau_ltcg *ltcg, u32 n,
+		     struct nouveau_mm_node **pnode)
+{
+	struct nvc0_ltcg_priv *priv = (struct nvc0_ltcg_priv *)ltcg;
+	int ret;
+
+	ret = nouveau_mm_head(&priv->tags, 1, n, n, 1, pnode);
+	if (ret)
+		*pnode = NULL;
+
+	return ret;
+}
+
+static void
+nvc0_ltcg_tags_free(struct nouveau_ltcg *ltcg, struct nouveau_mm_node **pnode)
+{
+	struct nvc0_ltcg_priv *priv = (struct nvc0_ltcg_priv *)ltcg;
+
+	nouveau_mm_free(&priv->tags, pnode);
+}
+
+static void
+nvc0_ltcg_tags_clear(struct nouveau_ltcg *ltcg, u32 first, u32 count)
+{
+	struct nvc0_ltcg_priv *priv = (struct nvc0_ltcg_priv *)ltcg;
+	u32 last = first + count - 1;
+	int p, i;
+
+	BUG_ON((first > last) || (last >= priv->num_tags));
+
+	nv_wr32(priv, 0x17e8cc, first);
+	nv_wr32(priv, 0x17e8d0, last);
+	nv_wr32(priv, 0x17e8c8, 0x4); /* trigger clear */
+
+	/* wait until it's finished with clearing */
+	for (p = 0; p < priv->part_nr; ++p) {
+		if (!(priv->part_mask & (1 << p)))
+			continue;
+		for (i = 0; i < priv->subp_nr; ++i)
+			nv_wait(priv, 0x1410c8 + p * 0x2000 + i * 0x400, ~0, 0);
+	}
+}
+
+/* TODO: Figure out tag memory details and drop the over-cautious allocation.
+ */
+static int
+nvc0_ltcg_init_tag_ram(struct nouveau_fb *pfb, struct nvc0_ltcg_priv *priv)
+{
+	u32 tag_size, tag_margin, tag_align;
+	int ret;
+
+	nv_wr32(priv, 0x17e8d8, priv->part_nr);
+
+	/* tags for 1/4 of VRAM should be enough (8192/4 per GiB of VRAM) */
+	priv->num_tags = (pfb->ram.size >> 17) / 4;
+	if (priv->num_tags > (1 << 17))
+		priv->num_tags = 1 << 17; /* we have 17 bits in PTE */
+	priv->num_tags = (priv->num_tags + 63) & ~63; /* round up to 64 */
+
+	tag_align = priv->part_nr * 0x800;
+	tag_margin = (tag_align < 0x6000) ? 0x6000 : tag_align;
+
+	/* 4 part 4 sub: 0x2000 bytes for 56 tags */
+	/* 3 part 4 sub: 0x6000 bytes for 168 tags */
+	/*
+	 * About 147 bytes per tag. Let's be safe and allocate x2, which makes
+	 * 0x4980 bytes for 64 tags, and round up to 0x6000 bytes for 64 tags.
+	 *
+	 * For 4 GiB of memory we'll have 8192 tags which makes 3 MiB, < 0.1 %.
+	 */
+	tag_size  = (priv->num_tags / 64) * 0x6000 + tag_margin;
+	tag_size += tag_align;
+	tag_size  = (tag_size + 0xfff) >> 12; /* round up */
+
+	ret = nouveau_mm_tail(&pfb->vram, 0, tag_size, tag_size, 1,
+	                      &priv->tag_ram);
+	if (ret) {
+		priv->num_tags = 0;
+	} else {
+		u64 tag_base = (priv->tag_ram->offset << 12) + tag_margin;
+
+		tag_base += tag_align - 1;
+		tag_base /= tag_align;
+
+		nv_wr32(priv, 0x17e8d4, tag_base);
+	}
+	ret = nouveau_mm_init(&priv->tags, 0, priv->num_tags, 1);
+
+	return ret;
+}
+
+static int
 nvc0_ltcg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	       struct nouveau_oclass *oclass, void *data, u32 size,
 	       struct nouveau_object **pobject)
 {
 	struct nvc0_ltcg_priv *priv;
+	struct nouveau_fb *pfb = nouveau_fb(parent);
 	int ret;
 
 	ret = nouveau_ltcg_create(parent, engine, oclass, &priv);
@@ -74,19 +174,44 @@ nvc0_ltcg_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	if (ret)
 		return ret;
 
-	priv->subp_nr = nv_rd32(priv, 0x17e8dc) >> 24;
+	priv->part_nr = nv_rd32(priv, 0x022438);
+	priv->part_mask = nv_rd32(priv, 0x022554);
+
+	priv->subp_nr = nv_rd32(priv, 0x17e8dc) >> 28;
+
 	nv_mask(priv, 0x17e820, 0x00100000, 0x00000000); /* INTR_EN &= ~0x10 */
 
+	ret = nvc0_ltcg_init_tag_ram(pfb, priv);
+	if (ret)
+		return ret;
+
+	priv->base.tags_alloc = nvc0_ltcg_tags_alloc;
+	priv->base.tags_free  = nvc0_ltcg_tags_free;
+	priv->base.tags_clear = nvc0_ltcg_tags_clear;
+
 	nv_subdev(priv)->intr = nvc0_ltcg_intr;
 	return 0;
 }
 
+static void
+nvc0_ltcg_dtor(struct nouveau_object *object)
+{
+	struct nouveau_ltcg *ltcg = (struct nouveau_ltcg *)object;
+	struct nvc0_ltcg_priv *priv = (struct nvc0_ltcg_priv *)ltcg;
+	struct nouveau_fb *pfb = nouveau_fb(ltcg->base.base.parent);
+
+	nouveau_mm_fini(&priv->tags);
+	nouveau_mm_free(&pfb->vram, &priv->tag_ram);
+
+	nouveau_ltcg_destroy(ltcg);
+}
+
 struct nouveau_oclass
 nvc0_ltcg_oclass = {
 	.handle = NV_SUBDEV(LTCG, 0xc0),
 	.ofuncs = &(struct nouveau_ofuncs) {
 		.ctor = nvc0_ltcg_ctor,
-		.dtor = _nouveau_ltcg_dtor,
+		.dtor = nvc0_ltcg_dtor,
 		.init = _nouveau_ltcg_init,
 		.fini = _nouveau_ltcg_fini,
 	},
diff --git a/drivers/gpu/drm/nouveau/core/subdev/vm/nvc0.c
b/drivers/gpu/drm/nouveau/core/subdev/vm/nvc0.c
index 30c61e6..4c3b0a2 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/vm/nvc0.c
+++ b/drivers/gpu/drm/nouveau/core/subdev/vm/nvc0.c
@@ -28,12 +28,54 @@
 #include <subdev/timer.h>
 #include <subdev/fb.h>
 #include <subdev/vm.h>
+#include <subdev/ltcg.h>
 
 struct nvc0_vmmgr_priv {
 	struct nouveau_vmmgr base;
 	spinlock_t lock;
 };
 
+
+/* Map from compressed to corresponding uncompressed storage type.
+ * The value 0xff represents an invalid storage type.
+ */
+const u8 nvc0_pte_storage_type_map[256] +{
+	0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0xff, 0x01, /* 0x00 */
+	0x01, 0x01, 0x01, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0x11, 0xff, 0xff, 0xff, 0xff, 0xff, 0x11, /* 0x10 */
+	0x11, 0x11, 0x11, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x26, 0x27, /* 0x20 */
+	0x28, 0x29, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x30 */
+	0xff, 0xff, 0x26, 0x27, 0x28, 0x29, 0x26, 0x27,
+	0x28, 0x29, 0xff, 0xff, 0xff, 0xff, 0x46, 0xff, /* 0x40 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0x46, 0x46, 0x46, 0x46, 0xff, 0xff, 0xff, /* 0x50 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x60 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x70 */
+	0xff, 0xff, 0xff, 0x7b, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x7b, 0x7b, /* 0x80 */
+	0x7b, 0x7b, 0xff, 0x8b, 0x8c, 0x8d, 0x8e, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0x90 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0x8b, 0x8c, 0x8d, 0x8e, 0xa7, /* 0xa0 */
+	0xa8, 0xa9, 0xaa, 0xff, 0xff, 0xff, 0xff, 0xff,
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0xb0 */
+	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xa7,
+	0xa8, 0xa9, 0xaa, 0xc3, 0xff, 0xff, 0xff, 0xff, /* 0xc0 */
+	0xff, 0xff, 0xff, 0xff, 0xfe, 0xfe, 0xc3, 0xc3,
+	0xc3, 0xc3, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, /* 0xd0 */
+	0xfe, 0xff, 0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe,
+	0xfe, 0xff, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xff, /* 0xe0 */
+	0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xfe, 0xff,
+	0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, /* 0xf0 */
+	0xfe, 0xfe, 0xfe, 0xfe, 0xff, 0xfd, 0xfe, 0xff
+};
+
+
 static void
 nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 index,
 		struct nouveau_gpuobj *pgt[2])
@@ -68,10 +110,20 @@ static void
 nvc0_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
 	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
 {
-	u32 next = 1 << (vma->node->type - 8);
+	u64 next = 1 << (vma->node->type - 8);
 
 	phys  = nvc0_vm_addr(vma, phys, mem->memtype, 0);
 	pte <<= 3;
+
+	if (mem->tag) {
+		struct nouveau_ltcg *ltcg +		
nouveau_ltcg(vma->vm->vmm->base.base.parent);
+		u32 tag = mem->tag->offset + (delta >> 17);
+		phys |= (u64)tag << (32 + 12);
+		next |= (u64)1   << (32 + 12);
+		ltcg->tags_clear(ltcg, tag, cnt);
+	}
+
 	while (cnt--) {
 		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
 		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
@@ -85,10 +137,12 @@ nvc0_vm_map_sg(struct nouveau_vma *vma, struct
nouveau_gpuobj *pgt,
 	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
 {
 	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 7 : 5;
+	/* compressed storage types are invalid for system memory */
+	u32 memtype = nvc0_pte_storage_type_map[mem->memtype & 0xff];
 
 	pte <<= 3;
 	while (cnt--) {
-		u64 phys = nvc0_vm_addr(vma, *list++, mem->memtype, target);
+		u64 phys = nvc0_vm_addr(vma, *list++, memtype, target);
 		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
 		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
 		pte += 8;
-- 
1.7.3.4

---
 drivers/gpu/drm/nouveau/core/engine/graph/nv40.c   |   10 ++++++++++
 drivers/gpu/drm/nouveau/core/engine/graph/nv50.c   |   10 ++++++++++
 drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c   |   15 +++++++++++++++
 drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h   |    2 ++
 drivers/gpu/drm/nouveau/core/engine/graph/nve0.c   |    2 ++
 .../gpu/drm/nouveau/core/include/engine/graph.h    |    4 ++++
 drivers/gpu/drm/nouveau/nouveau_abi16.c            |   12 ++++--------
 7 files changed, 47 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nv40.c
b/drivers/gpu/drm/nouveau/core/engine/graph/nv40.c
index 17049d5..193a5de 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nv40.c
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nv40.c
@@ -46,6 +46,14 @@ struct nv40_graph_chan {
 	struct nouveau_graph_chan base;
 };
 
+static u64
+nv40_graph_units(struct nouveau_graph *graph)
+{
+	struct nv40_graph_priv *priv = (void *)graph;
+
+	return nv_rd32(priv, 0x1540);
+}
+
 /*******************************************************************************
  * Graphics object classes
 
******************************************************************************/
@@ -359,6 +367,8 @@ nv40_graph_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	else
 		nv_engine(priv)->sclass = nv40_graph_sclass;
 	nv_engine(priv)->tile_prog = nv40_graph_tile_prog;
+
+	priv->base.units = nv40_graph_units;
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nv50.c
b/drivers/gpu/drm/nouveau/core/engine/graph/nv50.c
index f2b1a7a..1ac3611 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nv50.c
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nv50.c
@@ -48,6 +48,14 @@ struct nv50_graph_chan {
 	struct nouveau_graph_chan base;
 };
 
+static u64
+nv50_graph_units(struct nouveau_graph *graph)
+{
+	struct nv50_graph_priv *priv = (void *)graph;
+
+	return nv_rd32(priv, 0x1540);
+}
+
 /*******************************************************************************
  * Graphics object classes
 
******************************************************************************/
@@ -819,6 +827,8 @@ nv50_graph_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	nv_subdev(priv)->intr = nv50_graph_intr;
 	nv_engine(priv)->cclass = &nv50_graph_cclass;
 
+	priv->base.units = nv50_graph_units;
+
 	switch (nv_device(priv)->chipset) {
 	case 0x50:
 		nv_engine(priv)->sclass = nv50_graph_sclass;
diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c
b/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c
index 0de0dd7..5ce4941 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c
@@ -60,6 +60,19 @@ nvc8_graph_sclass[] = {
 	{}
 };
 
+u64
+nvc0_graph_units(struct nouveau_graph *graph)
+{
+	struct nvc0_graph_priv *priv = (void *)graph;
+	u64 cfg;
+
+	cfg  = (u32)priv->gpc_nr;
+	cfg |= (u32)priv->tpc_total << 8;
+	cfg |= (u64)priv->rop_nr << 32;
+
+	return cfg;
+}
+
 /*******************************************************************************
  * PGRAPH context
 
******************************************************************************/
@@ -529,6 +542,8 @@ nvc0_graph_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	nv_subdev(priv)->intr = nvc0_graph_intr;
 	nv_engine(priv)->cclass = &nvc0_graph_cclass;
 
+	priv->base.units = nvc0_graph_units;
+
 	if (nouveau_boolopt(device->cfgopt, "NvGrUseFW", false)) {
 		nv_info(priv, "using external firmware\n");
 		if (nvc0_graph_ctor_fw(priv, "fuc409c", &priv->fuc409c) ||
diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h
b/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h
index a1e78de..af033dc 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h
@@ -169,4 +169,6 @@ int  nvc0_graph_context_ctor(struct nouveau_object *, struct
nouveau_object *,
 			     struct nouveau_object **);
 void nvc0_graph_context_dtor(struct nouveau_object *);
 
+u64 nvc0_graph_units(struct nouveau_graph *);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
b/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
index 4857f91..4b45afb 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
@@ -217,6 +217,8 @@ nve0_graph_ctor(struct nouveau_object *parent, struct
nouveau_object *engine,
 	nv_engine(priv)->cclass = &nve0_graph_cclass;
 	nv_engine(priv)->sclass = nve0_graph_sclass;
 
+	priv->base.units = nvc0_graph_units;
+
 	if (nouveau_boolopt(device->cfgopt, "NvGrUseFW", false)) {
 		nv_info(priv, "using external firmware\n");
 		if (nvc0_graph_ctor_fw(priv, "fuc409c", &priv->fuc409c) ||
diff --git a/drivers/gpu/drm/nouveau/core/include/engine/graph.h
b/drivers/gpu/drm/nouveau/core/include/engine/graph.h
index 6943b40..5d39243 100644
--- a/drivers/gpu/drm/nouveau/core/include/engine/graph.h
+++ b/drivers/gpu/drm/nouveau/core/include/engine/graph.h
@@ -26,6 +26,10 @@ struct nouveau_graph_chan {
 
 struct nouveau_graph {
 	struct nouveau_engine base;
+
+	/* Returns chipset-specific counts of units packed into an u64.
+	 */
+	u64 (*units)(struct nouveau_graph *);
 };
 
 static inline struct nouveau_graph *
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c
b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index 3b6dc88..f19f0a4 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -30,6 +30,7 @@
 #include <subdev/fb.h>
 #include <subdev/timer.h>
 #include <subdev/instmem.h>
+#include <engine/graph.h>
 
 #include "nouveau_drm.h"
 #include "nouveau_dma.h"
@@ -168,6 +169,7 @@ nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS)
 	struct nouveau_drm *drm = nouveau_drm(dev);
 	struct nouveau_device *device = nv_device(drm->device);
 	struct nouveau_timer *ptimer = nouveau_timer(device);
+	struct nouveau_graph *graph = (void *)nouveau_engine(device, NVDEV_ENGINE_GR);
 	struct drm_nouveau_getparam *getparam = data;
 
 	switch (getparam->param) {
@@ -208,14 +210,8 @@ nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS)
 		getparam->value = 1;
 		break;
 	case NOUVEAU_GETPARAM_GRAPH_UNITS:
-		/* NV40 and NV50 versions are quite different, but register
-		 * address is the same. User is supposed to know the card
-		 * family anyway... */
-		if (device->chipset >= 0x40) {
-			getparam->value = nv_rd32(device, 0x001540);
-			break;
-		}
-		/* FALLTHRU */
+		getparam->value = graph->units ? graph->units(graph) : 0;
+		break;
 	default:
 		nv_debug(device, "unknown parameter %lld\n", getparam->param);
 		return -EINVAL;
-- 
1.7.3.4

---
 .../gpu/drm/nouveau/core/engine/software/nvc0.c    |   29 ++++++++++++++++++++
 1 files changed, 29 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/core/engine/software/nvc0.c
b/drivers/gpu/drm/nouveau/core/engine/software/nvc0.c
index a523eaa..d698e71 100644
--- a/drivers/gpu/drm/nouveau/core/engine/software/nvc0.c
+++ b/drivers/gpu/drm/nouveau/core/engine/software/nvc0.c
@@ -94,6 +94,32 @@ nvc0_software_mthd_flip(struct nouveau_object *object, u32
mthd,
 	return -EINVAL;
 }
 
+static int
+nvc0_software_mthd_mp_control(struct nouveau_object *object, u32 mthd,
+                              void *args, u32 size)
+{
+	struct nvc0_software_chan *chan = (void *)nv_engctx(object->parent);
+	struct nvc0_software_priv *priv = (void *)nv_object(chan)->engine;
+	u32 data = *(u32 *)args;
+
+	switch (mthd) {
+	case 0x600:
+		nv_wr32(priv, 0x419e00, data); /* MP.PM_UNK000 */
+		break;
+	case 0x644:
+		if (data & ~0x1ffffe)
+			return -EINVAL;
+		nv_wr32(priv, 0x419e44, data); /* MP.TRAP_WARP_ERROR_EN */
+		break;
+	case 0x6ac:
+		nv_wr32(priv, 0x419eac, data); /* MP.PM_UNK0AC */
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
 static struct nouveau_omthds
 nvc0_software_omthds[] = {
 	{ 0x0400, 0x0400, nvc0_software_mthd_vblsem_offset },
@@ -101,6 +127,9 @@ nvc0_software_omthds[] = {
 	{ 0x0408, 0x0408, nvc0_software_mthd_vblsem_value },
 	{ 0x040c, 0x040c, nvc0_software_mthd_vblsem_release },
 	{ 0x0500, 0x0500, nvc0_software_mthd_flip },
+	{ 0x0600, 0x0600, nvc0_software_mthd_mp_control },
+	{ 0x0644, 0x0644, nvc0_software_mthd_mp_control },
+	{ 0x06ac, 0x06ac, nvc0_software_mthd_mp_control },
 	{}
 };
 
-- 
1.7.3.4

---
 drivers/gpu/drm/nouveau/core/engine/graph/nve0.c |  230 ++++++++++++++++++++++
 1 files changed, 230 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
b/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
index 4b45afb..e411b18 100644
--- a/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
+++ b/drivers/gpu/drm/nouveau/core/engine/graph/nve0.c
@@ -77,11 +77,214 @@ nve0_graph_ctxctl_isr(struct nvc0_graph_priv *priv)
 	nv_wr32(priv, 0x409c20, ustat);
 }
 
+static const struct nouveau_enum nve0_mp_warp_error[] = {
+	{ 0x00, "(no error)" },
+	{ 0x01, "stack mismatch" },
+	{ 0x05, "misaligned PC" },
+	{ 0x08, "misaligned GPR" },
+	{ 0x09, "invalid opcode" },
+	{ 0x0d, "out of bounds GPR" },
+	{ 0x0e, "out of bounds l/s/a[]" },
+	{ 0x0f, "unaligned memory access" },
+	{ 0x11, "invalid param" },
+	{}
+};
+
+static const struct nouveau_enum nve0_mp_global_error[] = {
+	{ 2, "multiple warp errors" },
+	{ 3, "out of stack space" },
+	{}
+};
+
+static const struct nouveau_enum nve0_gpc_rop_error[] = {
+	{ 1, "RT pitch overrun" },
+	{ 4, "RT width overrun" },
+	{ 5, "RT height overrun" },
+	{ 7, "ZETA storage type mismatch" },
+	{ 8, "RT storage type mismatch" },
+	{ 10, "RT linear mismatch" },
+	{}
+};
+
+static const struct nouveau_enum nve0_sked_error[] = {
+	{ 7, "constant buffer size" },
+	{ 9, "local memory size pos" },
+	{ 10, "local memory size neg" },
+	{ 11, "warp cstack size" },
+	{ 12, "total temp size" },
+	{ 13, "register count" },
+	{ 18, "total threads" },
+	{ 20, "program offset" },
+	{ 21, "shared memory size" },
+	{ 25, "shared/l1 config" },
+	{ 26, "total register count" },
+	{}
+};
+
+static void
+nve0_graph_mp_trap(struct nvc0_graph_priv *priv, int gpc, int tp)
+{
+	int i;
+	u32 bpct = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x010));
+	u32 bpst = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x00c));
+	u32 werr = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x648));
+	u32 gerr = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x650));
+
+	nv_error(priv, "GPC%i/TP%i/MP trap:", gpc, tp);
+
+	for (i = 0; i <= 31; ++i) {
+		if (!(gerr & (1 << i)))
+			continue;
+		pr_cont(" <");
+		nouveau_enum_print(nve0_mp_global_error, i);
+		pr_cont(">");
+	}
+	if (werr) {
+		pr_cont(" <");
+		nouveau_enum_print(nve0_mp_warp_error, werr & 0xffff);
+		pr_cont(">");
+	}
+	pr_cont("\n");
+
+	nv_error(priv, "breakpoint control,status = %08x %08x\n", bpct,
bpst);
+
+	/* disable MP trap to avoid spam */
+	nv_mask(priv, TPC_UNIT(gpc, tp, 0x50c), 0x2, 0x0);
+
+	/* TODO: figure out how to resume after an MP trap */
+}
+
+static void
+nve0_graph_tp_trap(struct nvc0_graph_priv *priv, int gpc, int tp)
+{
+	u32 stat = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x508));
+
+	if (stat & 0x1) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x224));
+		nv_error(priv, "GPC%i/TP%i/TEX trap: %08x\n",
+			 gpc, tp, trap);
+
+		nv_wr32(priv, TPC_UNIT(gpc, tp, 0x224), 0xc0000000);
+		stat &= ~0x1;
+	}
+
+	if (stat & 0x2) {
+		nve0_graph_mp_trap(priv, gpc, tp);
+		stat &= ~0x2;
+	}
+
+	if (stat & 0x4) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x084));
+		nv_error(priv, "GPC%i/TP%i/POLY trap: %08x\n",
+			 gpc, tp, trap);
+
+		nv_wr32(priv, TPC_UNIT(gpc, tp, 0x084), 0xc0000000);
+		stat &= ~0x4;
+	}
+
+	if (stat & 0x8) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tp, 0x48c));
+		nv_error(priv, "GPC%i/TP%i/L1C trap: %08x\n",
+			 gpc, tp, trap);
+
+		nv_wr32(priv, TPC_UNIT(gpc, tp, 0x48c), 0xc0000000);
+		stat &= ~0x8;
+	}
+
+	if (stat) {
+		nv_error(priv, "GPC%i/TP%i: unknown stat %08x\n",
+			 gpc, tp, stat);
+	}
+}
+
+static void
+nve0_graph_gpc_trap(struct nvc0_graph_priv *priv)
+{
+	const u32 mask = nv_rd32(priv, 0x400118);
+	int gpc;
+
+	for (gpc = 0; gpc < 4; ++gpc) {
+		u32 stat;
+		int tp;
+
+		if (!(mask & (1 << gpc)))
+			continue;
+		stat = nv_rd32(priv, GPC_UNIT(gpc, 0x2c90));
+
+		if (stat & 0x0001) {
+			u32 trap[4];
+			int i;
+
+			trap[0] = nv_rd32(priv, GPC_UNIT(gpc, 0x0420));
+			trap[1] = nv_rd32(priv, GPC_UNIT(gpc, 0x0434));
+			trap[2] = nv_rd32(priv, GPC_UNIT(gpc, 0x0438));
+			trap[3] = nv_rd32(priv, GPC_UNIT(gpc, 0x043c));
+
+			nv_error(priv, "GPC%i/PROP trap:", gpc);
+			for (i = 0; i <= 29; ++i) {
+				if (!(trap[0] & (1 << i)))
+					continue;
+				pr_cont(" <");
+				nouveau_enum_print(nve0_gpc_rop_error, i);
+				pr_cont(">");
+			}
+			pr_cont("\n");
+
+			nv_error(priv, "x = %u, y = %u, "
+				 "format = %x, storage type = %x\n",
+				 trap[1] & 0xffff,
+				 trap[1] >> 16,
+				 (trap[2] >> 8) & 0x3f,
+				 trap[3] & 0xff);
+
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+			stat &= ~0x0001;
+		}
+
+		if (stat & 0x0002) {
+			u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x0900));
+			nv_error(priv, "GPC%i/ZCULL trap: %08x\n", gpc,
+				 trap);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+			stat &= ~0x0002;
+		}
+
+		if (stat & 0x0004) {
+			u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x1028));
+			nv_error(priv, "GPC%i/CCACHE trap: %08x\n", gpc,
+				 trap);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+			stat &= ~0x0004;
+		}
+
+		if (stat & 0x0008) {
+			u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x0824));
+			nv_error(priv, "GPC%i/ESETUP trap %08x\n", gpc,
+				 trap);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+			stat &= ~0x0008;
+		}
+
+		for (tp = 0; tp < 8; ++tp) {
+			if (stat & (1 << (16 + tp)))
+				nve0_graph_tp_trap(priv, gpc, tp);
+		}
+		stat &= ~0xff0000;
+
+		if (stat) {
+			nv_error(priv, "GPC%i: unknown stat %08x\n",
+				 gpc, stat);
+		}
+	}
+}
+
+
 static void
 nve0_graph_trap_isr(struct nvc0_graph_priv *priv, int chid, u64 inst,
 		struct nouveau_object *engctx)
 {
 	u32 trap = nv_rd32(priv, 0x400108);
+	int i;
 	int rop;
 
 	if (trap & 0x00000001) {
@@ -102,6 +305,33 @@ nve0_graph_trap_isr(struct nvc0_graph_priv *priv, int chid,
u64 inst,
 		trap &= ~0x00000010;
 	}
 
+	if (trap & 0x00000100) {
+		u32 stat = nv_rd32(priv, 0x407020);
+		nv_error(priv, "SKED ch %d [0x%010llx %s]:",
+			 chid, inst, nouveau_client_name(engctx));
+
+		for (i = 0; i <= 29; ++i) {
+			if (!(stat & (1 << i)))
+				continue;
+			pr_cont(" <");
+			nouveau_enum_print(nve0_sked_error, i);
+			pr_cont(">");
+		}
+		pr_cont("\n");
+
+		if (stat & 0x3fffffff)
+			nv_wr32(priv, 0x407020, 0x40000000);
+		nv_wr32(priv, 0x400108, 0x00000100);
+		trap &= ~0x00000100;
+	}
+
+	if (trap & 0x01000000) {
+		nv_error(priv, "GPC ch %d [0x%010llx %s]:\n",
+			 chid, inst, nouveau_client_name(engctx));
+		nve0_graph_gpc_trap(priv);
+		trap &= ~0x01000000;
+	}
+
 	if (trap & 0x02000000) {
 		for (rop = 0; rop < priv->rop_nr; rop++) {
 			u32 statz = nv_rd32(priv, ROP_UNIT(rop, 0x070));
-- 
1.7.3.4