drm/nouveau: tidy up and extend dma object creation interfaces
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 11b2370..0f33132 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -39,22 +39,22 @@
if (dev_priv->card_type >= NV_50) {
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
- dev_priv->vm_end, NV_DMA_ACCESS_RO,
- NV_DMA_TARGET_AGP, &pushbuf);
+ dev_priv->vm_end, NV_MEM_ACCESS_RO,
+ NV_MEM_TARGET_VM, &pushbuf);
chan->pushbuf_base = pb->bo.offset;
} else
if (pb->bo.mem.mem_type == TTM_PL_TT) {
- ret = nouveau_gpuobj_gart_dma_new(chan, 0,
- dev_priv->gart_info.aper_size,
- NV_DMA_ACCESS_RO, &pushbuf,
- NULL);
+ ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
+ dev_priv->gart_info.aper_size,
+ NV_MEM_ACCESS_RO,
+ NV_MEM_TARGET_GART, &pushbuf);
chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
} else
if (dev_priv->card_type != NV_04) {
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
dev_priv->fb_available_size,
- NV_DMA_ACCESS_RO,
- NV_DMA_TARGET_VIDMEM, &pushbuf);
+ NV_MEM_ACCESS_RO,
+ NV_MEM_TARGET_VRAM, &pushbuf);
chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
} else {
/* NV04 cmdbuf hack, from original ddx.. not sure of it's
@@ -62,11 +62,10 @@
* VRAM.
*/
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
- pci_resource_start(dev->pdev,
- 1),
+ pci_resource_start(dev->pdev, 1),
dev_priv->fb_available_size,
- NV_DMA_ACCESS_RO,
- NV_DMA_TARGET_PCI, &pushbuf);
+ NV_MEM_ACCESS_RO,
+ NV_MEM_TARGET_PCI, &pushbuf);
chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
}
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index d76d2c0..a52b1da 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -886,12 +886,14 @@
extern int nouveau_gpuobj_dma_new(struct nouveau_channel *, int class,
uint64_t offset, uint64_t size, int access,
int target, struct nouveau_gpuobj **);
-extern int nouveau_gpuobj_gart_dma_new(struct nouveau_channel *,
- uint64_t offset, uint64_t size,
- int access, struct nouveau_gpuobj **,
- uint32_t *o_ret);
extern int nouveau_gpuobj_gr_new(struct nouveau_channel *, int class,
struct nouveau_gpuobj **);
+extern int nv50_gpuobj_dma_new(struct nouveau_channel *, int class, u64 base,
+ u64 size, int target, int access, u32 type,
+ u32 comp, struct nouveau_gpuobj **pobj);
+extern void nv50_gpuobj_dma_init(struct nouveau_gpuobj *, u32 offset,
+ int class, u64 base, u64 size, int target,
+ int access, u32 type, u32 comp);
extern int nouveau_ioctl_grobj_alloc(struct drm_device *, void *data,
struct drm_file *);
extern int nouveau_ioctl_gpuobj_free(struct drm_device *, void *data,
@@ -1545,6 +1547,22 @@
dev->pdev->subsystem_device == sub_device;
}
+/* memory type/access flags, do not match hardware values */
+#define NV_MEM_ACCESS_RO 1
+#define NV_MEM_ACCESS_WO 2
+#define NV_MEM_ACCESS_RW (NV_MEM_ACCESS_RO | NV_MEM_ACCESS_WO)
+#define NV_MEM_ACCESS_VM 4
+
+#define NV_MEM_TARGET_VRAM 0
+#define NV_MEM_TARGET_PCI 1
+#define NV_MEM_TARGET_PCI_NOSNOOP 2
+#define NV_MEM_TARGET_VM 3
+#define NV_MEM_TARGET_GART 4
+
+#define NV_MEM_TYPE_VM 0x7f
+#define NV_MEM_COMP_VM 0x03
+
+/* NV_SW object class */
#define NV_SW 0x0000506e
#define NV_SW_DMA_SEMAPHORE 0x00000060
#define NV_SW_SEMAPHORE_OFFSET 0x00000064
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index 91aa6c5..2579fc6 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -459,8 +459,8 @@
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
mem->start << PAGE_SHIFT,
mem->size << PAGE_SHIFT,
- NV_DMA_ACCESS_RW,
- NV_DMA_TARGET_VIDMEM, &obj);
+ NV_MEM_ACCESS_RW,
+ NV_MEM_TARGET_VRAM, &obj);
if (ret)
return ret;
diff --git a/drivers/gpu/drm/nouveau/nouveau_notifier.c b/drivers/gpu/drm/nouveau/nouveau_notifier.c
index 2c5a1f6..a050b7b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_notifier.c
+++ b/drivers/gpu/drm/nouveau/nouveau_notifier.c
@@ -99,7 +99,6 @@
int size, uint32_t *b_offset)
{
struct drm_device *dev = chan->dev;
- struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *nobj = NULL;
struct drm_mm_node *mem;
uint32_t offset;
@@ -113,31 +112,15 @@
return -ENOMEM;
}
- offset = chan->notifier_bo->bo.mem.start << PAGE_SHIFT;
- if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_VRAM) {
- target = NV_DMA_TARGET_VIDMEM;
- } else
- if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_TT) {
- if (dev_priv->gart_info.type == NOUVEAU_GART_SGDMA &&
- dev_priv->card_type < NV_50) {
- ret = nouveau_sgdma_get_page(dev, offset, &offset);
- if (ret)
- return ret;
- target = NV_DMA_TARGET_PCI;
- } else {
- target = NV_DMA_TARGET_AGP;
- if (dev_priv->card_type >= NV_50)
- offset += dev_priv->vm_gart_base;
- }
- } else {
- NV_ERROR(dev, "Bad DMA target, mem_type %d!\n",
- chan->notifier_bo->bo.mem.mem_type);
- return -EINVAL;
- }
+ if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_VRAM)
+ target = NV_MEM_TARGET_VRAM;
+ else
+ target = NV_MEM_TARGET_GART;
+ offset = chan->notifier_bo->bo.mem.start << PAGE_SHIFT;
offset += mem->start;
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, offset,
- mem->size, NV_DMA_ACCESS_RW, target,
+ mem->size, NV_MEM_ACCESS_RW, target,
&nobj);
if (ret) {
drm_mm_put_block(mem);
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index e8c74de..924653c 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -404,115 +404,159 @@
The method below creates a DMA object in instance RAM and returns a handle
to it that can be used to set up context objects.
*/
-int
-nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class,
- uint64_t offset, uint64_t size, int access,
- int target, struct nouveau_gpuobj **gpuobj)
-{
- struct drm_device *dev = chan->dev;
- struct drm_nouveau_private *dev_priv = dev->dev_private;
- struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
- int ret;
- NV_DEBUG(dev, "ch%d class=0x%04x offset=0x%llx size=0x%llx\n",
- chan->id, class, offset, size);
- NV_DEBUG(dev, "access=%d target=%d\n", access, target);
+void
+nv50_gpuobj_dma_init(struct nouveau_gpuobj *obj, u32 offset, int class,
+ u64 base, u64 size, int target, int access,
+ u32 type, u32 comp)
+{
+ struct drm_nouveau_private *dev_priv = obj->dev->dev_private;
+ struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
+ u32 flags0;
+
+ flags0 = (comp << 29) | (type << 22) | class;
+ flags0 |= 0x00100000;
+
+ switch (access) {
+ case NV_MEM_ACCESS_RO: flags0 |= 0x00040000; break;
+ case NV_MEM_ACCESS_RW:
+ case NV_MEM_ACCESS_WO: flags0 |= 0x00080000; break;
+ default:
+ break;
+ }
switch (target) {
- case NV_DMA_TARGET_AGP:
- offset += dev_priv->gart_info.aper_base;
+ case NV_MEM_TARGET_VRAM:
+ flags0 |= 0x00010000;
+ break;
+ case NV_MEM_TARGET_PCI:
+ flags0 |= 0x00020000;
+ break;
+ case NV_MEM_TARGET_PCI_NOSNOOP:
+ flags0 |= 0x00030000;
+ break;
+ case NV_MEM_TARGET_GART:
+ base += dev_priv->vm_gart_base;
+ default:
+ flags0 &= ~0x00100000;
+ break;
+ }
+
+ /* convert to base + limit */
+ size = (base + size) - 1;
+
+ nv_wo32(obj, offset + 0x00, flags0);
+ nv_wo32(obj, offset + 0x04, lower_32_bits(size));
+ nv_wo32(obj, offset + 0x08, lower_32_bits(base));
+ nv_wo32(obj, offset + 0x0c, upper_32_bits(size) << 24 |
+ upper_32_bits(base));
+ nv_wo32(obj, offset + 0x10, 0x00000000);
+ nv_wo32(obj, offset + 0x14, 0x00000000);
+
+ pinstmem->flush(obj->dev);
+}
+
+int
+nv50_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base, u64 size,
+ int target, int access, u32 type, u32 comp,
+ struct nouveau_gpuobj **pobj)
+{
+ struct drm_device *dev = chan->dev;
+ int ret;
+
+ ret = nouveau_gpuobj_new(dev, chan, 24, 16, NVOBJ_FLAG_ZERO_ALLOC |
+ NVOBJ_FLAG_ZERO_FREE, pobj);
+ if (ret)
+ return ret;
+
+ nv50_gpuobj_dma_init(*pobj, 0, class, base, size, target,
+ access, type, comp);
+ return 0;
+}
+
+int
+nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base,
+ u64 size, int access, int target,
+ struct nouveau_gpuobj **pobj)
+{
+ struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+ struct drm_device *dev = chan->dev;
+ struct nouveau_gpuobj *obj;
+ u32 page_addr, flags0, flags2;
+ int ret;
+
+ if (dev_priv->card_type >= NV_50) {
+ u32 comp = (target == NV_MEM_TARGET_VM) ? NV_MEM_COMP_VM : 0;
+ u32 type = (target == NV_MEM_TARGET_VM) ? NV_MEM_TYPE_VM : 0;
+
+ return nv50_gpuobj_dma_new(chan, class, base, size,
+ target, access, type, comp, pobj);
+ }
+
+ if (target == NV_MEM_TARGET_GART) {
+ if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
+ target = NV_MEM_TARGET_PCI_NOSNOOP;
+ base += dev_priv->gart_info.aper_base;
+ } else
+ if (base != 0) {
+ ret = nouveau_sgdma_get_page(dev, base, &page_addr);
+ if (ret)
+ return ret;
+
+ target = NV_MEM_TARGET_PCI;
+ base = page_addr;
+ } else {
+ nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, pobj);
+ return 0;
+ }
+ }
+
+ flags0 = class;
+ flags0 |= 0x00003000; /* PT present, PT linear */
+ flags2 = 0;
+
+ switch (target) {
+ case NV_MEM_TARGET_PCI:
+ flags0 |= 0x00020000;
+ break;
+ case NV_MEM_TARGET_PCI_NOSNOOP:
+ flags0 |= 0x00030000;
break;
default:
break;
}
- ret = nouveau_gpuobj_new(dev, chan,
- nouveau_gpuobj_class_instmem_size(dev, class),
- 16, NVOBJ_FLAG_ZERO_ALLOC |
- NVOBJ_FLAG_ZERO_FREE, gpuobj);
- if (ret) {
- NV_ERROR(dev, "Error creating gpuobj: %d\n", ret);
+ switch (access) {
+ case NV_MEM_ACCESS_RO:
+ flags0 |= 0x00004000;
+ break;
+ case NV_MEM_ACCESS_WO:
+ flags0 |= 0x00008000;
+ default:
+ flags2 |= 0x00000002;
+ break;
+ }
+
+ flags0 |= (base & 0x00000fff) << 20;
+ flags2 |= (base & 0xfffff000);
+
+ ret = nouveau_gpuobj_new(dev, chan, (dev_priv->card_type >= NV_40) ?
+ 32 : 16, 16, NVOBJ_FLAG_ZERO_ALLOC |
+ NVOBJ_FLAG_ZERO_FREE, &obj);
+ if (ret)
return ret;
- }
- if (dev_priv->card_type < NV_50) {
- uint32_t frame, adjust, pte_flags = 0;
+ nv_wo32(obj, 0x00, flags0);
+ nv_wo32(obj, 0x04, size - 1);
+ nv_wo32(obj, 0x08, flags2);
+ nv_wo32(obj, 0x0c, flags2);
- if (access != NV_DMA_ACCESS_RO)
- pte_flags |= (1<<1);
- adjust = offset & 0x00000fff;
- frame = offset & ~0x00000fff;
-
- nv_wo32(*gpuobj, 0, ((1<<12) | (1<<13) | (adjust << 20) |
- (access << 14) | (target << 16) |
- class));
- nv_wo32(*gpuobj, 4, size - 1);
- nv_wo32(*gpuobj, 8, frame | pte_flags);
- nv_wo32(*gpuobj, 12, frame | pte_flags);
- } else {
- uint64_t limit = offset + size - 1;
- uint32_t flags0, flags5;
-
- if (target == NV_DMA_TARGET_VIDMEM) {
- flags0 = 0x00190000;
- flags5 = 0x00010000;
- } else {
- flags0 = 0x7fc00000;
- flags5 = 0x00080000;
- }
-
- nv_wo32(*gpuobj, 0, flags0 | class);
- nv_wo32(*gpuobj, 4, lower_32_bits(limit));
- nv_wo32(*gpuobj, 8, lower_32_bits(offset));
- nv_wo32(*gpuobj, 12, ((upper_32_bits(limit) & 0xff) << 24) |
- (upper_32_bits(offset) & 0xff));
- nv_wo32(*gpuobj, 20, flags5);
- }
-
- instmem->flush(dev);
-
- (*gpuobj)->engine = NVOBJ_ENGINE_SW;
- (*gpuobj)->class = class;
+ obj->engine = NVOBJ_ENGINE_SW;
+ obj->class = class;
+ *pobj = obj;
return 0;
}
-int
-nouveau_gpuobj_gart_dma_new(struct nouveau_channel *chan,
- uint64_t offset, uint64_t size, int access,
- struct nouveau_gpuobj **gpuobj,
- uint32_t *o_ret)
-{
- struct drm_device *dev = chan->dev;
- struct drm_nouveau_private *dev_priv = dev->dev_private;
- int ret;
-
- if (dev_priv->gart_info.type == NOUVEAU_GART_AGP ||
- (dev_priv->card_type >= NV_50 &&
- dev_priv->gart_info.type == NOUVEAU_GART_SGDMA)) {
- ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
- offset + dev_priv->vm_gart_base,
- size, access, NV_DMA_TARGET_AGP,
- gpuobj);
- if (o_ret)
- *o_ret = 0;
- } else
- if (dev_priv->gart_info.type == NOUVEAU_GART_SGDMA) {
- nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, gpuobj);
- if (offset & ~0xffffffffULL) {
- NV_ERROR(dev, "obj offset exceeds 32-bits\n");
- return -EINVAL;
- }
- if (o_ret)
- *o_ret = (uint32_t)offset;
- ret = (*gpuobj != NULL) ? 0 : -EINVAL;
- } else {
- NV_ERROR(dev, "Invalid GART type %d\n", dev_priv->gart_info.type);
- return -EINVAL;
- }
-
- return ret;
-}
-
/* Context objects in the instance RAM have the following structure.
* On NV40 they are 32 byte long, on NV30 and smaller 16 bytes.
@@ -806,8 +850,8 @@
if (dev_priv->card_type >= NV_50) {
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
0, dev_priv->vm_end,
- NV_DMA_ACCESS_RW,
- NV_DMA_TARGET_AGP, &vram);
+ NV_MEM_ACCESS_RW,
+ NV_MEM_TARGET_VM, &vram);
if (ret) {
NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
return ret;
@@ -815,8 +859,8 @@
} else {
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
0, dev_priv->fb_available_size,
- NV_DMA_ACCESS_RW,
- NV_DMA_TARGET_VIDMEM, &vram);
+ NV_MEM_ACCESS_RW,
+ NV_MEM_TARGET_VRAM, &vram);
if (ret) {
NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
return ret;
@@ -834,20 +878,13 @@
if (dev_priv->card_type >= NV_50) {
ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
0, dev_priv->vm_end,
- NV_DMA_ACCESS_RW,
- NV_DMA_TARGET_AGP, &tt);
- if (ret) {
- NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
- return ret;
- }
- } else
- if (dev_priv->gart_info.type != NOUVEAU_GART_NONE) {
- ret = nouveau_gpuobj_gart_dma_new(chan, 0,
- dev_priv->gart_info.aper_size,
- NV_DMA_ACCESS_RW, &tt, NULL);
+ NV_MEM_ACCESS_RW,
+ NV_MEM_TARGET_VM, &tt);
} else {
- NV_ERROR(dev, "Invalid GART type %d\n", dev_priv->gart_info.type);
- ret = -EINVAL;
+ ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
+ 0, dev_priv->gart_info.aper_size,
+ NV_MEM_ACCESS_RW,
+ NV_MEM_TARGET_GART, &tt);
}
if (ret) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_reg.h b/drivers/gpu/drm/nouveau/nouveau_reg.h
index df3a87e..04e8fb7 100644
--- a/drivers/gpu/drm/nouveau/nouveau_reg.h
+++ b/drivers/gpu/drm/nouveau/nouveau_reg.h
@@ -79,17 +79,6 @@
# define NV40_RAMHT_CONTEXT_ENGINE_SHIFT 20
# define NV40_RAMHT_CONTEXT_INSTANCE_SHIFT 0
-/* DMA object defines */
-#define NV_DMA_ACCESS_RW 0
-#define NV_DMA_ACCESS_RO 1
-#define NV_DMA_ACCESS_WO 2
-#define NV_DMA_TARGET_VIDMEM 0
-#define NV_DMA_TARGET_PCI 2
-#define NV_DMA_TARGET_AGP 3
-/* The following is not a real value used by the card, it's changed by
- * nouveau_object_dma_create */
-#define NV_DMA_TARGET_PCI_NONLINEAR 8
-
/* Some object classes we care about in the drm */
#define NV_CLASS_DMA_FROM_MEMORY 0x00000002
#define NV_CLASS_DMA_TO_MEMORY 0x00000003
diff --git a/drivers/gpu/drm/nouveau/nouveau_sgdma.c b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
index 54af760..db32644 100644
--- a/drivers/gpu/drm/nouveau/nouveau_sgdma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
@@ -247,14 +247,11 @@
*/
gpuobj->cinst = gpuobj->pinst;
- /* Maybe use NV_DMA_TARGET_AGP for PCIE? NVIDIA do this, and
- * confirmed to work on c51. Perhaps means NV_DMA_TARGET_PCIE
- * on those cards? */
nv_wo32(gpuobj, 0, NV_CLASS_DMA_IN_MEMORY |
(1 << 12) /* PT present */ |
(0 << 13) /* PT *not* linear */ |
- (NV_DMA_ACCESS_RW << 14) |
- (NV_DMA_TARGET_PCI << 16));
+ (0 << 14) /* RW */ |
+ (2 << 16) /* PCI */);
nv_wo32(gpuobj, 4, aper_size - 1);
for (i = 2; i < 2 + (aper_size >> 12); i++)
nv_wo32(gpuobj, i * 4, 0x00000000);
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 35b2840..e779e93 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -536,7 +536,7 @@
ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
0, dev_priv->vram_size,
- NV_DMA_ACCESS_RW, NV_DMA_TARGET_VIDMEM,
+ NV_MEM_ACCESS_RW, NV_MEM_TARGET_VRAM,
&gpuobj);
if (ret)
goto out_err;
@@ -546,9 +546,10 @@
if (ret)
goto out_err;
- ret = nouveau_gpuobj_gart_dma_new(dev_priv->channel, 0,
- dev_priv->gart_info.aper_size,
- NV_DMA_ACCESS_RW, &gpuobj, NULL);
+ ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
+ 0, dev_priv->gart_info.aper_size,
+ NV_MEM_ACCESS_RW, NV_MEM_TARGET_GART,
+ &gpuobj);
if (ret)
goto out_err;