Francisco Jerez
2009-Aug-19 17:03 UTC
[Nouveau] [PATCH] drm/nouveau: Add a MM for mappable VRAM that isn't usable as scanout.
Dynamically resizing the framebuffer on nv04 was like playing Russian
roulette (and it often happened gratuitously) because it seems unable
to scan out from buffers above 16MB. This patch splits the mappable
VRAM into two chunks when that's the case, and makes the higher one to
be used as well when applicable.
Signed-off-by: Francisco Jerez <currojerez at riseup.net>
---
drivers/gpu/drm/nouveau/nouveau_bo.c | 24 ++++++++++++++++++++++--
drivers/gpu/drm/nouveau/nouveau_drv.c | 2 ++
drivers/gpu/drm/nouveau/nouveau_gem.c | 15 ++++++++++-----
drivers/gpu/drm/nouveau/nouveau_mem.c | 26 +++++++++++++++++++++++++-
4 files changed, 59 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
b/drivers/gpu/drm/nouveau/nouveau_bo.c
index eba257a..06c2d13 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -68,8 +68,11 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel
*chan,
nvbo->tile_mode = tile_mode;
nvbo->tile_flags = tile_flags;
- if (!nvbo->mappable && (flags & TTM_PL_FLAG_VRAM))
- flags |= TTM_PL_FLAG_PRIV0;
+ if (flags & TTM_PL_FLAG_VRAM) {
+ flags |= TTM_PL_FLAG_PRIV1;
+ if (!nvbo->mappable)
+ flags |= TTM_PL_FLAG_PRIV0;
+ }
/* Some of the tile_flags have a periodic structure of 24*4096 bytes,
* align to to that as well as the page size. Overallocate memory to
@@ -271,6 +274,21 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev,
uint32_t type,
man->available_caching man->default_caching = 0;
break;
+ case TTM_PL_PRIV1: /* Mappable but unusable as scanout. */
+ {
+ man->flags = TTM_MEMTYPE_FLAG_FIXED |
+ TTM_MEMTYPE_FLAG_MAPPABLE |
+ TTM_MEMTYPE_FLAG_NEEDS_IOREMAP;
+ man->available_caching = TTM_PL_FLAG_UNCACHED |
+ TTM_PL_FLAG_WC;
+ man->default_caching = TTM_PL_FLAG_WC;
+
+ man->io_addr = NULL;
+ man->io_offset = drm_get_resource_start(dev, 1);
+ man->io_size = drm_get_resource_len(dev, 1);
+ man->gpu_offset = dev_priv->vm_vram_base;
+ break;
+ }
case TTM_PL_TT:
switch (dev_priv->gart_info.type) {
case NOUVEAU_GART_AGP:
@@ -564,6 +582,7 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo,
struct file *filp)
static uint32_t nouveau_mem_prios[] = {
TTM_PL_PRIV0,
+ TTM_PL_PRIV1,
TTM_PL_VRAM,
TTM_PL_TT,
TTM_PL_SYSTEM
@@ -571,6 +590,7 @@ static uint32_t nouveau_mem_prios[] = {
static uint32_t nouveau_busy_prios[] = {
TTM_PL_TT,
TTM_PL_PRIV0,
+ TTM_PL_PRIV1,
TTM_PL_VRAM,
TTM_PL_SYSTEM
};
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c
b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 65d884e..7d2a032 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -114,6 +114,8 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t
pm_state)
NV_INFO(dev, "Evicting buffers...\n");
ttm_bo_evict_mm(&dev_priv->ttm.bdev, TTM_PL_VRAM);
+ if (dev_priv->ttm.bdev.man[TTM_PL_PRIV1].has_type)
+ ttm_bo_evict_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV1);
NV_INFO(dev, "Idling channels...\n");
for (i = 0; i < engine->fifo.channels; i++) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c
b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 1c06801..f3baadf 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -206,7 +206,8 @@ nouveau_gem_set_domain(struct drm_gem_object *gem, uint32_t
read_domains,
if ((valid_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
(read_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
(bo->mem.mem_type == TTM_PL_VRAM ||
- bo->mem.mem_type == TTM_PL_PRIV0))
+ bo->mem.mem_type == TTM_PL_PRIV0 ||
+ bo->mem.mem_type == TTM_PL_PRIV1))
flags = TTM_PL_FLAG_VRAM;
else
if ((valid_domains & NOUVEAU_GEM_DOMAIN_GART) &&
@@ -221,8 +222,11 @@ nouveau_gem_set_domain(struct drm_gem_object *gem, uint32_t
read_domains,
flags = TTM_PL_FLAG_TT;
}
- if ((flags & TTM_PL_FLAG_VRAM) && !nvbo->mappable)
- flags |= TTM_PL_FLAG_PRIV0;
+ if (flags & TTM_PL_FLAG_VRAM) {
+ flags |= TTM_PL_FLAG_PRIV1;
+ if (!nvbo->mappable)
+ flags |= TTM_PL_FLAG_PRIV0;
+ }
bo->proposed_placement &= ~TTM_PL_MASK_MEM;
bo->proposed_placement |= flags;
@@ -352,7 +356,8 @@ retry:
if (nvbo->bo.offset == b->presumed_offset &&
(((nvbo->bo.mem.mem_type == TTM_PL_VRAM ||
- nvbo->bo.mem.mem_type == TTM_PL_PRIV0) &&
+ nvbo->bo.mem.mem_type == TTM_PL_PRIV0 ||
+ nvbo->bo.mem.mem_type == TTM_PL_PRIV1) &&
b->presumed_domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
(nvbo->bo.mem.mem_type == TTM_PL_TT &&
b->presumed_domain & NOUVEAU_GEM_DOMAIN_GART))) {
@@ -761,7 +766,7 @@ domain_to_ttm(struct nouveau_bo *nvbo, uint32_t domain)
uint32_t flags = 0;
if (domain & NOUVEAU_GEM_DOMAIN_VRAM) {
- flags |= TTM_PL_FLAG_VRAM;
+ flags |= TTM_PL_FLAG_VRAM | TTM_PL_FLAG_PRIV1;
if (!nvbo->mappable)
flags |= TTM_PL_FLAG_PRIV0;
}
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c
b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 01160fe..61336a2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -299,6 +299,8 @@ void nouveau_mem_close(struct drm_device *dev)
if (dev_priv->ttm.bdev.man[TTM_PL_PRIV0].has_type)
ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV0);
+ if (dev_priv->ttm.bdev.man[TTM_PL_PRIV1].has_type)
+ ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV1);
ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_VRAM);
ttm_bo_device_release(&dev_priv->ttm.bdev);
@@ -410,6 +412,17 @@ uint64_t nouveau_mem_fb_amount(struct drm_device *dev)
return 0;
}
+static uint64_t nouveau_mem_scanout_limit(struct drm_device *dev)
+{
+ struct drm_nouveau_private *dev_priv=dev->dev_private;
+ uint64_t vram_size = dev_priv->fb_available_size;
+
+ if (dev_priv->card_type == NV_04)
+ return min(16ull*1024*1024, vram_size);
+
+ return vram_size;
+}
+
static void nouveau_mem_reset_agp(struct drm_device *dev)
{
uint32_t saved_pci_nv_1, saved_pci_nv_19, pmc_enable;
@@ -482,7 +495,7 @@ nouveau_mem_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
- uint32_t vram_size, bar1_size, text_size;
+ uint32_t vram_size, bar1_size, text_size, scanout_size;
int ret, dma_bits = 32;
dev_priv->fb_phys = drm_get_resource_start(dev, 1);
@@ -517,6 +530,7 @@ nouveau_mem_init(struct drm_device *dev)
/* non-mappable vram */
dev_priv->fb_available_size = nouveau_mem_fb_amount(dev);
dev_priv->fb_available_size -= dev_priv->ramin_rsvd_vram;
+ scanout_size = nouveau_mem_scanout_limit(dev) >> PAGE_SHIFT;
vram_size = dev_priv->fb_available_size >> PAGE_SHIFT;
bar1_size = drm_get_resource_len(dev, 1) >> PAGE_SHIFT;
text_size = (256 * 1024) >> PAGE_SHIFT;
@@ -530,6 +544,16 @@ nouveau_mem_init(struct drm_device *dev)
vram_size = bar1_size;
}
+ /* mappable vram that's unusable as scanout. */
+ if (scanout_size < vram_size) {
+ if ((ret = ttm_bo_init_mm(bdev, TTM_PL_PRIV1, scanout_size,
+ vram_size - scanout_size))) {
+ NV_ERROR(dev, "Failed PRIV1 mm init: %d\n", ret);
+ return ret;
+ }
+ vram_size = scanout_size;
+ }
+
/* mappable vram */
if ((ret = ttm_bo_init_mm(bdev, TTM_PL_VRAM, text_size,
vram_size - text_size))) {
--
1.6.3.3
Francisco Jerez
2009-Aug-19 19:13 UTC
[Nouveau] [PATCHv2] drm/nouveau: Add a MM for mappable VRAM that isn't usable as scanout.
Dynamically resizing the framebuffer on nv04 was like playing Russian
roulette (and it often happened gratuitously) because it seems unable
to scan out from buffers above 16MB. This patch splits the mappable
VRAM into two chunks when that's the case, and makes the higher one to
be used as well when applicable.
Signed-off-by: Francisco Jerez <currojerez at riseup.net>
---
Rather don't allow userspace pinned buffers in PRIV1, even if it's
sometimes safe.
drivers/gpu/drm/nouveau/nouveau_bo.c | 24 ++++++++++++++++++++++--
drivers/gpu/drm/nouveau/nouveau_drv.c | 2 ++
drivers/gpu/drm/nouveau/nouveau_gem.c | 13 +++++++++----
drivers/gpu/drm/nouveau/nouveau_mem.c | 26 +++++++++++++++++++++++++-
4 files changed, 58 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 358a9aa..9f175ce 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -68,8 +68,11 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel
*chan,
nvbo->tile_mode = tile_mode;
nvbo->tile_flags = tile_flags;
- if (!nvbo->mappable && (flags & TTM_PL_FLAG_VRAM))
- flags |= TTM_PL_FLAG_PRIV0;
+ if (flags & TTM_PL_FLAG_VRAM) {
+ flags |= TTM_PL_FLAG_PRIV1;
+ if (!nvbo->mappable)
+ flags |= TTM_PL_FLAG_PRIV0;
+ }
/* Some of the tile_flags have a periodic structure of 24*4096 bytes,
* align to to that as well as the page size. Overallocate memory to
@@ -296,6 +299,21 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev,
uint32_t type,
man->available_caching man->default_caching = 0;
break;
+ case TTM_PL_PRIV1: /* Mappable but unusable as scanout. */
+ {
+ man->flags = TTM_MEMTYPE_FLAG_FIXED |
+ TTM_MEMTYPE_FLAG_MAPPABLE |
+ TTM_MEMTYPE_FLAG_NEEDS_IOREMAP;
+ man->available_caching = TTM_PL_FLAG_UNCACHED |
+ TTM_PL_FLAG_WC;
+ man->default_caching = TTM_PL_FLAG_WC;
+
+ man->io_addr = NULL;
+ man->io_offset = drm_get_resource_start(dev, 1);
+ man->io_size = drm_get_resource_len(dev, 1);
+ man->gpu_offset = dev_priv->vm_vram_base;
+ break;
+ }
case TTM_PL_TT:
switch (dev_priv->gart_info.type) {
case NOUVEAU_GART_AGP:
@@ -589,6 +607,7 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo,
struct file *filp)
static uint32_t nouveau_mem_prios[] = {
TTM_PL_PRIV0,
+ TTM_PL_PRIV1,
TTM_PL_VRAM,
TTM_PL_TT,
TTM_PL_SYSTEM
@@ -596,6 +615,7 @@ static uint32_t nouveau_mem_prios[] = {
static uint32_t nouveau_busy_prios[] = {
TTM_PL_TT,
TTM_PL_PRIV0,
+ TTM_PL_PRIV1,
TTM_PL_VRAM,
TTM_PL_SYSTEM
};
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c
b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 65d884e..7d2a032 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -114,6 +114,8 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t
pm_state)
NV_INFO(dev, "Evicting buffers...\n");
ttm_bo_evict_mm(&dev_priv->ttm.bdev, TTM_PL_VRAM);
+ if (dev_priv->ttm.bdev.man[TTM_PL_PRIV1].has_type)
+ ttm_bo_evict_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV1);
NV_INFO(dev, "Idling channels...\n");
for (i = 0; i < engine->fifo.channels; i++) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c
b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 1c06801..cbb174a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -206,7 +206,8 @@ nouveau_gem_set_domain(struct drm_gem_object *gem, uint32_t
read_domains,
if ((valid_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
(read_domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
(bo->mem.mem_type == TTM_PL_VRAM ||
- bo->mem.mem_type == TTM_PL_PRIV0))
+ bo->mem.mem_type == TTM_PL_PRIV0 ||
+ bo->mem.mem_type == TTM_PL_PRIV1))
flags = TTM_PL_FLAG_VRAM;
else
if ((valid_domains & NOUVEAU_GEM_DOMAIN_GART) &&
@@ -221,8 +222,11 @@ nouveau_gem_set_domain(struct drm_gem_object *gem, uint32_t
read_domains,
flags = TTM_PL_FLAG_TT;
}
- if ((flags & TTM_PL_FLAG_VRAM) && !nvbo->mappable)
- flags |= TTM_PL_FLAG_PRIV0;
+ if (flags & TTM_PL_FLAG_VRAM) {
+ flags |= TTM_PL_FLAG_PRIV1;
+ if (!nvbo->mappable)
+ flags |= TTM_PL_FLAG_PRIV0;
+ }
bo->proposed_placement &= ~TTM_PL_MASK_MEM;
bo->proposed_placement |= flags;
@@ -352,7 +356,8 @@ retry:
if (nvbo->bo.offset == b->presumed_offset &&
(((nvbo->bo.mem.mem_type == TTM_PL_VRAM ||
- nvbo->bo.mem.mem_type == TTM_PL_PRIV0) &&
+ nvbo->bo.mem.mem_type == TTM_PL_PRIV0 ||
+ nvbo->bo.mem.mem_type == TTM_PL_PRIV1) &&
b->presumed_domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
(nvbo->bo.mem.mem_type == TTM_PL_TT &&
b->presumed_domain & NOUVEAU_GEM_DOMAIN_GART))) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c
b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 01160fe..61336a2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -299,6 +299,8 @@ void nouveau_mem_close(struct drm_device *dev)
if (dev_priv->ttm.bdev.man[TTM_PL_PRIV0].has_type)
ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV0);
+ if (dev_priv->ttm.bdev.man[TTM_PL_PRIV1].has_type)
+ ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_PRIV1);
ttm_bo_clean_mm(&dev_priv->ttm.bdev, TTM_PL_VRAM);
ttm_bo_device_release(&dev_priv->ttm.bdev);
@@ -410,6 +412,17 @@ uint64_t nouveau_mem_fb_amount(struct drm_device *dev)
return 0;
}
+static uint64_t nouveau_mem_scanout_limit(struct drm_device *dev)
+{
+ struct drm_nouveau_private *dev_priv=dev->dev_private;
+ uint64_t vram_size = dev_priv->fb_available_size;
+
+ if (dev_priv->card_type == NV_04)
+ return min(16ull*1024*1024, vram_size);
+
+ return vram_size;
+}
+
static void nouveau_mem_reset_agp(struct drm_device *dev)
{
uint32_t saved_pci_nv_1, saved_pci_nv_19, pmc_enable;
@@ -482,7 +495,7 @@ nouveau_mem_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
- uint32_t vram_size, bar1_size, text_size;
+ uint32_t vram_size, bar1_size, text_size, scanout_size;
int ret, dma_bits = 32;
dev_priv->fb_phys = drm_get_resource_start(dev, 1);
@@ -517,6 +530,7 @@ nouveau_mem_init(struct drm_device *dev)
/* non-mappable vram */
dev_priv->fb_available_size = nouveau_mem_fb_amount(dev);
dev_priv->fb_available_size -= dev_priv->ramin_rsvd_vram;
+ scanout_size = nouveau_mem_scanout_limit(dev) >> PAGE_SHIFT;
vram_size = dev_priv->fb_available_size >> PAGE_SHIFT;
bar1_size = drm_get_resource_len(dev, 1) >> PAGE_SHIFT;
text_size = (256 * 1024) >> PAGE_SHIFT;
@@ -530,6 +544,16 @@ nouveau_mem_init(struct drm_device *dev)
vram_size = bar1_size;
}
+ /* mappable vram that's unusable as scanout. */
+ if (scanout_size < vram_size) {
+ if ((ret = ttm_bo_init_mm(bdev, TTM_PL_PRIV1, scanout_size,
+ vram_size - scanout_size))) {
+ NV_ERROR(dev, "Failed PRIV1 mm init: %d\n", ret);
+ return ret;
+ }
+ vram_size = scanout_size;
+ }
+
/* mappable vram */
if ((ret = ttm_bo_init_mm(bdev, TTM_PL_VRAM, text_size,
vram_size - text_size))) {
--
1.6.3.3
Possibly Parallel Threads
- [PATCH] drm/nouveau: always do buffer object moves on bo->channel
- [PATCH] drm/nv10/plane: add plane support for nv10-nv40
- A patch for extending pdf device to embed popup text and web links
- [PATCH v1 0/4] GPU Direct RDMA (P2P DMA) for Device Private Pages
- [Bug 76475] New: Nouveau fails to load due to unknown opcode 0x80