This small series introduces TTM helper functions as well as Nouveau hooks that are needed to ensure buffer coherency on ARM. Most of this series is a forward-port of some patches Lucas Stach sent last year and that are also needed for Nouveau GK20A support:
http://lists.freedesktop.org/archives/nouveau/2013-August/014026.html
Another patch takes care of flushing the CPU write-buffer when writing BOs through a non-BAR path.
Alexandre Courbot (1): drm/nouveau: introduce CPU cache flushing macro
Lucas Stach (3): drm/ttm: recognize ARM arch in ioprot handler drm/ttm: introduce dma cache sync helpers drm/nouveau: hook up cache sync functions
drivers/gpu/drm/nouveau/core/os.h | 17 +++++++++++++++ drivers/gpu/drm/nouveau/nouveau_bo.c | 40 +++++++++++++++++++++++++++++++++-- drivers/gpu/drm/nouveau/nouveau_bo.h | 20 ++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_gem.c | 8 ++++++- drivers/gpu/drm/ttm/ttm_bo_util.c | 2 +- drivers/gpu/drm/ttm/ttm_tt.c | 25 ++++++++++++++++++++++ include/drm/ttm/ttm_bo_driver.h | 28 ++++++++++++++++++++++++ 7 files changed, 136 insertions(+), 4 deletions(-)
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de Signed-off-by: Alexandre Courbot acourbot@nvidia.com --- drivers/gpu/drm/ttm/ttm_bo_util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index 1df856f78568..30e5d90cb7bc 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -500,7 +500,7 @@ pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp) pgprot_val(tmp) |= _PAGE_GUARDED; } #endif -#if defined(__ia64__) +#if defined(__ia64__) || defined(__arm__) if (caching_flags & TTM_PL_FLAG_WC) tmp = pgprot_writecombine(tmp); else
From: Lucas Stach dev@lynxeye.de
On arches with non-coherent PCI, we need to flush caches ourselfes at the appropriate places. Introduce two small helpers to make things easy for TTM based drivers.
Signed-off-by: Lucas Stach dev@lynxeye.de Signed-off-by: Alexandre Courbot acourbot@nvidia.com --- drivers/gpu/drm/ttm/ttm_tt.c | 25 +++++++++++++++++++++++++ include/drm/ttm/ttm_bo_driver.h | 28 ++++++++++++++++++++++++++++ 2 files changed, 53 insertions(+)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index 75f319090043..05a316b71ad1 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -38,6 +38,7 @@ #include <linux/swap.h> #include <linux/slab.h> #include <linux/export.h> +#include <linux/dma-mapping.h> #include <drm/drm_cache.h> #include <drm/drm_mem_util.h> #include <drm/ttm/ttm_module.h> @@ -248,6 +249,30 @@ void ttm_dma_tt_fini(struct ttm_dma_tt *ttm_dma) } EXPORT_SYMBOL(ttm_dma_tt_fini);
+void ttm_dma_tt_cache_sync_for_device(struct ttm_dma_tt *ttm_dma, + struct device *dev) +{ + int i; + + for (i = 0; i < ttm_dma->ttm.num_pages; i++) { + dma_sync_single_for_device(dev, ttm_dma->dma_address[i], + PAGE_SIZE, DMA_TO_DEVICE); + } +} +EXPORT_SYMBOL(ttm_dma_tt_cache_sync_for_device); + +void ttm_dma_tt_cache_sync_for_cpu(struct ttm_dma_tt *ttm_dma, + struct device *dev) +{ + int i; + + for (i = 0; i < ttm_dma->ttm.num_pages; i++) { + dma_sync_single_for_cpu(dev, ttm_dma->dma_address[i], + PAGE_SIZE, DMA_FROM_DEVICE); + } +} +EXPORT_SYMBOL(ttm_dma_tt_cache_sync_for_cpu); + void ttm_tt_unbind(struct ttm_tt *ttm) { int ret; diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index a5183da3ef92..52fb709568fc 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -41,6 +41,7 @@ #include <linux/fs.h> #include <linux/spinlock.h> #include <linux/reservation.h> +#include <linux/device.h>
struct ttm_backend_func { /** @@ -690,6 +691,33 @@ extern int ttm_tt_swapout(struct ttm_tt *ttm, */ extern void ttm_tt_unpopulate(struct ttm_tt *ttm);
+/** + * ttm_dma_tt_cache_sync_for_device: + * + * @ttm A struct ttm_tt of the type returned by ttm_dma_tt_init. + * @dev A struct device representing the device to which to sync. + * + * This function will flush the CPU caches on arches where snooping in the + * TT is not available. On fully coherent arches this will turn into an (almost) + * noop. This makes sure that data written by the CPU is visible to the device. + */ +extern void ttm_dma_tt_cache_sync_for_device(struct ttm_dma_tt *ttm_dma, + struct device *dev); + +/** + * ttm_dma_tt_cache_sync_for_cpu: + * + * @ttm A struct ttm_tt of the type returned by ttm_dma_tt_init. + * @dev A struct device representing the device from which to sync. + * + * This function will invalidate the CPU caches on arches where snooping in the + * TT is not available. On fully coherent arches this will turn into an (almost) + * noop. This makes sure that the CPU does not read any stale cached or + * prefetched data. + */ +extern void ttm_dma_tt_cache_sync_for_cpu(struct ttm_dma_tt *ttm_dma, + struct device *dev); + /* * ttm_bo.c */
On Mon, May 19, 2014 at 04:10:56PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
On arches with non-coherent PCI,
I guess since this applies to gk20a
we need to flush caches ourselfes at
"ourselves". Or perhaps even reword to something like: "..., caches need to be flushed and invalidated explicitly", since dma_sync_for_cpu() does invalidate rather than flush.
the appropriate places. Introduce two small helpers to make things easy for TTM based drivers.
Signed-off-by: Lucas Stach dev@lynxeye.de Signed-off-by: Alexandre Courbot acourbot@nvidia.com
drivers/gpu/drm/ttm/ttm_tt.c | 25 +++++++++++++++++++++++++ include/drm/ttm/ttm_bo_driver.h | 28 ++++++++++++++++++++++++++++ 2 files changed, 53 insertions(+)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
[...]
+void ttm_dma_tt_cache_sync_for_device(struct ttm_dma_tt *ttm_dma,
struct device *dev)+{
- int i;
This should probably be unsigned long to match the type of ttm_dma->ttm.num_pages.
Thierry
On Mon, May 19, 2014 at 5:33 PM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, May 19, 2014 at 04:10:56PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
On arches with non-coherent PCI,
I guess since this applies to gk20a
we need to flush caches ourselfes at
"ourselves". Or perhaps even reword to something like: "..., caches need to be flushed and invalidated explicitly", since dma_sync_for_cpu() does invalidate rather than flush.
Rephrased as "On arches for which access to GPU memory is non-coherent, caches need to be flushed and invalidated explicitly at the appropriate places."
the appropriate places. Introduce two small helpers to make things easy for TTM based drivers.
Signed-off-by: Lucas Stach dev@lynxeye.de Signed-off-by: Alexandre Courbot acourbot@nvidia.com
drivers/gpu/drm/ttm/ttm_tt.c | 25 +++++++++++++++++++++++++ include/drm/ttm/ttm_bo_driver.h | 28 ++++++++++++++++++++++++++++ 2 files changed, 53 insertions(+)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
[...]
+void ttm_dma_tt_cache_sync_for_device(struct ttm_dma_tt *ttm_dma,
struct device *dev)+{
int i;This should probably be unsigned long to match the type of ttm_dma->ttm.num_pages.
Fixed.
Thanks, Alex.
On Fri, May 23, 2014 at 02:49:40PM +0900, Alexandre Courbot wrote:
On Mon, May 19, 2014 at 5:33 PM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, May 19, 2014 at 04:10:56PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
On arches with non-coherent PCI,
I guess since this applies to gk20a
we need to flush caches ourselfes at
"ourselves". Or perhaps even reword to something like: "..., caches need to be flushed and invalidated explicitly", since dma_sync_for_cpu() does invalidate rather than flush.
Rephrased as "On arches for which access to GPU memory is non-coherent, caches need to be flushed and invalidated explicitly at the appropriate places."
Nit: s/arches/architectures/
Thierry
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com --- drivers/gpu/drm/nouveau/nouveau_bo.c | 32 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_bo.h | 20 ++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_gem.c | 8 +++++++- 3 files changed, 59 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index b6dc85c614be..0886f47e5244 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -407,6 +407,8 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible, { int ret;
+ nouveau_bo_sync_for_device(nvbo); + ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement, interruptible, no_wait_gpu); if (ret) @@ -487,6 +489,36 @@ nouveau_bo_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) return 0; }
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void +nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) +{ + struct nouveau_device *device; + struct ttm_tt *ttm = nvbo->bo.ttm; + + device = nouveau_dev(nouveau_bdev(ttm->bdev)->dev); + + if (nvbo->bo.ttm && nvbo->bo.ttm->caching_state == tt_cached) + ttm_dma_tt_cache_sync_for_cpu((struct ttm_dma_tt *)nvbo->bo.ttm, + nv_device_base(device)); +} + +void +nouveau_bo_sync_for_device(struct nouveau_bo *nvbo) +{ + struct ttm_tt *ttm = nvbo->bo.ttm; + + if (ttm && ttm->caching_state == tt_cached) { + struct nouveau_device *device; + + device = nouveau_dev(nouveau_bdev(ttm->bdev)->dev); + + ttm_dma_tt_cache_sync_for_device((struct ttm_dma_tt *)ttm, + nv_device_base(device)); + } +} +#endif + static int nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, struct ttm_mem_type_manager *man) diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h index ff17c1f432fc..ead214931223 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.h +++ b/drivers/gpu/drm/nouveau/nouveau_bo.h @@ -89,6 +89,26 @@ int nouveau_bo_vma_add(struct nouveau_bo *, struct nouveau_vm *, struct nouveau_vma *); void nouveau_bo_vma_del(struct nouveau_bo *, struct nouveau_vma *);
+#if IS_ENABLED(CONFIG_ARCH_TEGRA) +#define NOUVEAU_NEED_CACHE_SYNC +#endif + +#ifdef NOUVEAU_NEED_CACHE_SYNC +void nouveau_bo_sync_for_cpu(struct nouveau_bo *); +void nouveau_bo_sync_for_device(struct nouveau_bo *); +#else +static inline void +nouveau_bo_sync_for_cpu(struct nouveau_bo *) +{ +} + +static inline void +nouveau_bo_sync_for_device(struct nouveau_bo *) +{ +} +#endif + + /* TODO: submit equivalent to TTM generic API upstream? */ static inline void __iomem * nvbo_kmap_obj_iovirtual(struct nouveau_bo *nvbo) diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c index c90c0dc0afe8..b7e42fdc9634 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -897,7 +897,13 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data, ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait); spin_unlock(&nvbo->bo.bdev->fence_lock); drm_gem_object_unreference_unlocked(gem); - return ret; + + if (ret) + return ret; + + nouveau_bo_sync_for_cpu(nvbo); + + return 0; }
int
On Mon, May 19, 2014 at 04:10:57PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com
Perhaps having a propery commit message here would be good.
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void +nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) +{
- struct nouveau_device *device;
- struct ttm_tt *ttm = nvbo->bo.ttm;
- device = nouveau_dev(nouveau_bdev(ttm->bdev)->dev);
- if (nvbo->bo.ttm && nvbo->bo.ttm->caching_state == tt_cached)
ttm_dma_tt_cache_sync_for_cpu((struct ttm_dma_tt *)nvbo->bo.ttm,nv_device_base(device));
Can we be certain at this point that the struct ttm_tt is in fact a struct ttm_dma_tt?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
[...]
+#if IS_ENABLED(CONFIG_ARCH_TEGRA) +#define NOUVEAU_NEED_CACHE_SYNC +#endif
I know I gave this as an example myself when we discussed this offline, but I'm now thinking that this might actually be better off in Kconfig.
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void nouveau_bo_sync_for_cpu(struct nouveau_bo *); +void nouveau_bo_sync_for_device(struct nouveau_bo *); +#else +static inline void +nouveau_bo_sync_for_cpu(struct nouveau_bo *) +{ +}
+static inline void +nouveau_bo_sync_for_device(struct nouveau_bo *) +{ +} +#endif
There's a gratuituous blank line here.
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c index c90c0dc0afe8..b7e42fdc9634 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -897,7 +897,13 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data, ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait); spin_unlock(&nvbo->bo.bdev->fence_lock); drm_gem_object_unreference_unlocked(gem);
- return ret;
- if (ret)
return ret;- nouveau_bo_sync_for_cpu(nvbo);
- return 0;
}
This could be rewritten as:
if (!ret) nouveau_bo_sync_for_cpu(nvbo);
return ret;
Which would be slightly shorter.
On second thought, perhaps part of nouveau_gem_ioctl_cpu_prep() could be refactored into a separate function to make this more symmetric. If we put that in nouveau_bo.c and name it nouveau_bo_wait() for example, the dummies can go away and both nouveau_bo_sync_for_{cpu,device}() can be made static. I also think that's cleaner because it has both variants of the nouveau_bo_sync_for_*() calls in the same file.
Thierry
Am Montag, den 19.05.2014, 10:46 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:57PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com
Perhaps having a propery commit message here would be good.
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void +nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) +{
- struct nouveau_device *device;
- struct ttm_tt *ttm = nvbo->bo.ttm;
- device = nouveau_dev(nouveau_bdev(ttm->bdev)->dev);
- if (nvbo->bo.ttm && nvbo->bo.ttm->caching_state == tt_cached)
ttm_dma_tt_cache_sync_for_cpu((struct ttm_dma_tt *)nvbo->bo.ttm,nv_device_base(device));Can we be certain at this point that the struct ttm_tt is in fact a struct ttm_dma_tt?
Yes, for all cases except AGP, where things are mapped WC anyways (so caching_state is always != cached) nouveau_bo_driver uses nouveau_ttm_tt_create() as its ttm_tt_create callback. This in turn calls nouveau_sgdma_create_ttm() which uses ttm_dma_tt_init() unconditionally.
Regards, Lucas
On Mon, May 19, 2014 at 5:46 PM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, May 19, 2014 at 04:10:57PM +0900, Alexandre Courbot wrote:
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com
Perhaps having a propery commit message here would be good.
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void +nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) +{
struct nouveau_device *device;struct ttm_tt *ttm = nvbo->bo.ttm;device = nouveau_dev(nouveau_bdev(ttm->bdev)->dev);if (nvbo->bo.ttm && nvbo->bo.ttm->caching_state == tt_cached)ttm_dma_tt_cache_sync_for_cpu((struct ttm_dma_tt *)nvbo->bo.ttm,nv_device_base(device));Can we be certain at this point that the struct ttm_tt is in fact a struct ttm_dma_tt?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
[...]
+#if IS_ENABLED(CONFIG_ARCH_TEGRA) +#define NOUVEAU_NEED_CACHE_SYNC +#endif
I know I gave this as an example myself when we discussed this offline, but I'm now thinking that this might actually be better off in Kconfig.
+#ifdef NOUVEAU_NEED_CACHE_SYNC +void nouveau_bo_sync_for_cpu(struct nouveau_bo *); +void nouveau_bo_sync_for_device(struct nouveau_bo *); +#else +static inline void +nouveau_bo_sync_for_cpu(struct nouveau_bo *) +{ +}
+static inline void +nouveau_bo_sync_for_device(struct nouveau_bo *) +{ +} +#endif
There's a gratuituous blank line here.
Fixed.
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c index c90c0dc0afe8..b7e42fdc9634 100644 --- a/drivers/gpu/drm/nouveau/nouveau_gem.c +++ b/drivers/gpu/drm/nouveau/nouveau_gem.c @@ -897,7 +897,13 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data, ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait); spin_unlock(&nvbo->bo.bdev->fence_lock); drm_gem_object_unreference_unlocked(gem);
return ret;
if (ret)return ret;nouveau_bo_sync_for_cpu(nvbo);return 0;}
This could be rewritten as:
if (!ret) nouveau_bo_sync_for_cpu(nvbo); return ret;Which would be slightly shorter.
I prefer to have a clear, easy to read code flow here by keeping error-handling within conditions (and not the other way round). This kind of optimization is very well done by the compiler.
On second thought, perhaps part of nouveau_gem_ioctl_cpu_prep() could be refactored into a separate function to make this more symmetric. If we put that in nouveau_bo.c and name it nouveau_bo_wait() for example, the dummies can go away and both nouveau_bo_sync_for_{cpu,device}() can be made static. I also think that's cleaner because it has both variants of the nouveau_bo_sync_for_*() calls in the same file.
Yep, agreed. I will give it a try in the next version of the series.
Thanks, Alex.
Am Montag, den 19.05.2014, 16:10 +0900 schrieb Alexandre Courbot:
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com
drivers/gpu/drm/nouveau/nouveau_bo.c | 32 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_bo.h | 20 ++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_gem.c | 8 +++++++- 3 files changed, 59 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index b6dc85c614be..0886f47e5244 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -407,6 +407,8 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible, { int ret;
- nouveau_bo_sync_for_device(nvbo);
- ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement, interruptible, no_wait_gpu); if (ret)
@@ -487,6 +489,36 @@ nouveau_bo_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) return 0; }
+#ifdef NOUVEAU_NEED_CACHE_SYNC
I don't like this ifdef at all. I know calling this functions will add a little overhead to x86 where it isn't strictly required, but I think it's negligible.
When I looked at them the dma_sync_single_for_[device|cpu] functions which are called here map out to just a drain of the PCI store buffer on x86, which should be fast enough to be done unconditionally. They won't so any time-consuming cache synchronization on PCI coherent arches.
Regards, Lucas
On Mon, May 19, 2014 at 6:31 PM, Lucas Stach l.stach@pengutronix.de wrote:
Am Montag, den 19.05.2014, 16:10 +0900 schrieb Alexandre Courbot:
From: Lucas Stach dev@lynxeye.de
Signed-off-by: Lucas Stach dev@lynxeye.de [acourbot@nvidia.com: make conditional and platform-friendly] Signed-off-by: Alexandre Courbot acourbot@nvidia.com
drivers/gpu/drm/nouveau/nouveau_bo.c | 32 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_bo.h | 20 ++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_gem.c | 8 +++++++- 3 files changed, 59 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index b6dc85c614be..0886f47e5244 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -407,6 +407,8 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible, { int ret;
nouveau_bo_sync_for_device(nvbo);ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement, interruptible, no_wait_gpu); if (ret)@@ -487,6 +489,36 @@ nouveau_bo_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) return 0; }
+#ifdef NOUVEAU_NEED_CACHE_SYNC
I don't like this ifdef at all. I know calling this functions will add a little overhead to x86 where it isn't strictly required, but I think it's negligible.
When I looked at them the dma_sync_single_for_[device|cpu] functions which are called here map out to just a drain of the PCI store buffer on x86, which should be fast enough to be done unconditionally. They won't so any time-consuming cache synchronization on PCI coherent arches.
If Ben agrees with it I am also perfectly fine with getting rid of this #ifdef.
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
This patch introduces a macro that flushes the caches on ARM and translates to a no-op on other architectures, and uses it when writing to in-memory BOs. It will also be useful for implementations of instmem that access shared memory directly instead of going through PRAMIN.
Signed-off-by: Alexandre Courbot acourbot@nvidia.com --- drivers/gpu/drm/nouveau/core/os.h | 17 +++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_bo.c | 8 ++++++-- 2 files changed, 23 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/core/os.h b/drivers/gpu/drm/nouveau/core/os.h index d0ced94ca54c..274b4460bb03 100644 --- a/drivers/gpu/drm/nouveau/core/os.h +++ b/drivers/gpu/drm/nouveau/core/os.h @@ -38,4 +38,21 @@ #endif /* def __BIG_ENDIAN else */ #endif /* !ioread32_native */
+#if defined(__arm__) + +#define nv_cpu_cache_flush_area(va, size) \ +do { \ + phys_addr_t pa = virt_to_phys(va); \ + __cpuc_flush_dcache_area(va, size); \ + outer_flush_range(pa, pa + size); \ +} while (0) + +#else + +#define nv_cpu_cache_flush_area(va, size) \ +do { \ +} while (0) + +#endif /* defined(__arm__) */ + #endif diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem); - else + else { *mem = val; + nv_cpu_cache_flush_area(mem, 2); + } }
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem); - else + else { *mem = val; + nv_cpu_cache_flush_area(mem, 4); + } }
static struct ttm_tt *
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
Also: s/explicitely/explicitly/
This patch introduces a macro that flushes the caches on ARM and translates to a no-op on other architectures, and uses it when writing to in-memory BOs. It will also be useful for implementations of instmem that access shared memory directly instead of going through PRAMIN.
Presumably instmem can access shared memory on all architectures, so this doesn't seem like a property of the architecture but rather of the memory pool backing the instmem.
In that case I wonder if this shouldn't be moved into an operation that is implemented by the backing memory pool and be a noop where the cache doesn't need explicit flushing.
diff --git a/drivers/gpu/drm/nouveau/core/os.h b/drivers/gpu/drm/nouveau/core/os.h index d0ced94ca54c..274b4460bb03 100644 --- a/drivers/gpu/drm/nouveau/core/os.h +++ b/drivers/gpu/drm/nouveau/core/os.h @@ -38,4 +38,21 @@ #endif /* def __BIG_ENDIAN else */ #endif /* !ioread32_native */
+#if defined(__arm__)
+#define nv_cpu_cache_flush_area(va, size) \ +do { \
- phys_addr_t pa = virt_to_phys(va); \
- __cpuc_flush_dcache_area(va, size); \
- outer_flush_range(pa, pa + size); \
+} while (0)
Couldn't this be a static inline function?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
Thierry
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
Also: s/explicitely/explicitly/
This patch introduces a macro that flushes the caches on ARM and translates to a no-op on other architectures, and uses it when writing to in-memory BOs. It will also be useful for implementations of instmem that access shared memory directly instead of going through PRAMIN.
Presumably instmem can access shared memory on all architectures, so this doesn't seem like a property of the architecture but rather of the memory pool backing the instmem.
In that case I wonder if this shouldn't be moved into an operation that is implemented by the backing memory pool and be a noop where the cache doesn't need explicit flushing.
diff --git a/drivers/gpu/drm/nouveau/core/os.h b/drivers/gpu/drm/nouveau/core/os.h index d0ced94ca54c..274b4460bb03 100644 --- a/drivers/gpu/drm/nouveau/core/os.h +++ b/drivers/gpu/drm/nouveau/core/os.h @@ -38,4 +38,21 @@ #endif /* def __BIG_ENDIAN else */ #endif /* !ioread32_native */
+#if defined(__arm__)
+#define nv_cpu_cache_flush_area(va, size) \ +do { \
- phys_addr_t pa = virt_to_phys(va); \
- __cpuc_flush_dcache_area(va, size); \
- outer_flush_range(pa, pa + size); \
+} while (0)
Couldn't this be a static inline function?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
I don't think the explicit flushing for those things makes sense. I think it is a lot more effective to just map the BOs write-combined on PCI non-coherent arches. This way any writes will be buffered. Reads will be slow, but I don't think nouveau is reading back a lot from those buffers. Using the write-combining buffer doesn't need any additional synchronization as it will get flushed on pushbuf kickoff anyways.
Regards, Lucas
On Mon, May 19, 2014 at 11:22:11AM +0200, Lucas Stach wrote:
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
I was criticizing the wording in the commit message. Perhaps it could be enhanced with what you just said.
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
I don't think the explicit flushing for those things makes sense. I think it is a lot more effective to just map the BOs write-combined on PCI non-coherent arches. This way any writes will be buffered. Reads will be slow, but I don't think nouveau is reading back a lot from those buffers. Using the write-combining buffer doesn't need any additional synchronization as it will get flushed on pushbuf kickoff anyways.
Sounds good to me.
Thierry
On Mon, May 19, 2014 at 12:03:17PM +0200, Thierry Reding wrote:
On Mon, May 19, 2014 at 11:22:11AM +0200, Lucas Stach wrote:
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
I was criticizing the wording in the commit message. Perhaps it could be enhanced with what you just said.
Shouldn't this be done in the dma mapping layer? I know that i915 does all the cpu cache flushing itself, but that's because the x86 dma layer refuses to believe that there are non-coherent platforms on x86. But on arm it can cope.
This is somewhat important for dma-buf buffer sharing since if the cpu cache control is done in drivers you must do double-flushing on shared buffers. Atm you have to do that anyway, but at least this would make it easier. The other problem is that ttm reinvents half of the dma mapping functions.
Just my 2 cents. -Daniel
On Mon, May 19, 2014 at 7:03 PM, Thierry Reding thierry.reding@gmail.com wrote:
On Mon, May 19, 2014 at 11:22:11AM +0200, Lucas Stach wrote:
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
I was criticizing the wording in the commit message. Perhaps it could be enhanced with what you just said.
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
I don't think the explicit flushing for those things makes sense. I think it is a lot more effective to just map the BOs write-combined on PCI non-coherent arches. This way any writes will be buffered. Reads will be slow, but I don't think nouveau is reading back a lot from those buffers. Using the write-combining buffer doesn't need any additional synchronization as it will get flushed on pushbuf kickoff anyways.
Sounds good to me.
I will need to wrap my head around TTM some more to understand how to do this the right way, but it is true that brute-forcing in-memory BO mappings to be WC make the addition of nv_cpu_cache_flush_area() unneeded. Is that the direction we want to take with this?
On Mon, May 19, 2014 at 6:22 PM, Lucas Stach l.stach@pengutronix.de wrote:
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
Also: s/explicitely/explicitly/
This patch introduces a macro that flushes the caches on ARM and translates to a no-op on other architectures, and uses it when writing to in-memory BOs. It will also be useful for implementations of instmem that access shared memory directly instead of going through PRAMIN.
Presumably instmem can access shared memory on all architectures, so this doesn't seem like a property of the architecture but rather of the memory pool backing the instmem.
In that case I wonder if this shouldn't be moved into an operation that is implemented by the backing memory pool and be a noop where the cache doesn't need explicit flushing.
diff --git a/drivers/gpu/drm/nouveau/core/os.h b/drivers/gpu/drm/nouveau/core/os.h index d0ced94ca54c..274b4460bb03 100644 --- a/drivers/gpu/drm/nouveau/core/os.h +++ b/drivers/gpu/drm/nouveau/core/os.h @@ -38,4 +38,21 @@ #endif /* def __BIG_ENDIAN else */ #endif /* !ioread32_native */
+#if defined(__arm__)
+#define nv_cpu_cache_flush_area(va, size) \ +do { \
- phys_addr_t pa = virt_to_phys(va); \
- __cpuc_flush_dcache_area(va, size); \
- outer_flush_range(pa, pa + size); \
+} while (0)
Couldn't this be a static inline function?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
I don't think the explicit flushing for those things makes sense. I think it is a lot more effective to just map the BOs write-combined on PCI non-coherent arches. This way any writes will be buffered. Reads will be slow, but I don't think nouveau is reading back a lot from those buffers. Using the write-combining buffer doesn't need any additional synchronization as it will get flushed on pushbuf kickoff anyways.
I tried to go that way, and something interesting happened.
What I did: remove this patch and instead set the following caching parameters for the TTM_PL_TT case in nouveau_bo_init_mem_type():
man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC; man->default_caching = TTM_PL_FLAG_WC;
What happened: no runtime errors as what happened when caching is enabled. However, many of the vertex and texture buffers seem to be partially corrupted. In glmark2 the 3d models had many vertices (but not all) at the wrong position. Note that not all the scenes ended up being corrupted - in particular, when two consecutive scenes used the same model, the second instance would be uncorrupted.
Forcing the caching to TTM_PL_FLAG_UNCACHED led to the same result. What is interesting is that while data like vertices and textures got corrupted, pushbuffers and shader programs seem to be just fine, as I could not see any runtime error.
I don't really understand what kind of caching behavior could lead to that. If anyone has any idea, I'd love to hear.
Thanks, Alex.
On Mon, Jun 9, 2014 at 7:41 PM, Alexandre Courbot gnurou@gmail.com wrote:
On Mon, May 19, 2014 at 6:22 PM, Lucas Stach l.stach@pengutronix.de wrote:
Am Montag, den 19.05.2014, 11:02 +0200 schrieb Thierry Reding:
On Mon, May 19, 2014 at 04:10:58PM +0900, Alexandre Courbot wrote:
Some architectures (e.g. ARM) need the CPU buffers to be explicitely flushed for a memory write to take effect. Not doing so results in synchronization issues, especially after writing to BOs.
It seems to me that the above is generally true for all architectures, not just ARM.
No, on PCI coherent arches, like x86 and some PowerPCs, the GPU will snoop the CPU caches and therefore an explicit cache flush is not required.
Also: s/explicitely/explicitly/
This patch introduces a macro that flushes the caches on ARM and translates to a no-op on other architectures, and uses it when writing to in-memory BOs. It will also be useful for implementations of instmem that access shared memory directly instead of going through PRAMIN.
Presumably instmem can access shared memory on all architectures, so this doesn't seem like a property of the architecture but rather of the memory pool backing the instmem.
In that case I wonder if this shouldn't be moved into an operation that is implemented by the backing memory pool and be a noop where the cache doesn't need explicit flushing.
diff --git a/drivers/gpu/drm/nouveau/core/os.h b/drivers/gpu/drm/nouveau/core/os.h index d0ced94ca54c..274b4460bb03 100644 --- a/drivers/gpu/drm/nouveau/core/os.h +++ b/drivers/gpu/drm/nouveau/core/os.h @@ -38,4 +38,21 @@ #endif /* def __BIG_ENDIAN else */ #endif /* !ioread32_native */
+#if defined(__arm__)
+#define nv_cpu_cache_flush_area(va, size) \ +do { \
- phys_addr_t pa = virt_to_phys(va); \
- __cpuc_flush_dcache_area(va, size); \
- outer_flush_range(pa, pa + size); \
+} while (0)
Couldn't this be a static inline function?
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
[...]
index 0886f47e5244..b9c9729c5733 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -437,8 +437,10 @@ nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) mem = &mem[index]; if (is_iomem) iowrite16_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 2);- }
}
u32 @@ -461,8 +463,10 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) mem = &mem[index]; if (is_iomem) iowrite32_native(val, (void __force __iomem *)mem);
- else
- else { *mem = val;
nv_cpu_cache_flush_area(mem, 4);- }
This looks rather like a sledgehammer to me. Effectively this turns nvbo into an uncached buffer. With additional overhead of constantly flushing caches. Wouldn't it make more sense to locate the places where these are called and flush the cache after all the writes have completed?
I don't think the explicit flushing for those things makes sense. I think it is a lot more effective to just map the BOs write-combined on PCI non-coherent arches. This way any writes will be buffered. Reads will be slow, but I don't think nouveau is reading back a lot from those buffers. Using the write-combining buffer doesn't need any additional synchronization as it will get flushed on pushbuf kickoff anyways.
I tried to go that way, and something interesting happened.
What I did: remove this patch and instead set the following caching parameters for the TTM_PL_TT case in nouveau_bo_init_mem_type():
man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC; man->default_caching = TTM_PL_FLAG_WC;What happened: no runtime errors as what happened when caching is enabled. However, many of the vertex and texture buffers seem to be partially corrupted. In glmark2 the 3d models had many vertices (but not all) at the wrong position. Note that not all the scenes ended up being corrupted - in particular, when two consecutive scenes used the same model, the second instance would be uncorrupted.
Forcing the caching to TTM_PL_FLAG_UNCACHED led to the same result. What is interesting is that while data like vertices and textures got corrupted, pushbuffers and shader programs seem to be just fine, as I could not see any runtime error.
An interesting fact: if I change ttm_bo_kmap_ttm() such as kernel mappings of BOs are always performed write-combined, and leave the TTM_PL_TT default caching to TTM_PL_FLAG_CACHED so user-space mappings remain cached, the corruptions just vanish. It seems to be the fact of setting user-space mappings to anything non-cached that leads to this puzzling behavior. Certainly some subtlety of ARM mappings are getting over my head here.
If we need to implement different policies for kernel and user-space mappings, this might complicate things a bit, especially since support needs to be in TTM and not only Nouveau. I will submit a RFC tomorrow if I don't hear better ideas by then.
Alex.
dri-devel@lists.freedesktop.org
-
Alexandre Courbot -
Alexandre Courbot -
Daniel Vetter -
Lucas Stach -
Thierry Reding