* [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2
@ 2017-03-31 9:47 Christian König
2017-03-31 9:47 ` [PATCH 2/6] drm/ttm: add io_mem_pfn callback Christian König
` (2 more replies)
0 siblings, 3 replies; 15+ messages in thread
From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW,
dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
No need to implement the same logic twice. Also check if the busy placements
are identical to the already scanned placements before checking them.
v2: improve check even more as suggested by Michel.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/ttm/ttm_bo.c | 45 ++++++++++++++++++++++++--------------------
1 file changed, 25 insertions(+), 20 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 989b98b..348b17e 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1046,29 +1046,17 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
return ret;
}
-bool ttm_bo_mem_compat(struct ttm_placement *placement,
- struct ttm_mem_reg *mem,
- uint32_t *new_flags)
+static bool ttm_bo_places_compat(const struct ttm_place *places,
+ unsigned num_placement,
+ struct ttm_mem_reg *mem,
+ uint32_t *new_flags)
{
- int i;
+ unsigned i;
- for (i = 0; i < placement->num_placement; i++) {
- const struct ttm_place *heap = &placement->placement[i];
- if (mem->mm_node &&
- (mem->start < heap->fpfn ||
- (heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn)))
- continue;
+ for (i = 0; i < num_placement; i++) {
+ const struct ttm_place *heap = &places[i];
- *new_flags = heap->flags;
- if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) &&
- (*new_flags & mem->placement & TTM_PL_MASK_MEM))
- return true;
- }
-
- for (i = 0; i < placement->num_busy_placement; i++) {
- const struct ttm_place *heap = &placement->busy_placement[i];
- if (mem->mm_node &&
- (mem->start < heap->fpfn ||
+ if (mem->mm_node && (mem->start < heap->fpfn ||
(heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn)))
continue;
@@ -1077,6 +1065,23 @@ bool ttm_bo_mem_compat(struct ttm_placement *placement,
(*new_flags & mem->placement & TTM_PL_MASK_MEM))
return true;
}
+ return false;
+}
+
+bool ttm_bo_mem_compat(struct ttm_placement *placement,
+ struct ttm_mem_reg *mem,
+ uint32_t *new_flags)
+{
+ if (ttm_bo_places_compat(placement->placement, placement->num_placement,
+ mem, new_flags))
+ return true;
+
+ if ((placement->busy_placement != placement->placement ||
+ placement->num_busy_placement > placement->num_placement) &&
+ ttm_bo_places_compat(placement->busy_placement,
+ placement->num_busy_placement,
+ mem, new_flags))
+ return true;
return false;
}
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread* [PATCH 2/6] drm/ttm: add io_mem_pfn callback 2017-03-31 9:47 [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König @ 2017-03-31 9:47 ` Christian König 2017-03-31 9:47 ` [PATCH 4/6] drm/amdgpu: drop alpha support Christian König [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> 2 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw) To: amd-gfx, dri-devel From: Christian König <christian.koenig@amd.com> This allows the driver to handle io_mem mappings on their own. Signed-off-by: Christian König <christian.koenig@amd.com> Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 1 + drivers/gpu/drm/ast/ast_ttm.c | 1 + drivers/gpu/drm/bochs/bochs_mm.c | 1 + drivers/gpu/drm/cirrus/cirrus_ttm.c | 1 + drivers/gpu/drm/mgag200/mgag200_ttm.c | 1 + drivers/gpu/drm/nouveau/nouveau_bo.c | 1 + drivers/gpu/drm/qxl/qxl_ttm.c | 1 + drivers/gpu/drm/radeon/radeon_ttm.c | 1 + drivers/gpu/drm/ttm/ttm_bo_vm.c | 10 +++++++++- drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 + drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 + include/drm/ttm/ttm_bo_api.h | 11 +++++++++++ include/drm/ttm/ttm_bo_driver.h | 9 +++++++++ 13 files changed, 39 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 21876ee..7bf5ba7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1089,6 +1089,7 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int amdgpu_ttm_init(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/ast/ast_ttm.c b/drivers/gpu/drm/ast/ast_ttm.c index 50c910e..e879496 100644 --- a/drivers/gpu/drm/ast/ast_ttm.c +++ b/drivers/gpu/drm/ast/ast_ttm.c @@ -236,6 +236,7 @@ struct ttm_bo_driver ast_bo_driver = { .verify_access = ast_bo_verify_access, .io_mem_reserve = &ast_ttm_io_mem_reserve, .io_mem_free = &ast_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int ast_mm_init(struct ast_private *ast) diff --git a/drivers/gpu/drm/bochs/bochs_mm.c b/drivers/gpu/drm/bochs/bochs_mm.c index e4c1125..804afbc 100644 --- a/drivers/gpu/drm/bochs/bochs_mm.c +++ b/drivers/gpu/drm/bochs/bochs_mm.c @@ -205,6 +205,7 @@ struct ttm_bo_driver bochs_bo_driver = { .verify_access = bochs_bo_verify_access, .io_mem_reserve = &bochs_ttm_io_mem_reserve, .io_mem_free = &bochs_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int bochs_mm_init(struct bochs_device *bochs) diff --git a/drivers/gpu/drm/cirrus/cirrus_ttm.c b/drivers/gpu/drm/cirrus/cirrus_ttm.c index f53aa8f..93dbcd3 100644 --- a/drivers/gpu/drm/cirrus/cirrus_ttm.c +++ b/drivers/gpu/drm/cirrus/cirrus_ttm.c @@ -236,6 +236,7 @@ struct ttm_bo_driver cirrus_bo_driver = { .verify_access = cirrus_bo_verify_access, .io_mem_reserve = &cirrus_ttm_io_mem_reserve, .io_mem_free = &cirrus_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int cirrus_mm_init(struct cirrus_device *cirrus) diff --git a/drivers/gpu/drm/mgag200/mgag200_ttm.c b/drivers/gpu/drm/mgag200/mgag200_ttm.c index 657598b..565a217 100644 --- a/drivers/gpu/drm/mgag200/mgag200_ttm.c +++ b/drivers/gpu/drm/mgag200/mgag200_ttm.c @@ -236,6 +236,7 @@ struct ttm_bo_driver mgag200_bo_driver = { .verify_access = mgag200_bo_verify_access, .io_mem_reserve = &mgag200_ttm_io_mem_reserve, .io_mem_free = &mgag200_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int mgag200_mm_init(struct mga_device *mdev) diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index 3949a74..978a5e7 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -1570,6 +1570,7 @@ struct ttm_bo_driver nouveau_bo_driver = { .fault_reserve_notify = &nouveau_ttm_fault_reserve_notify, .io_mem_reserve = &nouveau_ttm_io_mem_reserve, .io_mem_free = &nouveau_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; struct nvkm_vma * diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c index 2955f91..28fa56e 100644 --- a/drivers/gpu/drm/qxl/qxl_ttm.c +++ b/drivers/gpu/drm/qxl/qxl_ttm.c @@ -394,6 +394,7 @@ static struct ttm_bo_driver qxl_bo_driver = { .verify_access = &qxl_verify_access, .io_mem_reserve = &qxl_ttm_io_mem_reserve, .io_mem_free = &qxl_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, .move_notify = &qxl_bo_move_notify, }; diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c index 6571384..d07ff84 100644 --- a/drivers/gpu/drm/radeon/radeon_ttm.c +++ b/drivers/gpu/drm/radeon/radeon_ttm.c @@ -873,6 +873,7 @@ static struct ttm_bo_driver radeon_bo_driver = { .fault_reserve_notify = &radeon_bo_fault_reserve_notify, .io_mem_reserve = &radeon_ttm_io_mem_reserve, .io_mem_free = &radeon_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; int radeon_ttm_init(struct radeon_device *rdev) diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index 750733a..be84e3dc 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -231,7 +231,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) */ for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) { if (bo->mem.bus.is_iomem) - pfn = ((bo->mem.bus.base + bo->mem.bus.offset) >> PAGE_SHIFT) + page_offset; + pfn = bdev->driver->io_mem_pfn(bo, page_offset); else { page = ttm->pages[page_offset]; if (unlikely(!page && i == 0)) { @@ -324,6 +324,14 @@ static struct ttm_buffer_object *ttm_bo_vm_lookup(struct ttm_bo_device *bdev, return bo; } +unsigned long ttm_bo_default_io_mem_pfn(struct ttm_buffer_object *bo, + unsigned long page_offset) +{ + return ((bo->mem.bus.base + bo->mem.bus.offset) >> PAGE_SHIFT) + + page_offset; +} +EXPORT_SYMBOL(ttm_bo_default_io_mem_pfn); + int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma, struct ttm_bo_device *bdev) { diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c index a8875a7..8b76bf9 100644 --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c @@ -432,6 +432,7 @@ static struct ttm_bo_driver virtio_gpu_bo_driver = { .verify_access = &virtio_gpu_verify_access, .io_mem_reserve = &virtio_gpu_ttm_io_mem_reserve, .io_mem_free = &virtio_gpu_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, .move_notify = &virtio_gpu_bo_move_notify, .swap_notify = &virtio_gpu_bo_swap_notify, }; diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c index 1de9669..ba2fa7b 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c @@ -859,4 +859,5 @@ struct ttm_bo_driver vmw_bo_driver = { .fault_reserve_notify = &vmw_ttm_fault_reserve_notify, .io_mem_reserve = &vmw_ttm_io_mem_reserve, .io_mem_free = &vmw_ttm_io_mem_free, + .io_mem_pfn = ttm_bo_default_io_mem_pfn, }; diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 2d0f63e..3b302a5 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -724,6 +724,17 @@ extern int ttm_fbdev_mmap(struct vm_area_struct *vma, struct ttm_buffer_object *bo); /** + * ttm_bo_default_iomem_pfn - get a pfn for a page offset + * + * @bo: the BO we need to look up the pfn for + * @page_offset: offset inside the BO to look up. + * + * Calculate the PFN for iomem based mappings during page fault + */ +unsigned long ttm_bo_default_io_mem_pfn(struct ttm_buffer_object *bo, + unsigned long page_offset); + +/** * ttm_bo_mmap - mmap out of the ttm device address space. * * @filp: filp as input from the mmap method. diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 7777d14..cb5072a 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -462,6 +462,15 @@ struct ttm_bo_driver { struct ttm_mem_reg *mem); void (*io_mem_free)(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem); + + /** + * Return the pfn for a given page_offset inside the BO. + * + * @bo: the BO to look up the pfn for + * @page_offset: the offset to look up + */ + unsigned long (*io_mem_pfn)(struct ttm_buffer_object *bo, + unsigned long page_offset); }; /** -- 2.5.0 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 4/6] drm/amdgpu: drop alpha support 2017-03-31 9:47 [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König 2017-03-31 9:47 ` [PATCH 2/6] drm/ttm: add io_mem_pfn callback Christian König @ 2017-03-31 9:47 ` Christian König [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> 2 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw) To: amd-gfx, dri-devel From: Christian König <christian.koenig@amd.com> We will probably never see this combination. Signed-off-by: Christian König <christian.koenig@amd.com> Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 25 ------------------------- 1 file changed, 25 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 7bf5ba7..524abca 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -538,31 +538,6 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ return -EINVAL; mem->bus.base = adev->mc.aper_base; mem->bus.is_iomem = true; -#ifdef __alpha__ - /* - * Alpha: use bus.addr to hold the ioremap() return, - * so we can modify bus.base below. - */ - if (mem->placement & TTM_PL_FLAG_WC) - mem->bus.addr = - ioremap_wc(mem->bus.base + mem->bus.offset, - mem->bus.size); - else - mem->bus.addr = - ioremap_nocache(mem->bus.base + mem->bus.offset, - mem->bus.size); - if (!mem->bus.addr) - return -ENOMEM; - - /* - * Alpha: Use just the bus offset plus - * the hose/domain memory base for bus.base. - * It then can be used to build PTEs for VRAM - * access, as done in ttm_bo_vm_fault(). - */ - mem->bus.base = (mem->bus.base & 0x0ffffffffUL) + - adev->ddev->hose->dense_mem_base; -#endif break; default: return -EINVAL; -- 2.5.0 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 15+ messages in thread
[parent not found: <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>]
* [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> @ 2017-03-31 9:47 ` Christian König 2017-04-03 16:27 ` Nicolai Hähnle 2017-03-31 9:47 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König ` (2 subsequent siblings) 3 siblings, 1 reply; 15+ messages in thread From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Christian König <christian.koenig@amd.com> This allows drivers to specify if they need a contiguous allocation or not. v2: use space instead of tab Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/ttm/ttm_bo.c | 4 +++- include/drm/ttm/ttm_placement.h | 1 + 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 348b17e..da37cdf 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -1062,7 +1062,9 @@ static bool ttm_bo_places_compat(const struct ttm_place *places, *new_flags = heap->flags; if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && - (*new_flags & mem->placement & TTM_PL_MASK_MEM)) + (*new_flags & mem->placement & TTM_PL_MASK_MEM) && + (!(*new_flags & TTM_PL_FLAG_CONTIGUOUS) || + (mem->placement & TTM_PL_FLAG_CONTIGUOUS))) return true; } return false; diff --git a/include/drm/ttm/ttm_placement.h b/include/drm/ttm/ttm_placement.h index 932be0c..e88a8e3 100644 --- a/include/drm/ttm/ttm_placement.h +++ b/include/drm/ttm/ttm_placement.h @@ -63,6 +63,7 @@ #define TTM_PL_FLAG_CACHED (1 << 16) #define TTM_PL_FLAG_UNCACHED (1 << 17) #define TTM_PL_FLAG_WC (1 << 18) +#define TTM_PL_FLAG_CONTIGUOUS (1 << 19) #define TTM_PL_FLAG_NO_EVICT (1 << 21) #define TTM_PL_FLAG_TOPDOWN (1 << 22) -- 2.5.0 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 15+ messages in thread
* Re: [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 2017-03-31 9:47 ` [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 Christian König @ 2017-04-03 16:27 ` Nicolai Hähnle 0 siblings, 0 replies; 15+ messages in thread From: Nicolai Hähnle @ 2017-04-03 16:27 UTC (permalink / raw) To: Christian König, amd-gfx, dri-devel On 31.03.2017 11:47, Christian König wrote: > From: Christian König <christian.koenig@amd.com> > > This allows drivers to specify if they need a contiguous allocation or not. > > v2: use space instead of tab > > Signed-off-by: Christian König <christian.koenig@amd.com> Patches 1-3: Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> Patch 4: Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com> > --- > drivers/gpu/drm/ttm/ttm_bo.c | 4 +++- > include/drm/ttm/ttm_placement.h | 1 + > 2 files changed, 4 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c > index 348b17e..da37cdf 100644 > --- a/drivers/gpu/drm/ttm/ttm_bo.c > +++ b/drivers/gpu/drm/ttm/ttm_bo.c > @@ -1062,7 +1062,9 @@ static bool ttm_bo_places_compat(const struct ttm_place *places, > > *new_flags = heap->flags; > if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && > - (*new_flags & mem->placement & TTM_PL_MASK_MEM)) > + (*new_flags & mem->placement & TTM_PL_MASK_MEM) && > + (!(*new_flags & TTM_PL_FLAG_CONTIGUOUS) || > + (mem->placement & TTM_PL_FLAG_CONTIGUOUS))) > return true; > } > return false; > diff --git a/include/drm/ttm/ttm_placement.h b/include/drm/ttm/ttm_placement.h > index 932be0c..e88a8e3 100644 > --- a/include/drm/ttm/ttm_placement.h > +++ b/include/drm/ttm/ttm_placement.h > @@ -63,6 +63,7 @@ > #define TTM_PL_FLAG_CACHED (1 << 16) > #define TTM_PL_FLAG_UNCACHED (1 << 17) > #define TTM_PL_FLAG_WC (1 << 18) > +#define TTM_PL_FLAG_CONTIGUOUS (1 << 19) > #define TTM_PL_FLAG_NO_EVICT (1 << 21) > #define TTM_PL_FLAG_TOPDOWN (1 << 22) > > -- Lerne, wie die Welt wirklich ist, Aber vergiss niemals, wie sie sein sollte. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 15+ messages in thread
* [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> 2017-03-31 9:47 ` [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 Christian König @ 2017-03-31 9:47 ` Christian König [not found] ` <1490953652-3703-5-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> 2017-03-31 9:47 ` [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers Christian König 2017-04-03 12:37 ` [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König 3 siblings, 1 reply; 15+ messages in thread From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Christian König <christian.koenig@amd.com> Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS instead of a placement limit. That allows us to better handle CPU accessible placements. Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- 2 files changed, 15 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index d6b2de9..387d190 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev, if (domain & AMDGPU_GEM_DOMAIN_VRAM) { unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT; - unsigned lpfn = 0; - - /* This forces a reallocation if the flag wasn't set before */ - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; places[c].fpfn = 0; - places[c].lpfn = lpfn; + places[c].lpfn = 0; places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM; + if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) places[c].lpfn = visible_pfn; else places[c].flags |= TTM_PL_FLAG_TOPDOWN; + + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; c++; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index d710226..af2d172 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, const struct ttm_place *place, struct ttm_mem_reg *mem) { - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); struct amdgpu_vram_mgr *mgr = man->priv; struct drm_mm *mm = &mgr->mm; struct drm_mm_node *nodes; @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (!lpfn) lpfn = man->size; - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || - place->lpfn || amdgpu_vram_page_split == -1) { + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || + amdgpu_vram_page_split == -1) { pages_per_node = ~0ul; num_nodes = 1; } else { @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, aflags = DRM_MM_CREATE_TOP; } + mem->start = 0; pages_left = mem->num_pages; spin_lock(&mgr->lock); for (i = 0; i < num_nodes; ++i) { unsigned long pages = min(pages_left, pages_per_node); uint32_t alignment = mem->page_alignment; + unsigned long start; if (pages == pages_per_node) alignment = pages_per_node; @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (unlikely(r)) goto error; + /* + * Calculate a virtual BO start address to easily check if + * everything is CPU accessible. + */ + start = nodes[i].start + nodes[i].size - mem->num_pages; + mem->start = max(mem->start, start); pages_left -= pages; } spin_unlock(&mgr->lock); - mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET; mem->mm_node = nodes; return 0; -- 2.5.0 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 15+ messages in thread
[parent not found: <1490953652-3703-5-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>]
* Re: [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS [not found] ` <1490953652-3703-5-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> @ 2017-04-03 16:22 ` Nicolai Hähnle [not found] ` <425dfa9e-c86b-6958-f7ff-31b91a0f9e21-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 0 siblings, 1 reply; 15+ messages in thread From: Nicolai Hähnle @ 2017-04-03 16:22 UTC (permalink / raw) To: Christian König, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW On 31.03.2017 11:47, Christian König wrote: > From: Christian König <christian.koenig@amd.com> > > Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS > instead of a placement limit. That allows us to better handle CPU > accessible placements. > > Signed-off-by: Christian König <christian.koenig@amd.com> > Acked-by: Michel Dänzer <michel.daenzer@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ > drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- > 2 files changed, 15 insertions(+), 10 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > index d6b2de9..387d190 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev, > > if (domain & AMDGPU_GEM_DOMAIN_VRAM) { > unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT; > - unsigned lpfn = 0; > - > - /* This forces a reallocation if the flag wasn't set before */ > - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) > - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; > > places[c].fpfn = 0; > - places[c].lpfn = lpfn; > + places[c].lpfn = 0; > places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | > TTM_PL_FLAG_VRAM; > + > if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) > places[c].lpfn = visible_pfn; > else > places[c].flags |= TTM_PL_FLAG_TOPDOWN; > + > + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) > + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; > c++; > } > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > index d710226..af2d172 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, > const struct ttm_place *place, > struct ttm_mem_reg *mem) > { > - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); > struct amdgpu_vram_mgr *mgr = man->priv; > struct drm_mm *mm = &mgr->mm; > struct drm_mm_node *nodes; > @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, > if (!lpfn) > lpfn = man->size; > > - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || > - place->lpfn || amdgpu_vram_page_split == -1) { > + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || > + amdgpu_vram_page_split == -1) { > pages_per_node = ~0ul; > num_nodes = 1; > } else { > @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, > aflags = DRM_MM_CREATE_TOP; > } > > + mem->start = 0; > pages_left = mem->num_pages; > > spin_lock(&mgr->lock); > for (i = 0; i < num_nodes; ++i) { > unsigned long pages = min(pages_left, pages_per_node); > uint32_t alignment = mem->page_alignment; > + unsigned long start; > > if (pages == pages_per_node) > alignment = pages_per_node; > @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, > if (unlikely(r)) > goto error; > > + /* > + * Calculate a virtual BO start address to easily check if > + * everything is CPU accessible. > + */ > + start = nodes[i].start + nodes[i].size - mem->num_pages; This might wrap around (be a signed negative number), completely breaking the max() logic below. > + mem->start = max(mem->start, start); > pages_left -= pages; > } > spin_unlock(&mgr->lock); > > - mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET; If we're going to abuse mem->start anyway, might I suggest just keeping track of max(nodes[i].start + nodes[i].size), and then setting mem->start to a magic (macro'd) constant based on whether everything is in visible VRAM or not? Then the check in amdgpu_ttm_io_mem_reserve could be simplified accordingly. Also, I think patches #6 and #5 should be exchanged, otherwise there's a temporary bug in handling split visible VRAM buffers. Cheers, Nicolai > mem->mm_node = nodes; > > return 0; > -- Lerne, wie die Welt wirklich ist, Aber vergiss niemals, wie sie sein sollte. _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 15+ messages in thread
[parent not found: <425dfa9e-c86b-6958-f7ff-31b91a0f9e21-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]
* Re: [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS [not found] ` <425dfa9e-c86b-6958-f7ff-31b91a0f9e21-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> @ 2017-04-04 11:33 ` Christian König 2017-04-04 14:24 ` Nicolai Hähnle 0 siblings, 1 reply; 15+ messages in thread From: Christian König @ 2017-04-04 11:33 UTC (permalink / raw) To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Am 03.04.2017 um 18:22 schrieb Nicolai Hähnle: > On 31.03.2017 11:47, Christian König wrote: >> From: Christian König <christian.koenig@amd.com> >> >> Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS >> instead of a placement limit. That allows us to better handle CPU >> accessible placements. >> >> Signed-off-by: Christian König <christian.koenig@amd.com> >> Acked-by: Michel Dänzer <michel.daenzer@amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ >> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- >> 2 files changed, 15 insertions(+), 10 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >> index d6b2de9..387d190 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >> @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct >> amdgpu_device *adev, >> >> if (domain & AMDGPU_GEM_DOMAIN_VRAM) { >> unsigned visible_pfn = adev->mc.visible_vram_size >> >> PAGE_SHIFT; >> - unsigned lpfn = 0; >> - >> - /* This forces a reallocation if the flag wasn't set before */ >> - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) >> - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; >> >> places[c].fpfn = 0; >> - places[c].lpfn = lpfn; >> + places[c].lpfn = 0; >> places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | >> TTM_PL_FLAG_VRAM; >> + >> if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) >> places[c].lpfn = visible_pfn; >> else >> places[c].flags |= TTM_PL_FLAG_TOPDOWN; >> + >> + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) >> + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; >> c++; >> } >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> index d710226..af2d172 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct >> ttm_mem_type_manager *man, >> const struct ttm_place *place, >> struct ttm_mem_reg *mem) >> { >> - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); >> struct amdgpu_vram_mgr *mgr = man->priv; >> struct drm_mm *mm = &mgr->mm; >> struct drm_mm_node *nodes; >> @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct >> ttm_mem_type_manager *man, >> if (!lpfn) >> lpfn = man->size; >> >> - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || >> - place->lpfn || amdgpu_vram_page_split == -1) { >> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || >> + amdgpu_vram_page_split == -1) { >> pages_per_node = ~0ul; >> num_nodes = 1; >> } else { >> @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct >> ttm_mem_type_manager *man, >> aflags = DRM_MM_CREATE_TOP; >> } >> >> + mem->start = 0; >> pages_left = mem->num_pages; >> >> spin_lock(&mgr->lock); >> for (i = 0; i < num_nodes; ++i) { >> unsigned long pages = min(pages_left, pages_per_node); >> uint32_t alignment = mem->page_alignment; >> + unsigned long start; >> >> if (pages == pages_per_node) >> alignment = pages_per_node; >> @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct >> ttm_mem_type_manager *man, >> if (unlikely(r)) >> goto error; >> >> + /* >> + * Calculate a virtual BO start address to easily check if >> + * everything is CPU accessible. >> + */ >> + start = nodes[i].start + nodes[i].size - mem->num_pages; > > This might wrap around (be a signed negative number), completely > breaking the max() logic below. Good point, going to fix that. > >> + mem->start = max(mem->start, start); >> pages_left -= pages; >> } >> spin_unlock(&mgr->lock); >> >> - mem->start = num_nodes == 1 ? nodes[0].start : >> AMDGPU_BO_INVALID_OFFSET; > > If we're going to abuse mem->start anyway, might I suggest just > keeping track of max(nodes[i].start + nodes[i].size), and then setting > mem->start to a magic (macro'd) constant based on whether everything > is in visible VRAM or not? > No, that would break in kernel mappings. > Then the check in amdgpu_ttm_io_mem_reserve could be simplified > accordingly. > > Also, I think patches #6 and #5 should be exchanged, otherwise there's > a temporary bug in handling split visible VRAM buffers. Hui? Why? Patch #6 enables the whole thing by not making the contiguous flag mandatory for CPU mappings any more. Switching those would cause problems with detecting when a BO is not in visible VRAM. Regards, Christian. _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS 2017-04-04 11:33 ` Christian König @ 2017-04-04 14:24 ` Nicolai Hähnle 0 siblings, 0 replies; 15+ messages in thread From: Nicolai Hähnle @ 2017-04-04 14:24 UTC (permalink / raw) To: Christian König, amd-gfx, dri-devel On 04.04.2017 13:33, Christian König wrote: > Am 03.04.2017 um 18:22 schrieb Nicolai Hähnle: >> On 31.03.2017 11:47, Christian König wrote: >>> From: Christian König <christian.koenig@amd.com> >>> >>> Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS >>> instead of a placement limit. That allows us to better handle CPU >>> accessible placements. >>> >>> Signed-off-by: Christian König <christian.koenig@amd.com> >>> Acked-by: Michel Dänzer <michel.daenzer@amd.com> >>> --- >>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ >>> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- >>> 2 files changed, 15 insertions(+), 10 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >>> index d6b2de9..387d190 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c >>> @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct >>> amdgpu_device *adev, >>> >>> if (domain & AMDGPU_GEM_DOMAIN_VRAM) { >>> unsigned visible_pfn = adev->mc.visible_vram_size >> >>> PAGE_SHIFT; >>> - unsigned lpfn = 0; >>> - >>> - /* This forces a reallocation if the flag wasn't set before */ >>> - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) >>> - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; >>> >>> places[c].fpfn = 0; >>> - places[c].lpfn = lpfn; >>> + places[c].lpfn = 0; >>> places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | >>> TTM_PL_FLAG_VRAM; >>> + >>> if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) >>> places[c].lpfn = visible_pfn; >>> else >>> places[c].flags |= TTM_PL_FLAG_TOPDOWN; >>> + >>> + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) >>> + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; >>> c++; >>> } >>> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> index d710226..af2d172 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct >>> ttm_mem_type_manager *man, >>> const struct ttm_place *place, >>> struct ttm_mem_reg *mem) >>> { >>> - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); >>> struct amdgpu_vram_mgr *mgr = man->priv; >>> struct drm_mm *mm = &mgr->mm; >>> struct drm_mm_node *nodes; >>> @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct >>> ttm_mem_type_manager *man, >>> if (!lpfn) >>> lpfn = man->size; >>> >>> - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || >>> - place->lpfn || amdgpu_vram_page_split == -1) { >>> + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || >>> + amdgpu_vram_page_split == -1) { >>> pages_per_node = ~0ul; >>> num_nodes = 1; >>> } else { >>> @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct >>> ttm_mem_type_manager *man, >>> aflags = DRM_MM_CREATE_TOP; >>> } >>> >>> + mem->start = 0; >>> pages_left = mem->num_pages; >>> >>> spin_lock(&mgr->lock); >>> for (i = 0; i < num_nodes; ++i) { >>> unsigned long pages = min(pages_left, pages_per_node); >>> uint32_t alignment = mem->page_alignment; >>> + unsigned long start; >>> >>> if (pages == pages_per_node) >>> alignment = pages_per_node; >>> @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct >>> ttm_mem_type_manager *man, >>> if (unlikely(r)) >>> goto error; >>> >>> + /* >>> + * Calculate a virtual BO start address to easily check if >>> + * everything is CPU accessible. >>> + */ >>> + start = nodes[i].start + nodes[i].size - mem->num_pages; >> >> This might wrap around (be a signed negative number), completely >> breaking the max() logic below. > > Good point, going to fix that. > >> >>> + mem->start = max(mem->start, start); >>> pages_left -= pages; >>> } >>> spin_unlock(&mgr->lock); >>> >>> - mem->start = num_nodes == 1 ? nodes[0].start : >>> AMDGPU_BO_INVALID_OFFSET; >> >> If we're going to abuse mem->start anyway, might I suggest just >> keeping track of max(nodes[i].start + nodes[i].size), and then setting >> mem->start to a magic (macro'd) constant based on whether everything >> is in visible VRAM or not? >> > > No, that would break in kernel mappings. > >> Then the check in amdgpu_ttm_io_mem_reserve could be simplified >> accordingly. >> >> Also, I think patches #6 and #5 should be exchanged, otherwise there's >> a temporary bug in handling split visible VRAM buffers. > > Hui? Why? Patch #6 enables the whole thing by not making the contiguous > flag mandatory for CPU mappings any more. Ah, I missed the fact that it's guarded by the check in amdgpu_bo_fault_reserve_notify. You're right, the order of patches is good. Cheers, Nicolai > > Switching those would cause problems with detecting when a BO is not in > visible VRAM. > > Regards, > Christian. > -- Lerne, wie die Welt wirklich ist, Aber vergiss niemals, wie sie sein sollte. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 15+ messages in thread
* [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> 2017-03-31 9:47 ` [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 Christian König 2017-03-31 9:47 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König @ 2017-03-31 9:47 ` Christian König 2017-04-03 16:25 ` Nicolai Hähnle 2017-04-03 12:37 ` [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König 3 siblings, 1 reply; 15+ messages in thread From: Christian König @ 2017-03-31 9:47 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Christian König <christian.koenig@amd.com> This avoids merging them together on page fault. Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 +--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 ++++++++++++---- 2 files changed, 13 insertions(+), 7 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 387d190..10237a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -927,8 +927,7 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo) size = bo->mem.num_pages << PAGE_SHIFT; offset = bo->mem.start << PAGE_SHIFT; /* TODO: figure out how to map scattered VRAM to the CPU */ - if ((offset + size) <= adev->mc.visible_vram_size && - (abo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)) + if ((offset + size) <= adev->mc.visible_vram_size) return 0; /* Can't move a pinned BO to visible VRAM */ @@ -936,7 +935,6 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo) return -EINVAL; /* hurrah the memory is not visible ! */ - abo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS; amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM); lpfn = adev->mc.visible_vram_size >> PAGE_SHIFT; for (i = 0; i < abo->placement.num_placement; i++) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 524abca..10b793a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -529,9 +529,6 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ case TTM_PL_TT: break; case TTM_PL_VRAM: - if (mem->start == AMDGPU_BO_INVALID_OFFSET) - return -EINVAL; - mem->bus.offset = mem->start << PAGE_SHIFT; /* check if it's visible */ if ((mem->bus.offset + mem->bus.size) > adev->mc.visible_vram_size) @@ -549,6 +546,17 @@ static void amdgpu_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_re { } +static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo, + unsigned long page_offset) +{ + struct drm_mm_node *mm = bo->mem.mm_node; + uint64_t size = mm->size; + + mm += page_offset / size; + page_offset %= size; + return (bo->mem.bus.base >> PAGE_SHIFT) + mm->start + page_offset; +} + /* * TTM backend functions. */ @@ -1064,7 +1072,7 @@ static struct ttm_bo_driver amdgpu_bo_driver = { .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_free = &amdgpu_ttm_io_mem_free, - .io_mem_pfn = ttm_bo_default_io_mem_pfn, + .io_mem_pfn = amdgpu_ttm_io_mem_pfn, }; int amdgpu_ttm_init(struct amdgpu_device *adev) -- 2.5.0 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 15+ messages in thread
* Re: [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers 2017-03-31 9:47 ` [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers Christian König @ 2017-04-03 16:25 ` Nicolai Hähnle 2017-04-04 11:30 ` Christian König 0 siblings, 1 reply; 15+ messages in thread From: Nicolai Hähnle @ 2017-04-03 16:25 UTC (permalink / raw) To: Christian König, amd-gfx, dri-devel On 31.03.2017 11:47, Christian König wrote: > From: Christian König <christian.koenig@amd.com> > > This avoids merging them together on page fault. > > Signed-off-by: Christian König <christian.koenig@amd.com> > Acked-by: Michel Dänzer <michel.daenzer@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 +--- > drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 ++++++++++++---- > 2 files changed, 13 insertions(+), 7 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > index 387d190..10237a8 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c > @@ -927,8 +927,7 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo) > size = bo->mem.num_pages << PAGE_SHIFT; > offset = bo->mem.start << PAGE_SHIFT; > /* TODO: figure out how to map scattered VRAM to the CPU */ > - if ((offset + size) <= adev->mc.visible_vram_size && > - (abo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)) > + if ((offset + size) <= adev->mc.visible_vram_size) > return 0; > > /* Can't move a pinned BO to visible VRAM */ > @@ -936,7 +935,6 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo) > return -EINVAL; > > /* hurrah the memory is not visible ! */ > - abo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS; > amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM); > lpfn = adev->mc.visible_vram_size >> PAGE_SHIFT; > for (i = 0; i < abo->placement.num_placement; i++) { > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > index 524abca..10b793a 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c > @@ -529,9 +529,6 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ > case TTM_PL_TT: > break; > case TTM_PL_VRAM: > - if (mem->start == AMDGPU_BO_INVALID_OFFSET) > - return -EINVAL; > - > mem->bus.offset = mem->start << PAGE_SHIFT; > /* check if it's visible */ > if ((mem->bus.offset + mem->bus.size) > adev->mc.visible_vram_size) I believe the various mem->bus members are now unused, aren't they? Unless I missed something, it's best to clean this up and no longer set them (or set them to some obvious poison values). Cheers, Nicolai > @@ -549,6 +546,17 @@ static void amdgpu_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_re > { > } > > +static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo, > + unsigned long page_offset) > +{ > + struct drm_mm_node *mm = bo->mem.mm_node; > + uint64_t size = mm->size; > + > + mm += page_offset / size; > + page_offset %= size; > + return (bo->mem.bus.base >> PAGE_SHIFT) + mm->start + page_offset; > +} > + > /* > * TTM backend functions. > */ > @@ -1064,7 +1072,7 @@ static struct ttm_bo_driver amdgpu_bo_driver = { > .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, > .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, > .io_mem_free = &amdgpu_ttm_io_mem_free, > - .io_mem_pfn = ttm_bo_default_io_mem_pfn, > + .io_mem_pfn = amdgpu_ttm_io_mem_pfn, > }; > > int amdgpu_ttm_init(struct amdgpu_device *adev) > -- Lerne, wie die Welt wirklich ist, Aber vergiss niemals, wie sie sein sollte. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers 2017-04-03 16:25 ` Nicolai Hähnle @ 2017-04-04 11:30 ` Christian König 0 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-04-04 11:30 UTC (permalink / raw) To: Nicolai Hähnle, amd-gfx, dri-devel Am 03.04.2017 um 18:25 schrieb Nicolai Hähnle: > On 31.03.2017 11:47, Christian König wrote: >> From: Christian König <christian.koenig@amd.com> >> [SNIP] >> mem->bus.offset = mem->start << PAGE_SHIFT; >> /* check if it's visible */ >> if ((mem->bus.offset + mem->bus.size) > >> adev->mc.visible_vram_size) > > I believe the various mem->bus members are now unused, aren't they? > Unless I missed something, it's best to clean this up and no longer > set them (or set them to some obvious poison values). No, they are still needed for in kernel mappings. I only handle split mappings from userspace with the new callback. Regards, Christian. _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 [not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> ` (2 preceding siblings ...) 2017-03-31 9:47 ` [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers Christian König @ 2017-04-03 12:37 ` Christian König 3 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-04-03 12:37 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Daenzer, Michel, Haehnle, Nicolai, Alex Deucher Ping! Any more comments on that set or can I push it? Thanks, Christian. Am 31.03.2017 um 11:47 schrieb Christian König: > From: Christian König <christian.koenig@amd.com> > > No need to implement the same logic twice. Also check if the busy placements > are identical to the already scanned placements before checking them. > > v2: improve check even more as suggested by Michel. > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/gpu/drm/ttm/ttm_bo.c | 45 ++++++++++++++++++++++++-------------------- > 1 file changed, 25 insertions(+), 20 deletions(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c > index 989b98b..348b17e 100644 > --- a/drivers/gpu/drm/ttm/ttm_bo.c > +++ b/drivers/gpu/drm/ttm/ttm_bo.c > @@ -1046,29 +1046,17 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo, > return ret; > } > > -bool ttm_bo_mem_compat(struct ttm_placement *placement, > - struct ttm_mem_reg *mem, > - uint32_t *new_flags) > +static bool ttm_bo_places_compat(const struct ttm_place *places, > + unsigned num_placement, > + struct ttm_mem_reg *mem, > + uint32_t *new_flags) > { > - int i; > + unsigned i; > > - for (i = 0; i < placement->num_placement; i++) { > - const struct ttm_place *heap = &placement->placement[i]; > - if (mem->mm_node && > - (mem->start < heap->fpfn || > - (heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn))) > - continue; > + for (i = 0; i < num_placement; i++) { > + const struct ttm_place *heap = &places[i]; > > - *new_flags = heap->flags; > - if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && > - (*new_flags & mem->placement & TTM_PL_MASK_MEM)) > - return true; > - } > - > - for (i = 0; i < placement->num_busy_placement; i++) { > - const struct ttm_place *heap = &placement->busy_placement[i]; > - if (mem->mm_node && > - (mem->start < heap->fpfn || > + if (mem->mm_node && (mem->start < heap->fpfn || > (heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn))) > continue; > > @@ -1077,6 +1065,23 @@ bool ttm_bo_mem_compat(struct ttm_placement *placement, > (*new_flags & mem->placement & TTM_PL_MASK_MEM)) > return true; > } > + return false; > +} > + > +bool ttm_bo_mem_compat(struct ttm_placement *placement, > + struct ttm_mem_reg *mem, > + uint32_t *new_flags) > +{ > + if (ttm_bo_places_compat(placement->placement, placement->num_placement, > + mem, new_flags)) > + return true; > + > + if ((placement->busy_placement != placement->placement || > + placement->num_busy_placement > placement->num_placement) && > + ttm_bo_places_compat(placement->busy_placement, > + placement->num_busy_placement, > + mem, new_flags)) > + return true; > > return false; > } _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 15+ messages in thread
* [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat @ 2017-03-29 17:43 Christian König 2017-03-29 17:43 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König 0 siblings, 1 reply; 15+ messages in thread From: Christian König @ 2017-03-29 17:43 UTC (permalink / raw) To: dri-devel From: Christian König <christian.koenig@amd.com> No need to implement the same logic twice. Also check if the busy placements are identical to the already scanned placements before checking them. Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/ttm/ttm_bo.c | 45 ++++++++++++++++++++++++-------------------- 1 file changed, 25 insertions(+), 20 deletions(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 989b98b..96b1450 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -1046,29 +1046,17 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo, return ret; } -bool ttm_bo_mem_compat(struct ttm_placement *placement, - struct ttm_mem_reg *mem, - uint32_t *new_flags) +static bool ttm_bo_places_compat(const struct ttm_place *places, + unsigned num_placement, + struct ttm_mem_reg *mem, + uint32_t *new_flags) { - int i; + unsigned i; - for (i = 0; i < placement->num_placement; i++) { - const struct ttm_place *heap = &placement->placement[i]; - if (mem->mm_node && - (mem->start < heap->fpfn || - (heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn))) - continue; + for (i = 0; i < num_placement; i++) { + const struct ttm_place *heap = &places[i]; - *new_flags = heap->flags; - if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) && - (*new_flags & mem->placement & TTM_PL_MASK_MEM)) - return true; - } - - for (i = 0; i < placement->num_busy_placement; i++) { - const struct ttm_place *heap = &placement->busy_placement[i]; - if (mem->mm_node && - (mem->start < heap->fpfn || + if (mem->mm_node && (mem->start < heap->fpfn || (heap->lpfn != 0 && (mem->start + mem->num_pages) > heap->lpfn))) continue; @@ -1077,6 +1065,23 @@ bool ttm_bo_mem_compat(struct ttm_placement *placement, (*new_flags & mem->placement & TTM_PL_MASK_MEM)) return true; } + return false; +} + +bool ttm_bo_mem_compat(struct ttm_placement *placement, + struct ttm_mem_reg *mem, + uint32_t *new_flags) +{ + if (ttm_bo_places_compat(placement->placement, placement->num_placement, + mem, new_flags)) + return true; + + if ((placement->busy_placement != placement->placement || + placement->num_busy_placement != placement->num_placement) && + ttm_bo_places_compat(placement->busy_placement, + placement->num_busy_placement, + mem, new_flags)) + return true; return false; } -- 2.5.0 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS 2017-03-29 17:43 [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat Christian König @ 2017-03-29 17:43 ` Christian König 0 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-03-29 17:43 UTC (permalink / raw) To: dri-devel From: Christian König <christian.koenig@amd.com> Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS instead of a placement limit. That allows us to better handle CPU accessible placements. Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- 2 files changed, 15 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index d6b2de9..387d190 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev, if (domain & AMDGPU_GEM_DOMAIN_VRAM) { unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT; - unsigned lpfn = 0; - - /* This forces a reallocation if the flag wasn't set before */ - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; places[c].fpfn = 0; - places[c].lpfn = lpfn; + places[c].lpfn = 0; places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM; + if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) places[c].lpfn = visible_pfn; else places[c].flags |= TTM_PL_FLAG_TOPDOWN; + + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; c++; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index d710226..af2d172 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, const struct ttm_place *place, struct ttm_mem_reg *mem) { - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); struct amdgpu_vram_mgr *mgr = man->priv; struct drm_mm *mm = &mgr->mm; struct drm_mm_node *nodes; @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (!lpfn) lpfn = man->size; - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || - place->lpfn || amdgpu_vram_page_split == -1) { + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || + amdgpu_vram_page_split == -1) { pages_per_node = ~0ul; num_nodes = 1; } else { @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, aflags = DRM_MM_CREATE_TOP; } + mem->start = 0; pages_left = mem->num_pages; spin_lock(&mgr->lock); for (i = 0; i < num_nodes; ++i) { unsigned long pages = min(pages_left, pages_per_node); uint32_t alignment = mem->page_alignment; + unsigned long start; if (pages == pages_per_node) alignment = pages_per_node; @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (unlikely(r)) goto error; + /* + * Calculate a virtual BO start address to easily check if + * everything is CPU accessible. + */ + start = nodes[i].start + nodes[i].size - mem->num_pages; + mem->start = max(mem->start, start); pages_left -= pages; } spin_unlock(&mgr->lock); - mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET; mem->mm_node = nodes; return 0; -- 2.5.0 _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel ^ permalink raw reply related [flat|nested] 15+ messages in thread
* CPU mapping of split VRAM buffers
@ 2017-03-29 12:55 Christian König
[not found] ` <1490792146-2218-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
0 siblings, 1 reply; 15+ messages in thread
From: Christian König @ 2017-03-29 12:55 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW; +Cc: maraeo-Re5JQEeQqe8AvxtiuMwx3w
Hi guys,
this set implements CPU mapping of split VRAM buffers and could
help with some performance issues regarding this.
Please review and/or comment,
Christian.
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread[parent not found: <1490792146-2218-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>]
* [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS [not found] ` <1490792146-2218-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> @ 2017-03-29 12:55 ` Christian König 0 siblings, 0 replies; 15+ messages in thread From: Christian König @ 2017-03-29 12:55 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW; +Cc: maraeo-Re5JQEeQqe8AvxtiuMwx3w From: Christian König <christian.koenig@amd.com> Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS instead of a placement limit. That allows us to better handle CPU accessible placements. Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 14 ++++++++++---- 2 files changed, 15 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index d6b2de9..387d190 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -122,20 +122,19 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev, if (domain & AMDGPU_GEM_DOMAIN_VRAM) { unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT; - unsigned lpfn = 0; - - /* This forces a reallocation if the flag wasn't set before */ - if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) - lpfn = adev->mc.real_vram_size >> PAGE_SHIFT; places[c].fpfn = 0; - places[c].lpfn = lpfn; + places[c].lpfn = 0; places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM; + if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) places[c].lpfn = visible_pfn; else places[c].flags |= TTM_PL_FLAG_TOPDOWN; + + if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) + places[c].flags |= TTM_PL_FLAG_CONTIGUOUS; c++; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index d710226..af2d172 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -93,7 +93,6 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, const struct ttm_place *place, struct ttm_mem_reg *mem) { - struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo); struct amdgpu_vram_mgr *mgr = man->priv; struct drm_mm *mm = &mgr->mm; struct drm_mm_node *nodes; @@ -107,8 +106,8 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (!lpfn) lpfn = man->size; - if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS || - place->lpfn || amdgpu_vram_page_split == -1) { + if (place->flags & TTM_PL_FLAG_CONTIGUOUS || + amdgpu_vram_page_split == -1) { pages_per_node = ~0ul; num_nodes = 1; } else { @@ -126,12 +125,14 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, aflags = DRM_MM_CREATE_TOP; } + mem->start = 0; pages_left = mem->num_pages; spin_lock(&mgr->lock); for (i = 0; i < num_nodes; ++i) { unsigned long pages = min(pages_left, pages_per_node); uint32_t alignment = mem->page_alignment; + unsigned long start; if (pages == pages_per_node) alignment = pages_per_node; @@ -145,11 +146,16 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man, if (unlikely(r)) goto error; + /* + * Calculate a virtual BO start address to easily check if + * everything is CPU accessible. + */ + start = nodes[i].start + nodes[i].size - mem->num_pages; + mem->start = max(mem->start, start); pages_left -= pages; } spin_unlock(&mgr->lock); - mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET; mem->mm_node = nodes; return 0; -- 2.5.0 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 15+ messages in thread
end of thread, other threads:[~2017-04-04 14:24 UTC | newest]
Thread overview: 15+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-03-31 9:47 [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König
2017-03-31 9:47 ` [PATCH 2/6] drm/ttm: add io_mem_pfn callback Christian König
2017-03-31 9:47 ` [PATCH 4/6] drm/amdgpu: drop alpha support Christian König
[not found] ` <1490953652-3703-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2017-03-31 9:47 ` [PATCH 3/6] drm/ttm: add TTM_PL_FLAG_CONTIGUOUS v2 Christian König
2017-04-03 16:27 ` Nicolai Hähnle
2017-03-31 9:47 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König
[not found] ` <1490953652-3703-5-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2017-04-03 16:22 ` Nicolai Hähnle
[not found] ` <425dfa9e-c86b-6958-f7ff-31b91a0f9e21-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2017-04-04 11:33 ` Christian König
2017-04-04 14:24 ` Nicolai Hähnle
2017-03-31 9:47 ` [PATCH 6/6] drm/amdgpu: handle CPU access for split VRAM buffers Christian König
2017-04-03 16:25 ` Nicolai Hähnle
2017-04-04 11:30 ` Christian König
2017-04-03 12:37 ` [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat v2 Christian König
-- strict thread matches above, loose matches on Subject: below --
2017-03-29 17:43 [PATCH 1/6] drm/ttm: cleanup and optimize ttm_bo_mem_compat Christian König
2017-03-29 17:43 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König
2017-03-29 12:55 CPU mapping of split VRAM buffers Christian König
[not found] ` <1490792146-2218-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2017-03-29 12:55 ` [PATCH 5/6] drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS Christian König
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.