* [PATCH 0/4] drm/amdgpu: pin_size fixes
@ 2018-06-15 14:42 Michel Dänzer
[not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>
0 siblings, 1 reply; 12+ messages in thread
From: Michel Dänzer @ 2018-06-15 14:42 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Michel Dänzer <michel.daenzer@amd.com>
I came up with the fixes in patches 1 & 3 while looking into
https://bugs.freedesktop.org/106872 . It's not clear that these will be
enough to fix that, but they are moving in the right direction anyway.
Patch 2 is refactoring in preparation for patch 3.
Patch 4 is separate so that patches 1-3 can be backported to older
kernels which don't have amdgpu_gmc_vram_full_visible.
Michel Dänzer (4):
drm/amdgpu: Update pin_size values before unpinning BO
drm/amdgpu: Refactor amdgpu_vram_mgr_bo_invisible_size helper
drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate
drm/amdgpu: Use gmc_vram_full_visible in vram_mgr_bo_invisible_size
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 24 ++++++-------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 38 ++++++++++++++++++++
3 files changed, 49 insertions(+), 14 deletions(-)
--
2.17.1
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 12+ messages in thread[parent not found: <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>]
* [PATCH 1/4] drm/amdgpu: Update pin_size values before unpinning BO [not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> @ 2018-06-15 14:42 ` Michel Dänzer 2018-06-15 14:42 ` [PATCH 2/4] drm/amdgpu: Refactor amdgpu_vram_mgr_bo_invisible_size helper Michel Dänzer ` (2 subsequent siblings) 3 siblings, 0 replies; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 14:42 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Michel Dänzer <michel.daenzer@amd.com> At least in theory, ttm_bo_validate may move the BO, in which case the pin_size accounting would be inconsistent with when the BO was pinned. Cc: stable@vger.kernel.org Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 18 ++++++++---------- 1 file changed, 8 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index f5b0b180a6cc..210e7631f4b7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -969,15 +969,6 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo) bo->pin_count--; if (bo->pin_count) return 0; - for (i = 0; i < bo->placement.num_placement; i++) { - bo->placements[i].lpfn = 0; - bo->placements[i].flags &= ~TTM_PL_FLAG_NO_EVICT; - } - r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); - if (unlikely(r)) { - dev_err(adev->dev, "%p validate failed for unpin\n", bo); - goto error; - } if (bo->tbo.mem.mem_type == TTM_PL_VRAM) { adev->vram_pin_size -= amdgpu_bo_size(bo); @@ -987,7 +978,14 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo) adev->gart_pin_size -= amdgpu_bo_size(bo); } -error: + for (i = 0; i < bo->placement.num_placement; i++) { + bo->placements[i].lpfn = 0; + bo->placements[i].flags &= ~TTM_PL_FLAG_NO_EVICT; + } + r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); + if (unlikely(r)) + dev_err(adev->dev, "%p validate failed for unpin\n", bo); + return r; } -- 2.17.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 2/4] drm/amdgpu: Refactor amdgpu_vram_mgr_bo_invisible_size helper [not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> 2018-06-15 14:42 ` [PATCH 1/4] drm/amdgpu: Update pin_size values before unpinning BO Michel Dänzer @ 2018-06-15 14:42 ` Michel Dänzer 2018-06-15 14:42 ` [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate Michel Dänzer 2018-06-15 14:42 ` [PATCH 4/4] drm/amdgpu: Use gmc_vram_full_visible in vram_mgr_bo_invisible_size Michel Dänzer 3 siblings, 0 replies; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 14:42 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Michel Dänzer <michel.daenzer@amd.com> Preparation for the following fix, no functional change intended. Cc: stable@vger.kernel.org Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 6 ++---- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 16 ++++++++++++++++ 3 files changed, 19 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 210e7631f4b7..512f59836436 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -918,8 +918,7 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain, domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type); if (domain == AMDGPU_GEM_DOMAIN_VRAM) { adev->vram_pin_size += amdgpu_bo_size(bo); - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) - adev->invisible_pin_size += amdgpu_bo_size(bo); + adev->invisible_pin_size += amdgpu_vram_mgr_bo_invisible_size(bo); } else if (domain == AMDGPU_GEM_DOMAIN_GTT) { adev->gart_pin_size += amdgpu_bo_size(bo); } @@ -972,8 +971,7 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo) if (bo->tbo.mem.mem_type == TTM_PL_VRAM) { adev->vram_pin_size -= amdgpu_bo_size(bo); - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) - adev->invisible_pin_size -= amdgpu_bo_size(bo); + adev->invisible_pin_size -= amdgpu_vram_mgr_bo_invisible_size(bo); } else if (bo->tbo.mem.mem_type == TTM_PL_TT) { adev->gart_pin_size -= amdgpu_bo_size(bo); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h index e969c879d87e..e5da4654b630 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h @@ -73,6 +73,7 @@ bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_mem_reg *mem); uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man); int amdgpu_gtt_mgr_recover(struct ttm_mem_type_manager *man); +u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo); uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man); uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index 9c47e860e5e6..491fd0a789f7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -96,6 +96,22 @@ static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, adev->gmc.visible_vram_size : end) - start; } +/** + * amdgpu_vram_mgr_bo_invisible_size - CPU invisible BO size + * + * @bo: &amdgpu_bo buffer object + * + * Returns: + * How much of the given &amdgpu_bo buffer object lies in CPU invisible VRAM. + */ +u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) +{ + if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) + return amdgpu_bo_size(bo); + + return 0; +} + /** * amdgpu_vram_mgr_new - allocate new ranges * -- 2.17.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> 2018-06-15 14:42 ` [PATCH 1/4] drm/amdgpu: Update pin_size values before unpinning BO Michel Dänzer 2018-06-15 14:42 ` [PATCH 2/4] drm/amdgpu: Refactor amdgpu_vram_mgr_bo_invisible_size helper Michel Dänzer @ 2018-06-15 14:42 ` Michel Dänzer [not found] ` <20180615144230.18480-4-michel-otUistvHUpPR7s880joybQ@public.gmane.org> 2018-06-15 14:42 ` [PATCH 4/4] drm/amdgpu: Use gmc_vram_full_visible in vram_mgr_bo_invisible_size Michel Dänzer 3 siblings, 1 reply; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 14:42 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Michel Dänzer <michel.daenzer@amd.com> Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least partially in CPU visible VRAM, in particular when all VRAM is visible. Cc: stable@vger.kernel.org Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 26 ++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index 491fd0a789f7..9b34df7017f0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -106,10 +106,32 @@ static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, */ u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) { - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + struct amdgpu_vram_mgr *mgr = adev->mman.bdev.man[TTM_PL_VRAM].priv; + struct ttm_mem_reg *mem = &bo->tbo.mem; + struct drm_mm_node *nodes = mem->mm_node; + unsigned pages = mem->num_pages; + u64 usage = 0; + + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) + return 0; + + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) return amdgpu_bo_size(bo); - return 0; + if (!nodes) + return 0; + + spin_lock(&mgr->lock); + while (pages) { + pages -= nodes->size; + usage += nodes->size << PAGE_SHIFT; + usage -= amdgpu_vram_mgr_vis_size(adev, nodes); + ++nodes; + } + spin_unlock(&mgr->lock); + + return usage; } /** -- 2.17.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 12+ messages in thread
[parent not found: <20180615144230.18480-4-michel-otUistvHUpPR7s880joybQ@public.gmane.org>]
* Re: [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <20180615144230.18480-4-michel-otUistvHUpPR7s880joybQ@public.gmane.org> @ 2018-06-15 15:11 ` Christian König [not found] ` <093bbd7a-02cd-068f-a53a-2d4f1c5b3084-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 2018-06-18 16:52 ` [PATCH v2 " Michel Dänzer 1 sibling, 1 reply; 12+ messages in thread From: Christian König @ 2018-06-15 15:11 UTC (permalink / raw) To: Michel Dänzer, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Am 15.06.2018 um 16:42 schrieb Michel Dänzer: > From: Michel Dänzer <michel.daenzer@amd.com> > > Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least > partially in CPU visible VRAM, in particular when all VRAM is visible. > > Cc: stable@vger.kernel.org > Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 26 ++++++++++++++++++-- > 1 file changed, 24 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > index 491fd0a789f7..9b34df7017f0 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > @@ -106,10 +106,32 @@ static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, > */ > u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) > { > - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) > + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); > + struct amdgpu_vram_mgr *mgr = adev->mman.bdev.man[TTM_PL_VRAM].priv; > + struct ttm_mem_reg *mem = &bo->tbo.mem; > + struct drm_mm_node *nodes = mem->mm_node; > + unsigned pages = mem->num_pages; > + u64 usage = 0; > + > + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) > + return 0; Andrey just added a helper for that into amdgpu_gmc.h. > + > + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) > return amdgpu_bo_size(bo); > > - return 0; > + if (!nodes) > + return 0; > + > + spin_lock(&mgr->lock); I actually don't see any need to grab the lock here. Apart from that looks like a good cleanup to me, Christian. > + while (pages) { > + pages -= nodes->size; > + usage += nodes->size << PAGE_SHIFT; > + usage -= amdgpu_vram_mgr_vis_size(adev, nodes); > + ++nodes; > + } > + spin_unlock(&mgr->lock); > + > + return usage; > } > > /** _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
[parent not found: <093bbd7a-02cd-068f-a53a-2d4f1c5b3084-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]
* Re: [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <093bbd7a-02cd-068f-a53a-2d4f1c5b3084-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> @ 2018-06-15 15:14 ` Michel Dänzer [not found] ` <098df480-3db3-485f-8e39-4ebf8201b3a0-otUistvHUpPR7s880joybQ@public.gmane.org> 0 siblings, 1 reply; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 15:14 UTC (permalink / raw) To: christian.koenig-5C7GfCeVMHo; +Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW On 2018-06-15 05:11 PM, Christian König wrote: > Am 15.06.2018 um 16:42 schrieb Michel Dänzer: >> From: Michel Dänzer <michel.daenzer@amd.com> >> >> Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least >> partially in CPU visible VRAM, in particular when all VRAM is visible. >> >> Cc: stable@vger.kernel.org >> Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 26 ++++++++++++++++++-- >> 1 file changed, 24 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> index 491fd0a789f7..9b34df7017f0 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >> @@ -106,10 +106,32 @@ static u64 amdgpu_vram_mgr_vis_size(struct >> amdgpu_device *adev, >> */ >> u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) >> { >> - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) >> + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); >> + struct amdgpu_vram_mgr *mgr = adev->mman.bdev.man[TTM_PL_VRAM].priv; >> + struct ttm_mem_reg *mem = &bo->tbo.mem; >> + struct drm_mm_node *nodes = mem->mm_node; >> + unsigned pages = mem->num_pages; >> + u64 usage = 0; >> + >> + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) >> + return 0; > > Andrey just added a helper for that into amdgpu_gmc.h. Yep, see patch 4 and the cover letter. :) >> + >> + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) >> return amdgpu_bo_size(bo); >> - return 0; >> + if (!nodes) >> + return 0; >> + >> + spin_lock(&mgr->lock); > > I actually don't see any need to grab the lock here. Ah, so amdgpu_vram_mgr_del only grabs it for drm_mm_remove_node? -- Earthling Michel Dänzer | http://www.amd.com Libre software enthusiast | Mesa and X developer _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
[parent not found: <098df480-3db3-485f-8e39-4ebf8201b3a0-otUistvHUpPR7s880joybQ@public.gmane.org>]
* Re: [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <098df480-3db3-485f-8e39-4ebf8201b3a0-otUistvHUpPR7s880joybQ@public.gmane.org> @ 2018-06-15 15:18 ` Christian König [not found] ` <b94d92e5-908d-534a-907a-ad97124b22ec-5C7GfCeVMHo@public.gmane.org> 0 siblings, 1 reply; 12+ messages in thread From: Christian König @ 2018-06-15 15:18 UTC (permalink / raw) To: Michel Dänzer; +Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Am 15.06.2018 um 17:14 schrieb Michel Dänzer: > On 2018-06-15 05:11 PM, Christian König wrote: >> Am 15.06.2018 um 16:42 schrieb Michel Dänzer: >>> From: Michel Dänzer <michel.daenzer@amd.com> >>> >>> Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least >>> partially in CPU visible VRAM, in particular when all VRAM is visible. >>> >>> Cc: stable@vger.kernel.org >>> Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> >>> --- >>> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 26 ++++++++++++++++++-- >>> 1 file changed, 24 insertions(+), 2 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> index 491fd0a789f7..9b34df7017f0 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c >>> @@ -106,10 +106,32 @@ static u64 amdgpu_vram_mgr_vis_size(struct >>> amdgpu_device *adev, >>> */ >>> u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) >>> { >>> - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) >>> + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); >>> + struct amdgpu_vram_mgr *mgr = adev->mman.bdev.man[TTM_PL_VRAM].priv; >>> + struct ttm_mem_reg *mem = &bo->tbo.mem; >>> + struct drm_mm_node *nodes = mem->mm_node; >>> + unsigned pages = mem->num_pages; >>> + u64 usage = 0; >>> + >>> + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) >>> + return 0; >> Andrey just added a helper for that into amdgpu_gmc.h. > Yep, see patch 4 and the cover letter. :) Yeah, seen that after hitting send :) >>> + >>> + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) >>> return amdgpu_bo_size(bo); >>> - return 0; >>> + if (!nodes) >>> + return 0; >>> + >>> + spin_lock(&mgr->lock); >> I actually don't see any need to grab the lock here. > Ah, so amdgpu_vram_mgr_del only grabs it for drm_mm_remove_node? Yes, exactly. Christian. _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
[parent not found: <b94d92e5-908d-534a-907a-ad97124b22ec-5C7GfCeVMHo@public.gmane.org>]
* Re: [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <b94d92e5-908d-534a-907a-ad97124b22ec-5C7GfCeVMHo@public.gmane.org> @ 2018-06-15 15:19 ` Michel Dänzer [not found] ` <3140ef24-f8c9-5863-ff44-4c83192eabb1-otUistvHUpPR7s880joybQ@public.gmane.org> 0 siblings, 1 reply; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 15:19 UTC (permalink / raw) To: Christian König; +Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW On 2018-06-15 05:18 PM, Christian König wrote: > Am 15.06.2018 um 17:14 schrieb Michel Dänzer: >> On 2018-06-15 05:11 PM, Christian König wrote: >>> Am 15.06.2018 um 16:42 schrieb Michel Dänzer: > >>>> + >>>> + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) >>>> return amdgpu_bo_size(bo); >>>> - return 0; >>>> + if (!nodes) >>>> + return 0; >>>> + >>>> + spin_lock(&mgr->lock); >>> I actually don't see any need to grab the lock here. >> Ah, so amdgpu_vram_mgr_del only grabs it for drm_mm_remove_node? > > Yes, exactly. Will drop it in v2, thanks! -- Earthling Michel Dänzer | http://www.amd.com Libre software enthusiast | Mesa and X developer _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
[parent not found: <3140ef24-f8c9-5863-ff44-4c83192eabb1-otUistvHUpPR7s880joybQ@public.gmane.org>]
* Re: [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <3140ef24-f8c9-5863-ff44-4c83192eabb1-otUistvHUpPR7s880joybQ@public.gmane.org> @ 2018-06-18 11:30 ` Christian König 0 siblings, 0 replies; 12+ messages in thread From: Christian König @ 2018-06-18 11:30 UTC (permalink / raw) To: Michel Dänzer, Christian König Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Am 15.06.2018 um 17:19 schrieb Michel Dänzer: > On 2018-06-15 05:18 PM, Christian König wrote: >> Am 15.06.2018 um 17:14 schrieb Michel Dänzer: >>> On 2018-06-15 05:11 PM, Christian König wrote: >>>> Am 15.06.2018 um 16:42 schrieb Michel Dänzer: >>>>> + >>>>> + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) >>>>> return amdgpu_bo_size(bo); >>>>> - return 0; >>>>> + if (!nodes) >>>>> + return 0; >>>>> + >>>>> + spin_lock(&mgr->lock); >>>> I actually don't see any need to grab the lock here. >>> Ah, so amdgpu_vram_mgr_del only grabs it for drm_mm_remove_node? >> Yes, exactly. > Will drop it in v2, thanks! Haven't seeen v2 yet, but with that lock removed the whole series is Reviewed-by: Christian König <christian.koenig@amd.com>. Regards, Christian. _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v2 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <20180615144230.18480-4-michel-otUistvHUpPR7s880joybQ@public.gmane.org> 2018-06-15 15:11 ` Christian König @ 2018-06-18 16:52 ` Michel Dänzer [not found] ` <20180618165233.5718-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> 1 sibling, 1 reply; 12+ messages in thread From: Michel Dänzer @ 2018-06-18 16:52 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW; +Cc: Christian König From: Michel Dänzer <michel.daenzer@amd.com> Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least partially in CPU visible VRAM, in particular when all VRAM is visible. v2: * Don't take VRAM mgr spinlock, not needed (Christian König) * Make loop logic simpler and clearer. Cc: stable@vger.kernel.org Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index ae0049c6c52c..b6333f92ba45 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -106,10 +106,26 @@ static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, */ u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) { - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + struct ttm_mem_reg *mem = &bo->tbo.mem; + struct drm_mm_node *nodes = mem->mm_node; + unsigned pages = mem->num_pages; + u64 usage = 0; + + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) + return 0; + + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) return amdgpu_bo_size(bo); - return 0; + while (nodes && pages) { + usage += nodes->size << PAGE_SHIFT; + usage -= amdgpu_vram_mgr_vis_size(adev, nodes); + pages -= nodes->size; + ++nodes; + } + + return usage; } /** -- 2.17.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 12+ messages in thread
[parent not found: <20180618165233.5718-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>]
* Re: [PATCH v2 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate [not found] ` <20180618165233.5718-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> @ 2018-06-18 17:04 ` Christian König 0 siblings, 0 replies; 12+ messages in thread From: Christian König @ 2018-06-18 17:04 UTC (permalink / raw) To: Michel Dänzer, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Am 18.06.2018 um 18:52 schrieb Michel Dänzer: > From: Michel Dänzer <michel.daenzer@amd.com> > > Even BOs with AMDGPU_GEM_CREATE_NO_CPU_ACCESS may end up at least > partially in CPU visible VRAM, in particular when all VRAM is visible. > > v2: > * Don't take VRAM mgr spinlock, not needed (Christian König) > * Make loop logic simpler and clearer. > > Cc: stable@vger.kernel.org > Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 20 ++++++++++++++++++-- > 1 file changed, 18 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > index ae0049c6c52c..b6333f92ba45 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c > @@ -106,10 +106,26 @@ static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev, > */ > u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) > { > - if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) > + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); > + struct ttm_mem_reg *mem = &bo->tbo.mem; > + struct drm_mm_node *nodes = mem->mm_node; > + unsigned pages = mem->num_pages; > + u64 usage = 0; > + > + if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) > + return 0; > + > + if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) > return amdgpu_bo_size(bo); > > - return 0; > + while (nodes && pages) { > + usage += nodes->size << PAGE_SHIFT; > + usage -= amdgpu_vram_mgr_vis_size(adev, nodes); > + pages -= nodes->size; > + ++nodes; > + } > + > + return usage; > } > > /** _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH 4/4] drm/amdgpu: Use gmc_vram_full_visible in vram_mgr_bo_invisible_size [not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org> ` (2 preceding siblings ...) 2018-06-15 14:42 ` [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate Michel Dänzer @ 2018-06-15 14:42 ` Michel Dänzer 3 siblings, 0 replies; 12+ messages in thread From: Michel Dänzer @ 2018-06-15 14:42 UTC (permalink / raw) To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW From: Michel Dänzer <michel.daenzer@amd.com> Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index 9b34df7017f0..9814e9aab317 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -113,7 +113,7 @@ u64 amdgpu_vram_mgr_bo_invisible_size(struct amdgpu_bo *bo) unsigned pages = mem->num_pages; u64 usage = 0; - if (adev->gmc.visible_vram_size == adev->gmc.real_vram_size) + if (amdgpu_gmc_vram_full_visible(&adev->gmc)) return 0; if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT) -- 2.17.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx ^ permalink raw reply related [flat|nested] 12+ messages in thread
end of thread, other threads:[~2018-06-18 17:04 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-06-15 14:42 [PATCH 0/4] drm/amdgpu: pin_size fixes Michel Dänzer
[not found] ` <20180615144230.18480-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>
2018-06-15 14:42 ` [PATCH 1/4] drm/amdgpu: Update pin_size values before unpinning BO Michel Dänzer
2018-06-15 14:42 ` [PATCH 2/4] drm/amdgpu: Refactor amdgpu_vram_mgr_bo_invisible_size helper Michel Dänzer
2018-06-15 14:42 ` [PATCH 3/4] drm/amdgpu: Make amdgpu_vram_mgr_bo_invisible_size always accurate Michel Dänzer
[not found] ` <20180615144230.18480-4-michel-otUistvHUpPR7s880joybQ@public.gmane.org>
2018-06-15 15:11 ` Christian König
[not found] ` <093bbd7a-02cd-068f-a53a-2d4f1c5b3084-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2018-06-15 15:14 ` Michel Dänzer
[not found] ` <098df480-3db3-485f-8e39-4ebf8201b3a0-otUistvHUpPR7s880joybQ@public.gmane.org>
2018-06-15 15:18 ` Christian König
[not found] ` <b94d92e5-908d-534a-907a-ad97124b22ec-5C7GfCeVMHo@public.gmane.org>
2018-06-15 15:19 ` Michel Dänzer
[not found] ` <3140ef24-f8c9-5863-ff44-4c83192eabb1-otUistvHUpPR7s880joybQ@public.gmane.org>
2018-06-18 11:30 ` Christian König
2018-06-18 16:52 ` [PATCH v2 " Michel Dänzer
[not found] ` <20180618165233.5718-1-michel-otUistvHUpPR7s880joybQ@public.gmane.org>
2018-06-18 17:04 ` Christian König
2018-06-15 14:42 ` [PATCH 4/4] drm/amdgpu: Use gmc_vram_full_visible in vram_mgr_bo_invisible_size Michel Dänzer
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.