From 66257db7a5276d7f1aff8850388a0eb675ffa2e1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicolai=20H=C3=A4hnle?= Date: Thu, 15 Dec 2016 17:23:49 +0100 Subject: drm/ttm: add evict parameter to ttm_bo_driver::move_notify MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Ensure that the driver can listen to evictions even when they don't take the path through ttm_bo_driver::move. This is crucial for amdgpu, which relies on an eviction counter to skip re-binding page tables when possible. Signed-off-by: Nicolai Hähnle Reviewed-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index bf79b73e1538..c29db99f9048 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -849,6 +849,7 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer, } void amdgpu_bo_move_notify(struct ttm_buffer_object *bo, + bool evict, struct ttm_mem_reg *new_mem) { struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev); -- cgit v1.2.3 From 661a7606c97b6e6a73f553b0a6b60f553d5afa77 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicolai=20H=C3=A4hnle?= Date: Thu, 15 Dec 2016 17:26:42 +0100 Subject: drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This catches evictions of shadow page tables from the GART. Since shadow page tables are always stored in system memory, amdgpu_bo_move is never called for them. This fixes a crash during command submission that occurs when only a shadow page table and no other BOs were evicted since the last submission. Fixes: 1baa439fb2f4e586 ("drm/amdgpu: allocate shadow for pd/pt bo V2") Signed-off-by: Nicolai Hähnle Reviewed-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index c29db99f9048..d94cdeff0617 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -862,6 +862,10 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo, abo = container_of(bo, struct amdgpu_bo, tbo); amdgpu_vm_bo_invalidate(adev, abo); + /* remember the eviction */ + if (evict) + atomic64_inc(&adev->num_evictions); + /* update statistics */ if (!new_mem) return; -- cgit v1.2.3 From 36ea83d1512008ab6631c54971e8de8cf2f0a7b4 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicolai=20H=C3=A4hnle?= Date: Tue, 10 Jan 2017 19:06:00 +0100 Subject: drm/amd/amdgpu: lock reservation object while creating shadow bo MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit ttm_bo_init checks that the reservation object is locked. This is the caller's responsibility when resv != NULL. Otherwise, the inline reservation object of the newly allocated buffer is used and must explicitly be locked. Uninterruptible w/w locks without an acquire context are always successful. v2: use ww_mutex_lock Signed-off-by: Nicolai Hähnle Reviewed-by: Edward O'Callaghan (v1) Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index d94cdeff0617..ac6b074b8821 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -472,7 +472,16 @@ int amdgpu_bo_create(struct amdgpu_device *adev, return r; if (amdgpu_need_backup(adev) && (flags & AMDGPU_GEM_CREATE_SHADOW)) { + if (!resv) { + r = ww_mutex_lock(&(*bo_ptr)->tbo.resv->lock, NULL); + WARN_ON(r != 0); + } + r = amdgpu_bo_create_shadow(adev, size, byte_align, (*bo_ptr)); + + if (!resv) + ww_mutex_unlock(&(*bo_ptr)->tbo.resv->lock); + if (r) amdgpu_bo_unref(bo_ptr); } -- cgit v1.2.3 From f1543f58cbc7f62c4ea0cca57bbb4a95dc752264 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicolai=20H=C3=A4hnle?= Date: Tue, 10 Jan 2017 20:36:56 +0100 Subject: drm/amd/amdgpu: fix locking in bo creation error path MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Unlock the resv lock only if we were the ones to lock it in the first place. Signed-off-by: Nicolai Hähnle Reviewed-by: Edward O'Callaghan Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index ac6b074b8821..7aecf2667f9d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -408,7 +408,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, return 0; fail_unreserve: - ww_mutex_unlock(&bo->tbo.resv->lock); + if (!resv) + ww_mutex_unlock(&bo->tbo.resv->lock); amdgpu_bo_unref(&bo); return r; } -- cgit v1.2.3 From e1f055b30187204439b6693a118b9dde9cda04ae Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= Date: Tue, 10 Jan 2017 17:27:49 +0100 Subject: drm/amdgpu: user BO priority instead of self coding it (v2) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Keeping groups of BOs on the LRU is to time consuming on command submission. Instead use the newly added BO priority to give a certain eviction order. v2: agd: trivial warning fix Signed-off-by: Christian König Reviewed-by: Roger.He Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 7aecf2667f9d..c43e8c534899 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, if (unlikely(r != 0)) return r; + bo->tbo.priority = ilog2(bo->tbo.num_pages); + bo->tbo.priority = min(bo->tbo.priority, (unsigned)(TTM_MAX_BO_PRIORITY - 1)); + if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED && bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) { struct dma_fence *fence; -- cgit v1.2.3 From 373308a5f5f3f82656567481f688350887abb771 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= Date: Mon, 23 Jan 2017 16:28:06 -0500 Subject: drm/amdgpu: double the priority of kernel allocations MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Give kernel allocations a higher priority cause it is often more work to swap them back in. Signed-off-by: Christian König Reviewed-by: Roger.He Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index c43e8c534899..a97d88314b1b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -387,6 +387,8 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, return r; bo->tbo.priority = ilog2(bo->tbo.num_pages); + if (kernel) + bo->tbo.priority *= 2; bo->tbo.priority = min(bo->tbo.priority, (unsigned)(TTM_MAX_BO_PRIORITY - 1)); if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED && -- cgit v1.2.3 From a2e2f29970aa1e53219e3499382c41d61c8959f8 Mon Sep 17 00:00:00 2001 From: Nils Holland Date: Sun, 22 Jan 2017 20:15:27 +0100 Subject: drm/amdgpu: Bring bo creation in line with radeon driver (v2) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Add the bo creation changes that have been done to the radeon driver in recent times, e.g. disable GTT WC on 32 bit because it is broken there, and also disable it generally (and print a warning message) when CONFIG_X86_PAT is not set. v2: agd: fix warning in defined(CONFIG_X86) && !defined(CONFIG_X86_PAT) case Reviewed-by: Michel Dänzer Reviewed-by: Christian König Signed-off-by: Nils Holland Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index a97d88314b1b..d1aa291b2638 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -363,11 +363,31 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, bo->flags = flags; +#ifdef CONFIG_X86_32 + /* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit + * See https://bugs.freedesktop.org/show_bug.cgi?id=84627 + */ + bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC; +#elif defined(CONFIG_X86) && !defined(CONFIG_X86_PAT) + /* Don't try to enable write-combining when it can't work, or things + * may be slow + * See https://bugs.freedesktop.org/show_bug.cgi?id=88758 + */ + +#warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance \ + thanks to write-combining + + if (bo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) + DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for " + "better performance thanks to write-combining\n"); + bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC; +#else /* For architectures that don't support WC memory, * mask out the WC flag from the BO */ if (!drm_arch_can_wc_memory()) bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC; +#endif amdgpu_fill_placement_to_bo(bo, placement); /* Kernel allocation are uninterruptible */ -- cgit v1.2.3 From 31bb90f1cd084e0ca1359455eaf74bb64b5c5c82 Mon Sep 17 00:00:00 2001 From: Arnd Bergmann Date: Wed, 1 Feb 2017 16:59:21 +0100 Subject: drm/amdgpu: shut up #warning for compile testing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit My randconfig tests on linux-next showed a newly introduced warning: drivers/gpu/drm/amd/amdgpu/amdgpu_object.c: In function 'amdgpu_bo_create_restricted': drivers/gpu/drm/amd/amdgpu/amdgpu_object.c:377:2: error: #warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance thanks to write-combining [-Werror=cpp] Generally speaking, warnings about bad kernel configuration are not particularly helpful. We could enforce the selection of X86_PAT through Kconfig, so the driver cannot even be used unless it is enabled, or we could just rely on the runtime warning that is also there. In this version, I'm making the warning conditional on CONFIG_COMPILE_TEST, which shuts it up for me, but not people that may actually want to run the kernel as a compromize. Fixes: a2e2f29970aa ("drm/amdgpu: Bring bo creation in line with radeon driver (v2)") Reviewed-by: Michel Dänzer Signed-off-by: Arnd Bergmann Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index d1aa291b2638..da3a53a39d0b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -374,8 +374,10 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, * See https://bugs.freedesktop.org/show_bug.cgi?id=88758 */ +#ifndef CONFIG_COMPILE_TEST #warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance \ thanks to write-combining +#endif if (bo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for " -- cgit v1.2.3 From 38fc4856ad98f230bc91da0421dec69e4aee40f8 Mon Sep 17 00:00:00 2001 From: Samuel Pitoiset Date: Thu, 9 Feb 2017 11:33:36 +0100 Subject: drm/amdgpu: fix a potential deadlock in amdgpu_bo_create_restricted() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When ttm_bo_init() fails, the reservation mutex should be unlocked. In debug build, the kernel reported "possible recursive locking detected" in this codepath. For debugging purposes, I also added a "WARN_ON(ww_mutex_is_locked())" when ttm_bo_init() fails and the mutex was locked as expected. This should fix (random) GPU hangs. The easy way to reproduce the issue is to change the "Super Sampling" option from 1.0 to 2.0 in Hitman. It will create a huge buffer, evict a bunch of buffers (around ~5k) and deadlock. This regression has been introduced pretty recently. v2: only release the mutex if resv is NULL Fixes: 12a852219583 ("drm/amdgpu: improve AMDGPU_GEM_CREATE_VRAM_CLEARED handling (v2)") Reviewed-by: Christian König Signed-off-by: Samuel Pitoiset Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index da3a53a39d0b..993c003a42e4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -405,8 +405,11 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, &bo->placement, page_align, !kernel, NULL, acc_size, sg, resv ? resv : &bo->tbo.ttm_resv, &amdgpu_ttm_bo_destroy); - if (unlikely(r != 0)) + if (unlikely(r != 0)) { + if (!resv) + ww_mutex_unlock(&bo->tbo.resv->lock); return r; + } bo->tbo.priority = ilog2(bo->tbo.num_pages); if (kernel) -- cgit v1.2.3 From fad061270ac43ff9eed315f0eae7c40b694592de Mon Sep 17 00:00:00 2001 From: Samuel Pitoiset Date: Thu, 9 Feb 2017 11:33:37 +0100 Subject: drm/amdgpu: report the number of bytes moved at buffer creation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Like ttm_bo_validate(), ttm_bo_init() might need to move BO and the number of bytes moved by TTM should be reported. This can help the throttle buffer migration mechanism to make a better decision. v2: fix computation Reviewed-by: Christian König Signed-off-by: Samuel Pitoiset Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_object.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 993c003a42e4..be80a4a68d7b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -323,6 +323,7 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, struct amdgpu_bo *bo; enum ttm_bo_type type; unsigned long page_align; + u64 initial_bytes_moved; size_t acc_size; int r; @@ -401,10 +402,15 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, locked = ww_mutex_trylock(&bo->tbo.ttm_resv.lock); WARN_ON(!locked); } + + initial_bytes_moved = atomic64_read(&adev->num_bytes_moved); r = ttm_bo_init(&adev->mman.bdev, &bo->tbo, size, type, &bo->placement, page_align, !kernel, NULL, acc_size, sg, resv ? resv : &bo->tbo.ttm_resv, &amdgpu_ttm_bo_destroy); + amdgpu_cs_report_moved_bytes(adev, + atomic64_read(&adev->num_bytes_moved) - initial_bytes_moved); + if (unlikely(r != 0)) { if (!resv) ww_mutex_unlock(&bo->tbo.resv->lock); -- cgit v1.2.3