From a86170290c38ca1dad556a42f1d57f39dd62af45 Mon Sep 17 00:00:00 2001 From: Roger He Date: Fri, 19 Jan 2018 14:34:16 +0800 Subject: drm/ttm: fix missing parameter change for ttm_bo_cleanup_refs Missed in the patche: dc94777 drm/ttm: enable swapout for reserved BOs during allocation. don't unreserve the BO if it is not reserved by itself. Signed-off-by: Roger He Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/ttm/ttm_bo.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 893003fc76a1..2fef09a56d16 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -1727,7 +1727,7 @@ int ttm_bo_swapout(struct ttm_bo_global *glob, struct ttm_operation_ctx *ctx) kref_get(&bo->list_kref); if (!list_empty(&bo->ddestroy)) { - ret = ttm_bo_cleanup_refs(bo, false, false, true); + ret = ttm_bo_cleanup_refs(bo, false, false, locked); kref_put(&bo->list_kref, ttm_bo_release_list); return ret; } -- cgit v1.2.3 From 95244db2d3f743f37e69446a2807dd1a42750542 Mon Sep 17 00:00:00 2001 From: Tom St Denis Date: Fri, 26 Jan 2018 09:32:29 -0500 Subject: drm/ttm: Fix 'buf' pointer update in ttm_bo_vm_access_kmap() (v2) The buf pointer was not being incremented inside the loop meaning the same block of data would be read or written repeatedly. (v2) Change 'buf' pointer to uint8_t* type Cc: stable@vger.kernel.org Fixes: 09ac4fcb3f25 ("drm/ttm: Implement vm_operations_struct.access v2") Signed-off-by: Tom St Denis Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/ttm/ttm_bo_vm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index 08a3c324242e..60fcef1593dd 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -316,7 +316,7 @@ static void ttm_bo_vm_close(struct vm_area_struct *vma) static int ttm_bo_vm_access_kmap(struct ttm_buffer_object *bo, unsigned long offset, - void *buf, int len, int write) + uint8_t *buf, int len, int write) { unsigned long page = offset >> PAGE_SHIFT; unsigned long bytes_left = len; @@ -345,6 +345,7 @@ static int ttm_bo_vm_access_kmap(struct ttm_buffer_object *bo, ttm_bo_kunmap(&map); page++; + buf += bytes; bytes_left -= bytes; offset = 0; } while (bytes_left); -- cgit v1.2.3 From 59fd27cd2fd762b08a1f54de8bbe3b82aba06a13 Mon Sep 17 00:00:00 2001 From: Huang Rui Date: Fri, 15 Dec 2017 09:33:21 +0800 Subject: drm/amdgpu: use queue 0 for kiq ring It must use queue id 0, because CGPG_IDLE/SAVE/LOAD/RUN only can be issued on queue 0. Signed-off-by: Huang Rui Acked-by: Hawking Zhang Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index bb40d2529a30..239bf2a4b3c6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -179,8 +179,12 @@ static int amdgpu_gfx_kiq_acquire(struct amdgpu_device *adev, amdgpu_gfx_bit_to_queue(adev, queue_bit, &mec, &pipe, &queue); - /* Using pipes 2/3 from MEC 2 seems cause problems */ - if (mec == 1 && pipe > 1) + /* + * 1. Using pipes 2/3 from MEC 2 seems cause problems. + * 2. It must use queue id 0, because CGPG_IDLE/SAVE/LOAD/RUN + * only can be issued on queue 0. + */ + if ((mec == 1 && pipe > 1) || queue != 0) continue; ring->me = mec + 1; -- cgit v1.2.3 From d3aab672de4902ce25e9858fbb5bf67e8254b23c Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 24 Jan 2018 14:57:02 +0100 Subject: drm/amdgpu: fix another potential cause of VM faults The root PD can be evicted directly after allocating it, just validate it on first use. Signed-off-by: Christian König Acked-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 32 ++++++++++++-------------------- 1 file changed, 12 insertions(+), 20 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 6fc16eecf2dc..5afbc5e714d0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -2262,12 +2262,12 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, AMDGPU_VM_PTE_COUNT(adev) * 8); + uint64_t init_pde_value = 0, flags; unsigned ring_instance; struct amdgpu_ring *ring; struct drm_sched_rq *rq; + unsigned long size; int r, i; - u64 flags; - uint64_t init_pde_value = 0; vm->va = RB_ROOT_CACHED; for (i = 0; i < AMDGPU_MAX_VMHUBS; i++) @@ -2318,29 +2318,21 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, flags |= (AMDGPU_GEM_CREATE_NO_CPU_ACCESS | AMDGPU_GEM_CREATE_SHADOW); - r = amdgpu_bo_create(adev, - amdgpu_vm_bo_size(adev, adev->vm_manager.root_level), - align, true, - AMDGPU_GEM_DOMAIN_VRAM, - flags, - NULL, NULL, init_pde_value, &vm->root.base.bo); + size = amdgpu_vm_bo_size(adev, adev->vm_manager.root_level); + r = amdgpu_bo_create(adev, size, align, true, AMDGPU_GEM_DOMAIN_VRAM, + flags, NULL, NULL, init_pde_value, + &vm->root.base.bo); if (r) goto error_free_sched_entity; + r = amdgpu_bo_reserve(vm->root.base.bo, true); + if (r) + goto error_free_root; + vm->root.base.vm = vm; list_add_tail(&vm->root.base.bo_list, &vm->root.base.bo->va); - INIT_LIST_HEAD(&vm->root.base.vm_status); - - if (vm->use_cpu_for_update) { - r = amdgpu_bo_reserve(vm->root.base.bo, false); - if (r) - goto error_free_root; - - r = amdgpu_bo_kmap(vm->root.base.bo, NULL); - amdgpu_bo_unreserve(vm->root.base.bo); - if (r) - goto error_free_root; - } + list_add_tail(&vm->root.base.vm_status, &vm->evicted); + amdgpu_bo_unreserve(vm->root.base.bo); if (pasid) { unsigned long flags; -- cgit v1.2.3 From 153b9e1b75441cd51f69e88b4748c23166de0d64 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 24 Jan 2018 13:37:48 +0100 Subject: drm/amdgpu: fix locking in vega10_ih_prescreen_iv The vm pointer can become invalid as soon as the lock is released. Signed-off-by: Christian König Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/vega10_ih.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c index b69ceafb7888..746748e6b731 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c @@ -278,9 +278,9 @@ static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev) /* Track retry faults in per-VM fault FIFO. */ spin_lock(&adev->vm_manager.pasid_lock); vm = idr_find(&adev->vm_manager.pasid_idr, pasid); - spin_unlock(&adev->vm_manager.pasid_lock); if (WARN_ON_ONCE(!vm)) { /* VM not found, process it normally */ + spin_unlock(&adev->vm_manager.pasid_lock); amdgpu_ih_clear_fault(adev, key); return true; } @@ -288,9 +288,11 @@ static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev) r = kfifo_put(&vm->faults, key); if (!r) { /* FIFO is full. Ignore it until there is space */ + spin_unlock(&adev->vm_manager.pasid_lock); amdgpu_ih_clear_fault(adev, key); goto ignore_iv; } + spin_unlock(&adev->vm_manager.pasid_lock); /* It's the first fault for this address, process it normally */ return true; -- cgit v1.2.3 From 147f255884a7d9bd3707d4674a8bbacf24efbcf5 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 24 Jan 2018 13:38:52 +0100 Subject: drm/amdgpu: remove WARN_ON when VM isn't found v2 It can easily be that the VM is already destroyed when this runs. v2: fix test inversion Signed-off-by: Christian König Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/vega10_ih.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c index 746748e6b731..ee14d78be2a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c @@ -278,7 +278,7 @@ static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev) /* Track retry faults in per-VM fault FIFO. */ spin_lock(&adev->vm_manager.pasid_lock); vm = idr_find(&adev->vm_manager.pasid_idr, pasid); - if (WARN_ON_ONCE(!vm)) { + if (!vm) { /* VM not found, process it normally */ spin_unlock(&adev->vm_manager.pasid_lock); amdgpu_ih_clear_fault(adev, key); -- cgit v1.2.3 From 3a61b527b4e1f285d21b6e9e623dc45cf8bb391f Mon Sep 17 00:00:00 2001 From: Julia Lawall Date: Sat, 27 Jan 2018 15:28:15 +0100 Subject: drm/radeon: adjust tested variable Check the variable that was most recently initialized. The semantic match that finds this problem is as follows: (http://coccinelle.lip6.fr/) // @@ expression x, y, f, g, e, m; statement S1,S2,S3,S4; @@ x = f(...); if (\(<+...x...+>\&e\)) S1 else S2 ( x = g(...); | m = g(...,&x,...); | y = g(...); *if (e) S3 else S4 ) // Signed-off-by: Julia Lawall Reviewed-by: Christian König Signed-off-by: Alex Deucher Cc: stable@vger.kernel.org --- drivers/gpu/drm/radeon/radeon_uvd.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c b/drivers/gpu/drm/radeon/radeon_uvd.c index d34d1cf33895..95f4db70dd22 100644 --- a/drivers/gpu/drm/radeon/radeon_uvd.c +++ b/drivers/gpu/drm/radeon/radeon_uvd.c @@ -995,7 +995,7 @@ int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev, /* calc dclk divider with current vco freq */ dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk, pd_min, pd_even); - if (vclk_div > pd_max) + if (dclk_div > pd_max) break; /* vco is too big, it has to stop */ /* calc score with current vco freq */ -- cgit v1.2.3 From 7e24a3ea825e546487c3fc47f8cbf6512f6c9e8c Mon Sep 17 00:00:00 2001 From: Shirish S Date: Fri, 19 Jan 2018 14:17:22 +0530 Subject: drm/amdgpu: disable coarse grain clockgating for ST The CGCG feature on Stoney is causing GFX related issues such as freezes and blank outs. Signed-off-by: Shirish S Reviewed-by: Arindam Nath Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/vi.c | 1 - 1 file changed, 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index da2b99c2d95f..5bd10f52ca8a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -1022,7 +1022,6 @@ static int vi_common_early_init(void *handle) AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_CGTS | AMD_CG_SUPPORT_GFX_CGTS_LS | - AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS | AMD_CG_SUPPORT_BIF_LS | AMD_CG_SUPPORT_HDP_MGCG | -- cgit v1.2.3 From fb4bbba2775399149ca902f31eb5c46cc7a8d8b8 Mon Sep 17 00:00:00 2001 From: Shirish S Date: Mon, 5 Feb 2018 09:23:00 +0530 Subject: drm/amdgpu: re-enable CGCG on CZ and disable on ST The CGCG feature on Stoney is causing GFX related issues such as freezes and blank outs. Signed-off-by: Shirish S Reviewed-by: Arindam Nath Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/vi.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index 5bd10f52ca8a..1e3e05a11f7a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -1022,6 +1022,7 @@ static int vi_common_early_init(void *handle) AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_CGTS | AMD_CG_SUPPORT_GFX_CGTS_LS | + AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS | AMD_CG_SUPPORT_BIF_LS | AMD_CG_SUPPORT_HDP_MGCG | @@ -1048,7 +1049,6 @@ static int vi_common_early_init(void *handle) AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_CGTS | AMD_CG_SUPPORT_GFX_CGTS_LS | - AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS | AMD_CG_SUPPORT_BIF_LS | AMD_CG_SUPPORT_HDP_MGCG | -- cgit v1.2.3