diff options
| author | Chris Wilson <chris@chris-wilson.co.uk> | 2019-08-14 21:24:01 +0300 |
|---|---|---|
| committer | Chris Wilson <chris@chris-wilson.co.uk> | 2019-08-16 14:40:58 +0300 |
| commit | b016cd6ed4b772759804e0d6082bd1f5ca63b8ee (patch) | |
| tree | 278f8b1426b40316f5744cd6155ec51ecc9dd6c7 /drivers/gpu/drm | |
| parent | dc2e1e5b279966affbd11ff7cfef52eb634ca2c9 (diff) | |
| download | linux-b016cd6ed4b772759804e0d6082bd1f5ca63b8ee.tar.xz | |
dma-buf: Restore seqlock around dma_resv updates
This reverts
67c97fb79a7f ("dma-buf: add reservation_object_fences helper")
dd7a7d1ff2f1 ("drm/i915: use new reservation_object_fences helper")
0e1d8083bddb ("dma-buf: further relax reservation_object_add_shared_fence")
5d344f58da76 ("dma-buf: nuke reservation_object seq number")
The scenario that defeats simply grabbing a set of shared/exclusive
fences and using them blissfully under RCU is that any of those fences
may be reallocated by a SLAB_TYPESAFE_BY_RCU fence slab cache. In this
scenario, while keeping the rcu_read_lock we need to establish that no
fence was changed in the dma_resv after a read (or full) memory barrier.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190814182401.25009-1-chris@chris-wilson.co.uk
Diffstat (limited to 'drivers/gpu/drm')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/i915/gem/i915_gem_busy.c | 24 |
2 files changed, 23 insertions, 8 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index bc4ec6b20a87..76e3516484e7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -251,7 +251,12 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, new->shared_max = old->shared_max; new->shared_count = k; - rcu_assign_pointer(resv->fence, new); + /* Install the new fence list, seqcount provides the barriers */ + preempt_disable(); + write_seqcount_begin(&resv->seq); + RCU_INIT_POINTER(resv->fence, new); + write_seqcount_end(&resv->seq); + preempt_enable(); /* Drop the references to the removed fences or move them to ef_list */ for (i = j, k = 0; i < old->shared_count; ++i) { diff --git a/drivers/gpu/drm/i915/gem/i915_gem_busy.c b/drivers/gpu/drm/i915/gem/i915_gem_busy.c index a2aff1d8290e..3d4f5775a4ba 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_busy.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_busy.c @@ -83,8 +83,7 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, struct drm_i915_gem_busy *args = data; struct drm_i915_gem_object *obj; struct dma_resv_list *list; - unsigned int i, shared_count; - struct dma_fence *excl; + unsigned int seq; int err; err = -ENOENT; @@ -110,18 +109,29 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, * to report the overall busyness. This is what the wait-ioctl does. * */ - dma_resv_fences(obj->base.resv, &excl, &list, &shared_count); +retry: + seq = raw_read_seqcount(&obj->base.resv->seq); /* Translate the exclusive fence to the READ *and* WRITE engine */ - args->busy = busy_check_writer(excl); + args->busy = + busy_check_writer(rcu_dereference(obj->base.resv->fence_excl)); /* Translate shared fences to READ set of engines */ - for (i = 0; i < shared_count; ++i) { - struct dma_fence *fence = rcu_dereference(list->shared[i]); + list = rcu_dereference(obj->base.resv->fence); + if (list) { + unsigned int shared_count = list->shared_count, i; - args->busy |= busy_check_reader(fence); + for (i = 0; i < shared_count; ++i) { + struct dma_fence *fence = + rcu_dereference(list->shared[i]); + + args->busy |= busy_check_reader(fence); + } } + if (args->busy && read_seqcount_retry(&obj->base.resv->seq, seq)) + goto retry; + err = 0; out: rcu_read_unlock(); |