drm/amdgpu: keep all fences in an RCU protected array v2

Just keep all HW fences in a RCU protected array as a
first step to replace the wait queue.

v2: update commit message, move fixes into separate patch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 26b137801aeaed8b0fbc1c4edaac782b189b91ea..2fea597f8b6cdf8058cd673c5d16af200c684659 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -393,6 +393,8 @@ struct amdgpu_fence_driver {
        unsigned                        irq_type;
        struct timer_list               fallback_timer;
        wait_queue_head_t               fence_queue;
+       unsigned                        num_fences_mask;
+       struct fence                    **fences;
 };
 
 /* some special values for the owner field */

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 4654113ecfae2f772383d2eeaccf901fc3b0468e..44eac91163eb5c2b4a7736aef634d7ab57f86aa0 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -124,6 +124,8 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **f)
 {
        struct amdgpu_device *adev = ring->adev;
        struct amdgpu_fence *fence;
+       struct fence *old, **ptr;
+       unsigned idx;
 
        fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL);
        if (fence == NULL)
@@ -137,7 +139,21 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **f)
                   fence->seq);
        amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
                               fence->seq, AMDGPU_FENCE_FLAG_INT);
+
+       idx = fence->seq & ring->fence_drv.num_fences_mask;
+       ptr = &ring->fence_drv.fences[idx];
+       /* This function can't be called concurrently anyway, otherwise
+        * emitting the fence would mess up the hardware ring buffer.
+        */
+       old = rcu_dereference_protected(*ptr, 1);
+
+       rcu_assign_pointer(*ptr, fence_get(&fence->base));
+
+       BUG_ON(old && !fence_is_signaled(old));
+       fence_put(old);
+
        *f = &fence->base;
+
        return 0;
 }
 
@@ -380,6 +396,11 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
                    (unsigned long)ring);
 
        init_waitqueue_head(&ring->fence_drv.fence_queue);
+       ring->fence_drv.num_fences_mask = num_hw_submission - 1;
+       ring->fence_drv.fences = kcalloc(num_hw_submission, sizeof(void *),
+                                        GFP_KERNEL);
+       if (!ring->fence_drv.fences)
+               return -ENOMEM;
 
        timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
        if (timeout == 0) {
@@ -441,10 +462,9 @@ int amdgpu_fence_driver_init(struct amdgpu_device *adev)
  */
 void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
 {
-       int i, r;
+       unsigned i, j;
+       int r;
 
-       if (atomic_dec_and_test(&amdgpu_fence_slab_ref))
-               kmem_cache_destroy(amdgpu_fence_slab);
        for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
                struct amdgpu_ring *ring = adev->rings[i];
 
@@ -460,8 +480,14 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
                               ring->fence_drv.irq_type);
                amd_sched_fini(&ring->sched);
                del_timer_sync(&ring->fence_drv.fallback_timer);
+               for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
+                       fence_put(ring->fence_drv.fences[i]);
+               kfree(ring->fence_drv.fences);
                ring->fence_drv.initialized = false;
        }
+
+       if (atomic_dec_and_test(&amdgpu_fence_slab_ref))
+               kmem_cache_destroy(amdgpu_fence_slab);
 }
 
 /**