leandrof/linux
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

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

Browse Source 
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>
This commit is contained in:
Christian König 2016-03-13 19:19:48 +01:00 committed by Alex Deucher
parent e6151a08bb
commit c89377d10a
2 changed files with 31 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
drivers/gpu/drm/amd/amdgpu/amdgpu.h
View File

@ -393,6 +393,8 @@ struct amdgpu_fence_driver {
unsigned irq_type; unsigned irq_type;
struct timer_list fallback_timer; struct timer_list fallback_timer;
wait_queue_head_t fence_queue; wait_queue_head_t fence_queue;
unsigned num_fences_mask;
struct fence **fences;
}; };
/* some special values for the owner field */ /* some special values for the owner field */

32
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
View File

@ -124,6 +124,8 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **f)
{ {
struct amdgpu_device *adev = ring->adev; struct amdgpu_device *adev = ring->adev;
struct amdgpu_fence *fence; struct amdgpu_fence *fence;
struct fence *old, **ptr;
unsigned idx;
fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL); fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL);
if (fence == NULL) if (fence == NULL)
@ -137,7 +139,21 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **f)
fence->seq); fence->seq);
amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
fence->seq, AMDGPU_FENCE_FLAG_INT); 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; *f = &fence->base;
return 0; return 0;
} }
@ -380,6 +396,11 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
(unsigned long)ring); (unsigned long)ring);
init_waitqueue_head(&ring->fence_drv.fence_queue); 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); timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
if (timeout == 0) { 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) 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++) { for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
struct amdgpu_ring *ring = adev->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); ring->fence_drv.irq_type);
amd_sched_fini(&ring->sched); amd_sched_fini(&ring->sched);
del_timer_sync(&ring->fence_drv.fallback_timer); 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; ring->fence_drv.initialized = false;
} }
if (atomic_dec_and_test(&amdgpu_fence_slab_ref))
kmem_cache_destroy(amdgpu_fence_slab);
} }
/** /**