drm/amdgpu: protect fence_process from multiple context

fence_process may be called from kthread, user thread and interrupt context.
it is possible to called concurrently, then will wake up fence queue multiple times.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

3 files changed, 7 insertions, 2 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 1b8d05ff88e7..0703fbfd5130 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -869,6 +869,7 @@ struct amdgpu_ring {
         struct amdgpu_fence_driver      fence_drv;
         struct amd_gpu_scheduler        *scheduler;
 
+        spinlock_t              fence_lock;
         struct mutex            *ring_lock;
         struct amdgpu_bo        *ring_obj;
         volatile uint32_t       *ring;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 1580d8d7a3bf..b0e15b574606 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -295,6 +295,7 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
         uint64_t seq, last_seq, last_emitted;
         unsigned count_loop = 0;
         bool wake = false;
+        unsigned long irqflags;
 
         /* Note there is a scenario here for an infinite loop but it's
          * very unlikely to happen. For it to happen, the current polling
@@ -317,6 +318,7 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
          * have temporarly set the last_seq not to the true real last
          * seq but to an older one.
          */
+        spin_lock_irqsave(&ring->fence_lock, irqflags);
         last_seq = atomic64_read(&ring->fence_drv.last_seq);
         do {
                 last_emitted = ring->fence_drv.sync_seq[ring->idx];
@@ -355,7 +357,7 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
                         if (handled_seq == latest_seq) {
                                 DRM_ERROR("ring %d, EOP without seq update (lastest_seq=%llu)\n",
                                           ring->idx, latest_seq);
-                                return;
+                                goto exit;
                         }
                         do {
                                 amd_sched_isr(ring->scheduler);
@@ -364,6 +366,8 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
 
                 wake_up_all(&ring->adev->fence_queue);
         }
+exit:
+        spin_unlock_irqrestore(&ring->fence_lock, irqflags);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 855e2196657a..1e68a561bbfe 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -367,7 +367,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
         }
         ring->next_rptr_gpu_addr = adev->wb.gpu_addr + (ring->next_rptr_offs * 4);
         ring->next_rptr_cpu_addr = &adev->wb.wb[ring->next_rptr_offs];
-
+        spin_lock_init(&ring->fence_lock);
         r = amdgpu_fence_driver_start_ring(ring, irq_src, irq_type);
         if (r) {
                 dev_err(adev->dev, "failed initializing fences (%d).\n", r);