OSDN Git Service

drm/radeon: Add radeon_test_syncing function v2
authorChristian König <deathsimple@vodafone.de>
Tue, 27 Sep 2011 10:31:00 +0000 (12:31 +0200)
committerDave Airlie <airlied@redhat.com>
Tue, 20 Dec 2011 19:50:19 +0000 (19:50 +0000)
Tests syncing between all rings by using
semaphores and fences.

v2: use radeon_testing as a bit flag rather than on/off switch
    this allow to test for one thing at a time (bo_move or semaphore
    test). It kind of break the usage if user wheren't using 1
    for bo move test but as it's a test feature i believe it's ok.

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/radeon/radeon.h
drivers/gpu/drm/radeon/radeon_device.c
drivers/gpu/drm/radeon/radeon_test.c

index 76c58e9..fbe902b 100644 (file)
@@ -913,6 +913,10 @@ void radeon_benchmark(struct radeon_device *rdev, int test_number);
  * Testing
  */
 void radeon_test_moves(struct radeon_device *rdev);
+void radeon_test_ring_sync(struct radeon_device *rdev,
+                          struct radeon_cp *cpA,
+                          struct radeon_cp *cpB);
+void radeon_test_syncing(struct radeon_device *rdev);
 
 
 /*
index 023c156..fa36b53 100644 (file)
@@ -823,9 +823,12 @@ int radeon_device_init(struct radeon_device *rdev,
                if (r)
                        return r;
        }
-       if (radeon_testing) {
+       if ((radeon_testing & 1)) {
                radeon_test_moves(rdev);
        }
+       if ((radeon_testing & 2)) {
+               radeon_test_syncing(rdev);
+       }
        if (radeon_benchmarking) {
                radeon_benchmark(rdev, radeon_benchmarking);
        }
index 160e7df..5f4d31e 100644 (file)
@@ -234,3 +234,91 @@ out_cleanup:
                printk(KERN_WARNING "Error while testing BO move.\n");
        }
 }
+
+void radeon_test_ring_sync(struct radeon_device *rdev,
+                          struct radeon_cp *cpA,
+                          struct radeon_cp *cpB)
+{
+       struct radeon_fence *fence = NULL;
+       struct radeon_semaphore *semaphore = NULL;
+       int ringA = radeon_ring_index(rdev, cpA);
+       int ringB = radeon_ring_index(rdev, cpB);
+       int r;
+
+       r = radeon_fence_create(rdev, &fence, ringA);
+       if (r) {
+               DRM_ERROR("Failed to create sync fence\n");
+               goto out_cleanup;
+       }
+
+       r = radeon_semaphore_create(rdev, &semaphore);
+       if (r) {
+               DRM_ERROR("Failed to create semaphore\n");
+               goto out_cleanup;
+       }
+
+       r = radeon_ring_lock(rdev, cpA, 64);
+       if (r) {
+               DRM_ERROR("Failed to lock ring %d\n", ringA);
+               goto out_cleanup;
+       }
+       radeon_semaphore_emit_wait(rdev, ringA, semaphore);
+       radeon_fence_emit(rdev, fence);
+       radeon_ring_unlock_commit(rdev, cpA);
+
+       mdelay(1000);
+
+       if (radeon_fence_signaled(fence)) {
+               DRM_ERROR("Fence signaled without waiting for semaphore.\n");
+               goto out_cleanup;
+       }
+
+       r = radeon_ring_lock(rdev, cpB, 64);
+       if (r) {
+               DRM_ERROR("Failed to lock ring %d\n", ringB);
+               goto out_cleanup;
+       }
+       radeon_semaphore_emit_signal(rdev, ringB, semaphore);
+       radeon_ring_unlock_commit(rdev, cpB);
+
+       r = radeon_fence_wait(fence, false);
+       if (r) {
+               DRM_ERROR("Failed to wait for sync fence\n");
+               goto out_cleanup;
+       }
+
+       DRM_INFO("Syncing between rings %d and %d seems to work.\n", ringA, ringB);
+
+out_cleanup:
+       if (semaphore)
+               radeon_semaphore_free(rdev, semaphore);
+
+       if (fence)
+               radeon_fence_unref(&fence);
+
+       if (r)
+               printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
+}
+
+void radeon_test_syncing(struct radeon_device *rdev)
+{
+       int i, j;
+
+       for (i = 1; i < RADEON_NUM_RINGS; ++i) {
+               struct radeon_cp *cpA = &rdev->cp[i];
+               if (!cpA->ready)
+                       continue;
+
+               for (j = 0; j < i; ++j) {
+                       struct radeon_cp *cpB = &rdev->cp[j];
+                       if (!cpB->ready)
+                               continue;
+
+                       DRM_INFO("Testing syncing between rings %d and %d\n", i, j);
+                       radeon_test_ring_sync(rdev, cpA, cpB);
+
+                       DRM_INFO("Testing syncing between rings %d and %d\n", j, i);
+                       radeon_test_ring_sync(rdev, cpB, cpA);
+               }
+       }
+}