drm/radeon: implement ring saving on reset v4

Try to save whatever is on the rings when
we encounter an lockup.

v2: Fix spelling error. Free saved ring data if reset fails.
    Add documentation for the new functions.
v3: Some more spelling fixes
v4: It doesn't make sense to save anything if all fences
    are signaled

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/radeon/radeon.h

@@ -768,6 +768,10 @@ int radeon_ring_test(struct radeon_device *rdev, struct radeon_ring *cp);
 void radeon_ring_force_activity(struct radeon_device *rdev, struct radeon_ring *ring);
 void radeon_ring_lockup_update(struct radeon_ring *ring);
 bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring);
+unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
+			    uint32_t **data);
+int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
+			unsigned size, uint32_t *data);
 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *cp, unsigned ring_size,
 		     unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
 		     u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop);

drivers/gpu/drm/radeon/radeon_device.c

@@ -996,7 +996,12 @@ int radeon_resume_kms(struct drm_device *dev)
 
 int radeon_gpu_reset(struct radeon_device *rdev)
 {
-	int r;
+	unsigned ring_sizes[RADEON_NUM_RINGS];
+	uint32_t *ring_data[RADEON_NUM_RINGS];
+
+	bool saved = false;
+
+	int i, r;
 	int resched;
 
 	down_write(&rdev->exclusive_lock);
@@ -1005,20 +1010,47 @@ int radeon_gpu_reset(struct radeon_device *rdev)
 	resched = ttm_bo_lock_delayed_workqueue(&rdev->mman.bdev);
 	radeon_suspend(rdev);
 
-	r = radeon_asic_reset(rdev);
-	if (!r) {
-		dev_info(rdev->dev, "GPU reset succeed\n");
-		radeon_resume(rdev);
-
-		r = radeon_ib_ring_tests(rdev);
-		if (r)
-			DRM_ERROR("ib ring test failed (%d).\n", r);
-
-		radeon_restore_bios_scratch_regs(rdev);
-		drm_helper_resume_force_mode(rdev->ddev);
-		ttm_bo_unlock_delayed_workqueue(&rdev->mman.bdev, resched);
+	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+		ring_sizes[i] = radeon_ring_backup(rdev, &rdev->ring[i],
+						   &ring_data[i]);
+		if (ring_sizes[i]) {
+			saved = true;
+			dev_info(rdev->dev, "Saved %d dwords of commands "
+				 "on ring %d.\n", ring_sizes[i], i);
+		}
 	}
 
+retry:
+	r = radeon_asic_reset(rdev);
+	if (!r) {
+		dev_info(rdev->dev, "GPU reset succeeded, trying to resume\n");
+		radeon_resume(rdev);
+	}
+
+	radeon_restore_bios_scratch_regs(rdev);
+	drm_helper_resume_force_mode(rdev->ddev);
+
+	if (!r) {
+		for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+			radeon_ring_restore(rdev, &rdev->ring[i],
+					    ring_sizes[i], ring_data[i]);
+		}
+
+		r = radeon_ib_ring_tests(rdev);
+		if (r) {
+			dev_err(rdev->dev, "ib ring test failed (%d).\n", r);
+			if (saved) {
+				radeon_suspend(rdev);
+				goto retry;
+			}
+		}
+	} else {
+		for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+			kfree(ring_data[i]);
+		}
+	}
+
+	ttm_bo_unlock_delayed_workqueue(&rdev->mman.bdev, resched);
 	if (r) {
 		/* bad news, how to tell it to userspace ? */
 		dev_info(rdev->dev, "GPU reset failed\n");

drivers/gpu/drm/radeon/radeon_ring.c

@@ -362,6 +362,88 @@ bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *rin
 	return false;
 }
 
+/**
+ * radeon_ring_backup - Back up the content of a ring
+ *
+ * @rdev: radeon_device pointer
+ * @ring: the ring we want to back up
+ *
+ * Saves all unprocessed commits from a ring, returns the number of dwords saved.
+ */
+unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
+			    uint32_t **data)
+{
+	unsigned size, ptr, i;
+	int ridx = radeon_ring_index(rdev, ring);
+
+	/* just in case lock the ring */
+	mutex_lock(&rdev->ring_lock);
+	*data = NULL;
+
+	if (ring->ring_obj == NULL || !ring->rptr_save_reg) {
+		mutex_unlock(&rdev->ring_lock);
+		return 0;
+	}
+
+	/* it doesn't make sense to save anything if all fences are signaled */
+	if (!radeon_fence_count_emitted(rdev, ridx)) {
+		mutex_unlock(&rdev->ring_lock);
+		return 0;
+	}
+
+	/* calculate the number of dw on the ring */
+	ptr = RREG32(ring->rptr_save_reg);
+	size = ring->wptr + (ring->ring_size / 4);
+	size -= ptr;
+	size &= ring->ptr_mask;
+	if (size == 0) {
+		mutex_unlock(&rdev->ring_lock);
+		return 0;
+	}
+
+	/* and then save the content of the ring */
+	*data = kmalloc(size * 4, GFP_KERNEL);
+	for (i = 0; i < size; ++i) {
+		(*data)[i] = ring->ring[ptr++];
+		ptr &= ring->ptr_mask;
+	}
+
+	mutex_unlock(&rdev->ring_lock);
+	return size;
+}
+
+/**
+ * radeon_ring_restore - append saved commands to the ring again
+ *
+ * @rdev: radeon_device pointer
+ * @ring: ring to append commands to
+ * @size: number of dwords we want to write
+ * @data: saved commands
+ *
+ * Allocates space on the ring and restore the previously saved commands.
+ */
+int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
+			unsigned size, uint32_t *data)
+{
+	int i, r;
+
+	if (!size || !data)
+		return 0;
+
+	/* restore the saved ring content */
+	r = radeon_ring_lock(rdev, ring, size);
+	if (r)
+		return r;
+
+	for (i = 0; i < size; ++i) {
+		radeon_ring_write(ring, data[i]);
+	}
+
+	radeon_ring_unlock_commit(rdev, ring);
+	kfree(data);
+	return 0;
+}
+
 int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
 		     unsigned rptr_offs, unsigned rptr_reg, unsigned wptr_reg,
 		     u32 ptr_reg_shift, u32 ptr_reg_mask, u32 nop)