drm/nouveau: Synchronize with the user channel before GPU object destruction.

There have been reports of PFIFO cache errors during context take down
(fdo bug 31637). They are caused by some GPU objects being taken out
while the channel is still potentially processing commands. Make sure
that all the previous rendering has landed before releasing a GPU
object.

Reported-by: Grzesiek Sójka <pld@pfu.pl>
Reported-by: Patrice Mandin <patmandin@gmail.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Acked-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/nouveau_channel.c

@@ -284,7 +284,6 @@ nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
 	unsigned long flags;
-	int ret;
 
 	/* decrement the refcount, and we're done if there's still refs */
 	if (likely(!atomic_dec_and_test(&chan->users))) {
@@ -297,19 +296,7 @@ nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 	nouveau_debugfs_channel_fini(chan);
 
 	/* give it chance to idle */
-	nouveau_fence_update(chan);
-	if (chan->fence.sequence != chan->fence.sequence_ack) {
-		struct nouveau_fence *fence = NULL;
-
-		ret = nouveau_fence_new(chan, &fence, true);
-		if (ret == 0) {
-			ret = nouveau_fence_wait(fence, false, false);
-			nouveau_fence_unref(&fence);
-		}
-
-		if (ret)
-			NV_ERROR(dev, "Failed to idle channel %d.\n", chan->id);
-	}
+	nouveau_channel_idle(chan);
 
 	/* ensure all outstanding fences are signaled.  they should be if the
 	 * above attempts at idling were OK, but if we failed this'll tell TTM
@@ -388,6 +375,27 @@ nouveau_channel_ref(struct nouveau_channel *chan,
 	*pchan = chan;
 }
 
+void
+nouveau_channel_idle(struct nouveau_channel *chan)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_fence *fence = NULL;
+	int ret;
+
+	nouveau_fence_update(chan);
+
+	if (chan->fence.sequence != chan->fence.sequence_ack) {
+		ret = nouveau_fence_new(chan, &fence, true);
+		if (!ret) {
+			ret = nouveau_fence_wait(fence, false, false);
+			nouveau_fence_unref(&fence);
+		}
+
+		if (ret)
+			NV_ERROR(dev, "Failed to idle channel %d.\n", chan->id);
+	}
+}
+
 /* cleans up all the fifos from file_priv */
 void
 nouveau_channel_cleanup(struct drm_device *dev, struct drm_file *file_priv)

drivers/gpu/drm/nouveau/nouveau_drv.c

@@ -197,22 +197,10 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 
 	NV_INFO(dev, "Idling channels...\n");
 	for (i = 0; i < pfifo->channels; i++) {
-		struct nouveau_fence *fence = NULL;
-
 		chan = dev_priv->channels.ptr[i];
-		if (!chan || !chan->pushbuf_bo)
-			continue;
-
-		ret = nouveau_fence_new(chan, &fence, true);
-		if (ret == 0) {
-			ret = nouveau_fence_wait(fence, false, false);
-			nouveau_fence_unref(&fence);
-		}
 
-		if (ret) {
-			NV_ERROR(dev, "Failed to idle channel %d for suspend\n",
-				 chan->id);
-		}
+		if (chan && chan->pushbuf_bo)
+			nouveau_channel_idle(chan);
 	}
 
 	pgraph->fifo_access(dev, false);

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -847,6 +847,7 @@ extern void nouveau_channel_put_unlocked(struct nouveau_channel **);
 extern void nouveau_channel_put(struct nouveau_channel **);
 extern void nouveau_channel_ref(struct nouveau_channel *chan,
 				struct nouveau_channel **pchan);
+extern void nouveau_channel_idle(struct nouveau_channel *chan);
 
 /* nouveau_object.c */
 #define NVOBJ_CLASS(d,c,e) do {                                                \

drivers/gpu/drm/nouveau/nouveau_object.c

@@ -1017,6 +1017,9 @@ int nouveau_ioctl_gpuobj_free(struct drm_device *dev, void *data,
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
 
+	/* Synchronize with the user channel */
+	nouveau_channel_idle(chan);
+
 	ret = nouveau_ramht_remove(chan, objfree->handle);
 	nouveau_channel_put(&chan);
 	return ret;