drm/nouveau/fifo/gf100-: protect channel preempt with subdev mutex

This avoids an issue that occurs when we're attempting to preempt multiple
channels simultaneously.  HW seems to ignore preempt requests while it's
still processing a previous one, which, well, makes sense.

Fixes random "fifo: SCHED_ERROR 0d []" + GPCCS page faults during parallel
piglit runs on (at least) GM107.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Cc: stable@vger.kernel.org

2 files changed, 11 insertions, 6 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
index cbc67f262322..12d964260a29 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
@@ -60,6 +60,7 @@ gf100_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *base,
         struct nvkm_gpuobj *inst = chan->base.inst;
         int ret = 0;
 
+        mutex_lock(&subdev->mutex);
         nvkm_wr32(device, 0x002634, chan->base.chid);
         if (nvkm_msec(device, 2000,
                 if (nvkm_rd32(device, 0x002634) == chan->base.chid)
@@ -67,10 +68,12 @@ gf100_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *base,
         ) < 0) {
                 nvkm_error(subdev, "channel %d [%s] kick timeout\n",
                            chan->base.chid, chan->base.object.client->name);
-                ret = -EBUSY;
-                if (suspend)
-                        return ret;
+                ret = -ETIMEDOUT;
         }
+        mutex_unlock(&subdev->mutex);
+
+        if (ret && suspend)
+                return ret;
 
         if (offset) {
                 nvkm_kmap(inst);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
index ed4351032ed6..a2df4f3e7763 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
@@ -40,7 +40,9 @@ gk104_fifo_gpfifo_kick(struct gk104_fifo_chan *chan)
         struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
         struct nvkm_device *device = subdev->device;
         struct nvkm_client *client = chan->base.object.client;
+        int ret = 0;
 
+        mutex_lock(&subdev->mutex);
         nvkm_wr32(device, 0x002634, chan->base.chid);
         if (nvkm_msec(device, 2000,
                 if (!(nvkm_rd32(device, 0x002634) & 0x00100000))
@@ -48,10 +50,10 @@ gk104_fifo_gpfifo_kick(struct gk104_fifo_chan *chan)
         ) < 0) {
                 nvkm_error(subdev, "channel %d [%s] kick timeout\n",
                            chan->base.chid, client->name);
-                return -EBUSY;
+                ret = -ETIMEDOUT;
         }
-
-        return 0;
+        mutex_unlock(&subdev->mutex);
+        return ret;
 }
 
 static u32