Re: [PATCH 1/5] virtiofs: Do not end request in submission context

On Mon, Oct 21, 2019 at 10:03:39AM +0200, Miklos Szeredi wrote:

[..]

quoted

 static void virtio_fs_hiprio_dispatch_work(struct work_struct *work)

@@ -502,6 +522,7 @@ static int virtio_fs_setup_vqs(struct virtio_device *vdev,
        names[VQ_HIPRIO] = fs->vqs[VQ_HIPRIO].name;
        INIT_WORK(&fs->vqs[VQ_HIPRIO].done_work, virtio_fs_hiprio_done_work);
        INIT_LIST_HEAD(&fs->vqs[VQ_HIPRIO].queued_reqs);
+       INIT_LIST_HEAD(&fs->vqs[VQ_HIPRIO].end_reqs);
        INIT_DELAYED_WORK(&fs->vqs[VQ_HIPRIO].dispatch_work,
                        virtio_fs_hiprio_dispatch_work);
        spin_lock_init(&fs->vqs[VQ_HIPRIO].lock);

@@ -511,8 +532,9 @@ static int virtio_fs_setup_vqs(struct virtio_device *vdev,
                spin_lock_init(&fs->vqs[i].lock);
                INIT_WORK(&fs->vqs[i].done_work, virtio_fs_requests_done_work);
                INIT_DELAYED_WORK(&fs->vqs[i].dispatch_work,
-                                       virtio_fs_dummy_dispatch_work);
+                                 virtio_fs_request_dispatch_work);
                INIT_LIST_HEAD(&fs->vqs[i].queued_reqs);
+               INIT_LIST_HEAD(&fs->vqs[i].end_reqs);
                snprintf(fs->vqs[i].name, sizeof(fs->vqs[i].name),
                         "requests.%u", i - VQ_REQUEST);
                callbacks[i] = virtio_fs_vq_done;

@@ -918,6 +940,7 @@ __releases(fiq->lock)
        struct fuse_conn *fc;
        struct fuse_req *req;
        struct fuse_pqueue *fpq;
+       struct virtio_fs_vq *fsvq;
        int ret;

        WARN_ON(list_empty(&fiq->pending));

@@ -951,7 +974,8 @@ __releases(fiq->lock)
        smp_mb__after_atomic();

 retry:
-       ret = virtio_fs_enqueue_req(&fs->vqs[queue_id], req);
+       fsvq = &fs->vqs[queue_id];
+       ret = virtio_fs_enqueue_req(fsvq, req);
        if (ret < 0) {
                if (ret == -ENOMEM || ret == -ENOSPC) {
                        /* Virtqueue full. Retry submission */

@@ -965,7 +989,13 @@ __releases(fiq->lock)
                clear_bit(FR_SENT, &req->flags);
                list_del_init(&req->list);
                spin_unlock(&fpq->lock);
-               fuse_request_end(fc, req);
+
+               /* Can't end request in submission context. Use a worker */
+               spin_lock(&fsvq->lock);
+               list_add_tail(&req->list, &fsvq->end_reqs);
+               schedule_delayed_work(&fsvq->dispatch_work,
+                                     msecs_to_jiffies(1));

What's the reason to delay by one msec?  If this is purely for
deadlock avoidance, then a zero delay would work better, no?

Hi Miklos,

I have no good reason to do that. Will change it to zero delay.

Thanks
Vivek