Re: [PATCH v9 2/4] vhost: Reintroduce kthread mode support in vhost

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Apr 21, 2025 at 11:39:14AM +0800, Jason Wang wrote:
> On Mon, Apr 21, 2025 at 10:45 AM Cindy Lu <lulu@redhat.com> wrote:
> >
> > This patch reintroduces kthread mode support in vhost,
> > It also introduces struct vhost_worker_ops to abstract
> > worker create/stop/wakeup operations.
> >
> > * Bring back the original vhost_worker() implementation,
> >   and renamed to vhost_run_work_kthread_list().
> >
> > * Add cgroup support for the kthread
> >
> > * Introduce struct vhost_worker_ops:
> >   - Encapsulates create / stop / wake‑up callbacks.
> >   - vhost_worker_create() selects the proper ops according to
> >     inherit_owner.
> >
> > This partially reverts or improves upon:
> > commit 6e890c5d5021 ("vhost: use vhost_tasks for worker threads")
> > commit 1cdaafa1b8b4 ("vhost: replace single worker pointer with xarray")
> >
> > Signed-off-by: Cindy Lu <lulu@redhat.com>
> > ---
> >  drivers/vhost/vhost.c | 188 ++++++++++++++++++++++++++++++++++++++----
> >  drivers/vhost/vhost.h |  12 +++
> >  2 files changed, 182 insertions(+), 18 deletions(-)
> >
> > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
> > index 250dc43f1786..be97028a8baf 100644
> > --- a/drivers/vhost/vhost.c
> > +++ b/drivers/vhost/vhost.c
> > @@ -22,6 +22,7 @@
> >  #include <linux/slab.h>
> >  #include <linux/vmalloc.h>
> >  #include <linux/kthread.h>
> > +#include <linux/cgroup.h>
> >  #include <linux/module.h>
> >  #include <linux/sort.h>
> >  #include <linux/sched/mm.h>
> > @@ -242,7 +243,7 @@ static void vhost_worker_queue(struct vhost_worker *worker,
> >                  * test_and_set_bit() implies a memory barrier.
> >                  */
> >                 llist_add(&work->node, &worker->work_list);
> > -               vhost_task_wake(worker->vtsk);
> > +               worker->ops->wakeup(worker);
> >         }
> >  }
> >
> > @@ -388,6 +389,44 @@ static void vhost_vq_reset(struct vhost_dev *dev,
> >         __vhost_vq_meta_reset(vq);
> >  }
> >
> > +static int vhost_run_work_kthread_list(void *data)
> > +{
> > +       struct vhost_worker *worker = data;
> > +       struct vhost_work *work, *work_next;
> > +       struct vhost_dev *dev = worker->dev;
> > +       struct llist_node *node;
> > +
> > +       kthread_use_mm(dev->mm);
> > +
> > +       for (;;) {
> > +               /* mb paired w/ kthread_stop */
> > +               set_current_state(TASK_INTERRUPTIBLE);
> > +
> > +               if (kthread_should_stop()) {
> > +                       __set_current_state(TASK_RUNNING);
> > +                       break;
> > +               }
> > +               node = llist_del_all(&worker->work_list);
> > +               if (!node)
> > +                       schedule();
> > +
> > +               node = llist_reverse_order(node);
> > +               /* make sure flag is seen after deletion */
> > +               smp_wmb();
> > +               llist_for_each_entry_safe(work, work_next, node, node) {
> > +                       clear_bit(VHOST_WORK_QUEUED, &work->flags);
> > +                       __set_current_state(TASK_RUNNING);
> > +                       kcov_remote_start_common(worker->kcov_handle);
> > +                       work->fn(work);
> > +                       kcov_remote_stop();
> > +                       cond_resched();
> > +               }
> > +       }
> > +       kthread_unuse_mm(dev->mm);
> > +
> > +       return 0;
> > +}
> > +
> >  static bool vhost_run_work_list(void *data)
> >  {
> >         struct vhost_worker *worker = data;
> > @@ -582,6 +621,46 @@ long vhost_dev_check_owner(struct vhost_dev *dev)
> >  }
> >  EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
> >
> > +struct vhost_attach_cgroups_struct {
> > +       struct vhost_work work;
> > +       struct task_struct *owner;
> > +       int ret;
> > +};
> > +
> > +static void vhost_attach_cgroups_work(struct vhost_work *work)
> > +{
> > +       struct vhost_attach_cgroups_struct *s;
> > +
> > +       s = container_of(work, struct vhost_attach_cgroups_struct, work);
> > +       s->ret = cgroup_attach_task_all(s->owner, current);
> > +}
> > +
> > +static int vhost_attach_task_to_cgroups(struct vhost_worker *worker)
> > +{
> > +       struct vhost_attach_cgroups_struct attach;
> > +       int saved_cnt;
> > +
> > +       attach.owner = current;
> > +
> > +       vhost_work_init(&attach.work, vhost_attach_cgroups_work);
> > +       vhost_worker_queue(worker, &attach.work);
> > +
> > +       mutex_lock(&worker->mutex);
> > +
> > +       /*
> > +        * Bypass attachment_cnt check in __vhost_worker_flush:
> > +        * Temporarily change it to INT_MAX to bypass the check
> > +        */
> > +       saved_cnt = worker->attachment_cnt;
> > +       worker->attachment_cnt = INT_MAX;
> > +       __vhost_worker_flush(worker);
> > +       worker->attachment_cnt = saved_cnt;
> 
> I wonder if it's easier to re-introduce the flush that was used before
> vhost kthread to avoid the tricks here. We can have flush ops for
> example.
> 
> Thanks

Nah we do not need ops, __vhost_worker_flush is just an internal
function. Refactor it so we can call the part without the check.

-- 
MST