On Mon, Apr 21, 2025 at 10:45 AM Cindy Lu <l...@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 <l...@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
