Re: [PATCH 4/5] io_uring: add a helper for poll clean

[Hao Xu]

在 2022/5/7 上午12:22, Pavel Begunkov 写道:
> On 5/6/22 08:01, Hao Xu wrote:
> > From: Hao Xu <howeyxu@xxxxxxxxxxx>
> >
> > Add a helper for poll clean, it will be used in the multishot accept in
> > the later patches.
> >
> > Signed-off-by: Hao Xu <howeyxu@xxxxxxxxxxx>
> > ---
> >   fs/io_uring.c | 23 ++++++++++++++++++-----
> >   1 file changed, 18 insertions(+), 5 deletions(-)
> >
> > diff --git a/fs/io_uring.c b/fs/io_uring.c
> > index d33777575faf..0a83ecc457d1 100644
> > --- a/fs/io_uring.c
> > +++ b/fs/io_uring.c
> > @@ -5711,6 +5711,23 @@ static int io_accept_prep(struct io_kiocb *req, 
> > const struct io_uring_sqe *sqe)
> >       return 0;
> >   }
> > +static inline void __io_poll_clean(struct io_kiocb *req)
> > +{
> > +    struct io_ring_ctx *ctx = req->ctx;
> > +
> > +    io_poll_remove_entries(req);
> > +    spin_lock(&ctx->completion_lock);
> > +    hash_del(&req->hash_node);
> > +    spin_unlock(&ctx->completion_lock);
> > +}
> > +
> > +#define REQ_F_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
> > +static inline void io_poll_clean(struct io_kiocb *req)
> > +{
> > +    if ((req->flags & REQ_F_APOLL_MULTI_POLLED) == 
> > REQ_F_APOLL_MULTI_POLLED)
>
> So it triggers for apoll multishot only when REQ_F_POLLED is _not_ set,
> but if it's not set it did never go through arm_poll / etc. and there is
> nothing to clean up. What is the catch?

No, it is triggered for apoll multishot only when REQ_F_POLLED is set..
> btw, don't see the function used in this patch, better to add it later
> or at least mark with attribute unused, or some may get build failures.
Gotcha.
> > +        __io_poll_clean(req);
> > +}
> > +
> >   static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
> >   {
> >       struct io_accept *accept = &req->accept;
> > @@ -6041,17 +6058,13 @@ static void io_poll_task_func(struct io_kiocb 
> > *req, bool *locked)
> >   static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
> >   {
> > -    struct io_ring_ctx *ctx = req->ctx;
> >       int ret;
> >       ret = io_poll_check_events(req, locked);
> >       if (ret > 0)
> >           return;
> > -    io_poll_remove_entries(req);
> > -    spin_lock(&ctx->completion_lock);
> > -    hash_del(&req->hash_node);
> > -    spin_unlock(&ctx->completion_lock);
> > +    __io_poll_clean(req);
> >       if (!ret)
> >           io_req_task_submit(req, locked);