Re: [PATCH 1/2] io_uring: avoid ring quiesce while registering/unregistering eventfd

[Usama Arif]

On 03/02/2022 15:48, Pavel Begunkov wrote:
> On 2/3/22 15:11, Usama Arif wrote:
> > This is done by creating a new RCU data structure (io_ev_fd) as part of
> > io_ring_ctx that holds the eventfd_ctx.
> >
> > The function io_eventfd_signal is executed under rcu_read_lock with a
> > single rcu_dereference to io_ev_fd so that if another thread unregisters
> > the eventfd while io_eventfd_signal is still being executed, the
> > eventfd_signal for which io_eventfd_signal was called completes
> > successfully.
> >
> > The process of registering/unregistering eventfd is done under a lock
> > so multiple threads don't enter a race condition while
> > registering/unregistering eventfd.
> >
> > With the above approach ring quiesce can be avoided which is much more
> > expensive then using RCU lock. On the system tested, io_uring_reigster 
> > with
> > IORING_REGISTER_EVENTFD takes less than 1ms with RCU lock, compared to 
> > 15ms
> > before with ring quiesce.
> >
> > Signed-off-by: Usama Arif <usama.arif@xxxxxxxxxxxxx>
> > ---
> >   fs/io_uring.c | 103 +++++++++++++++++++++++++++++++++++++++-----------
> >   1 file changed, 80 insertions(+), 23 deletions(-)
> >
> > diff --git a/fs/io_uring.c b/fs/io_uring.c
> > index 2e04f718319d..f07cfbb387a6 100644
> > --- a/fs/io_uring.c
> > +++ b/fs/io_uring.c
> > @@ -326,6 +326,12 @@ struct io_submit_state {
> >       struct blk_plug        plug;
> >   };
>
> > -static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
> > +static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx, 
> > struct io_ev_fd *ev_fd)
> >   {
> > -    if (likely(!ctx->cq_ev_fd))
> > +    if (likely(!ev_fd))
> >           return false;
> >       if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
> >           return false;
> >       return !ctx->eventfd_async || io_wq_current_is_worker();
> >   }
> > +static void io_eventfd_signal(struct io_ring_ctx *ctx)
> > +{
> > +    struct io_ev_fd *ev_fd;
> > +
> > +    rcu_read_lock();
>
> Please always think about the fast path, which is not set eventfd.
> We don't want extra overhead here.

I guess this should be ok now from v3?

Thanks,
Usama
> if (ctx->ev_fd) {
>      rcu_read_lock();
>          ev_fd = rcu_deref(...);
>          ...
>          rcu_read_unlock();
> }