Re: [PATCH v2] epoll: add nsec timeout support

[Willem de Bruijn]

On Mon, Nov 16, 2020 at 12:11 PM David Laight <David.Laight@xxxxxxxxxx> wrote:
>
> From: Willem de Bruijn
> > Sent: 16 November 2020 17:01
> >
> > On Mon, Nov 16, 2020 at 11:19 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > >
> > > On Mon, Nov 16, 2020 at 11:10:01AM -0500, Willem de Bruijn wrote:
> > > > diff --git a/include/uapi/linux/eventpoll.h b/include/uapi/linux/eventpoll.h
> > > > index 8a3432d0f0dc..f6ef9c9f8ac2 100644
> > > > --- a/include/uapi/linux/eventpoll.h
> > > > +++ b/include/uapi/linux/eventpoll.h
> > > > @@ -21,6 +21,7 @@
> > > >
> > > >  /* Flags for epoll_create1.  */
> > > >  #define EPOLL_CLOEXEC O_CLOEXEC
> > > > +#define EPOLL_NSTIMEO 0x1
> > > >
> > > >  /* Valid opcodes to issue to sys_epoll_ctl() */
> > > >  #define EPOLL_CTL_ADD 1
> > >
> > > Not a problem with your patch, but this concerns me.  O_CLOEXEC is
> > > defined differently for each architecture, so we need to stay out of
> > > several different bits when we define new flags for EPOLL_*.  Maybe
> > > this:
> > >
> > > /*
> > >  * Flags for epoll_create1.  O_CLOEXEC may be different bits, depending
> > >  * on the CPU architecture.  Reserve the known ones.
> > >  */
> > > #define EPOLL_CLOEXEC           O_CLOEXEC
> > > #define EPOLL_RESERVED_FLAGS    0x00680000
> > > #define EPOLL_NSTIMEO           0x00000001
> >
> > Thanks. Good point, I'll add that in v3.
>
> You could also add a compile assert to check that the flag is reserved.

Like this?

        /* Check the EPOLL_* constant for consistency.  */
        BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC);
+        BUILD_BUG_ON(EPOLL_NSTIMEO & EPOLL_RESERVED_FLAGS);