Re: [RFC PATCH for-next 2/4] epoll: epoll() syscall declaration

[Eric Wong <normalperson@yhbt.net>]

Nathaniel Yazdani <n1ght.4nd.d4y@gmail.com> wrote:
> On Sun, Feb 23, 2014 at 9:32 PM, Eric Wong <normalperson@yhbt.net> wrote:
> > Nathaniel Yazdani <n1ght.4nd.d4y@gmail.com> wrote:
> >> +asmlinkage long sys_epoll(int ep, struct epoll __user *in,
> >> +                       unsigned int inc, struct epoll __user *out,
> >> +                       unsigned int outc, int timeout);
> >
> > I can understand using the new struct for 'in', but 'out' could just be
> > "struct epoll_event *" like sys_epoll_wait, right?
> >
> >>  asmlinkage long sys_epoll_wait(int epfd, struct epoll_event __user *events,
> 
> Yeah and I went back and forth on that, it just seemed to me that the
> inconsistency could be confusing to others... maybe instead of defining a new
> struct to begin with it might make me sense to just have an 'infd' array of file
> descriptors in addition to an 'in' array of epoll_event struct
> (obviously the length
> of these would be identical).

I don't think a separate array for in is a good idea, too error prone
and you lose locality.

For output, some users either end up allocating more memory/retrieve
fewer items with the larger struct for *out.