Re: [PATCH] Rework hrtimer_nanosleep to make sys_compat_nanosleep easier

[Arnd Bergmann <arnd@arndb.de>]

On Monday 15 October 2007, Anton Blanchard wrote:
> Pull the copy_to_user out of hrtimer_nanosleep and into the callers
> (common_nsleep, sys_nanosleep) in preparation for converting
> compat_sys_nanosleep to use hrtimers.

Looks good, except for two micro-optimization:

> Signed-off-by: Anton Blanchard <anton@samba.org>

Acked-by: Arnd Bergmann <arnd@arndb.de>

> @@ -1361,7 +1356,14 @@ sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
>         if (!timespec_valid(&tu))
>                 return -EINVAL;
>  
> -       return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
> +       ret = hrtimer_nanosleep(&tu, &rmt, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
> +
> +       if (ret && rmtp) {
> +               if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
> +                       return -EFAULT;
> +       }
> +
> +       return ret;
>  }
>  

If it's common to call sys_nanosleep with a NULL rmtp argument, we could save a
few cycles using 

	return hrtimer_nanosleep(&tu, rmtp ? &rmp : NULL, HRTIMER_MODE_REL,
								 CLOCK_MONOTONIC);

> diff --git a/kernel/posix-timers.c b/kernel/posix-timers.c
> index 57efe04..cce8c75 100644
> --- a/kernel/posix-timers.c
> +++ b/kernel/posix-timers.c
> @@ -980,9 +980,19 @@ sys_clock_getres(const clockid_t which_clock, struct timespec __user *tp)
>  static int common_nsleep(const clockid_t which_clock, int flags,
>                          struct timespec *tsave, struct timespec __user *rmtp)
>  {
> -       return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
> +       struct timespec rmt;
> +       int ret;
> +
> +       ret = hrtimer_nanosleep(tsave, &rmt, flags & TIMER_ABSTIME ?
>                                  HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
>                                  which_clock);
> +
> +       if (ret && rmtp) {
> +               if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
> +                       return -EFAULT;
> +       }
> +
> +       return ret;
>  }

I think it would be better here to propagate the move to a kernel *rmtp
down to sys_clock_nanosleep so we get the same optimization in
compat_sys_clock_nanosleep. That should probably also be a separate
patch. I can do one if you don't do it first.

	Arnd <><