On Wed, Aug 17, 2011 at 10:17:52AM -0700, David Daney wrote:
> >diff --git a/arch/mips/kernel/scall64-o32.S b/arch/mips/kernel/scall64-o32.S
> >index 46c4763..f48b18e 100644
> >--- a/arch/mips/kernel/scall64-o32.S
> >+++ b/arch/mips/kernel/scall64-o32.S
> >@@ -441,7 +441,7 @@ sys_call_table:
> > PTR sys_fremovexattr /* 4235 */
> > PTR sys_tkill
> > PTR sys_sendfile64
> >- PTR compat_sys_futex
> >+ PTR sys_32_futex
>
> This change is redundant, scall64-o32.S already does the right thing
> so additional zero extending is not needed and is just extra
> instructions to execute for no reason.
Compat_sys_futex is a syscall entry point and for some configurations
such as CONFIG_FTRACE_SYSCALLS SYSCALL_DEFINE*() will do additional work
beyond cleaning up the arguments. The 3 unnecessary shifts are the
overhead we just gotta live with.
> > PTR compat_sys_sched_setaffinity
> > PTR compat_sys_sched_getaffinity /* 4240 */
> > PTR compat_sys_io_setup
>
> But really I think this patch fixes things at the wrong level. Each
> architecture potentially needs a similar patch. What would happen if
> we did something like:
> +++ b/kernel/futex_compat.c
> @@ -180,9 +180,9 @@ err_unlock:
> return ret;
> }
>
> -asmlinkage long compat_sys_futex(u32 __user *uaddr, int op, u32 val,
> - struct compat_timespec __user *utime, u32 __user *uaddr2,
> - u32 val3)
> +SYSCALL_DEFINE6(compat_sys_futex, u32 __user *, uaddr, int , op, u32, val,
> + struct compat_timespec __user *, utime, u32 __user *, uaddr2,
> + u32, val3)
> {
> struct timespec ts;
> ktime_t t, *tp = NULL;
>
> Obviously the function name is wrong, but a varient of
> SYSCALL_DEFINE*() could be created so the proper function names are
> produced.
Right now none of the the generic compat_ functions is wrapped in
SYSCALL_DEFINE* because for some architectures a further wrapper function
is needed. It seems some architectures call compat_ calls directly
without SYSCALL_DEFINE* which with CONFIG_FTRACE_SYSCALLS is a bug ...
Ralf
