Re: [PATCH 2/5] hurd: Microoptimize sigreturn

[Samuel Thibault]

Sergey Bugaev, le mar. 18 avril 2023 13:21:50 +0300, a ecrit:
> (Git managed to eat my email -- it both sent an empty one *and*
> truncated my local file (what?); so retyping and resending this.)
> 
> On Tue, Apr 18, 2023 at 2:23 AM Samuel Thibault <samuel.thibault@gnu.org> 
> wrote:
> > That doesn't seem to be true, we get segfaults with this patch applied
> > in these tests:
> >
> > ./htl/tst-kill6.test-result:FAIL: htl/tst-kill6
> > ./htl/tst-pt-tls2.test-result:FAIL: htl/tst-pt-tls2
> >
> > #0  0x0105a7c2 in sigstate_is_global_rcv (ss=0x0) at hurdsig.c:183
> > 183              && (ss->actions[0].sa_handler == SIG_IGN);
> > (gdb) bt
> > #0  0x0105a7c2 in sigstate_is_global_rcv (ss=0x0) at hurdsig.c:183
> > #1  __GI__hurd_sigstate_lock (ss=0x0) at hurdsig.c:192
> > #2  0x0108bb17 in __sigreturn (scp=0x2104bec) at 
> > ../sysdeps/mach/hurd/i386/sigreturn.c:94
> > #3  0x010602f6 in trampoline () from ./libc.so.0.3
> 
> Interesting; so while the sigstate for the thread must have been
> initialized, it seems it has never been installed into the
> tcb->_hurd_sigstate slot, because the thread has never called
> _hurd_self_sigstate (). So we have to make the full call to
> _hurd_self_sigstate () the first time, to find the sigstate and
> install it into the slot.
> 
> Please try this:

That one works fine, applied, thanks!


> Sergey
> 
> ---- >8 ----
> 
> 
> From 94fc9499390b4de9fbe10c3014d473ac42878c6a Mon Sep 17 00:00:00 2001
> From: Sergey Bugaev <bugaevc@gmail.com>
> Date: Fri, 14 Apr 2023 16:32:45 +0300
> Subject: [PATCH] hurd: Microoptimize sigreturn
> 
> Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
> ---
>  sysdeps/mach/hurd/i386/sigreturn.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/sysdeps/mach/hurd/i386/sigreturn.c 
> b/sysdeps/mach/hurd/i386/sigreturn.c
> index a0fc8891..1686734d 100644
> --- a/sysdeps/mach/hurd/i386/sigreturn.c
> +++ b/sysdeps/mach/hurd/i386/sigreturn.c
> @@ -29,7 +29,12 @@ static void
>  __sigreturn2 (int *usp, struct sigcontext *scp)
>  {
>    mach_port_t reply_port;
> -  struct hurd_sigstate *ss = _hurd_self_sigstate ();
> +  struct hurd_sigstate *ss;
> +
> +  /* We know the sigstate must be initialized by the call below, but the
> +     compiler does not.  Help it out a little bit by eliding the check that
> +     _hurd_self_sigstate makes otherwise.  */
> +  ss = THREAD_GETMEM (THREAD_SELF, _hurd_sigstate);
>    _hurd_sigstate_unlock (ss);
>  
>    /* Destroy the MiG reply port used by the signal handler, and restore the
> @@ -44,7 +49,7 @@ __sigreturn2 (int *usp, struct sigcontext *scp)
>       do.  */
>    reply_port = THREAD_GETMEM (THREAD_SELF, reply_port);
>    THREAD_SETMEM (THREAD_SELF, reply_port, MACH_PORT_DEAD);
> -  if (MACH_PORT_VALID (reply_port))
> +  if (__glibc_likely (MACH_PORT_VALID (reply_port)))
>      (void) __mach_port_mod_refs (__mach_task_self (), reply_port,
>                                   MACH_PORT_RIGHT_RECEIVE, -1);
>    THREAD_SETMEM (THREAD_SELF, reply_port, scp->sc_reply_port);
> @@ -69,6 +74,7 @@ __sigreturn2 (int *usp, struct sigcontext *scp)
>    /* Firewall.  */
>    A (hlt);
>  #undef A
> +  __builtin_unreachable ();
>  }
>  
>  int
> @@ -77,7 +83,7 @@ __sigreturn (struct sigcontext *scp)
>    struct hurd_sigstate *ss;
>    struct hurd_userlink *link = (void *) &scp[1];
>  
> -  if (scp == NULL || (scp->sc_mask & _SIG_CANT_MASK))
> +  if (__glibc_unlikely (scp == NULL || (scp->sc_mask & _SIG_CANT_MASK)))
>      {
>        errno = EINVAL;
>        return -1;
> -- 
> 2.39.2
> 

-- 
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.