On Wed, Apr 25, 2012 at 07:02:30PM +0200, Oleg Nesterov wrote:
> And there is another case:
>
> syscall interrupted, it returns ERESTARTSYS.
>
> do_signal() sees no signal, restarts the syscall
>
> !SA_RESTART comes before we return to user-mode
>
> This doesn't really differ from sigsuspend/ERESTARTNOHAND we discussed
> before.
... and dealt with the same way:
NEED_RESTART flag is set the first time around
no handler for the first signal, so we do nothing else
handler is found for the second signal
NEED_RESTART flag is present, so
we clear NEED_RESTART
we look at errno and see ERESTARTSYS
we check sa_flags and see !SA_RESTART
we set return value to -EINTR and proceed to set sigframe up
on the exit to userland we see NEED_RESTART not set, so off we go
I think I hadn't been clear enough; here's pseudocode for what I meant
do_signal()
{
if (we have any business doing restarts)
// note: we won't get here on subsequent calls of do_signal()
// due to the checks above; same logics that currently prevents
// double restarts
set NEED_RESTART flag
sig = get_signal_to_deliver(...)
if (sig) {
if (NEED_RESTART set) {
clear NEED_RESTART
same thing we do at that spot now - restart or EINTR
handle_signal(...)
...
return;
}
}
/* no handler */
if (test_and_clear_...(RESTORE_SIGMASK))
set_current_blocked(¤t->saved_sigmask);
}
and in asm glue, *after* checking for SIGPENDING/NOTIFY_RESUME, check
NEED_RESTART and if it's set do what we currently do for restarts on
handlerless signal.
BTW, if we combine that with ERESTART_RESTARTBLOCK trick on arm/parisc/unicore,
there won't be much work to do for restarts in that case; in x86ese it
would be simply
if (regs->ax == -ERESTART_RESTARTBLOCK)
set_thread_flag(TIF_EMULATE_RESTART_SYSCALL)
regs->ax = regs->orig_ax;
regs->ip -= 2;
all of that conditional on TS_NEED_RESTART...
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
