Re: [uml-devel] [patch 1/1] uml: replace pause with sigsuspend

[Bodo S. <bst...@fu...>]

Blaisorblade wrote:
> On Tuesday 17 May 2005 12:14, Bodo Stroesser wrote:
> 
>>I didn't have the time to test the patch, but I guess, it won't work.
>>
>>Maybe, the man pages for sigsuspend are somewhat missleading, saying:
>>
>>        The sigsuspend call temporarily replaces the signal mask for
>>        the process with that given by mask and then suspends the
>>        process until a signal is received.
> 
> In fact, I assumed that was correct...
> 
>>Reading the code of sys_sigsuspend() in arch/i386/kernel/signal.c,
>>you'll find that it will return to user only, if do_signal() returns 1.
>>This will happen, if there was a signal to deliver. Ignored signals
>>will not be delivered, so sys_sigsupend will not return on SIGWINCH,
>>if SIGWINCH handler is set to SIG_IGN.
> 
> Doh! You're right, actually.
> 
>>So, for UML's winch_handler, there is no difference between pause()
>>and sigsuspend(), because sigsuspend's feature of accepting a sigmask
>>for use while waiting, in fact isn't needed here as the same mask is
>>set already before with sigprocmask().
> 
> Ok, seems like I'll drop this patch. Thanks for the review.

Maybe, we really should use sigsuspend() instead of pause(), but the
reason for this isn't to remove the winch_handler.
The current loop in winch_thread() might miss a SIGWINCH, if a SIGWINCH
comes in while winch_thread() isn't waiting in wait().
So I think, winch_thread() should block all signals including SIGWINCH.
In its loop it should call sigsuspend() with a mask as argument, that
unblocks SIGWINCH while sigsuspend() waits.

I've attached a patch (tested a bit only).

	Bodo