Re: [Qemu-devel] Re: [5578] Increase default IO timeout from 10ms to 5s

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 2217 bytes --]

M. Warner Losh wrote:
> In message: <491081EB.5070905@web.de>
>             Jan Kiszka <jan.kiszka@web.de> writes:
> : M. Warner Losh wrote:
> : > In message: <20081104113204.GA32125@shareable.org>
> : >             Jamie Lokier <jamie@shareable.org> writes:
> : > : andrzej zaborowski wrote:
> : > : > > My man page even warns that the Linux
> : > : > > kernel is not implementing it yet, though I don't think this still
> : > : > > applies to recent 2.6.2x kernels.
> : > : > 
> : > : > According to the man page it moved to kernel at 2.6.16 but the glibc
> : > : > wrapper should be ok too.
> : > : 
> : > : If there's a glibc wrapper, it cannot be reliable...
> : > : 
> : > : *Looks at glibc source*
> : > : 
> : > : That's right.  The glibc pselect() wrapper has the same race condition
> : > : which prompted this QEMU bug.  If the signal arrives after unmasking
> : > : and before select() in the wrapper, then blocks.
> : > : 
> : > : In other words, don't use pselect() if you might run on a kernel older
> : > : than 2.6.16, or on a host architecture which adds pselect() in a later
> : > : kernel version.  Also, I wouldn't be surprised if older versions of
> : > : some BSDs have similar dodgy wrappers.
> : > 
> : > Which ones have a good kernel implementation of it?  FreeBSD's is
> : > currently approximately:
> : > 
> : > 	if (!mask)
> : > 		_sigprocmask(mask, &oldmask);
> : > 	/* here */
> : > 	select();
> : > 	if (!mask)
> : > 		_sigprocmask(oldmask, NULL);
> : > 
> : > I'm assuming that the problem is due to a signal arriving at /* here */.
> : 
> : I guess those things happen under some kind of preemption lock,
> : otherwise it would be a really poor implementation.
> 
> Why?  There's races there anyway.  IF you have a mutex to prevent
> multiple calls, doesn't that serialize calls to select?

[ sorting my mind ] What I meant was that some care will likely be taken
to prevent signal delivery _to userspace_ between the unmasking and the
select() call. At least Linux does delivery on syscall return, so should
FreeBSD do. I guess that signals arriving around "/* here */" will
simply prevent select() to block (or even run).

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 258 bytes --]