Re: another subtle signals issue

From: Daniel Jacobowitz (
Date: Tue Feb 11 2003 - 22:05:27 EST

On Tue, Feb 11, 2003 at 06:45:19PM -0800, Linus Torvalds wrote:
> On Tue, 11 Feb 2003, Roland McGrath wrote:
> >
> > I think sys_semop would be closer to right if it used ERESTARTSYS instead
> > of EINTR.
> You probably mean ERESTARTSYSNOHAND.
> There are lots of system calls that simply are not restartable. So
> TIF_SIGPENDING in general should be set only if required, and not "because
> it's easier".
> > The reason I am concerned about this is that I think any case that is
> > broken by the lack of the optimization in the patch below must also be
> > broken vis a vis the semantics of stop signals and SIGCONT (when SIG_DFL,
> > SIG_IGN, or blocked). POSIX says that when a process is stopped by
> > e.g. SIGSTOP, and then continued by SIGCONT, any functions that were in
> > progress at the time of stop are unaffected unless SIGCONT runs a handler.
> > That is, nobody returns EINTR because of the stop/continue.
> This is what ERESTARTNOHAND does, but quite often if you get interrupted
> you have to return _partial_ results, which is quite inefficient and
> sometimes breaks programs (ie you get things like a read() from a pipe
> that returns a partial result because you resized the window, and a
> SIGWINCH happened - and that is _bad_).
> The old code tried rather hard to make signals that were truly ignored
> (SIGSTOP/SIGCONT is not of that kind) be total non-events because of
> things like this.

For things with a timeout, shouldn't they be converted to use
ERESTART_RESTARTBLOCK? The situation Roland is describing is just
about the same as the original problem with nanosleep.

Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Sat Feb 15 2003 - 22:00:36 EST