Re: Incorrect cv_wait_sig() return values?

From: Kostik Belousov <kostikbel_at_gmail.com>
Date: Wed, 4 Aug 2010 19:04:04 +0300
On Wed, Aug 04, 2010 at 06:46:04PM +0300, Kostik Belousov wrote:
> On Wed, Aug 04, 2010 at 10:36:24AM -0400, Andrew Gallatin wrote:
> > Hi,
> > 
> > I recently noticed that cv_wait_sig() will return -1
> > rather than EINTR when a SIGINT is delivered.  This is
> > in contrast to CONDVAR(9) which states:
> > 
> > <...>
> >       cv_wait_sig() and cv_timedwait_sig() return prematurely with a
> >       value of EINTR or ERESTART if a signal is caught
> > <...>
> > 
> > To demonstrate the problem outside my out-of-tree driver, I
> > took the skeleton driver from
> > http://www.captain.at/programming/freebsd/
> > and added the following function, invoked at module
> > load:
> > 
> > static struct mtx m;
> > static struct cv c;
> > 
> > static void
> > cv_test(void)
> > {
> >         int rc;
> > 
> >         mtx_init(&m, "skel_m", MTX_DEF, MTX_DEF);
> >         cv_init(&c, "skel_c");
> >         mtx_lock(&m);
> >         rc = cv_wait_sig(&c, &m);
> >         mtx_unlock(&m);
> >         printf("cv_wait_sig returned %d\n", rc);
> >         cv_destroy(&c);
> >         mtx_destroy(&m);
> > }
> > 
> > 
> > I load the module, and I ^C kldload after a few seconds
> > to break out of the cv_wait_sig(), which results in this
> > output on console:
> > 
> > 	Skeleton KLD loaded.
> > 	cv_wait_sig returned -1
> > 
> > Am I doing something wrong, or are condvars broken?
> > I've tried to track this down with dtrace, but failed..
> 
> What version of the system do you use ? I cannot confirm this on
> the HEAD from several hours ago with the following test module.
> On the SIGINT I get `4' printed, which is EINTR.

BTW, -1 is ERESTART, so if you have SIGINT catched with SA_RESTART
flag in the process that initiated kldload(2) syscall, then -1
is the right return code for cv_wait_sig.

Received on Wed Aug 04 2010 - 14:04:12 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:05 UTC