Re: [LTP] signal usage bug in fcntl syscall tests

[Robert W. <ro...@us...>]

"David S. Miller" <da...@da...> wrote on 02/24/2005 01:20:45 PM:

> On Thu, 24 Feb 2005 12:03:04 -0500
> Robert Williamson <ro...@us...> wrote:
>
> > These tests were passing for me on both i386 and ppc64 platforms.  I
> > applied patch and they still passed, so I applied to the CVS tree for
the
> > next release.
>
> On your systems where they pass, what arguments are passed to
> sigaction() when signal() is called from a test program?
> signal() is just a wrapper in glibc that makes the appropriate
> sigaction() call with the appropriate args.  Every system I have
> tried passes SA_RESTART in such sigaction() calls which is the
> source of the problems.
>
> Perhaps the version of glibc on your machine is older and does
> not pass this flag in.
>
> I don't like how this "mysteriously" works for other people, and
> we need to get down to the details of why this is.

The test you provided does hang as expected.  I ran fcntl14, fcntl15, and
fcntl16 on a RHEL4 Power5 machine and they all passed.  Interestingly, if
I strace the three tests, fcntl14 hangs.  I've attached the strace output
from all three runs.

(See attached file: fcntl16.strace)(See attached file: fcntl14.strace)(See
attached file: fcntl15.strace)