segfault in test-strcmp

[Geoff Keating]

Roland McGrath <roland@redhat.com> writes:

> I looked at the test-strchr crash before the test-strcmp one, but I suspect
> the problem is similar.  The sysdeps/powerpc/powerpc32/strchr.S code is
> accessing the word following the word containing the search character.
> This faults because the test string is aligned on the end of a page so that
> the following word cannot be accessed (this happens on iteration 157 in
> test-strchr with no srandom).  The test program is arguably invoking
> undefined behavior, because the string is actually not terminated: the
> final word of the page contains the search character and three other
> nonzero bytes.  So a stupid but valid implementation of strchr that called
> strlen would crash.

Yes, ISO C requires that strings be zero-terminated (C99 section 7.1.1
paragraph 1), and requires that the argument to strchr be a string
(section 7.21.5.2 paragraph 2).  For non-zero-terminated character
arrays, use memchr instead.

> It may be quicker for someone who knows powerpc assembly better than I do
> to try to fix this.

I think the testcase should be fixed instead.

-- 
- Geoffrey Keating <geoffk@geoffk.org>