On 9/10/07, Dan Fandrich <dan_at_coneharvesters.com> wrote:
> Many of the 500 series tests in Tor's Linux 2.6 i686 icc 9.1 autobuilds are
> failing with segmentation faults, e.g.
> http://curl.haxx.se/auto/log.cgi?id=20070910074756-8790#prob113
>
> Is this an icc/environment problem or a genuine curl code or build
> problem? Tor, can you try running some of those tests under gdb or
> with libSegFault to get an idea what's up? Thanks.

Quick check:
Core was generated by
`/home/tor/build/test/tests/libtest/.libs/lt-lib540
http://test.remote.server.co'.
Program terminated with signal 11, Segmentation fault.
#0 0xee4ff2b7 in ?? ()
(gdb) where
#0 0xee4ff2b7 in ?? ()
#1 0xb7f1baaa in curl_multi_init () at multi.c:358
#2 0x08049062 in test (URL=0xbffdbc6f
"http://test.remote.server.com/path/540") at lib540.c:123
#3 0x0804927f in main (argc=5, argv=0xbffdb2e4) at first.c:85
(gdb) up
#1 0xb7f1baaa in curl_multi_init () at multi.c:358
358 struct Curl_multi *multi = (void *)calloc(sizeof(struct
Curl_multi), 1);

It isn't immediately obvious what fails. That line is the first line
in a function with no parameters. calloc apparently segfaults when
called with an argument of 120,1. If things like this segfaults it's
usually because of a corrupted/overwritten stack, in my experience.

I'll try to find time to look more deeply into this later.

-Tor
Received on 2007-09-11