On 2/26/2014 9:26 AM, Helge Deller wrote:
I'm now on 3.13.2 and things are similar. While I did an initial test
with
100 invocations of ssh which went flawlessly I now see random process
errors again. Examples can be found here:
http://my.cdash.org/index.php?project=Qsmtp&date=2014-02-11
Normally the reports should have arrived much earlier, but for an
unknown
reason the resolv.conf (generated by DHCP) was empty, I assume that the
DHCP process has failed during while renewing the IP address. The
segfaults during the build have not happened with the older kernel.
I can now confirm that the cache coherency fixes are _not_ the reason for
those failures I see. I have rebuild 3.13.2 without them and still see
random segfaults in my dashboards. I'll try the mmap() patch next.
After 2,5 days with 3.13.2 and 0576da2c08e3d332f1b0653030d28ab804585ab6 I
have not yet seen a single random segfault.
A short update on this: I am still running on that kernel and have not seen a
single random process crash since then. I'm willing to help in debugging this
if anyone tells me how, but I will now just revert that patch from every new
kernel to get a stable machine.
On a debian-kernel linux-image-3.13-1-parisc64-smp (3.13.4-1) and the latest libc6
which Dave uploaded, I'm now seeing similiar issues which are most likely related:
lsXX:~# dpkg-reconfigure locales
Generating locales (this might take a while)...
de_DE.ISO-8859-1...cannot map archive header: Invalid argument
In my opinion, this is a glibc bug. The issue has been discussed before.
I'm not seeing random segvs with 3.13.2 and later in general use. I
haven't reverted anything.
The only random program failures that I see are thread related. There
are a bunch of failures
in the glibc testsuite, but these are not new. There was a message on
the console this morning about
reaching maximum lock depth 1024. This was probably from glibc
testsuite. There is a known
issue with locking in glib2.0 which sometimes causes dot to segv.
If you are testing programs that use threads intensively, then I could
see that you might see problems.
A TLS bug affecting thread local variables was recently fixed in gcc-4.8.
Dave
--
John David Anglin dave.anglin@xxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
