06.08.2013 18:39, Edward J. Shornock wrote:
* Michael Tokarev <m...@tls.msk.ru> [06-08-2013 13:21 EEST]:
I just tried wheezy version, -- it fails in exactly the same way in
exactly the same place.
So I don't see how this works for you and how this can be a reguression.
Perhaps I just "lucked out" the time that it worked. In retrospect I
should have tried multiple times before claiming success with the wheezy
version. Maybe running within vbox made it act differently?
Might be. The current error is always reproducible, and has been
reproduced by qemu developers as well. Gdb confirms it fails at
exactly the same place in 1.1 version too, with similar instructions
(give or take slightly different code generated by 1.1 vs 1.6-tobe).
I can say that some of the tcg errors, such as the ones that can be
triggered when update-ca-certificates runs when emulating ARM, do not
fail consistently. Perhaps it's the same situation here?
Does update-ca-certificates run any multi-threaded apps?
I'm still referring to multi-threaded apps issues. Which, on
a modern system, can happen with non-multithreaded apps too
just because of the way how these run other apps - using
clone() instead of fork() (which is what bash does!).
The issue is that multi-threaded emulation in qemu-user is
known broken, with multiple steps and patches needed to fix
different parts of it, and with time the situation becomes
worse because more and more "guest" userspace use new, broken
ways.
If that's the case and this ARM stuff fails due to this MT
issue, it should be a bit random, because there are many race
conditions in the code which are triggered when running in
MT mode, so sometimes you will hit these, and sometimes not.
It is almost NEVER happens when you pin the whole process
on host into a single CPU/core on host (running it with
taskset).
Anyhow, sorry for the incorrect report.
It isn't incorrect, it's perfectly fine bugreport. And if your
version actually did not break on wheezy, that might point to
a diffrent issue or a reason where you were lucky not hitting
it. The bug is still there and can be reproduced by multiple
people.
Thanks,
/mjt
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
