Xen devel - Oct 2007 - Weird bug: system instability after cat /proc/kcore

Ok, the existence of this bug just blew my mind. It appears that after
coppying /proc/kcore to disk the system starts to behave strangely.
Processes start to sometimes segfault randomly or die with a glibc
double free error. The system is perfectly fine prior to catting
/proc/kcore. I can reproduce this problem on the 2.6.18 kernel from
xen 3.1.0 and 3.1.1-rc2. To trigger it I just have to run ''cat
/proc/kcore >coredump'' and then start hitting the system with
something like a gdb compile.  I usually see a segfault or some
strange failure before the compile finishes.

-- 
Michael Marineau
Oregon State University
mike@marineau.org

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

So the crashes you have been seeing only occur after accessing /proc/kcore?

 -- Keir

On 4/10/07 00:03, "Michael Marineau" <mike@marineau.org> wrote:
> Ok, the existence of this bug just blew my mind. It appears that after
> coppying /proc/kcore to disk the system starts to behave strangely.
> Processes start to sometimes segfault randomly or die with a glibc
> double free error. The system is perfectly fine prior to catting
> /proc/kcore. I can reproduce this problem on the 2.6.18 kernel from
> xen 3.1.0 and 3.1.1-rc2. To trigger it I just have to run ''cat
> /proc/kcore >coredump'' and then start hitting the system with
> something like a gdb compile.  I usually see a segfault or some
> strange failure before the compile finishes.


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 10/3/07, Keir Fraser <Keir.Fraser@cl.cam.ac.uk>
wrote:
> So the crashes you have been seeing only occur after accessing /proc/kcore?
>
>  -- Keir
There are no crashes associated with this kcore problem. I didn''t even
have kcore enabled until I made another attempt at digging into the
oops I reported earlier. All that happens with this issue are some
occasional segfaults and sometimes a glibc double free error. There
isn''t a kernel oops or other error message. The sagfaults can take
some time to trigger so I don''t have a lot of info yet but this is a
quick rundown of what I was able to find today:

* I usually hit segfaults during an ''emerge gdb'' but one time
it took
a second compile before I hit it. Other loads should also work, but I
just stuck with one to keep things simple.
* If I do a compile round before reading kcore I had no problem (but
that might be luck)
* I had to use cat /proc/kcore > somefile, cat /proc/kcore >/dev/null
didn''t trigger it as far as I could tell.
* This is on an x86 box w/ gcc 3.4. I haven''t tried amd64 or other gcc
versions yet.

This is pretty weird and obscure so unless someone has any ideas I''m
not going to spend much time on it right now.
>
> On 4/10/07 00:03, "Michael Marineau" <mike@marineau.org>
wrote:
>
> > Ok, the existence of this bug just blew my mind. It appears that after
> > coppying /proc/kcore to disk the system starts to behave strangely.
> > Processes start to sometimes segfault randomly or die with a glibc
> > double free error. The system is perfectly fine prior to catting
> > /proc/kcore. I can reproduce this problem on the 2.6.18 kernel from
> > xen 3.1.0 and 3.1.1-rc2. To trigger it I just have to run
''cat
> > /proc/kcore >coredump'' and then start hitting the system
with
> > something like a gdb compile.  I usually see a segfault or some
> > strange failure before the compile finishes.


-- 
Michael Marineau
Oregon State University
mike@marineau.org

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel