Xen devel - Oct 2003 - segfault bug in xeno-unstable...

The actual version of xeno-unstable (Changeset 1.506) segfaults with some
standard commands - at least with my 
RedHat V8 (I haven''t tested RedHat V9 yet) installation.

If you enter "vi" or "vim" or "rpm -qa" as root or
non-root user,
xeno-linux just segfaults with no further comments (the next line of
output just shows "Segmentation fault", nothing more).

BTW: Especially for production use, it would be nicer if xen could
somehow pass up "printk"-output in xen to dom0''s kernel
logger,
so that you don''t need to configure and run a second machine for
the serial line debug output all the time. While checking the segfault
bug, I have not connected a debug machine to the serial line, so I
don''t know if it would have showed something helpful :-(

After a lot of tests with bk pull and bk undo, I was able to find out,
that this bug was introduced with Changeset 1.478. Changesets
up to 1.477 are working OK, without the segfaults, but starting
with CSet 1.478 you get the segfault.
I wasn''t able and didn''t try to find the cause for that,
because a
lot of sophisticated stuff got checked in with CSet 1.478.
Perhaps somebody from the Xen Project team can check it 
out - and also if they get segfaults too when using the above 
commands on newer xeno-unstable versions ;-)
I am sure there are also other command / programs segfaulting 
(e.g. "gdm" also did it) so this is not just an issue restricted
to vi and rpm ;-)


Another thing I realized while compiling and trying to load the 
HyperSCSI kernel module:
When loading the HyperSCSI client kernel module (which was 
compiled against xeno-linux), it complains about 2 undefined 
Kernel Symbols:

do_hypervisior_callback      and
HYPERVISOR_shared_info

After adding 2 EXPORT_SYMBOL lines for these symbols in
../arch/xeno/kernel/i386_ksyms.c and
recompiling the kernel, everything loaded and worked OK. 
I hope it was not against xen design principles to add the
2 EXPORT_SYMBOL lines ;-)


I had not yet the time to check the new vbd patches together 
with HyperSCSI, because the segfault bug was delaying me; 
perhaps I''ll do it on monday or tuesday.


Regards + many thanks to the Xen project team for this nice 
and professional software :))

Sven

-- 
NEU FÜR ALLE - GMX MediaCenter - für Fotos, Musik, Dateien...
Fotoalbum, File Sharing, MMS, Multimedia-Gruß, GMX FotoService

Jetzt kostenlos anmelden unter http://www.gmx.net

+++ GMX - die erste Adresse für Mail, Message, More! +++



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
SourceForge.net hosts over 70,000 Open Source Projects.
See the people who have HELPED US provide better services:
Click here: http://sourceforge.net/supporters.php
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

> After a lot of tests with bk pull and bk undo, I was able to find out,
> that this bug was introduced with Changeset 1.478. Changesets
> up to 1.477 are working OK, without the segfaults, but starting
> with CSet 1.478 you get the segfault.
> I wasn''t able and didn''t try to find the cause for that,
because a
> lot of sophisticated stuff got checked in with CSet 1.478.
> Perhaps somebody from the Xen Project team can check it 
> out - and also if they get segfaults too when using the above 
> commands on newer xeno-unstable versions ;-)
Cheers for working out which changeset causes the problem. I''ll try to
reproduce the bug today and get it fixed. It''s almost certainly some
fairly stupid context-switching bug.
> After adding 2 EXPORT_SYMBOL lines for these symbols in
> ../arch/xeno/kernel/i386_ksyms.c and
> recompiling the kernel, everything loaded and worked OK. 
> I hope it was not against xen design principles to add the
> 2 EXPORT_SYMBOL lines ;-)
No. I''ll check those two lines into the repository. Thanks!

 -- Keir


-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
SourceForge.net hosts over 70,000 Open Source Projects.
See the people who have HELPED US provide better services:
Click here: http://sourceforge.net/supporters.php
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

> The actual version of xeno-unstable (Changeset 1.506) segfaults with some
> standard commands - at least with my 
> RedHat V8 (I haven''t tested RedHat V9 yet) installation.
> If you enter "vi" or "vim" or "rpm -qa" as
root or non-root user,
> xeno-linux just segfaults with no further comments (the next line of
> output just shows "Segmentation fault", nothing more).
> After a lot of tests with bk pull and bk undo, I was able to find out,
> that this bug was introduced with Changeset 1.478. 
Oops. 

I''ve just tried "rpm -qa" and "vi" under 1.506 on a
RH7.2
installation, and it seems to work. I''ll try setting up a RH8
install and retrying. My guess is that this is something to do
with our per-domain ldt switching code. The pthreads library uses
fs/gs for accessing thread local storage, and the exact mechanism
seems to change with every libc and RH version. However, its not
clear that rpm and vi use pthreads.

1.478 was a major change, so I''m not entirely
surprised. Shouldn''t be hard to fix, though.
 
> BTW: Especially for production use, it would be nicer if xen could
> somehow pass up "printk"-output in xen to dom0''s kernel
logger,
> so that you don''t need to configure and run a second machine for
> the serial line debug output all the time. While checking the segfault
> bug, I have not connected a debug machine to the serial line, so I
> don''t know if it would have showed something helpful :-(
Yes, having Xen output (including boot messages) passed up to
domain0 would certainly be helpful. 
Just writing into a ring buffer and then mapping it from dom0
and having a daemon dump it to syslog should work. Might be worth
looking to see how klogd does it.

 > Another thing I realized while compiling and trying to load the
> HyperSCSI kernel module:
> When loading the HyperSCSI client kernel module (which was 
> compiled against xeno-linux), it complains about 2 undefined 
> Kernel Symbols:
> 
> do_hypervisior_callback      and
> HYPERVISOR_shared_info
> 
> After adding 2 EXPORT_SYMBOL lines for these symbols in
> ../arch/xeno/kernel/i386_ksyms.c and
> recompiling the kernel, everything loaded and worked OK. 
> I hope it was not against xen design principles to add the
> 2 EXPORT_SYMBOL lines ;-)
We generally don''t use modules much, so I''m afraid we
hadn''t
spotted this. We''ll add the exports into the master tree, thanks.
 
> Regards + many thanks to the Xen project team for this nice 
> and professional software :))
Thanks.

Ian



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
SourceForge.net hosts over 70,000 Open Source Projects.
See the people who have HELPED US provide better services:
Click here: http://sourceforge.net/supporters.php
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

> The actual version of xeno-unstable (Changeset 1.506) segfaults with some
> standard commands - at least with my 
> RedHat V8 (I haven''t tested RedHat V9 yet) installation.
> 
> If you enter "vi" or "vim" or "rpm -qa" as
root or non-root user,
> xeno-linux just segfaults with no further comments (the next line of
> output just shows "Segmentation fault", nothing more).
I''ve checked in a fix to xeno-unstable.bk which sorts this problem out
for us. Please pull and check.
 
> After adding 2 EXPORT_SYMBOL lines for these symbols in
> ../arch/xeno/kernel/i386_ksyms.c and
> recompiling the kernel, everything loaded and worked OK. 
> I hope it was not against xen design principles to add the
> 2 EXPORT_SYMBOL lines ;-)
This is also checked in.

 Cheers,
 Keir


-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
SourceForge.net hosts over 70,000 Open Source Projects.
See the people who have HELPED US provide better services:
Click here: http://sourceforge.net/supporters.php
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel