Skip to contentSkip to conversation

Kernel hang on multi-threaded X process crash

DISCUSSION Manuel McLure 4 participants 11 messages Apr 23, 2001 – Apr 24, 2001

lore
Share & tools
lore

Manuel McLure

author
#0 lore raw

Well, this is what I get for being on the cutting edge... :-(

I'm now running into problems where the machine will totally hang (no network, no SysRq, no nothin') regularly. The triggers seem to be running aviplay or mozilla.

Symptoms will be that I am running aviplay or mozilla, and the machine will suddenly hang and need to be hard-reset. I can trigger it 100% by telling aviplay to zoom 2x.

I finally managed to reproduce it while I was on a console (I telneted in from another machine, and ran aviplay on the X display that was on console 7 while the machine was displaying console 1) - the only message before the hang was "Trying to vfree() nonexistent vm area (d0992000)" - no Oops was shown.

Whenever this happens, the e2fsck step at reboot shows a
"Entry 'core.XXXX' in <dir> (XXXXXX) had deleted/unused inode XXXXXX. CLEARED" message. The core file is always in whatever directory I was running the process that seems to cause the crash. It seems like either the core is a symptom of the underlying problem, or the process coredumping is causing the hang.

The machine is an Athlon Thunderbird 900MHz with 256M of PC133 DRAM on an MSI K7T Turbo R motherboard. I am running 2.4.3-ac12 currently, 2.4.3-ac11 and 2.4.3-ac5 hung the same way at least once each before I started tracking this down. I am running Red Hat 7.1, and am using the XFree86-4.0.3 RPMs that come with RH71 with the CVS DRI trunk installed over it. The kernel was built with kgcc, a gcc-2.96 built kernel has the same problem.

Any ideas?

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
author
#1 lore raw
Quoted 8 lines · On 2001.04.22 18:55 Manuel McLure wrote:

The machine is an Athlon Thunderbird 900MHz with 256M of PC133 DRAM on an MSI K7T Turbo R motherboard. I am running 2.4.3-ac12 currently, 2.4.3-ac11
and 2.4.3-ac5 hung the same way at least once each before I started tracking this down. I am running Red Hat 7.1, and am using the XFree86-4.0.3 RPMs that come with RH71 with the CVS DRI trunk installed over it. The kernel was built with kgcc, a gcc-2.96 built kernel has the same problem.

Following up on myself, I replaced the contents of /usr/X11R6 server with the standard 4.0.3 RPMs that come with RH 7.1 and it made no difference. Also, if it's important my video card is a Voodoo 5 5500.

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
author
#2 lore raw
In reply to Manuel McLure
Quoted 11 lines · On 2001.04.22 19:27 Manuel McLure wrote:
Quoted 1 line · On 2001.04.22 18:55 Manuel McLure wrote:

The machine is an Athlon Thunderbird 900MHz with 256M of PC133 DRAM on

an

Quoted 6 lines

MSI K7T Turbo R motherboard. I am running 2.4.3-ac12 currently, 2.4.3-ac11
and 2.4.3-ac5 hung the same way at least once each before I started tracking this down. I am running Red Hat 7.1, and am using the XFree86-4.0.3 RPMs that come with RH71 with the CVS DRI trunk installed over it. The kernel was built with kgcc, a gcc-2.96 built kernel has

Quoted 6 lines

the

Quoted 1 line

same problem.

Following up on myself, I replaced the contents of /usr/X11R6 server with the standard 4.0.3 RPMs that come with RH 7.1 and it made no difference. Also, if it's important my video card is a Voodoo 5 5500.

To follow up on my followup, I can now reproduce this 100% and get the "Trying to vfree()..." message on the console. To do this I start Mozilla, switch to a text console, and do a "killall -QUIT mozilla". A couple of "Trying to vfree()..." messages later, it's big red switch time.

I'm going to try this with standard 2.4.3 as well as the 2.4.2 that comes with RH71 - hopefully my filesystem will handle all the fscks :-(

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
#3 lore raw
Quoted 11 lines
Quoted 3 lines

Following up on myself, I replaced the contents of /usr/X11R6 server with the standard 4.0.3 RPMs that come with RH 7.1 and it made no difference. Also, if it's important my video card is a Voodoo 5 5500.

To follow up on my followup, I can now reproduce this 100% and get the "Trying to vfree()..." message on the console. To do this I start Mozilla, switch to a text console, and do a "killall -QUIT mozilla". A couple of "Trying to vfree()..." messages later, it's big red switch time.

I'm going to try this with standard 2.4.3 as well as the 2.4.2 that comes with RH71 - hopefully my filesystem will handle all the fscks :-(

Do you have DRI enabled and if so does disabling DRI help ?

author
#4 lore raw
In reply to Manuel McLure
Quoted 20 lines · On 2001.04.22 20:17 Manuel McLure wrote:
Quoted 4 lines · On 2001.04.22 19:27 Manuel McLure wrote:
Quoted 1 line · On 2001.04.22 18:55 Manuel McLure wrote:

The machine is an Athlon Thunderbird 900MHz with 256M of PC133 DRAM

on

Quoted 6 lines

an

Quoted 5 lines

MSI K7T Turbo R motherboard. I am running 2.4.3-ac12 currently, 2.4.3-ac11
and 2.4.3-ac5 hung the same way at least once each before I started tracking this down. I am running Red Hat 7.1, and am using the XFree86-4.0.3 RPMs that come with RH71 with the CVS DRI trunk

installed

Quoted 5 lines
Quoted 1 line

over it. The kernel was built with kgcc, a gcc-2.96 built kernel has

the

Quoted 1 line

same problem.

Following up on myself, I replaced the contents of /usr/X11R6 server

with

Quoted 8 lines
Quoted 1 line

the standard 4.0.3 RPMs that come with RH 7.1 and it made no

difference.

Quoted 1 line

Also, if it's important my video card is a Voodoo 5 5500.

To follow up on my followup, I can now reproduce this 100% and get the "Trying to vfree()..." message on the console. To do this I start Mozilla,
switch to a text console, and do a "killall -QUIT mozilla". A couple of

                                                    ^^^^^^^

Make that "killall -QUIT mozilla-bin"...

Quoted 4 lines

"Trying to vfree()..." messages later, it's big red switch time.

I'm going to try this with standard 2.4.3 as well as the 2.4.2 that comes with RH71 - hopefully my filesystem will handle all the fscks :-(

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
#5 lore raw
In reply to Manuel McLure
Quoted 9 lines · Manuel McLure
Quoted 5 lines · On 2001.04.22 20:17 Manuel McLure wrote:

To follow up on my followup, I can now reproduce this 100% and get the "Trying to vfree()..." message on the console. To do this I start Mozilla,
switch to a text console, and do a "killall -QUIT mozilla". A couple of

Make that "killall -QUIT mozilla-bin"...

I'll see if mine does this too. I've been having hard locks intermittantly too, sometimes starting X, mostly with mozilla. My system's a dual Pmmx200 with ATI rage pro and Matrox Mill 2 PCI.

-- Ferret

author
#6 lore raw
In reply to Manuel McLure
Quoted 33 lines · On 2001.04.22 20:23 Manuel McLure wrote:
Quoted 6 lines · On 2001.04.22 20:17 Manuel McLure wrote:
Quoted 4 lines · On 2001.04.22 19:27 Manuel McLure wrote:
Quoted 1 line · On 2001.04.22 18:55 Manuel McLure wrote:

The machine is an Athlon Thunderbird 900MHz with 256M of PC133 DRAM

Quoted 9 lines

on

Quoted 6 lines

an

Quoted 5 lines

MSI K7T Turbo R motherboard. I am running 2.4.3-ac12 currently, 2.4.3-ac11
and 2.4.3-ac5 hung the same way at least once each before I started tracking this down. I am running Red Hat 7.1, and am using the XFree86-4.0.3 RPMs that come with RH71 with the CVS DRI trunk

installed

Quoted 1 line
Quoted 1 line

over it. The kernel was built with kgcc, a gcc-2.96 built kernel

has

Quoted 5 lines
Quoted 4 lines

the

Quoted 1 line

same problem.

Following up on myself, I replaced the contents of /usr/X11R6 server

with

Quoted 8 lines
Quoted 1 line

the standard 4.0.3 RPMs that come with RH 7.1 and it made no

difference.

Quoted 1 line

Also, if it's important my video card is a Voodoo 5 5500.

To follow up on my followup, I can now reproduce this 100% and get the "Trying to vfree()..." message on the console. To do this I start Mozilla,
switch to a text console, and do a "killall -QUIT mozilla". A couple of

                                                    ^^^^^^^
Quoted 8 lines

Make that "killall -QUIT mozilla-bin"...

Quoted 3 lines

"Trying to vfree()..." messages later, it's big red switch time.

I'm going to try this with standard 2.4.3 as well as the 2.4.2 that

comes

Quoted 1 line

with RH71 - hopefully my filesystem will handle all the fscks :-(

At Andrew Morton's suggestion, I added a BUG() to mm/vmalloc.c:vfree() and copied down the stack trace. According to gdb on vmlinux, the symbols are:

vfree
release_segments+50
exit_mmap+17
elf_core_dump
elf_core_dump
mmput+38
do_exit+157
do_invalid_op
die+79
do_invalid_op+127
vfree+110
__call_console_drivers+59
_call_console_drivers+87
call_console_drivers+228
release_console_sem+21
error_code+52
elf_core_dump
vfree+110
release_segments+50
exit_mmap+17
elf_core_dump
elf_core_dump
mmput+38
do_coredump+556
collect_signal+168
elf_core_dump
do_signal+418
do_general_protection
force_sig_info+121
do_general_protection
force_sig+17
do_general_protection+53
error_code+52
signal_return+20

(at least one more BUG() output scrolled off the console).

I also tested this on the 2.4.2-2 kernel that comes with RH 7.1 and I did not see the problem. Also, if with 2.4.3-ac12 I used Alt-Sysrq-s and Alt-Sysrq-u to sync and mount my filesystems read-only before doing the killall, the problem did not occur either. Could the core dumping code be causing the problem?

I'll now try with the base 2.4.3 to see what happens.

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
#7 lore raw

Strange trace but it looks like a bug in the -ac experimental multithreaded core dump patches. I've got a couple of other reports consistent with them being broken somewhere

Does it have to be something like mozilla (xmms also probably breaks it) that does this. If so I suspect its specific to multithreaded apps and its a bug in the core dump changes.

If so I guess I revert them

author
#8 lore raw
In reply to Manuel McLure
Quoted 6 lines · On 2001.04.23 02:38 Alan Cox wrote:

Strange trace but it looks like a bug in the -ac experimental multithreaded
core dump patches. I've got a couple of other reports consistent with them
being broken somewhere

Quoted 7 lines

Does it have to be something like mozilla (xmms also probably breaks it) that
does this. If so I suspect its specific to multithreaded apps and its a bug
in the core dump changes.

If so I guess I revert them

Both mozilla and aviplay (which are both multithreaded) trigger this - I haven't tried with xmms. Simpler programs like xclock or cat don't trigger it.

To answer the question in your other email, I don't have DRI enabled (since tdfx.o won't load for me due to rwsem fixes - see other thread).

Thanks for your help.

Signature · 3 lines
Manuel A. McLure KE6TAW | ...for in Ulthar, according to an ancient
<[email protected]>     | and significant law, no man may kill a cat.
<http://www.mclure.org> |             -- H.P. Lovecraft
#9 lore raw
Quoted 3 lines

Both mozilla and aviplay (which are both multithreaded) trigger this - I haven't tried with xmms. Simpler programs like xclock or cat don't trigger it.

Thanks. I'll revert the core dump stuff for 2.4.4-ac unless anyone cares to fix the fix

#10 lore raw
In reply to Manuel McLure

Alan-

I certainly care to fix it (since I wrote the patch). Since `aviplay' seems to be the easy way to trigger it I'll look into it.

Quoted 11 lines · Alan Cox
Quoted 3 lines

Both mozilla and aviplay (which are both multithreaded) trigger this - I haven't tried with xmms. Simpler programs like xclock or cat don't trigger it.

Thanks. I'll revert the core dump stuff for 2.4.4-ac unless anyone cares to fix the fix
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 4 lines
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
[email protected]
Ph: 303/938-9838