Can't use VMware with recent Fedora Core devel gcc and kernel packages
Jonathan Kamens
packages from Fedora Core development, in particular,
kernel-smp-2.6.11-1.1353_FC4 and gcc-4.0.0-8, the bridge module fails
to load and causes a kernel null pointer dereference that looks like
this:
Unable to handle kernel NULL pointer dereference at virtual address 00000069
printing eip:
c02a4eb9
*pde = 36f04001
Oops: 0000 [#1]
SMP
Modules linked in: vmnet(U) vmmon(U) nfs lockd md5 ipv6 parport_pc lp parport autofs4 sunrpc softdog dm_mod video button battery ac uhci_hcd ehci_hcd shpchp i2c_viapro i2c_core snd_via82xx gameport snd_ac97_codec snd_seq_dummy snd_seq_oss snd_seq_midi_event snd_seq snd_pcm_oss snd_mixer_oss snd_pcm snd_timer snd_page_alloc snd_mpu401_uart snd_rawmidi snd_seq_device snd soundcore via_rhine mii floppy ext3 jbd sata_via libata sd_mod scsi_mod
CPU: 1
EIP: 0060:[<c02a4eb9>] Tainted: P VLI
EFLAGS: 00010282 (2.6.11-1.1353_FC4smp)
EIP is at sk_alloc+0x10/0x146
eax: 00000010 ebx: 00000220 ecx: 00000001 edx: 00000220
esi: f54d86dc edi: f54d86e1 ebp: f7a2d2c0 esp: f568ae08
ds: 007b es: 007b ss: 0068
Process vmnet-bridge (pid: 3544, threadinfo=f568a000 task=f7935020)
Stack: badc0ded 00000001 00000010 f7deb39c c014bcf3 f54d86d0 f54d86dc f54d86e1
f7a2d2c0 f8af3500 00000000 f7a2d2c0 c0121b72 f8af41bc f54d86d0 f7a2d2c5
f8af36c3 f8af41bc f54d86dc 00000000 00000001 f54d86dc f7a2d2c0 f54d86d0
Call Trace:
[<c014bcf3>] check_poison_obj+0x24/0x189
[<f8af3500>] VNetBridgeUp+0xd0/0x184 [vmnet]
[<c0121b72>] printk+0x1b/0x1f
[<f8af36c3>] VNetBridgeNotify+0xad/0x13f [vmnet]
[<c02ab7a6>] register_netdevice_notifier+0x3c/0x58
[<f8af2eb2>] VNetBridge_Create+0xaa/0x2a4 [vmnet]
[<f8af06bb>] VNetFileOpIoctl+0x2e4/0x512 [vmnet]
[<c011963a>] do_page_fault+0x269/0x6a7
[<c016b300>] chrdev_open+0x1e/0x170
[<c0308335>] schedule+0x405/0xc5e
[<c0308361>] schedule+0x431/0xc5e
[<c014baed>] poison_obj+0x20/0x3d
[<c0173f44>] do_ioctl+0x64/0x6d
[<c01740a7>] vfs_ioctl+0x56/0x1d0
[<c017427e>] sys_ioctl+0x5d/0x6b
[<c0104025>] syscall_call+0x7/0xb
Code: c7 84 24 84 00 00 00 00 00 00 00 c7 84 24 88 00 00 00 00 00 00 00 e9 ac fc ff ff 55 57 56 53 83 ec 10 89 44 24 04 89 d3 89 0c 24 <8b> 69 68 85 ed 0f 84 fc 00 00 00 89 e8 e8 19 8b ea ff 89 c6 85
Has anybody else seen anything like this and/or have any suggestions
for what do to about it?
Bob Goddard
> When I attempt to configure VMware against kernel 2.6 and gcc 4.0
> packages from Fedora Core development, in particular,
> kernel-smp-2.6.11-1.1353_FC4 and gcc-4.0.0-8, the bridge module fails
> to load and causes a kernel null pointer dereference that looks like
> this:
[...]
> Has anybody else seen anything like this and/or have any suggestions
> for what do to about it?
I would have thought the answer was rather obvious.... do not use GCC4.
Try GCC3 and see if that works. I mean, comeon... GCC4 has only been out
for 5 weeks.
B
Jonathan Kamens
>I would have thought the answer was rather obvious.... do not use GCC4.
>Try GCC3 and see if that works. I mean, comeon... GCC4 has only been out
>for 5 weeks.
I believe Fedora has had prerelease versions of gcc4 for
significantly longer than 5 weeks.
What you've described is certainly an option, but an awkward
one, since the kernel I'm running was compiled with gcc4, and
VMware won't let you mix and match, for good reason.
I could back down some of my packages to get back to gcc3,
but I prefer to keep up-to-date with FC devel to help catch
bugs. I could also hack vmware-config.pl to let me use the
compat gcc32 compiler for the modules even though the kernel
is compiled with gcc4, and that might work, or it might be a
catastrophe.
Or I could post here and let VMware know about the problem
and hope that someone with more experience in this area and
time to work on it will come up with a solution soon (since
they'll have to solve it eventually anyway).
Right now, I'm opting for the last approach. I'll eventually
give up and try one of the other ones.
Bob Goddard
> Bob Goddard <vmware...@mailtrap.org.uk> writes:
>>I would have thought the answer was rather obvious.... do not use
>>GCC4. Try GCC3 and see if that works. I mean, comeon... GCC4 has only
>>been out for 5 weeks.
>
> I believe Fedora has had prerelease versions of gcc4 for
> significantly longer than 5 weeks.
Yes, and remember when RH went to GCC3.whatever and what a fsckup
that was.
> What you've described is certainly an option, but an awkward
> one, since the kernel I'm running was compiled with gcc4, and
> VMware won't let you mix and match, for good reason.
CC=<path to GCC3> ./vmware-install.pl
but I doubt that would work as I would imagine the object headers
are different.
> I could back down some of my packages to get back to gcc3,
> but I prefer to keep up-to-date with FC devel to help catch
> bugs. I could also hack vmware-config.pl to let me use the
> compat gcc32 compiler for the modules even though the kernel
> is compiled with gcc4, and that might work, or it might be a
> catastrophe.
Remember that FC are all regarded as beta releases for Enterprise.
> Or I could post here and let VMware know about the problem
> and hope that someone with more experience in this area and
> time to work on it will come up with a solution soon (since
> they'll have to solve it eventually anyway).
Yes but I'm sure they will be well aware of the problems already.
I would be surprised if they did not enable support of it at least
for a couple of months.
Newsgroups are only an informal approach to support. You would
be better off logging a support call via the web. They seem
to handle calls better than some companies I could name.
B
Jonathan Kamens
>Newsgroups are only an informal approach to support. You would
>be better off logging a support call via the web. They seem
>to handle calls better than some companies I could name.
I plan on doing that, but I wanted to post here about it as
well, both to see if anyone else has encountered the problem
and knows a solution and to alert people who haven't yet
encountered it to save them from it.
Russell East
I'm mostly just a "fly ion the wall" in this newsgroup, but I find it
useful to see what other people are experiencing. It is after all about
"experimental" issues with VMware for Linux, right?
-- Russell
Bob Goddard
No. It's supposed to be about the experimental releases of vmware.
B
Thomas M Steenholdt
Really, this has nothing to do with GCC4. It's failing similarly on my
FC3 installation with GCC 3.4.3. Rather the problem seems to be changes
in kernel functions between 2.6.11 and 2.6.12 somewhere - possibly even
by redhat specific patches. I haven't located all the problems. One
thing is certain, though. the sk_alloc function has changed, which is
the reason the there are warnings when building the vmnet module under
the newer kernels. Fixing this alone doesn't seem to fix the problem,
though, however there's probably some other things that are changed as
well. I don't think this should be too difficult for the vmnet module
maintainer to fix.
The problem seems to be a kernel NULL pointer dereference somewhere in
VNetBridgeUp function in vmnet.o
/Thomas
Jonathan Kamens
>Try getting the vmware-any-any-92 updates from here
>http://ftp.cvut.cz/vmware/.
>
>Did the trick for me!
Me too! Thanks.
jik
Thomas M Steenholdt
Hi Jonathan,
Try getting the vmware-any-any-92 updates from here
http://ftp.cvut.cz/vmware/.
Did the trick for me!
Hope it helps,
Thomas