kvm-kmod-2.6.33 (or 2.6.32) messes up pages on guest exit

kvm-kmod-2.6.33 (or 2.6.32) messes up pages on guest exit

[Henrik Holst]

Hi,

 I'm running a few Debian Lenny host machines with kernel 2.6.26, in
production we use kvm-kmod-2.6.31.5 without any problems. Today I
tested to change to kvm-kmod-2.6.33 and everything went just fine up
to the moment when a guest exited and when it did the kernel started
to log thousands of rows about page errors on the host.

modprobe -r kvm-intel (and kvm) and modprobe of the 2.6.31.5 version
made the problems go away again. Could it be that 2.6.26 is a little
too old kernel to run as host for the newer kvm?

Mar  7 19:30:47 prince kernel: [12310060.335805] br0: port 2(tap0)
entering disabled state
Mar  7 19:30:47 prince kernel: [12310060.576672] device tap0 left
promiscuous mode
Mar  7 19:30:47 prince kernel: [12310060.576676] br0: port 2(tap0)
entering disabled state
Mar  7 19:30:47 prince kernel: [12310060.688303] Pid: 10813, comm:
grep Not tainted 2.6.26-1-amd64 #1
Mar  7 19:30:47 prince kernel: [12310060.688303]
Mar  7 19:30:47 prince kernel: [12310060.688303] Call Trace:
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff80274f7d>]
bad_page+0x6b/0x95
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff802763b2>]
get_page_from_freelist+0x3e1/0x603
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff80276856>]
__alloc_pages_internal+0xd6/0x3bf
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff80281637>]
handle_mm_fault+0x244/0x867
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff802851c8>]
mmap_region+0x35e/0x439
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff80221fac>]
do_page_fault+0x5d8/0x9c8
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff80285829>]
do_mmap_pgoff+0x2ff/0x364
Mar  7 19:30:47 prince kernel: [12310060.688303]  [<ffffffff8042a0b9>]
error_exit+0x0/0x60
Mar  7 19:30:47 prince kernel: [12310060.688303]
Mar  7 19:30:47 prince kernel: [12310060.704319] Pid: 10811, comm:
net.agent Tainted: G    B     2.6.26-1-amd64 #1
Mar  7 19:30:47 prince kernel: [12310060.704319]
Mar  7 19:30:47 prince kernel: [12310060.704319] Call Trace:
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff80274f7d>]
bad_page+0x6b/0x95
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff802763b2>]
get_page_from_freelist+0x3e1/0x603
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff80276856>]
__alloc_pages_internal+0xd6/0x3bf
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff80281637>]
handle_mm_fault+0x244/0x867
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8027ee2c>]
follow_page+0x60/0x2e5
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff80281f73>]
get_user_pages+0x319/0x417
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8029e366>]
get_arg_page+0x46/0xa5
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8029e591>]
copy_strings+0xf0/0x1cd
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8029e68f>]
copy_strings_kernel+0x21/0x33
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8029fa05>]
do_execve+0x110/0x215
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8020a439>]
sys_execve+0x35/0x4c
Mar  7 19:30:47 prince kernel: [12310060.704319]  [<ffffffff8020c30a>]
stub_execve+0x6a/0xc0

Regards, Henrik Holst
Witsbits AB

Re: kvm-kmod-2.6.33 (or 2.6.32) messes up pages on guest exit

[Jan Kiszka]

Henrik Holst wrote:
> Hi,
> 
>  I'm running a few Debian Lenny host machines with kernel 2.6.26, in
> production we use kvm-kmod-2.6.31.5 without any problems. Today I
> tested to change to kvm-kmod-2.6.33 and everything went just fine up
> to the moment when a guest exited and when it did the kernel started
> to log thousands of rows about page errors on the host.
> 
> modprobe -r kvm-intel (and kvm) and modprobe of the 2.6.31.5 version
> made the problems go away again. Could it be that 2.6.26 is a little
> too old kernel to run as host for the newer kvm?

Maybe. I'm only testing against 2.6.27 as oldest host, down to 2.6.24 is
solely build-tested. Maybe the missing MMU notfiers in <= 2.6.26 cause
troubles, though this used to work before.

Can't promise that I find the time to look into this (such old kernels
are out of official scope). If I managed to, I would try to bisect over
kvm-kmod-2.6.32 what import from kvm.git or what kvm-kmod wrapping
brought us the breakage. But maybe someone else finds the time, setup
support would be provided...

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: kvm-kmod-2.6.33 (or 2.6.32) messes up pages on guest exit

[Henrik Holst]

2010/3/8 Jan Kiszka <jan.kiszka@siemens.com>:
> Henrik Holst wrote:
>> Hi,
>>
>>  I'm running a few Debian Lenny host machines with kernel 2.6.26, in
>> production we use kvm-kmod-2.6.31.5 without any problems. Today I
>> tested to change to kvm-kmod-2.6.33 and everything went just fine up
>> to the moment when a guest exited and when it did the kernel started
>> to log thousands of rows about page errors on the host.
>>
>> modprobe -r kvm-intel (and kvm) and modprobe of the 2.6.31.5 version
>> made the problems go away again. Could it be that 2.6.26 is a little
>> too old kernel to run as host for the newer kvm?
>
> Maybe. I'm only testing against 2.6.27 as oldest host, down to 2.6.24 is
> solely build-tested. Maybe the missing MMU notfiers in <= 2.6.26 cause
> troubles, though this used to work before.
>
> Can't promise that I find the time to look into this (such old kernels
> are out of official scope). If I managed to, I would try to bisect over
> kvm-kmod-2.6.32 what import from kvm.git or what kvm-kmod wrapping
> brought us the breakage. But maybe someone else finds the time, setup
> support would be provided...
I figured as much, I'll try to bisect when I get the time.

/Henrik Holst