Xen devel - May 2005 - Linux-XenU kernel panic in allocs after live migration

Hi,

While testing the live migration I run into a Linux-xenU kernel panic 
that is basically a show stopper. So far I have tested both with 2.0.5 
and testing that is a few days old. It seems that (all?) kernel allocs 
after live migration fail as  e.g. ping started before migration still 
runs stable after migration, but on the other hand starting a new 
instance of ping results in a panic. Process crash stack traces, the 
kernel throws to the console, seem always to end a kernel alloc, 
version or another. Regular save and restore for xenU work fine.

My setup is a vanilla "xen-make-world" with the exception that I 
enabled devfs support for the xenU. Moreover, compressed ram disk 
support is compiled in, but the panics occur with cramfs ramdisks and 
regular ext2fs ramdisks. My xenU mounts its root over NFS.

Is this a SUE, a feature, or a bug? :-)

An example of numerous panics:

<1>invalid operand: 0000 [#3]
PREEMPT
Modules linked in:
CPU:    0
EIP:    0061:[<c016a4ab>]    Not tainted VLI
EFLAGS: 00010292   (2.6.10-xenU)
EIP is at new_inode+0x9/0x80
eax: c031b1a0   ebx: 00000001   ecx: 00000001   edx: 00000000
esi: 00000001   edi: b7faac40   ebp: ffffff9f   esp: c04dff1c
ds: 007b   es: 007b   ss: 0069
Process login (pid: 566, threadinfo=c04de000 task=c00f15a0)
Stack: 0000000e 00000001 c02261e1 c0453e00 00000001 c0226f3e c0147966 
00030002
        00000001 fbffc000 b7faac40 c04de000 c02271a5 00000001 00000001 
00000000
        c04dff78 00000000 c0227204 00000001 00000001 00000000 c04dff78 
fbffc000
Call Trace:
  [<c02261e1>] sock_alloc+0x14/0x6c
  [<c0226f3e>] __sock_create+0x65/0x29d
  [<c0147966>] anon_vma_unlink+0x66/0x74
  [<c02271a5>] sock_create+0x2f/0x33
  [<c0227204>] sys_socket+0x28/0x55
  [<c022803e>] sys_socketcall+0x89/0x251
  [<c011c375>] sys_time+0x15/0x4a
  [<c0109044>] syscall_call+0x7/0xb
Code: f6 81 04 01 00 00 30 74 0a 89 0c 24 e8 60 0d 00 00 eb c3 85 c9 75 
da 89 f6 8d bc 27 00 00 00 00 eb cd 53 b8 a0 b1 31 c0 83 ec 04 <0f> 0d 
08 90 8b 44 24 0c 89 04 24 e8 45 f6 ff ff 85 c0 89 c3 74

Teemu

--


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

On May 10, 2005, at 0:04, Teemu Koponen wrote:
> While testing the live migration I run into a Linux-xenU kernel panic 
> that is basically a show stopper. So far I have tested both with 2.0.5 
> and testing that is a few days old. It seems that (all?) kernel allocs 
> after live migration fail as  e.g. ping started before migration still 
> runs stable after migration, but on the other hand starting a new 
> instance of ping results in a panic. Process crash stack traces, the 
> kernel throws to the console, seem always to end a kernel alloc, 
> version or another. Regular save and restore for xenU work fine.
>
> My setup is a vanilla "xen-make-world" with the exception that I 
> enabled devfs support for the xenU. Moreover, compressed ram disk 
> support is compiled in, but the panics occur with cramfs ramdisks and 
> regular ext2fs ramdisks. My xenU mounts its root over NFS.
>
> Is this a SUE, a feature, or a bug? :-)
I wonder how sensitive Linux kernel is to underlying CPU changes, even 
though it is compiled to support only of a certain type CPU? In this 
specific case I''m trying to migrate between Athlon XP and Pentium-M 
domain-0 hosts, with a Linux domain-U kernel that is compiled to 
support Pentium-MMX class processor. Is this really a dead-end 
(translation: SUE) or should it work as I initially assumed?

Teemu

--


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

Based on what you''ve reported it should work. I know someone who
should be testing this today.



       -Kip
 
> I wonder how sensitive Linux kernel is to underlying CPU changes, even
> though it is compiled to support only of a certain type CPU? In this
> specific case I''m trying to migrate between Athlon XP and
Pentium-M
> domain-0 hosts, with a Linux domain-U kernel that is compiled to
> support Pentium-MMX class processor. Is this really a dead-end
> (translation: SUE) or should it work as I initially assumed?
> 
> Teemu
> 
> --
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On May 11, 2005, at 20:51, Kip Macy wrote:

Kip,
> Based on what you''ve reported it should work. I know someone who
> should be testing this today.
I got paranoid and tested with two identical hosts. Everything works 
smoothly (this time between two Pentium-M hosts). The same binaries, 
Athlon XP <-> Pentium-M, no luck.

Teemu

--


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