2.5.66 x86_64 oops when swapoff via IA32 emulation

2.5.66 x86_64 oops when swapoff via IA32 emulation

[mikpe]

I got this oops from swapoff when shutting down 2.5.66 on
x86_64 with 32-bit user-space (RH8.0). This was from the first
run with 2.5.66, I wasn't able to reproduce it in later runs.

/Mikael

Kernel BUG at page_alloc:687
invalid operand: 0000 [#1]
CPU 0 
Pid: 364, comm: swapoff Not tainted
RIP: 0010:[<ffffffff80143c64>] <ffffffff80143c64>{__free_pages+20}
RSP: 0000:000001000f4a5f30  EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000019 RCX: 0000000000101000
RDX: 0000000000000018 RSI: 0000000000000000 RDI: 0000010001074952
RBP: 000001000fea2bc0 R08: 000000000000fbff R09: 00000000ffffdb68
R10: 000001000f4a4000 R11: 0000000000000000 R12: 0000000000000001
R13: 000001000f62f280 R14: 0000010004360cc0 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffffffff802cd5c0(0000) knlGS:00000000a0047080
CS:  0010 DS: 002b ES: 002b CR0: 0000000080050033
CR2: 00000000420d40b0 CR3: 0000000000101000 CR4: 00000000000006a0
Process swapoff (pid: 364, stackpage=10001605100)
Stack: ffffffff801531d9 ffffff0000000000 ffffffff802a18a0 0000000000000000 
       ffffffff80155627 000000000804b328 00000000ffffdb68 0000000000000000 
       0000000000000000 0000000000000000 
Call Trace:<ffffffff801531d9>{__vunmap+169} <ffffffff80155627>{sys_swapoff+631} 
       <ffffffff801179fe>{ia32_do_syscall+30} 

Code: 0f 0b ab f3 23 80 ff ff ff ff af 02 ff 4f 08 0f 94 c0 84 c0 

Re: 2.5.66 x86_64 oops when swapoff via IA32 emulation

[Andi Kleen]

On Thu, Apr 03, 2003 at 04:56:31PM +0200, mikpe@csd.uu.se wrote:
> I got this oops from swapoff when shutting down 2.5.66 on
> x86_64 with 32-bit user-space (RH8.0). This was from the first
> run with 2.5.66, I wasn't able to reproduce it in later runs.

Is this on the Simulator or on a real Opteron/Athlon64? 

On the real CPU there is a known bug I wasn't able to track down yet - 
pci_free_consistent() seems to corrupt memory. When you do a shutdown
this is usually invoked in some driver's cleanup function (e.g. ifconfig
down on a network driver) and causes something later to crash.

-Andi

Re: 2.5.66 x86_64 oops when swapoff via IA32 emulation

[mikpe]

Andi Kleen writes:
 > On Thu, Apr 03, 2003 at 04:56:31PM +0200, mikpe@csd.uu.se wrote:
 > > I got this oops from swapoff when shutting down 2.5.66 on
 > > x86_64 with 32-bit user-space (RH8.0). This was from the first
 > > run with 2.5.66, I wasn't able to reproduce it in later runs.
 > 
 > Is this on the Simulator or on a real Opteron/Athlon64? 

Simics.

/Mikael

Re: 2.5.66 x86_64 oops when swapoff via IA32 emulation

[Andi Kleen]

On Thu, Apr 03, 2003 at 06:54:41PM +0200, mikpe@csd.uu.se wrote:
> Andi Kleen writes:
>  > On Thu, Apr 03, 2003 at 04:56:31PM +0200, mikpe@csd.uu.se wrote:
>  > > I got this oops from swapoff when shutting down 2.5.66 on
>  > > x86_64 with 32-bit user-space (RH8.0). This was from the first
>  > > run with 2.5.66, I wasn't able to reproduce it in later runs.
>  > 
>  > Is this on the Simulator or on a real Opteron/Athlon64? 
> 
> Simics.

Don't know what it could be then. Can you reproduce it ? 

-Andi

Re: 2.5.66 x86_64 oops when swapoff via IA32 emulation

[mikpe]

Andi Kleen writes:
 > On Thu, Apr 03, 2003 at 06:54:41PM +0200, mikpe@csd.uu.se wrote:
 > > Andi Kleen writes:
 > >  > On Thu, Apr 03, 2003 at 04:56:31PM +0200, mikpe@csd.uu.se wrote:
 > >  > > I got this oops from swapoff when shutting down 2.5.66 on
 > >  > > x86_64 with 32-bit user-space (RH8.0). This was from the first
 > >  > > run with 2.5.66, I wasn't able to reproduce it in later runs.
 > >  > 
 > >  > Is this on the Simulator or on a real Opteron/Athlon64? 
 > > 
 > > Simics.
 > 
 > Don't know what it could be then. Can you reproduce it ? 

Eventually I was able to reproduce it. The oops was the same as before,
from swapoff at shutdown.

What I did this time was to boot with mem=32M, and just run a silly
'cd /; tar cf - . --totals > /dev/null' command.
/proc/swaps showed no swap usage before shutdown, but apparently
something bad happended anyway.

(Running under Simics is slow enough that I don't build the kernels
or anything else on the "box" itself. So it only sees brief boot-and-
see-if-it-works-then-shutdown usage.)

/Mikael