On Tue, 09 Jun 2009 21:30:43 +0300
Avi Kivity <[email protected]> wrote:

> Dan Williams wrote:
> > On Mon, Jun 8, 2009 at 5:04 PM, Stephen Hemminger<[email protected]> 
> > wrote:
> >   
> >> I did some more debugging, and the problem appears to be that
> >> booting with nosmp still leaves all the CPU's present in the cpu
> >> possible mask. So code like dma engine (or loopback device), that
> >> use alloc_percpu get broken.
> >>
> >> Not sure why all the cpu's are still showing up in the possible mask.
> >> Rusty? do you have an answer.
> >>     
> >
> > Would this recent commit also fix this issue?
> >
> > commit a4c0364be3f43d3e17fe19270f8b3d64881606e6
> > Author: Avi Kivity <[email protected]>
> > Date:   Sat Jun 6 12:34:39 2009 +0300
> >
> >     KVM: Explicity initialize cpus_hardware_enabled
> >
> >     Under CONFIG_MAXSMP, cpus_hardware_enabled is allocated from the heap 
> > and
> >     not statically initialized.  This causes a crash on reboot when kvm 
> > thinks
> >     vmx is enabled on random nonexistent cpus and accesses nonexistent 
> > percpu
> >     lists.
> >
> >     Fix by explicitly clearing the variable.
> >
> >     Cc: [email protected]
> >     Reported-and-tested-by: Yinghai Lu <[email protected]>
> >     Signed-off-by: Avi Kivity <[email protected]>
> >   
> 
> This only affects reboots.  But yes, the cpumask code has bugs.  Does 
> the .config involved select MAXSMP?
> 

Patch does not fix the problem. Not sure if it is KVM only problem.
Look like possible cpumask always equals what ACPI / SMP tables report.
So currently looking into alloc_percpu which may not handle the possible
but never used cpus.

It is x86-64 config and does not have MAXSMP set. It does have NR_CPUS set.

-- 
--
To unsubscribe from this list: send the line "unsubscribe kernel-testers" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to