Re: [uml-user] Soft lockup?

[Joel P. <joe...@mi...>]

So what do I do? Might enabling pre-emption help?

   // Joel


On Thu, 24 Nov 2005, Jeff Dike wrote:

> On Wed, Nov 23, 2005 at 07:53:23PM +0100, Joel Palmius wrote:
>> The below happens more often than not (but not always). The host is
>> athlon64 running in 32-bit mode. Host kernel is 2.6.14-skas3-v9-pre7.
>> Guest kernel is 2.6.14-bs1. Same thing happens in skas0 and skas3.
>>
>> When the machine boots correctly it seems stable though.
>
> It's not really a problem in the sense that it's a crash.  It's more
> of a latenciy issue.  UML isn't getting a timer interrupt in 10 sec
> (!), which is 1000 times longer than it should.
>
> The cases I've seen are when the host is doing heavy I/O or UML is
> going in and out of the kernel very frequently.  The UML timer is
> based on SIGVALRM, which fires based on userspace time.
>
> Blaisorblade made a suggestion a while back to use SIGPROF instead,
> which fires based on when the process is in the kernel, as well.  This
> would likely at least alleviate the problem, but I'm suspicious that
> it would be covering up the real underlying problem.  I don't think
> that UML spending time in the kernel should cause a 1000-fold decrease
> in VTALRM delivery frequency.
>
> 				Jeff
>