Re: [BUG] set_mempolicy(MPOL_INTERLEAV) cause kernel panic

[KAMEZAWA Hiroyuki]

On Mon, 27 Jul 2009 17:38:30 -0700 (PDT)
David Rientjes <rientjes@xxxxxxxxxx> wrote:

> On Tue, 28 Jul 2009, KAMEZAWA Hiroyuki wrote:
> 
> > Because we dont' update, task->mems_allowed need to be initilaized as
> > N_POSSIBLE_NODES. At usual thinking,  it should be N_HIGH_MEMORY or
> > N_ONLINE_NODES, as my patch does.
> > 
> 

> On MEM_OFFLINE, cpusets calls scan_for_empty_cpusets() which will 
> intersect each system cpuset's mems_allowed with N_HIGH_MEMORY.  It then 
> calls update_tasks_nodemask() which will update task->mems_allowed for 
> each task assigned to those cpusets.  This has a callback into the 
> mempolicy code to rebind the policy with the new mems.
> 
> So there's no apparent issue with memory hotplug in dealing with cpuset 
> mems, although I suggested that this be done for MEM_GOING_OFFLINE instead 
> of waiting until the mem is actually offline.
>
I _wrote_ this is just a side story to bug.
online/offline isn't related to this bug.

> The problem originally reported here doesn't appear to have anything to do 
> with hotplug, it looks like it is the result of Lee's observation that 
> ia64 defaults top_cpuset's mems to N_POSSIBLE, which _should_ have been 
> updated by cpuset_init_smp(). 
cpuset_init_smp() just updates cpuset's mask.
init's task->mems_allowed is intizialized independently from cpuset's mask.

Could you teach me a pointer for Lee's observation ?

> So it makes me believe that N_HIGH_MEMORY 
> isn't actually ready by the time do_basic_setup() is called to be useful.
> 
N_HIGH_MEMORY should be ready when zonelist is built. If not, it's bug.


Thanks,
-Kame

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/