Re: [PATCH -v2] x86: allocate cpumask during check irq vectors

[Ingo Molnar <mingo@kernel.org>]

* Prarit Bhargava <prarit@redhat.com> wrote:

> 
> 
> On 01/26/2014 08:32 AM, Ingo Molnar wrote:
> > 
> > * Prarit Bhargava <prarit@redhat.com> wrote:
> > 
> >> On 01/25/2014 03:02 AM, Ingo Molnar wrote:
> >>>
> >>> * Yinghai Lu <yinghai@kernel.org> wrote:
> >>>
> >>>> Fix warning:
> >>>> arch/x86/kernel/irq.c: In function check_irq_vectors_for_cpu_disable:
> >>>> arch/x86/kernel/irq.c:337:1: warning: the frame size of 2052 bytes is larger than 2048 bytes
> >>>>
> >>>> when NR_CPUS=8192
> >>>>
> >>>> We should use zalloc_cpumask_var() instead.
> >>>>
> >>>> -v2: update to GFP_ATOMIC instead and free the allocated cpumask at last.
> >>>>
> >>>> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> >>>> Cc: Prarit Bhargava <prarit@redhat.com>
> >>>>
> >>>> ---
> >>>>  arch/x86/kernel/irq.c |   24 +++++++++++++++++-------
> >>>>  1 file changed, 17 insertions(+), 7 deletions(-)
> >>>>
> >>>> Index: linux-2.6/arch/x86/kernel/irq.c
> >>>> ===================================================================
> >>>> --- linux-2.6.orig/arch/x86/kernel/irq.c
> >>>> +++ linux-2.6/arch/x86/kernel/irq.c
> >>>> @@ -277,11 +277,18 @@ int check_irq_vectors_for_cpu_disable(vo
> >>>>  	unsigned int this_cpu, vector, this_count, count;
> >>>>  	struct irq_desc *desc;
> >>>>  	struct irq_data *data;
> >>>> -	struct cpumask affinity_new, online_new;
> >>>> +	cpumask_var_t affinity_new, online_new;
> >>>> +
> >>>> +	if (!alloc_cpumask_var(&affinity_new, GFP_ATOMIC))
> >>>> +		return -ENOMEM;
> >>>> +	if (!alloc_cpumask_var(&online_new, GFP_ATOMIC)) {
> >>>> +		free_cpumask_var(affinity_new);
> >>>> +		return -ENOMEM;
> >>>> +	}
> >>>
> >>> Atomic allocations can fail easily if the system is under duress.
> >>
> >> Then the hotplug attempt will fail which IMO is okay. [...]
> > 
> > Which is not OK at all for reliable operation, if the system has 
> > otherwise gobs of RAM, which just don't happen to be atomic 
> > allocatable!
> 
> Ingo, I'm really not sure what other option there is here.  Care to 
> suggest one?

Since only ever a single instance of this code will run, can it simply 
be a global cpumask_t variable?

Thanks,

	Ingo