Re: [PATCH 10/11] cpuset: allow writing offlined masks to cpuset.cpus/mems

[Li Zefan <lizefan-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>]

On 2013/8/21 22:18, Tejun Heo wrote:
> Hello,
> 
> On Wed, Aug 21, 2013 at 06:01:21PM +0800, Li Zefan wrote:
>> -		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
>> +		if (!cpumask_subset(trialcs->cpus_allowed,
>> +				    top_cpuset.cpus_allowed))
> 
> Hmmm... top_cpuset.cpus_allowed is filled using cpumask_setall(),
> which may include more bits than cpu_possible_mask, which is kinda
> weird.  We probably wanna initialize it with cpu_possible_mask and
> also maybe using cpu_possible_mask in the above would be clearer?

In cpuset_init(), all the bits in cpus_allowed are set. Then in
cpuset_init_smp(), it's set to cpu_active_mask.

so we should set top_cpuset.cpus_allowed to possible_mask if mount
with sane_behavior, otherwise set it to active_mask.

> 
> Also, shouldn't this be dependent upon sane_behavior?
> 

We already treat top_cpus.cpus_allowed differently depending on
save_behavior, so the if statement works correctly.