Re: [PATCH -tip v2] sched: more sched_domain iterations fix

[Xiaotian Feng <dfeng@redhat.com>]

On 04/26/2011 05:27 PM, Peter Zijlstra wrote:
> On Fri, 2011-04-22 at 18:53 +0800, Xiaotian feng wrote:
>> From: Xiaotian Feng<dfeng@redhat.com>
>>
>> sched_domain iterations needs to be protected by rcu_read_lock() now,
>> this patch adds another two places which needs the rcu lock, which is
>> spotted by following suspicious rcu_dereference_check() usage warnings.
>>
>> kernel/sched_rt.c:1244 invoked rcu_dereference_check() without protection!
>> kernel/sched_stats.h:41 invoked rcu_dereference_check() without protection!
>
> Much better, one worry:
>
>> Signed-off-by: Xiaotian Feng<dfeng@redhat.com>
>> Cc: Ingo Molnar<mingo@elte.hu>
>> Cc: Peter Zijlstra<peterz@infradead.org>
>> ---
>
>> diff --git a/kernel/sched_stats.h b/kernel/sched_stats.h
>> index 48ddf43..331e01b 100644
>> --- a/kernel/sched_stats.h
>> +++ b/kernel/sched_stats.h
>> @@ -37,7 +37,7 @@ static int show_schedstat(struct seq_file *seq, void *v)
>>
>>   #ifdef CONFIG_SMP
>>   		/* domain-specific stats */
>> -		preempt_disable();
>> +		rcu_read_lock();
>>   		for_each_domain(cpu, sd) {
>>   			enum cpu_idle_type itype;
>>
>> @@ -64,7 +64,7 @@ static int show_schedstat(struct seq_file *seq, void *v)
>>   			    sd->ttwu_wake_remote, sd->ttwu_move_affine,
>>   			    sd->ttwu_move_balance);
>>   		}
>> -		preempt_enable();
>> +		rcu_read_unlock();
>>   #endif
>>   	}
>>   	kfree(mask_str);
>
> Did you indeed validate that the preempt_disable() wasn't needed for
> anything else? Your changelog doesn't mention and I didn't check, just
> noticed the possibility on the first posting.
>
Sorry, I just checked them, preempt_disable/enable was introduced by 
commit 674311d,
the rcu_read_lock_sched is not existed at that time.

btw, as for_each_domain is protected by rcu_read_lock() and 
preempt_disable is not suffice
any more, could we also update comments in for_each_domain?

/*
  * The domain tree (rq->sd) is protected by RCU's quiescent state 
transition.
  * See detach_destroy_domains: synchronize_sched for details.
  *
  * The domain tree of any CPU may only be accessed from within
  * preempt-disabled sections.
  */