Re: [dm-devel] bug: using smp_processor_id() in preemptible code in rr_select_path()

[Jens Axboe <axboe@kernel.dk>]

On 08/08/2016 10:32 AM, Benjamin Block wrote:
> On 12:06 Fri 05 Aug     , Mike Snitzer wrote:
>> On Fri, Aug 05 2016 at 11:54am -0400,
>> Jens Axboe <axboe@kernel.dk> wrote:
>>
>>> On 08/05/2016 09:42 AM, Mike Snitzer wrote:
>>>> On Fri, Aug 05 2016 at 11:33P -0400,
>>>> Jens Axboe <axboe@kernel.dk> wrote:
>>>>
>>>>> On 08/05/2016 09:27 AM, Mike Snitzer wrote:
>>>>>> On Wed, Aug 03 2016 at 11:35am -0400,
>>>>>> Benjamin Block <bblock@linux.vnet.ibm.com> wrote:
>>>>>>
>>>>>>> Hej Mike,
>>>>>>>
>>>>>>> when running a debug-kernel today with several multipath-devices using
>>>>>>> the round-robin path selector I noticed that the kernel throws these
>>>>>>> warnings here:
>>>>>>>
>>>>>>> BUG: using smp_processor_id() in preemptible [00000000] code: kdmwork-252:0/881
>>>>>>> caller is rr_select_path+0x36/0x108 [dm_round_robin]
>>>>>>> CPU: 1 PID: 881 Comm: kdmwork-252:0 Not tainted 4.7.0-debug #4
>>>>>>>      00000000617679b8 0000000061767a48 0000000000000002 0000000000000000
>>>>>>>      0000000061767ae8 0000000061767a60 0000000061767a60 00000000001145d0
>>>>>>>      0000000000000000 0000000000b962ae 0000000000bb291e 000000000000000b
>>>>>>>      0000000061767aa8 0000000061767a48 0000000000000000 0000000000000000
>>>>>>>      0700000000b962ae 00000000001145d0 0000000061767a48 0000000061767aa8
>>>>>>> Call Trace:
>>>>>>> ([<00000000001144a2>] show_trace+0x8a/0xe0)
>>>>>>> ([<0000000000114586>] show_stack+0x8e/0xf0)
>>>>>>> ([<00000000006c7fdc>] dump_stack+0x9c/0xe0)
>>>>>>> ([<00000000006fbbc0>] check_preemption_disabled+0x108/0x130)
>>>>>>> ([<000003ff80268646>] rr_select_path+0x36/0x108 [dm_round_robin])
>>>>>>> ([<000003ff80259a42>] choose_path_in_pg+0x42/0xc8 [dm_multipath])
>>>>>>> ([<000003ff80259b62>] choose_pgpath+0x9a/0x1a0 [dm_multipath])
>>>>>>> ([<000003ff8025b51a>] __multipath_map.isra.5+0x72/0x228 [dm_multipath])
>>>>>>> ([<000003ff8025b75e>] multipath_map+0x3e/0x50 [dm_multipath])
>>>>>>> ([<000003ff80225eb6>] map_request+0x66/0x458 [dm_mod])
>>>>>>> ([<000003ff802262ec>] map_tio_request+0x44/0x70 [dm_mod])
>>>>>>> ([<000000000016835a>] kthread_worker_fn+0xf2/0x1d8)
>>>>>>> ([<00000000001681da>] kthread+0x112/0x120)
>>>>>>> ([<000000000098378a>] kernel_thread_starter+0x6/0xc)
>>>>>>> ([<0000000000983784>] kernel_thread_starter+0x0/0xc)
>>>>>>> no locks held by kdmwork-252:0/881.
>>>>>>>
> [:snip:]
>>>
>>> I always forget the details (if this confuses lockdep or not), but you
>>> could potentially turn it into:
>>>
>>> local_irq_save(flags);
>>> x = this_cpu_ptr();
>>> [...]
>>>
>>> spin_lock(&s->lock);
>>> [...]
>>>
>>> instead.
>>
>> Cool, I've coded up the patch (compile tested only).
>>
>> Benjamin, any chance you could test this against your v4.7 kernel and
>> report back?
>>
>> Thanks,
>> Mike
>>
>> diff --git a/drivers/md/dm-round-robin.c b/drivers/md/dm-round-robin.c
>> index 4ace1da..ed446f8 100644
>> --- a/drivers/md/dm-round-robin.c
>> +++ b/drivers/md/dm-round-robin.c
>> @@ -210,14 +210,17 @@ static struct dm_path *rr_select_path(struct path_selector *ps, size_t nr_bytes)
>>  	struct path_info *pi = NULL;
>>  	struct dm_path *current_path = NULL;
>>
>> +	local_irq_save(flags);
>>  	current_path = *this_cpu_ptr(s->current_path);
>>  	if (current_path) {
>>  		percpu_counter_dec(&s->repeat_count);
>> -		if (percpu_counter_read_positive(&s->repeat_count) > 0)
>> +		if (percpu_counter_read_positive(&s->repeat_count) > 0) {
>> +			local_irq_restore(flags);
>>  			return current_path;
>> +		}
>>  	}
>>
>> -	spin_lock_irqsave(&s->lock, flags);
>> +	spin_lock(&s->lock);
>>  	if (!list_empty(&s->valid_paths)) {
>>  		pi = list_entry(s->valid_paths.next, struct path_info, list);
>>  		list_move_tail(&pi->list, &s->valid_paths);
>> @@ -225,7 +228,8 @@ static struct dm_path *rr_select_path(struct path_selector *ps, size_t nr_bytes)
>>  		set_percpu_current_path(s, pi->path);
>>  		current_path = pi->path;
>>  	}
>> -	spin_unlock_irqrestore(&s->lock, flags);
>> +	spin_unlock(&s->lock);
>> +	local_irq_restore(flags);
>>
>>  	return current_path;
>>  }
>>
>
> Ok, this works as far as the warnings don't appear anymore. But while
> applying the patch and thinking about it, why local_irq_save() and not
> preempt_disable()? "Sounds" like this is the function you want, and I
> also stumbled across this in Documentation/preempt-locking.txt:
>
>   But keep in mind that 'irqs disabled' is a fundamentally unsafe way of
>   disabling preemption - any spin_unlock() decreasing the preemption
>   count to 0 might trigger a reschedule.
>
> The spinlock would do an other nested preempt_disable(), but those even
> out.

local_irq_save(), since we need to grab the lock irq safe very shortly
anyway. As long as they nest properly, the approach is fine, and it's
more efficient than first doing a preempt_disable(), then still needing
a irq safe spinlock.

-- 
Jens Axboe