Re: [PATCH BUGFIX] block, bfq: fix delayed stable merge check

["Holger Hoffstätte" <holger@applied-asynchrony.com>]

On 2021-05-24 19:41, Paolo Valente wrote:
> 
> 
>> Il giorno 24 mag 2021, alle ore 19:13, Holger Hoffstätte <holger@applied-asynchrony.com> ha scritto:
>>
>> On 2021-05-24 18:57, Paolo Valente wrote:
>>>> Il giorno 20 mag 2021, alle ore 09:15, Holger Hoffstätte <holger@applied-asynchrony.com> ha scritto:
>>>>
>>>> On 2021-05-18 12:43, Luca Mariotti wrote:
>>>>> When attempting to schedule a merge of a given bfq_queue with the currently
>>>>> in-service bfq_queue or with a cooperating bfq_queue among the scheduled
>>>>> bfq_queues, delayed stable merge is checked for rotational or non-queueing
>>>>> devs. For this stable merge to be performed, some conditions must be met.
>>>>> If the current bfq_queue underwent some split from some merged bfq_queue,
>>>>> one of these conditions is that two hundred milliseconds must elapse from
>>>>> split, otherwise this condition is always met.
>>>>> Unfortunately, by mistake, time_is_after_jiffies() was written instead of
>>>>> time_is_before_jiffies() for this check, verifying that less than two
>>>>> hundred milliseconds have elapsed instead of verifying that at least two
>>>>> hundred milliseconds have elapsed.
>>>>> Fix this issue by replacing time_is_after_jiffies() with
>>>>> time_is_before_jiffies().
>>>>> Signed-off-by: Luca Mariotti <mariottiluca1@hotmail.it>
>>>>> Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
>>>>> Signed-off-by: Pietro Pedroni <pedroni.pietro.96@gmail.com>
>>>>> ---
>>>>>   block/bfq-iosched.c | 2 +-
>>>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>>> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
>>>>> index acd1f881273e..2adb1e69c9d2 100644
>>>>> --- a/block/bfq-iosched.c
>>>>> +++ b/block/bfq-iosched.c
>>>>> @@ -2697,7 +2697,7 @@ bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
>>>>>   	if (unlikely(!bfqd->nonrot_with_queueing)) {
>>>>>   		if (bic->stable_merge_bfqq &&
>>>>>   		    !bfq_bfqq_just_created(bfqq) &&
>>>>> -		    time_is_after_jiffies(bfqq->split_time +
>>>>> +		    time_is_before_jiffies(bfqq->split_time +
>>>>>   					  msecs_to_jiffies(200))) {
>>>>>   			struct bfq_queue *stable_merge_bfqq =
>>>>>   				bic->stable_merge_bfqq;
>>>>
>>>> Not sure why but with this patch I quickly got a division-by-zero in BFQ and
>>>> complete system halt. Unfortunately I couldn't capture the exact stack trace,
>>>> but it read something like bfq_calc_weight() or something ike that.
>>>> I looked through the code and found bfq_delta(), so maybe weight got
>>>> reduced to 0?
>>>>
>>> Hi Holger,
>>> is this (easily) reproducible for you?  If so, I'd like to propose you
>>> a candidate fix.
>>
>> Yes, it's easily reproducible (should be reproducible on 5.13-rc as well).
>> Simple read/write I/O on a cold FS (rotational disk obviously) will crash
>> pretty much immediately; without it everything works fine, likely because the
>> bug (in the recent queue merging patches?) is never triggered due to the
>> accidentally-wrong time calculation.
> 
> Exactly!
> 
> Unfortunately, no crash happens on my systems.  Or, actually, crashes
> stopped after the attached fix.
> 
>> Will gladly test your patch! :)
>>
> 
> Here it is!
> 
> I'll make a proper commit after your early tests.
> 
> Crossing my fingers,
> Paolo

That did it - it now survived a bunch of heavy read/write/mixed I/O that
would previously crash right away. Maybe it's because btrfs uses several
workers and so different IOs got mixed together? Anyway:

Fixes: 430a67f9d616 ("block, bfq: merge bursts of newly-created queues")
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>

Thanks!
Holger