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

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

[Luca Mariotti]

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;
-- 
2.27.0

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

[Holger Hoffstätte]

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?

-h

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

[Holger Hoffstätte]

On 2021-05-20 09:15, Holger Hoffstätte wrote:
> 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?

Tried again, another boom. This time I managed to capture a stack trace
(scrolled out at the top, but it's the same as before and easily reproducible):

https://imgur.com/a/sU1pDaF

This is a heavily patched 5.10.x, but it's been perfectly stable so far
until I added this last patch; the one before was avoid-circular-stable-merges.
Maybe an unintentional side effect? In any case all I see is bfq_delta() inlined
into bfq_calc_finish() and exploding since entity->weight is apparently 0.

Hope this helps.

-h

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

[Paolo Valente]

> 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.

Thanks,
Paolo

> -h

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

[Holger Hoffstätte]

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.
Will gladly test your patch! :)

cheers
Holger

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

[Paolo Valente]

[-- Attachment #1: Type: text/plain, Size: 3217 bytes --]



> 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


[-- Attachment #2: 0001-block-bfq-avoid-delayed-merge-of-async-queues.patch.gz --]
[-- Type: application/x-gzip, Size: 697 bytes --]

[-- Attachment #3: Type: text/plain, Size: 20 bytes --]


> cheers
> Holger

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

[Holger Hoffstätte]

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

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

[Paolo Valente]

> Il giorno 24 mag 2021, alle ore 20:45, Holger Hoffstätte <holger@applied-asynchrony.com> ha scritto:
> 
> 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>
> 

Great!

Thank you very much!

I will put this fix in an upcoming patch series.

Paolo

> Thanks!
> Holger