mbox series

[BUGFIX,0/2] bfq: fix unbalanced decrements causing loss of throughput

Message ID 20181206181819.11148-1-paolo.valente@linaro.org
Headers show
Series bfq: fix unbalanced decrements causing loss of throughput | expand

Message

Paolo Valente Dec. 6, 2018, 6:18 p.m. UTC
Hi Jens,
the first patch in this series fixes an error in the decrementing of
the counter of the number of groups with pending I/O. This wrong
decrement caused loss of throughput or, less likely, of control on
I/O. The second patch is a fix of some wrong comments, which somehow
contributed to making the above bug more difficult to find.

Thanks,
Paolo

Paolo Valente (2):
  block, bfq: fix decrement of num_active_groups
  block, bfq: fix comments on __bfq_deactivate_entity

 block/bfq-iosched.c | 76 +++++++++++++++++++++++++++++++++++++----------------
 block/bfq-iosched.h | 51 +++++++++++++++++++++++++++++++++--
 block/bfq-wf2q.c    | 16 ++++++-----
 3 files changed, 112 insertions(+), 31 deletions(-)

--
2.16.1

Comments

Jens Axboe Dec. 7, 2018, 2:23 a.m. UTC | #1
On 12/6/18 11:18 AM, Paolo Valente wrote:
> Hi Jens,

> the first patch in this series fixes an error in the decrementing of

> the counter of the number of groups with pending I/O. This wrong

> decrement caused loss of throughput or, less likely, of control on

> I/O. The second patch is a fix of some wrong comments, which somehow

> contributed to making the above bug more difficult to find.


Are you fine with this going into 4.21? I can't quite tell what your
intent is. The first patch has a Fixes for something that went into
this series, but then patch 2 is a comment update that would not
normally be something to be applied at this stage.

-- 
Jens Axboe
Paolo Valente Dec. 7, 2018, 10:01 a.m. UTC | #2
> Il giorno 7 dic 2018, alle ore 03:23, Jens Axboe <axboe@kernel.dk> ha scritto:

> 

> On 12/6/18 11:18 AM, Paolo Valente wrote:

>> Hi Jens,

>> the first patch in this series fixes an error in the decrementing of

>> the counter of the number of groups with pending I/O. This wrong

>> decrement caused loss of throughput or, less likely, of control on

>> I/O. The second patch is a fix of some wrong comments, which somehow

>> contributed to making the above bug more difficult to find.

> 

> Are you fine with this going into 4.21? I can't quite tell what your

> intent is. The first patch has a Fixes for something


yep, that fixes a serious error.

> that went into

> this series, but then patch 2 is a comment update that would not

> normally be something to be applied at this stage.

> 


and yes, only comments changed by the second one

May it make sense to apply them in two steps, one in the 4.20 and the other one in the 4.21?

Thanks,
Paolo

> -- 

> Jens Axboe

>
Jens Axboe Dec. 7, 2018, 2:40 p.m. UTC | #3
On 12/7/18 3:01 AM, Paolo Valente wrote:
> 

> 

>> Il giorno 7 dic 2018, alle ore 03:23, Jens Axboe <axboe@kernel.dk> ha scritto:

>>

>> On 12/6/18 11:18 AM, Paolo Valente wrote:

>>> Hi Jens,

>>> the first patch in this series fixes an error in the decrementing of

>>> the counter of the number of groups with pending I/O. This wrong

>>> decrement caused loss of throughput or, less likely, of control on

>>> I/O. The second patch is a fix of some wrong comments, which somehow

>>> contributed to making the above bug more difficult to find.

>>

>> Are you fine with this going into 4.21? I can't quite tell what your

>> intent is. The first patch has a Fixes for something

> 

> yep, that fixes a serious error.

> 

>> that went into

>> this series, but then patch 2 is a comment update that would not

>> normally be something to be applied at this stage.

>>

> 

> and yes, only comments changed by the second one

> 

> May it make sense to apply them in two steps, one in the 4.20 and the other one in the 4.21?


I think so, I'll do that.

-- 
Jens Axboe
Paolo Valente Jan. 14, 2019, 9:09 a.m. UTC | #4
> Il giorno 7 dic 2018, alle ore 15:40, Jens Axboe <axboe@kernel.dk> ha scritto:

> 

> On 12/7/18 3:01 AM, Paolo Valente wrote:

>> 

>> 

>>> Il giorno 7 dic 2018, alle ore 03:23, Jens Axboe <axboe@kernel.dk> ha scritto:

>>> 

>>> On 12/6/18 11:18 AM, Paolo Valente wrote:

>>>> Hi Jens,

>>>> the first patch in this series fixes an error in the decrementing of

>>>> the counter of the number of groups with pending I/O. This wrong

>>>> decrement caused loss of throughput or, less likely, of control on

>>>> I/O. The second patch is a fix of some wrong comments, which somehow

>>>> contributed to making the above bug more difficult to find.

>>> 

>>> Are you fine with this going into 4.21? I can't quite tell what your

>>> intent is. The first patch has a Fixes for something

>> 

>> yep, that fixes a serious error.

>> 

>>> that went into

>>> this series, but then patch 2 is a comment update that would not

>>> normally be something to be applied at this stage.

>>> 

>> 

>> and yes, only comments changed by the second one

>> 

>> May it make sense to apply them in two steps, one in the 4.20 and the other one in the 4.21?

> 

> I think so, I'll do that.


Hi Jens,
is the second patch still queued?

Thanks,
Paolo

> 

> -- 

> Jens Axboe
Jens Axboe Jan. 14, 2019, 1:26 p.m. UTC | #5
On 1/14/19 2:09 AM, Paolo Valente wrote:
> 

> 

>> Il giorno 7 dic 2018, alle ore 15:40, Jens Axboe <axboe@kernel.dk> ha scritto:

>>

>> On 12/7/18 3:01 AM, Paolo Valente wrote:

>>>

>>>

>>>> Il giorno 7 dic 2018, alle ore 03:23, Jens Axboe <axboe@kernel.dk> ha scritto:

>>>>

>>>> On 12/6/18 11:18 AM, Paolo Valente wrote:

>>>>> Hi Jens,

>>>>> the first patch in this series fixes an error in the decrementing of

>>>>> the counter of the number of groups with pending I/O. This wrong

>>>>> decrement caused loss of throughput or, less likely, of control on

>>>>> I/O. The second patch is a fix of some wrong comments, which somehow

>>>>> contributed to making the above bug more difficult to find.

>>>>

>>>> Are you fine with this going into 4.21? I can't quite tell what your

>>>> intent is. The first patch has a Fixes for something

>>>

>>> yep, that fixes a serious error.

>>>

>>>> that went into

>>>> this series, but then patch 2 is a comment update that would not

>>>> normally be something to be applied at this stage.

>>>>

>>>

>>> and yes, only comments changed by the second one

>>>

>>> May it make sense to apply them in two steps, one in the 4.20 and the other one in the 4.21?

>>

>> I think so, I'll do that.

> 

> Hi Jens,

> is the second patch still queued?


That got dropped for some reason, I've applied it now.

-- 
Jens Axboe