lkml.org 
[lkml]   [2021]   [May]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH v2] block: fix trace completion for chained bio
From
Date

On 4/23/2021 4:04 PM, Edward Hsieh wrote:
> On 3/23/2021 5:22 AM, NeilBrown wrote:
>> On Wed, Mar 03 2021, edwardh wrote:
>>
>>> From: Edward Hsieh <edwardh@synology.com>
>>>
>>> For chained bio, trace_block_bio_complete in bio_endio is currently
>>> called
>>> only by the parent bio once upon all chained bio completed.
>>> However, the sector and size for the parent bio are modified in
>>> bio_split.
>>> Therefore, the size and sector of the complete events might not match
>>> the
>>> queue events in blktrace.
>>>
>>> The original fix of bio completion trace <fbbaf700e7b1> ("block: trace
>>> completion of all bios.") wants multiple complete events to correspond
>>> to one queue event but missed this.
>>>
>>> md/raid5 read with bio cross chunks can reproduce this issue.
>>>
>>> To fix, move trace completion into the loop for every chained bio to
>>> call.
>>
>> Thanks.  I think this is correct as far as tracing goes.
>> However the code still looks a bit odd.
>>
>> The comment for the handling of bio_chain_endio suggests that the *only*
>> purpose for that is to avoid deep recursion.  That suggests it should be
>> at the end of the function.
>> As it is blk_throtl_bio_endio() and bio_unint() are only called on the
>> last bio in a chain.
>> That seems wrong.
>>
>> I'd be more comfortable if the patch moved the bio_chain_endio()
>> handling to the end, after all of that.
>> So the function would end.
>>
>> if (bio->bi_end_io == bio_chain_endio) {
>>     bio = __bio_chain_endio(bio);
>>     goto again;
>> } else if (bio->bi_end_io)
>>     bio->bi_end_io(bio);
>>
>> Jens:  can you see any reason why that functions must only be called on
>> the last bio in the chain?
>>
>> Thanks,
>> NeilBrown
>>
>
> Hi Neil and Jens,
>
> From the commit message, bio_uninit is put here for bio allocated in
> special ways (e.g., on stack), that will not be release by bio_free. For
> chained bio, __bio_chain_endio invokes bio_put and release the
> resources, so it seems that we don't need to call bio_uninit for chained
> bio.
>
> The blk_throtl_bio_endio is used to update the latency for the throttle
> group. I think the latency should only be updated after the whole bio is
> finished?
>
> To make sense for the "tail call optimization" in the comment, I'll
> suggest to wrap the whole statement with an else. What do you think?
>
> if (bio->bi_end_io == bio_chain_endio) {
>     bio = __bio_chain_endio(bio);
>     goto again;
> } else {
>     blk_throtl_bio_endio(bio);
>     /* release cgroup info */
>     bio_uninit(bio);
>     if (bio->bi_end_io)
>         bio->bi_end_io(bio);
> }
>
> Thanks,
> Edward Hsieh

Hi Neil and Jens,

Any feedback on this one?

Thank you,
Edward Hsieh

\
 
 \ /
  Last update: 2021-05-10 04:07    [W:0.078 / U:0.616 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site