On 03/12/2020 06:49, Ming Lei wrote:
> On Mon, Nov 30, 2020 at 01:36:25AM +0000, Pavel Begunkov wrote:
>> blk_mq_get_tag() selects tag_offset in the beginning and doesn't update
>> it even though the tag set it may have changed hw queue during waiting.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
>> ---
>> block/blk-mq-tag.c | 7 +++++--
>> 1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
>> index 9c92053e704d..dbbf11edf9a6 100644
>> --- a/block/blk-mq-tag.c
>> +++ b/block/blk-mq-tag.c
>> @@ -101,10 +101,8 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)
>> return BLK_MQ_NO_TAG;
>> }
>> bt = tags->breserved_tags;
>> - tag_offset = 0;
>> } else {
>> bt = tags->bitmap_tags;
>> - tag_offset = tags->nr_reserved_tags;
>> }
>>
>> tag = __blk_mq_get_tag(data, bt);
>> @@ -167,6 +165,11 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)
>> sbitmap_finish_wait(bt, ws, &wait);
>>
>> found_tag:
>> + if (data->flags & BLK_MQ_REQ_RESERVED)
>> + tag_offset = 0;
>> + else
>> + tag_offset = tags->nr_reserved_tags;
>> +
>
> So far, the tag offset is tag-set wide, and 'tag_offset' is same for all tags,
> looks no need to add the extra check.
Thanks for clearing that. I still think that's a good thing to do as more
apparent to me and shouldn't be slower -- it unconditionally stores offset
on-stack, even for no-wait path (gcc 9.2). With it's optimised more like
if (!(data->flags & BLK_MQ_REQ_RESERVED))
tag += tags->nr_reserved_tags;
I like assembly a bit more, but not like that matters much.
--
Pavel Begunkov
