Re: [PATCH 3/4] block: move queue enter logic into blk_mq_submit_bio()

[Jens Axboe <axboe@kernel.dk>]

On 11/4/21 11:32 AM, Christoph Hellwig wrote:
> On Thu, Nov 04, 2021 at 06:14:30AM -0600, Jens Axboe wrote:
>> To player it safer, I would suggest we fold in something like the
>> below. That keeps the submit_checks() under the queue enter.
> 
> Yes, this looks much better.  Nit below:
> 
>>  	struct block_device *bdev = bio->bi_bdev;
>>  	struct request_queue *q = bdev_get_queue(bdev);
>> @@ -868,14 +868,13 @@ static void __submit_bio(struct bio *bio)
>>  {
>>  	struct gendisk *disk = bio->bi_bdev->bd_disk;
>>  
>> -	if (!submit_bio_checks(bio) || !blk_crypto_bio_prep(&bio))
>> -		return;
>>  	if (!disk->fops->submit_bio) {
>>  		blk_mq_submit_bio(bio);
>>  	} else {
>>  		if (unlikely(bio_queue_enter(bio) != 0))
>>  			return;
>> -		disk->fops->submit_bio(bio);
>> +		if (submit_bio_checks(bio) && blk_crypto_bio_prep(&bio))
>> +			disk->fops->submit_bio(bio);
>>  		blk_queue_exit(disk->queue);
>>  	}
> 
> A this point moving the whole ->submit_bio based branch into a
> helper probably makes sense as well.

Sure, I can do that.

>> +	if (unlikely(!blk_crypto_bio_prep(&bio)))
>> +		return;
>> +
>>  	blk_queue_bounce(q, &bio);
>>  	if (blk_may_split(q, bio))
>>  		__blk_queue_split(q, &bio, &nr_segs);
>> @@ -2551,6 +2554,8 @@ void blk_mq_submit_bio(struct bio *bio)
>>  
>>  		if (unlikely(!blk_mq_queue_enter(q, bio)))
>>  			return;
>> +		if (unlikely(!submit_bio_checks(bio)))
>> +			goto put_exit;
> 
> This now skips the checks for the cached request case, doesn't it?

It did, I did add that when folding it in though.

-- 
Jens Axboe