Re: [PATCH BUGFIX] block, bfq: postpone rq preparation to insert or merge

[Paolo Valente]

> Il giorno 07 mag 2018, alle ore 07:56, Mike Galbraith <efault@xxxxxx> ha scritto:
> 
> On Sun, 2018-05-06 at 09:42 +0200, Paolo Valente wrote:
>> 
>> diff --git a/block/bfq-mq-iosched.c b/block/bfq-mq-iosched.c
>> index 118f319af7c0..6662efe29b69 100644
>> --- a/block/bfq-mq-iosched.c
>> +++ b/block/bfq-mq-iosched.c
>> @@ -525,8 +525,13 @@ static void bfq_limit_depth(unsigned int op, struct blk_mq_alloc_data *data)
>>        if (unlikely(bfqd->sb_shift != bt->sb.shift))
>>                bfq_update_depths(bfqd, bt);
>> 
>> +#if 0
>>        data->shallow_depth =
>>                bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];
>                                                            ^^^^^^^^^^^^^
> 
> Q: why doesn't the top of this function look like so?
> 
> ---
> block/bfq-iosched.c |    2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> --- a/block/bfq-iosched.c
> +++ b/block/bfq-iosched.c
> @@ -539,7 +539,7 @@ static void bfq_limit_depth(unsigned int
> 	struct bfq_data *bfqd = data->q->elevator->elevator_data;
> 	struct sbitmap_queue *bt;
> 
> -	if (op_is_sync(op) && !op_is_write(op))
> +	if (!op_is_write(op))
> 		return;
> 
> 	if (data->flags & BLK_MQ_REQ_RESERVED) {
> 
> It looks a bit odd that these elements exist...
> 
> +       /*
> +        * no more than 75% of tags for sync writes (25% extra tags
> +        * w.r.t. async I/O, to prevent async I/O from starving sync
> +        * writes)
> +        */
> +       bfqd->word_depths[0][1] = max(((1U<<bfqd->sb_shift) * 3)>>2, 1U);
> 
> +       /* no more than ~37% of tags for sync writes (~20% extra tags) */
> +       bfqd->word_depths[1][1] = max(((1U<<bfqd->sb_shift) * 6)>>4, 1U);
> 
> ...yet we index via and log a guaranteed zero.
> 

I'm not sure I got your point, so, to help you help me quickly, I'll
repeat what I expect the code you highlight to do:

- sync reads must have no limitation, and the lines
if (op_is_sync(op) && !op_is_write(op))
	return;
make sure they don't

- sync writes must be limited, and the code you pasted above computes
those limits

- for sync writes, for which op_is_sync(op) is true (but the condition
"op_is_sync(op) && !op_is_write(op)" is false), the line:
	bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];
becomes
	bfqd->word_depths[!!bfqd->wr_busy_queues][1];
e yields the right limit for sync writes, depending on bfqd->wr_busy_queues.

Where is the bug?

Thanks,
Paolo


> 	-Mike
> 
>