Re: [PATCH V6 6/8] block, bfq: retrieve independent access ranges from request queue

[Paolo Valente <paolo.valente@linaro.org>]

> Il giorno 6 dic 2022, alle ore 10:02, Damien Le Moal <damien.lemoal@opensource.wdc.com> ha scritto:
> 
> On 12/6/22 17:41, Paolo Valente wrote:
>> 
>> 
>>> Il giorno 6 dic 2022, alle ore 09:29, Damien Le Moal <damien.lemoal@opensource.wdc.com> ha scritto:
>>> 
>>> On 12/6/22 17:06, Paolo Valente wrote:
>>>> 
>>>> 
>>>>> Il giorno 21 nov 2022, alle ore 02:01, Damien Le Moal <damien.lemoal@opensource.wdc.com> ha scritto:
>>>>> 
>>>> 
>>>> ...
>>>> 
>>>>> 
>>>>>> }
>>>>>> 
>>>>>> static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
>>>>>> @@ -7144,6 +7159,8 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
>>>>>> {
>>>>>> 	struct bfq_data *bfqd;
>>>>>> 	struct elevator_queue *eq;
>>>>>> +	unsigned int i;
>>>>>> +	struct blk_independent_access_ranges *ia_ranges = q->disk->ia_ranges;
>>>>>> 
>>>>>> 	eq = elevator_alloc(q, e);
>>>>>> 	if (!eq)
>>>>>> @@ -7187,10 +7204,31 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
>>>>>> 	bfqd->queue = q;
>>>>>> 
>>>>>> 	/*
>>>>>> -	 * Multi-actuator support not complete yet, default to single
>>>>>> -	 * actuator for the moment.
>>>>>> +	 * If the disk supports multiple actuators, we copy the independent
>>>>>> +	 * access ranges from the request queue structure.
>>>>>> 	 */
>>>>>> -	bfqd->num_actuators = 1;
>>>>>> +	spin_lock_irq(&q->queue_lock);
>>>>>> +	if (ia_ranges) {
>>>>>> +		/*
>>>>>> +		 * Check if the disk ia_ranges size exceeds the current bfq
>>>>>> +		 * actuator limit.
>>>>>> +		 */
>>>>>> +		if (ia_ranges->nr_ia_ranges > BFQ_MAX_ACTUATORS) {
>>>>>> +			pr_crit("nr_ia_ranges higher than act limit: iars=%d, max=%d.\n",
>>>>>> +				ia_ranges->nr_ia_ranges, BFQ_MAX_ACTUATORS);
>>>>>> +			pr_crit("Falling back to single actuator mode.\n");
>>>>>> +			bfqd->num_actuators = 0;
>>>>>> +		} else {
>>>>>> +			bfqd->num_actuators = ia_ranges->nr_ia_ranges;
>>>>>> +
>>>>>> +			for (i = 0; i < bfqd->num_actuators; i++)
>>>>>> +				bfqd->ia_ranges[i] = ia_ranges->ia_range[i];
>>>>>> +		}
>>>>>> +	} else {
>>>>>> +		bfqd->num_actuators = 0;
>>>>> 
>>>>> That is very weird. The default should be 1 actuator.
>>>>> ia_ranges->nr_ia_ranges is 0 when the disk does not provide any range
>>>>> information, meaning it is a regular disk with a single actuator.
>>>> 
>>>> Actually, IIUC this assignment to 0 seems to be done exactly when you
>>>> say that it should be done, i.e., when the disk does not provide any
>>>> range information (ia_ranges is NULL). Am I missing something else?
>>> 
>>> No ranges reported means no extra actuators, so a single actuator an
>>> single LBA range for the entire device.
>> 
>> I'm still confused, sorry.  Where will I read sector ranges from, if
>> no sector range information is available (ia_ranges is NULL)?
> 
> start = 0 and nr_sectors = bdev_nr_sectors(bdev).
> No ia_ranges to read.
> 

ok, thanks

>> 
>>> In that case, bfq should process
>>> all IOs using bfqd->ia_ranges[0]. The get range function will always
>>> return that range. That makes the code clean and avoids different path for
>>> nr_ranges == 1 and nr_ranges > 1. No ?
>> 
>> Apart from the above point, for which maybe there is some other
>> source of information for getting ranges, I see the following issue.
>> 
>> What you propose is to save sector information and trigger the
>> range-checking for loop also for the above single-actuator case.  Yet
>> txecuting (one iteration of) that loop will will always result in
>> getting a 0 as index.  So, what's the point is saving data and
>> executing code on each IO, for getting a static result that we already
>> know we will get?
> 
> Surely, you can add an "if (bfqd->num_actuators ==1)" optimization in
> strategic places to optimize for regular devices with a single actuator,
> which bfqd->num_actuators == 1 *exactly* describes. Having
> "bfqd->num_actuators = 0" makes no sense to me.
> 

Ok, I see your point at last, sorry.  I'll check the code, but I think
that there is no problem in moving from 0 to 1 actuators for the case
ia_ranges == NULL.  I meant to separate the case "single actuator with
ia_ranges available" (num_actuators = 1), from the case "no ia_ranges
available" (num_actuators = 0).  But evidently things don't work as I
thought, and using the same value (1) is ok.

Just, let me avoid setting the fields bfqd->sector and
bfqd->nr_sectors for a case where we don't use them.

Thanks,
Paolo

> But if you feel strongly about this, feel free to ignore this.
> 
>> 
>> Thanks,
>> Paolo
>> 
>>> 
>>>> 
>>>> Once again, all other suggestions applied. I'm about to submit a V7.
>>>> 
>>>> Thanks,
>>>> Paolo
>>>> 
>>> 
>>> -- 
>>> Damien Le Moal
>>> Western Digital Research
>> 
> 
> -- 
> Damien Le Moal
> Western Digital Research