On 11/4/19 12:30 PM, Keith Busch wrote:
> On Mon, Nov 04, 2019 at 10:05:43AM -0800, Christoph Hellwig wrote:
>> diff --git a/block/blk-merge.c b/block/blk-merge.c
>> index 48e6725b32ee..06eb38357b41 100644
>> --- a/block/blk-merge.c
>> +++ b/block/blk-merge.c
>> @@ -293,7 +293,7 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>> void __blk_queue_split(struct request_queue *q, struct bio **bio,
>> unsigned int *nr_segs)
>> {
>> - struct bio *split;
>> + struct bio *split = NULL;
>>
>> switch (bio_op(*bio)) {
>> case REQ_OP_DISCARD:
>> @@ -309,6 +309,19 @@ void __blk_queue_split(struct request_queue *q, struct bio **bio,
>> nr_segs);
>> break;
>> default:
>> + /*
>> + * All drivers must accept single-segments bios that are <=
>> + * PAGE_SIZE. This is a quick and dirty check that relies on
>> + * the fact that bi_io_vec[0] is always valid if a bio has data.
>> + * The check might lead to occasional false negatives when bios
>> + * are cloned, but compared to the performance impact of cloned
>> + * bios themselves the loop below doesn't matter anyway.
>> + */
>> + if ((*bio)->bi_vcnt == 1 &&
>> + (*bio)->bi_io_vec[0].bv_len <= PAGE_SIZE) {
>> + *nr_segs = 1;
>> + break;
>> + }
>> split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
>> break;
>> }
>
> If the device advertises a chunk boundary and this small IO happens to
> cross it, skipping the split is going to harm performance.
Does anyone do that, that isn't the first gen intel weirdness? Honest question,
but always seemed to me that this spec addition was driven entirely by that
one device.
And if they do, do they align on non-4k?
--
Jens Axboe
