On 2/18/22 10:41, Melanie Plageman (Microsoft) wrote:
Currently a single blk_mq_tag_set is associated with a Scsi_Host. When SCSI
controllers are limited, attaching multiple devices to the same controller is
required. In cloud environments with relatively high latency persistent storage,
requiring all devices on a controller to share a single blk_mq_tag_set
negatively impacts performance.
For example: a device provisioned with high IOPS and BW limits on the same
controller as a smaller and slower device can starve the slower device of tags.
This is especially noticeable when the slower device's workload has low I/O
depth tasks.
The Cc-list for this patch series is way too long. Cc-ing linux-scsi and
the most active SCSI contributors would have been sufficient.
Is the reported behavior reproducible with an upstream Linux kernel? I'm
asking this because I think that the following block layer code should
prevent the reported starvation:
/*
* For shared tag users, we track the number of currently active users
* and attempt to provide a fair share of the tag depth for each of them.
*/
static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
struct sbitmap_queue *bt)
{
unsigned int depth, users;
if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED))
return true;
/*
* Don't try dividing an ant
*/
if (bt->sb.depth == 1)
return true;
if (blk_mq_is_shared_tags(hctx->flags)) {
struct request_queue *q = hctx->queue;
if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags))
return true;
} else {
if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
return true;
}
users = atomic_read(&hctx->tags->active_queues);
if (!users)
return true;
/*
* Allow at least some tags
*/
depth = max((bt->sb.depth + users - 1) / users, 4U);
return __blk_mq_active_requests(hctx) < depth;
}
Bart.
