[PATCH v17 04/14] block: Support allocating from a specific software queue

[Bart Van Assche <bvanassche@xxxxxxx>]

A later patch will preserve the order of pipelined zoned writes by
submitting all zoned writes per zone to the same software queue as
previously submitted zoned writes. Hence support allocating a request
from a specific software queue.

Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Damien Le Moal <dlemoal@xxxxxxxxxx>
Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
---
 block/blk-mq.c | 18 ++++++++++++++----
 block/blk-mq.h |  1 +
 2 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 666e6e6ba143..4262c85be206 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -495,6 +495,7 @@ __blk_mq_alloc_requests_batch(struct blk_mq_alloc_data *data)
 static struct request *__blk_mq_alloc_requests(struct blk_mq_alloc_data *data)
 {
 	struct request_queue *q = data->q;
+	int swq_cpu = data->swq_cpu;
 	u64 alloc_time_ns = 0;
 	struct request *rq;
 	unsigned int tag;
@@ -507,7 +508,8 @@ static struct request *__blk_mq_alloc_requests(struct blk_mq_alloc_data *data)
 		data->flags |= BLK_MQ_REQ_NOWAIT;
 
 retry:
-	data->ctx = blk_mq_get_ctx(q);
+	data->swq_cpu = swq_cpu >= 0 ? swq_cpu : raw_smp_processor_id();
+	data->ctx = __blk_mq_get_ctx(q, data->swq_cpu);
 	data->hctx = blk_mq_map_queue(q, data->cmd_flags, data->ctx);
 
 	if (q->elevator) {
@@ -587,6 +589,7 @@ static struct request *blk_mq_rq_cache_fill(struct request_queue *q,
 		.cmd_flags	= opf,
 		.nr_tags	= plug->nr_ios,
 		.cached_rqs	= &plug->cached_rqs,
+		.swq_cpu	= -1,
 	};
 	struct request *rq;
 
@@ -648,6 +651,7 @@ struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
 			.flags		= flags,
 			.cmd_flags	= opf,
 			.nr_tags	= 1,
+			.swq_cpu	= -1,
 		};
 		int ret;
 
@@ -2964,12 +2968,14 @@ static bool blk_mq_attempt_bio_merge(struct request_queue *q,
 }
 
 static struct request *blk_mq_get_new_requests(struct request_queue *q,
+					       int swq_cpu,
 					       struct blk_plug *plug,
 					       struct bio *bio,
 					       unsigned int nsegs)
 {
 	struct blk_mq_alloc_data data = {
 		.q		= q,
+		.swq_cpu	= swq_cpu,
 		.nr_tags	= 1,
 		.cmd_flags	= bio->bi_opf,
 	};
@@ -2993,7 +2999,8 @@ static struct request *blk_mq_get_new_requests(struct request_queue *q,
  * Check if there is a suitable cached request and return it.
  */
 static struct request *blk_mq_peek_cached_request(struct blk_plug *plug,
-		struct request_queue *q, blk_opf_t opf)
+						  struct request_queue *q,
+						  int swq_cpu, blk_opf_t opf)
 {
 	enum hctx_type type = blk_mq_get_hctx_type(opf);
 	struct request *rq;
@@ -3003,6 +3010,8 @@ static struct request *blk_mq_peek_cached_request(struct blk_plug *plug,
 	rq = rq_list_peek(&plug->cached_rqs);
 	if (!rq || rq->q != q)
 		return NULL;
+	if (swq_cpu >= 0 && rq->mq_ctx->cpu != swq_cpu)
+		return NULL;
 	if (type != rq->mq_hctx->type &&
 	    (type != HCTX_TYPE_READ || rq->mq_hctx->type != HCTX_TYPE_DEFAULT))
 		return NULL;
@@ -3061,6 +3070,7 @@ void blk_mq_submit_bio(struct bio *bio)
 	struct blk_mq_hw_ctx *hctx;
 	unsigned int nr_segs;
 	struct request *rq;
+	int swq_cpu = -1;
 	blk_status_t ret;
 
 	/*
@@ -3109,7 +3119,7 @@ void blk_mq_submit_bio(struct bio *bio)
 		goto queue_exit;
 
 new_request:
-	rq = blk_mq_peek_cached_request(plug, q, bio->bi_opf);
+	rq = blk_mq_peek_cached_request(plug, q, swq_cpu, bio->bi_opf);
 	if (rq) {
 		blk_mq_use_cached_rq(rq, plug, bio);
 		/*
@@ -3119,7 +3129,7 @@ void blk_mq_submit_bio(struct bio *bio)
 		 */
 		blk_queue_exit(q);
 	} else {
-		rq = blk_mq_get_new_requests(q, plug, bio, nr_segs);
+		rq = blk_mq_get_new_requests(q, swq_cpu, plug, bio, nr_segs);
 		if (unlikely(!rq)) {
 			if (bio->bi_opf & REQ_NOWAIT)
 				bio_wouldblock_error(bio);
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 44979e92b79f..d5536dcf2182 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -158,6 +158,7 @@ struct blk_mq_alloc_data {
 	struct rq_list *cached_rqs;
 
 	/* input & output parameter */
+	int swq_cpu;
 	struct blk_mq_ctx *ctx;
 	struct blk_mq_hw_ctx *hctx;
 };