blk-mq: rework flush sequencing logic
Witch to using a preallocated flush_rq for blk-mq similar to what's done
with the old request path. This allows us to set up the request properly
with a tag from the actually allowed range and ->rq_disk as needed by
some drivers. To make life easier we also switch to dynamic allocation
of ->flush_rq for the old path.
This effectively reverts most of
"blk-mq: fix for flush deadlock"
and
"blk-mq: Don't reserve a tag for flush request"
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 14c8f35..a59b0565 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -194,27 +194,9 @@
}
static struct request *__blk_mq_alloc_request(struct blk_mq_hw_ctx *hctx,
- gfp_t gfp, bool reserved,
- int rw)
+ gfp_t gfp, bool reserved)
{
- struct request *req;
- bool is_flush = false;
- /*
- * flush need allocate a request, leave at least one request for
- * non-flush IO to avoid deadlock
- */
- if ((rw & REQ_FLUSH) && !(rw & REQ_FLUSH_SEQ)) {
- if (atomic_inc_return(&hctx->pending_flush) >=
- hctx->queue_depth - hctx->reserved_tags - 1) {
- atomic_dec(&hctx->pending_flush);
- return NULL;
- }
- is_flush = true;
- }
- req = blk_mq_alloc_rq(hctx, gfp, reserved);
- if (!req && is_flush)
- atomic_dec(&hctx->pending_flush);
- return req;
+ return blk_mq_alloc_rq(hctx, gfp, reserved);
}
static struct request *blk_mq_alloc_request_pinned(struct request_queue *q,
@@ -227,7 +209,7 @@
struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
struct blk_mq_hw_ctx *hctx = q->mq_ops->map_queue(q, ctx->cpu);
- rq = __blk_mq_alloc_request(hctx, gfp & ~__GFP_WAIT, reserved, rw);
+ rq = __blk_mq_alloc_request(hctx, gfp & ~__GFP_WAIT, reserved);
if (rq) {
blk_mq_rq_ctx_init(q, ctx, rq, rw);
break;
@@ -244,15 +226,14 @@
return rq;
}
-struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
- gfp_t gfp, bool reserved)
+struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp)
{
struct request *rq;
if (blk_mq_queue_enter(q))
return NULL;
- rq = blk_mq_alloc_request_pinned(q, rw, gfp, reserved);
+ rq = blk_mq_alloc_request_pinned(q, rw, gfp, false);
if (rq)
blk_mq_put_ctx(rq->mq_ctx);
return rq;
@@ -276,7 +257,7 @@
/*
* Re-init and set pdu, if we have it
*/
-static void blk_mq_rq_init(struct blk_mq_hw_ctx *hctx, struct request *rq)
+void blk_mq_rq_init(struct blk_mq_hw_ctx *hctx, struct request *rq)
{
blk_rq_init(hctx->queue, rq);
@@ -290,9 +271,6 @@
const int tag = rq->tag;
struct request_queue *q = rq->q;
- if ((rq->cmd_flags & REQ_FLUSH) && !(rq->cmd_flags & REQ_FLUSH_SEQ))
- atomic_dec(&hctx->pending_flush);
-
blk_mq_rq_init(hctx, rq);
blk_mq_put_tag(hctx->tags, tag);
@@ -946,14 +924,14 @@
hctx = q->mq_ops->map_queue(q, ctx->cpu);
trace_block_getrq(q, bio, rw);
- rq = __blk_mq_alloc_request(hctx, GFP_ATOMIC, false, bio->bi_rw);
+ rq = __blk_mq_alloc_request(hctx, GFP_ATOMIC, false);
if (likely(rq))
- blk_mq_rq_ctx_init(q, ctx, rq, bio->bi_rw);
+ blk_mq_rq_ctx_init(q, ctx, rq, rw);
else {
blk_mq_put_ctx(ctx);
trace_block_sleeprq(q, bio, rw);
- rq = blk_mq_alloc_request_pinned(q, bio->bi_rw,
- __GFP_WAIT|GFP_ATOMIC, false);
+ rq = blk_mq_alloc_request_pinned(q, rw, __GFP_WAIT|GFP_ATOMIC,
+ false);
ctx = rq->mq_ctx;
hctx = q->mq_ops->map_queue(q, ctx->cpu);
}
@@ -1230,9 +1208,7 @@
hctx->queue_num = i;
hctx->flags = reg->flags;
hctx->queue_depth = reg->queue_depth;
- hctx->reserved_tags = reg->reserved_tags;
hctx->cmd_size = reg->cmd_size;
- atomic_set(&hctx->pending_flush, 0);
blk_mq_init_cpu_notifier(&hctx->cpu_notifier,
blk_mq_hctx_notify, hctx);
@@ -1412,9 +1388,14 @@
blk_mq_init_flush(q);
blk_mq_init_cpu_queues(q, reg->nr_hw_queues);
- if (blk_mq_init_hw_queues(q, reg, driver_data))
+ q->flush_rq = kzalloc(round_up(sizeof(struct request) + reg->cmd_size,
+ cache_line_size()), GFP_KERNEL);
+ if (!q->flush_rq)
goto err_hw;
+ if (blk_mq_init_hw_queues(q, reg, driver_data))
+ goto err_flush_rq;
+
blk_mq_map_swqueue(q);
mutex_lock(&all_q_mutex);
@@ -1422,6 +1403,9 @@
mutex_unlock(&all_q_mutex);
return q;
+
+err_flush_rq:
+ kfree(q->flush_rq);
err_hw:
kfree(q->mq_map);
err_map:
