blk-mq: Don't reserve a tag for flush request

Reserving a tag (request) for flush to avoid dead lock is a overkill. A
tag is valuable resource. We can track the number of flush requests and
disallow having too many pending flush requests allocated. With this
patch, blk_mq_alloc_request_pinned() could do a busy nop (but not a dead
loop) if too many pending requests are allocated and new flush request
is allocated. But this should not be a problem, too many pending flush
requests are very rare case.

I verified this can fix the deadlock caused by too many pending flush
requests.

Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

2 files changed, 35 insertions, 19 deletions

diff --git a/block/blk-flush.c b/block/blk-flush.c
index 9288aaf35c21..9143e85226c7 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -284,9 +284,8 @@ static void mq_flush_work(struct work_struct *work)
 
         q = container_of(work, struct request_queue, mq_flush_work);
 
-        /* We don't need set REQ_FLUSH_SEQ, it's for consistency */
         rq = blk_mq_alloc_request(q, WRITE_FLUSH|REQ_FLUSH_SEQ,
-                __GFP_WAIT|GFP_ATOMIC, true);
+                __GFP_WAIT|GFP_ATOMIC, false);
         rq->cmd_type = REQ_TYPE_FS;
         rq->end_io = flush_end_io;
 
@@ -408,8 +407,11 @@ void blk_insert_flush(struct request *rq)
         /*
          * @policy now records what operations need to be done.  Adjust
          * REQ_FLUSH and FUA for the driver.
+         * We keep REQ_FLUSH for mq to track flush requests. For !FUA,
+         * we never dispatch the request directly.
          */
-        rq->cmd_flags &= ~REQ_FLUSH;
+        if (rq->cmd_flags & REQ_FUA)
+                rq->cmd_flags &= ~REQ_FLUSH;
         if (!(fflags & REQ_FUA))
                 rq->cmd_flags &= ~REQ_FUA;
 
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 57039fcd9c93..9072d0ab184f 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -194,9 +194,27 @@ static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
 }
 
 static struct request *__blk_mq_alloc_request(struct blk_mq_hw_ctx *hctx,
-                                              gfp_t gfp, bool reserved)
+                                              gfp_t gfp, bool reserved,
+                                              int rw)
 {
-        return blk_mq_alloc_rq(hctx, gfp, 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;
 }
 
 static struct request *blk_mq_alloc_request_pinned(struct request_queue *q,
@@ -209,7 +227,7 @@ static struct request *blk_mq_alloc_request_pinned(struct request_queue *q,
                 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);
+                rq = __blk_mq_alloc_request(hctx, gfp & ~__GFP_WAIT, reserved, rw);
                 if (rq) {
                         blk_mq_rq_ctx_init(q, ctx, rq, rw);
                         break;
@@ -272,6 +290,9 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
         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);
 
@@ -900,14 +921,14 @@ static void blk_mq_make_request(struct request_queue *q, struct bio *bio)
         hctx = q->mq_ops->map_queue(q, ctx->cpu);
 
         trace_block_getrq(q, bio, rw);
-        rq = __blk_mq_alloc_request(hctx, GFP_ATOMIC, false);
+        rq = __blk_mq_alloc_request(hctx, GFP_ATOMIC, false, bio->bi_rw);
         if (likely(rq))
-                blk_mq_rq_ctx_init(q, ctx, rq, rw);
+                blk_mq_rq_ctx_init(q, ctx, rq, bio->bi_rw);
         else {
                 blk_mq_put_ctx(ctx);
                 trace_block_sleeprq(q, bio, rw);
-                rq = blk_mq_alloc_request_pinned(q, rw, __GFP_WAIT|GFP_ATOMIC,
-                                                        false);
+                rq = blk_mq_alloc_request_pinned(q, bio->bi_rw,
+                                __GFP_WAIT|GFP_ATOMIC, false);
                 ctx = rq->mq_ctx;
                 hctx = q->mq_ops->map_queue(q, ctx->cpu);
         }
@@ -1184,7 +1205,9 @@ static int blk_mq_init_hw_queues(struct request_queue *q,
                 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);
@@ -1309,15 +1332,6 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_reg *reg,
                 reg->queue_depth = BLK_MQ_MAX_DEPTH;
         }
 
-        /*
-         * Set aside a tag for flush requests.  It will only be used while
-         * another flush request is in progress but outside the driver.
-         *
-         * TODO: only allocate if flushes are supported
-         */
-        reg->queue_depth++;
-        reg->reserved_tags++;
-
         if (reg->queue_depth < (reg->reserved_tags + BLK_MQ_TAG_MIN))
                 return ERR_PTR(-EINVAL);