Ok, I looked into this a bit more, and while I'm still fine with the
patch I think it's only half of what we should do here. There really
is no point in doing the first non-blocking path in blk_mq_map_request
either as bt_get itself already does the non-blocking pass, and also
runs the queue when scheduling in the later loop as well. So to get
towards what I had in my tree we also need something like this:
---
>From c69bf02929d9c37d193b004a4c3c85c1142fa996 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@xxxxxx>
Date: Thu, 22 Sep 2016 11:38:23 -0700
Subject: blk-mq: remove non-blocking pass in blk_mq_map_request
bt_get already does a non-blocking pass as well as running the queue
when scheduling internally, no need to duplicate it.
Signed-off-by: Christoph Hellwig <hch@xxxxxx>
---
block/blk-mq.c | 14 +-------------
1 file changed, 1 insertion(+), 13 deletions(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c
index eae2f12..e9ebe98 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1210,20 +1210,8 @@ static struct request *blk_mq_map_request(struct request_queue *q,
op_flags |= REQ_SYNC;
trace_block_getrq(q, bio, op);
- blk_mq_set_alloc_data(&alloc_data, q, BLK_MQ_REQ_NOWAIT, ctx, hctx);
+ blk_mq_set_alloc_data(&alloc_data, q, 0, ctx, hctx);
rq = __blk_mq_alloc_request(&alloc_data, op, op_flags);
- if (unlikely(!rq)) {
- blk_mq_run_hw_queue(hctx, false);
- blk_mq_put_ctx(ctx);
- trace_block_sleeprq(q, bio, op);
-
- ctx = blk_mq_get_ctx(q);
- hctx = q->mq_ops->map_queue(q, ctx->cpu);
- blk_mq_set_alloc_data(&alloc_data, q, 0, ctx, hctx);
- rq = __blk_mq_alloc_request(&alloc_data, op, op_flags);
- ctx = alloc_data.ctx;
- hctx = alloc_data.hctx;
- }
hctx->queued++;
data->hctx = hctx;
--
2.1.4
--
To unsubscribe from this list: send the line "unsubscribe linux-block" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
