nvmet: don't use BLK_MQ_REQ_NOWAIT for passthru

By default, we set the passthru request allocation flag such that it
returns the error in the following code path and we fail the I/O when
BLK_MQ_REQ_NOWAIT is used for request allocation :-

nvme_alloc_request()
 blk_mq_alloc_request()
  blk_mq_queue_enter()
   if (flag & BLK_MQ_REQ_NOWAIT)
        return -EBUSY; <-- return if busy.

On some controllers using BLK_MQ_REQ_NOWAIT ends up in I/O error where
the controller is perfectly healthy and not in a degraded state.

Block layer request allocation does allow us to wait instead of
immediately returning the error when we BLK_MQ_REQ_NOWAIT flag is not
used. This has shown to fix the I/O error problem reported under
heavy random write workload.

Remove the BLK_MQ_REQ_NOWAIT parameter for passthru request allocation
which resolves this issue.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

diff --git a/drivers/nvme/target/passthru.c b/drivers/nvme/target/passthru.c
index 0814cba8298a7ef86337dfa65e910730ea738336..8ee94f056898324cca7495adda2cc9469d97cd6d 100644 (file)
--- a/drivers/nvme/target/passthru.c
+++ b/drivers/nvme/target/passthru.c
@@ -244,7 +244,7 @@ static void nvmet_passthru_execute_cmd(struct nvmet_req *req)
                q = ns->queue;
        }
 
-       rq = nvme_alloc_request(q, req->cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
+       rq = nvme_alloc_request(q, req->cmd, 0, NVME_QID_ANY);
        if (IS_ERR(rq)) {
                status = NVME_SC_INTERNAL;
                goto out_put_ns;