[PATCH 3/3] nvme-tcp: per-controller I/O workqueues

[Hannes Reinecke]

From: Hannes Reinecke <hare at suse.de>

Implement per-controller I/O workqueues to reduce workqueue contention
during I/O and improve I/O performance.

Performance comparison:
               baseline rx/tx    blk-mq   multiple workqueues
4k seq write:  449MiB/s 480MiB/s 524MiB/s 540MiB/s
4k rand write: 410MiB/s 481MiB/s 524MiB/s 539MiB/s
4k seq read:   478MiB/s 481MiB/s 566MiB/s 582MiB/s
4k rand read:  547MiB/s 480MiB/s 511MiB/s 633MiB/s

Signed-off-by: Hannes Reinecke <hare at kernel.org>
---
 drivers/nvme/host/tcp.c | 34 ++++++++++++++++++----------------
 1 file changed, 18 insertions(+), 16 deletions(-)

diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index a5c42a7b4bee..fc8f682f686d 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -194,6 +194,7 @@ struct nvme_tcp_ctrl {
 	struct sockaddr_storage src_addr;
 	struct nvme_ctrl	ctrl;
 
+	struct workqueue_struct	*io_wq;
 	struct work_struct	err_work;
 	struct delayed_work	connect_work;
 	struct nvme_tcp_request async_req;
@@ -202,7 +203,6 @@ struct nvme_tcp_ctrl {
 
 static LIST_HEAD(nvme_tcp_ctrl_list);
 static DEFINE_MUTEX(nvme_tcp_ctrl_mutex);
-static struct workqueue_struct *nvme_tcp_wq;
 static const struct blk_mq_ops nvme_tcp_mq_ops;
 static const struct blk_mq_ops nvme_tcp_admin_mq_ops;
 static int nvme_tcp_try_send(struct nvme_tcp_queue *queue);
@@ -410,7 +410,7 @@ static inline void nvme_tcp_queue_request(struct nvme_tcp_request *req,
 	}
 
 	if (last && nvme_tcp_queue_has_pending(queue))
-		queue_work_on(queue->io_cpu, nvme_tcp_wq, &queue->io_work);
+		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);
 }
 
 static void nvme_tcp_process_req_list(struct nvme_tcp_queue *queue)
@@ -987,7 +987,7 @@ static void nvme_tcp_data_ready(struct sock *sk)
 	queue = sk->sk_user_data;
 	if (likely(queue && queue->rd_enabled) &&
 	    !test_bit(NVME_TCP_Q_POLLING, &queue->flags))
-		queue_work_on(queue->io_cpu, nvme_tcp_wq, &queue->io_work);
+		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);
 	read_unlock_bh(&sk->sk_callback_lock);
 }
 
@@ -999,7 +999,7 @@ static void nvme_tcp_write_space(struct sock *sk)
 	queue = sk->sk_user_data;
 	if (likely(queue && sk_stream_is_writeable(sk))) {
 		clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
-		queue_work_on(queue->io_cpu, nvme_tcp_wq, &queue->io_work);
+		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);
 	}
 	read_unlock_bh(&sk->sk_callback_lock);
 }
@@ -1325,7 +1325,7 @@ static void nvme_tcp_io_work(struct work_struct *w)
 		dev_dbg(queue->ctrl->ctrl.device, "queue %d: queue stall (%u msecs)\n",
 			nvme_tcp_queue_id(queue), jiffies_to_msecs(overrun));
 
-	queue_work_on(queue->io_cpu, nvme_tcp_wq, &queue->io_work);
+	queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);
 }
 
 static void nvme_tcp_free_crypto(struct nvme_tcp_queue *queue)
@@ -2486,6 +2486,8 @@ static void nvme_tcp_free_ctrl(struct nvme_ctrl *nctrl)
 
 	nvmf_free_options(nctrl->opts);
 free_ctrl:
+	destroy_workqueue(ctrl->io_wq);
+
 	kfree(ctrl->queues);
 	kfree(ctrl);
 }
@@ -2676,7 +2678,7 @@ static void nvme_tcp_commit_rqs(struct blk_mq_hw_ctx *hctx)
 	struct nvme_tcp_queue *queue = hctx->driver_data;
 
 	if (!llist_empty(&queue->req_list))
-		queue_work_on(queue->io_cpu, nvme_tcp_wq, &queue->io_work);
+		queue_work_on(queue->io_cpu, queue->ctrl->io_wq, &queue->io_work);
 }
 
 static blk_status_t nvme_tcp_queue_rq(struct blk_mq_hw_ctx *hctx,
@@ -2812,6 +2814,7 @@ static struct nvme_tcp_ctrl *nvme_tcp_alloc_ctrl(struct device *dev,
 		struct nvmf_ctrl_options *opts)
 {
 	struct nvme_tcp_ctrl *ctrl;
+	unsigned int wq_flags = WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_SYSFS;
 	int ret;
 
 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
@@ -2883,6 +2886,15 @@ static struct nvme_tcp_ctrl *nvme_tcp_alloc_ctrl(struct device *dev,
 	if (ret)
 		goto out_kfree_queues;
 
+	if (wq_unbound)
+		wq_flags |= WQ_UNBOUND;
+	ctrl->io_wq = alloc_workqueue("nvme_tcp_wq_%d", wq_flags, 0,
+				      ctrl->ctrl.instance);
+	if (!ctrl->io_wq) {
+		nvme_put_ctrl(&ctrl->ctrl);
+		return ERR_PTR(-ENOMEM);
+	}
+
 	return ctrl;
 out_kfree_queues:
 	kfree(ctrl->queues);
@@ -2948,7 +2960,6 @@ static struct nvmf_transport_ops nvme_tcp_transport = {
 
 static int __init nvme_tcp_init_module(void)
 {
-	unsigned int wq_flags = WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_SYSFS;
 	int cpu;
 
 	BUILD_BUG_ON(sizeof(struct nvme_tcp_hdr) != 8);
@@ -2960,13 +2971,6 @@ static int __init nvme_tcp_init_module(void)
 	BUILD_BUG_ON(sizeof(struct nvme_tcp_icresp_pdu) != 128);
 	BUILD_BUG_ON(sizeof(struct nvme_tcp_term_pdu) != 24);
 
-	if (wq_unbound)
-		wq_flags |= WQ_UNBOUND;
-
-	nvme_tcp_wq = alloc_workqueue("nvme_tcp_wq", wq_flags, 0);
-	if (!nvme_tcp_wq)
-		return -ENOMEM;
-
 	for_each_possible_cpu(cpu)
 		atomic_set(&nvme_tcp_cpu_queues[cpu], 0);
 
@@ -2985,8 +2989,6 @@ static void __exit nvme_tcp_cleanup_module(void)
 		nvme_delete_ctrl(&ctrl->ctrl);
 	mutex_unlock(&nvme_tcp_ctrl_mutex);
 	flush_workqueue(nvme_delete_wq);
-
-	destroy_workqueue(nvme_tcp_wq);
 }
 
 module_init(nvme_tcp_init_module);
-- 
2.35.3