nvme-tcp: optimize queue io_cpu assignment for multiple queue maps

Currently, queue io_cpu assignment is done sequentially for default,
read and poll queues based on queue id. This causes miss-alignment between
context of CPU initiating I/O and the I/O worker thread processing
queued requests or completions.

Change to modify queue io_cpu assignment to take into account queue
maps offset. Each queue io_cpu will start at zero for each queue map.
This essentially aligns read/poll queues to start over the same range as
default queues.

Testing performed by Mark with:
- ram device (nvmet)
- single CPU core (pinned)
- 100% 4k reads
- engine io_uring (not using sq_thread option)
- hipri flag set

Micro-benchmark results show a net gain of:
- increase of 18%-29% in IOPs
- reduction of 16%-22% in average latency
- reduction of 7%-23% in 99.99% latency

Baseline:
========
QDepth/Batch	| IOPs [k]	| Avg. Lat [us]	| 99.99% Lat [us]
-----------------------------------------------------------------
1/1 		| 32.4		| 30.11		| 50.94
32/8		| 179		| 168.20	| 371

CPU alignment:
=============
QDepth/Batch	| IOPs [k]	| Avg. Lat [us]	| 99.99% Lat [us]
-----------------------------------------------------------------
1/1 		| 38.5		|   25.18	| 39.16
32/8		| 231		|   130.75	| 343
Reported-by: Mark Wunderlich <mark.wunderlich@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>

drivers/nvme/host/tcp.c

@@ -1258,13 +1258,67 @@ static int nvme_tcp_init_connection(struct nvme_tcp_queue *queue)
 	return ret;
 }
 
+static bool nvme_tcp_admin_queue(struct nvme_tcp_queue *queue)
+{
+	return nvme_tcp_queue_id(queue) == 0;
+}
+
+static bool nvme_tcp_default_queue(struct nvme_tcp_queue *queue)
+{
+	struct nvme_tcp_ctrl *ctrl = queue->ctrl;
+	int qid = nvme_tcp_queue_id(queue);
+
+	return !nvme_tcp_admin_queue(queue) &&
+		qid < 1 + ctrl->io_queues[HCTX_TYPE_DEFAULT];
+}
+
+static bool nvme_tcp_read_queue(struct nvme_tcp_queue *queue)
+{
+	struct nvme_tcp_ctrl *ctrl = queue->ctrl;
+	int qid = nvme_tcp_queue_id(queue);
+
+	return !nvme_tcp_admin_queue(queue) &&
+		!nvme_tcp_default_queue(queue) &&
+		qid < 1 + ctrl->io_queues[HCTX_TYPE_DEFAULT] +
+			  ctrl->io_queues[HCTX_TYPE_READ];
+}
+
+static bool nvme_tcp_poll_queue(struct nvme_tcp_queue *queue)
+{
+	struct nvme_tcp_ctrl *ctrl = queue->ctrl;
+	int qid = nvme_tcp_queue_id(queue);
+
+	return !nvme_tcp_admin_queue(queue) &&
+		!nvme_tcp_default_queue(queue) &&
+		!nvme_tcp_read_queue(queue) &&
+		qid < 1 + ctrl->io_queues[HCTX_TYPE_DEFAULT] +
+			  ctrl->io_queues[HCTX_TYPE_READ] +
+			  ctrl->io_queues[HCTX_TYPE_POLL];
+}
+
+static void nvme_tcp_set_queue_io_cpu(struct nvme_tcp_queue *queue)
+{
+	struct nvme_tcp_ctrl *ctrl = queue->ctrl;
+	int qid = nvme_tcp_queue_id(queue);
+	int n = 0;
+
+	if (nvme_tcp_default_queue(queue))
+		n = qid - 1;
+	else if (nvme_tcp_read_queue(queue))
+		n = qid - ctrl->io_queues[HCTX_TYPE_DEFAULT] - 1;
+	else if (nvme_tcp_poll_queue(queue))
+		n = qid - ctrl->io_queues[HCTX_TYPE_DEFAULT] -
+				ctrl->io_queues[HCTX_TYPE_READ] - 1;
+	queue->io_cpu = cpumask_next_wrap(n - 1, cpu_online_mask, -1, false);
+}
+
 static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl,
 		int qid, size_t queue_size)
 {
 	struct nvme_tcp_ctrl *ctrl = to_tcp_ctrl(nctrl);
 	struct nvme_tcp_queue *queue = &ctrl->queues[qid];
 	struct linger sol = { .l_onoff = 1, .l_linger = 0 };
-	int ret, opt, rcv_pdu_size, n;
+	int ret, opt, rcv_pdu_size;
 
 	queue->ctrl = ctrl;
 	INIT_LIST_HEAD(&queue->send_list);
@@ -1343,11 +1397,7 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl,
 	}
 
 	queue->sock->sk->sk_allocation = GFP_ATOMIC;
-	if (!qid)
-		n = 0;
-	else
-		n = (qid - 1) % num_online_cpus();
-	queue->io_cpu = cpumask_next_wrap(n - 1, cpu_online_mask, -1, false);
+	nvme_tcp_set_queue_io_cpu(queue);
 	queue->request = NULL;
 	queue->data_remaining = 0;
 	queue->ddgst_remaining = 0;