RDMA/bnxt_re: use shadow qd while posting non blocking rcfw command

Whenever there is a fast path IO and create/destroy resources from
the slow path is happening in parallel, we may notice high latency
of slow path command completion.

Introduces a shadow queue depth to prevent the outstanding requests
to the FW. Driver will not allow more than #RCFW_CMD_NON_BLOCKING_SHADOW_QD
non-blocking commands to the Firmware.

Shadow queue depth is a soft limit only for non-blocking
commands. Blocking commands will be posted to the firmware
as long as there is a free slot.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-8-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

2 files changed, 61 insertions, 2 deletions

diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index 3c4f72aa5fee..f7d12388ef31 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -281,8 +281,21 @@ done:
 	return 0;
 }
 
-int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
-				 struct bnxt_qplib_cmdqmsg *msg)
+/**
+ * __bnxt_qplib_rcfw_send_message   -	qplib interface to send
+ * and complete rcfw command.
+ * @rcfw      -   rcfw channel instance of rdev
+ * @msg      -    qplib message internal
+ *
+ * This function does not account shadow queue depth. It will send
+ * all the command unconditionally as long as send queue is not full.
+ *
+ * Returns:
+ * 0 if command completed by firmware.
+ * Non zero if the command is not completed by firmware.
+ */
+static int __bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
+					  struct bnxt_qplib_cmdqmsg *msg)
 {
 	struct creq_qp_event *evnt = (struct creq_qp_event *)msg->resp;
 	u16 cookie;
@@ -331,6 +344,48 @@ int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
 
 	return rc;
 }
+
+/**
+ * bnxt_qplib_rcfw_send_message   -	qplib interface to send
+ * and complete rcfw command.
+ * @rcfw      -   rcfw channel instance of rdev
+ * @msg      -    qplib message internal
+ *
+ * Driver interact with Firmware through rcfw channel/slow path in two ways.
+ * a. Blocking rcfw command send. In this path, driver cannot hold
+ * the context for longer period since it is holding cpu until
+ * command is not completed.
+ * b. Non-blocking rcfw command send. In this path, driver can hold the
+ * context for longer period. There may be many pending command waiting
+ * for completion because of non-blocking nature.
+ *
+ * Driver will use shadow queue depth. Current queue depth of 8K
+ * (due to size of rcfw message there can be actual ~4K rcfw outstanding)
+ * is not optimal for rcfw command processing in firmware.
+ *
+ * Restrict at max #RCFW_CMD_NON_BLOCKING_SHADOW_QD Non-Blocking rcfw commands.
+ * Allow all blocking commands until there is no queue full.
+ *
+ * Returns:
+ * 0 if command completed by firmware.
+ * Non zero if the command is not completed by firmware.
+ */
+int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
+				 struct bnxt_qplib_cmdqmsg *msg)
+{
+	int ret;
+
+	if (!msg->block) {
+		down(&rcfw->rcfw_inflight);
+		ret = __bnxt_qplib_rcfw_send_message(rcfw, msg);
+		up(&rcfw->rcfw_inflight);
+	} else {
+		ret = __bnxt_qplib_rcfw_send_message(rcfw, msg);
+	}
+
+	return ret;
+}
+
 /* Completions */
 static int bnxt_qplib_process_func_event(struct bnxt_qplib_rcfw *rcfw,
 					 struct creq_func_event *func_event)
@@ -932,6 +987,7 @@ int bnxt_qplib_enable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw,
 		return rc;
 	}
 
+	sema_init(&rcfw->rcfw_inflight, RCFW_CMD_NON_BLOCKING_SHADOW_QD);
 	bnxt_qplib_start_rcfw(rcfw);
 
 	return 0;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
index 32e5b6767cc8..862bfbfa7884 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
@@ -66,6 +66,8 @@ static inline void bnxt_qplib_rcfw_cmd_prep(struct cmdq_base *req,
 	req->cmd_size = cmd_size;
 }
 
+/* Shadow queue depth for non blocking command */
+#define RCFW_CMD_NON_BLOCKING_SHADOW_QD	64
 #define RCFW_CMD_WAIT_TIME_MS		20000 /* 20 Seconds timeout */
 
 /* CMDQ elements */
@@ -197,6 +199,7 @@ struct bnxt_qplib_rcfw {
 	u64 oos_prev;
 	u32 init_oos_stats;
 	u32 cmdq_depth;
+	struct semaphore rcfw_inflight;
 };
 
 struct bnxt_qplib_cmdqmsg {