nvme_rdma: Support TOS for RDMA initiator

The spdk_nvme_ctrlr_opts now supports a transport_tos option
that allows setting of the 'type of service' value in the IPv4 header.

This is needed to support lossless RoCE setups.

Note: Only RDMA is supported at this point.

Change-Id: I21825fc197c60f539a7d2d651a970ea380d8b56d
Signed-off-by: Michael Haeuptle <michael.haeuptle@hpe.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/15908
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>

CHANGELOG.md

@@ -110,6 +110,11 @@ tags will appear in SPDK's log at the warn level. As the SPDK application exits,
 will log a summary of how many times `SPDK_LOG_DEPRECATED()` was called for each
 tag that was logged at least once.
 
+### nvme
+
+Added `transport_tos` to `spdk_nvme_ctrlr_opts` to support setting of the "type of service"
+value in the IPv4 header. Only RDMA is supported at this time.
+
 ## v22.09
 
 ### accel

CONFIG

@@ -92,6 +92,7 @@ CONFIG_FIO_SOURCE_DIR=/usr/src/fio
 CONFIG_RDMA=n
 CONFIG_RDMA_SEND_WITH_INVAL=n
 CONFIG_RDMA_SET_ACK_TIMEOUT=n
+CONFIG_RDMA_SET_TOS=n
 CONFIG_RDMA_PROV=verbs
 
 # Enable NVMe Character Devices.

configure

@@ -846,6 +846,15 @@ than or equal to 4.14 will see significantly reduced performance.
 		echo "RDMA_OPTION_ID_ACK_TIMEOUT is not supported"
 	fi
 
+	if echo -e '#include <rdma/rdma_cma.h>\n' \
+		'int main(void) { return !!RDMA_OPTION_ID_TOS; }\n' \
+		| "${BUILD_CMD[@]}" -c - 2> /dev/null; then
+		CONFIG[RDMA_SET_TOS]="y"
+	else
+		CONFIG[RDMA_SET_TOS]="n"
+		echo "RDMA_OPTION_ID_TOS is not supported"
+	fi
+
 	if [ "${CONFIG[RDMA_PROV]}" == "mlx5_dv" ]; then
 		if ! echo -e '#include <spdk/stdinc.h>\n' \
 			'#include <infiniband/mlx5dv.h>\n' \

include/spdk/nvme.h

@@ -275,8 +275,15 @@ struct spdk_nvme_ctrlr_opts {
 	 * 12B (header) + 2B (hash) + 176B (base64 for 1024b + crc32) + 3B (colons) + 1B (NULL) + 6B (extra space for future)
 	 */
 	char psk[200];
+
+	/**
+	 * It is used for RDMA transport.
+	 *
+	 * Set the IP protocol type of service value for RDMA transport. Default is 0, which means that the TOS will not be set.
+	 */
+	uint8_t transport_tos;
 } __attribute__((packed));
-SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_ctrlr_opts) == 817, "Incorrect size");
+SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_ctrlr_opts) == 818, "Incorrect size");
 
 /**
  * NVMe acceleration operation callback.

lib/nvme/nvme_internal.h

@@ -171,6 +171,8 @@ extern pid_t g_spdk_nvme_pid;
 #define SPDK_NVME_TRANSPORT_ACK_TIMEOUT_DISABLED	(0)
 #define SPDK_NVME_DEFAULT_TRANSPORT_ACK_TIMEOUT	SPDK_NVME_TRANSPORT_ACK_TIMEOUT_DISABLED
 
+#define SPDK_NVME_TRANSPORT_TOS_DISABLED	(0)
+
 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS	(10000)
 
 /* We want to fit submission and completion rings each in a single 2MB

lib/nvme/nvme_rdma.c

@@ -1023,6 +1023,18 @@ nvme_rdma_addr_resolved(struct nvme_rdma_qpair *rqpair, int ret)
 #endif
 	}
 
+	if (rqpair->qpair.ctrlr->opts.transport_tos != SPDK_NVME_TRANSPORT_TOS_DISABLED) {
+#ifdef SPDK_CONFIG_RDMA_SET_TOS
+		uint8_t tos = rqpair->qpair.ctrlr->opts.transport_tos;
+		ret = rdma_set_option(rqpair->cm_id, RDMA_OPTION_ID, RDMA_OPTION_ID_TOS, &tos, sizeof(tos));
+		if (ret) {
+			SPDK_NOTICELOG("Can't apply RDMA_OPTION_ID_TOS %u, ret %d\n", tos, ret);
+		}
+#else
+		SPDK_DEBUGLOG(nvme, "transport_tos is not supported\n");
+#endif
+	}
+
 	ret = rdma_resolve_route(rqpair->cm_id, NVME_RDMA_TIME_OUT_IN_MS);
 	if (ret) {
 		SPDK_ERRLOG("rdma_resolve_route\n");