[PATCH rfc 2/2] nvmet-tcp: set SO_PRIORITY for network sockets

[PATCH rfc 2/2] nvmet-tcp: set SO_PRIORITY for network sockets

[Wunderlich, Mark]

Enable ability to associate all sockets related to NVMf TCP
traffic to a priority group that will perform optimized
network processing for this traffic class. Maintain initial
default behavior of using priority of zero.

Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Mark Wunderlich <mark.wunderlich@intel.com>
---
 drivers/nvme/target/tcp.c |   25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
index a232ef5..10c8722 100644
--- a/drivers/nvme/target/tcp.c
+++ b/drivers/nvme/target/tcp.c
@@ -19,6 +19,16 @@
 
 #define NVMET_TCP_DEF_INLINE_DATA_SIZE	(4 * PAGE_SIZE)
 
+/* Define the socket priority to use for connections were it is desirable
+ * that the NIC consider performing optimized packet processing or filtering.
+ * A non-zero value being sufficient to indicate general consideration of any
+ * possible optimization.  Making it a module param allows for alternative
+ * values that may be unique for some NIC implementations.
+ */
+static int nvmet_tcp_so_priority;
+module_param(nvmet_tcp_so_priority, int, 0644);
+MODULE_PARM_DESC(nvmet_tcp_so_priority, "nvmet tcp socket optimize priority");
+
 #define NVMET_TCP_RECV_BUDGET		8
 #define NVMET_TCP_SEND_BUDGET		8
 #define NVMET_TCP_IO_WORK_BUDGET	64
@@ -1461,6 +1471,12 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)
 	if (ret)
 		return ret;
 
+	ret = kernel_setsockopt(sock, SOL_SOCKET, SO_PRIORITY,
+			(char *)&nvmet_tcp_so_priority,
+			sizeof(nvmet_tcp_so_priority));
+	if (ret)
+		return ret;
+
 	/* Set socket type of service */
 	if (inet->rcv_tos > 0) {
 		int tos = inet->rcv_tos;
@@ -1650,6 +1666,15 @@ static int nvmet_tcp_add_port(struct nvmet_port *nport)
 		goto err_sock;
 	}
 
+	ret = kernel_setsockopt(port->sock, SOL_SOCKET, SO_PRIORITY,
+			(char *)&nvmet_tcp_so_priority,
+			sizeof(nvmet_tcp_so_priority));
+	if (ret) {
+		pr_err("failed to set listen socket SO_PRIORITY, ret %d\n",
+			ret);
+		goto err_sock;
+	}
+
 	ret = kernel_bind(port->sock, (struct sockaddr *)&port->addr,
 			sizeof(port->addr));
 	if (ret) {


_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Re: [PATCH rfc 2/2] nvmet-tcp: set SO_PRIORITY for network sockets

[Sagi Grimberg]

> +/* Define the socket priority to use for connections were it is desirable
> + * that the NIC consider performing optimized packet processing or filtering.
> + * A non-zero value being sufficient to indicate general consideration of any
> + * possible optimization.  Making it a module param allows for alternative
> + * values that may be unique for some NIC implementations.
> + */
> +static int nvmet_tcp_so_priority;
> +module_param(nvmet_tcp_so_priority, int, 0644);
> +MODULE_PARM_DESC(nvmet_tcp_so_priority, "nvmet tcp socket optimize priority");
> +

No need for nvmet_tcp prefix I assume.

Should it be a bool if a non-zero value is sufficient?

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

RE: [PATCH rfc 2/2] nvmet-tcp: set SO_PRIORITY for network sockets

[Wunderlich, Mark]

>> +/* Define the socket priority to use for connections were it is 
>> +desirable
>> + * that the NIC consider performing optimized packet processing or filtering.
>> + * A non-zero value being sufficient to indicate general 
>> +consideration of any
>> + * possible optimization.  Making it a module param allows for 
>> +alternative
>> + * values that may be unique for some NIC implementations.
>> + */
>> +static int nvmet_tcp_so_priority;
>> +module_param(nvmet_tcp_so_priority, int, 0644); 
>> +MODULE_PARM_DESC(nvmet_tcp_so_priority, "nvmet tcp socket optimize 
>> +priority");
>> +

>No need for nvmet_tcp prefix I assume.

Sure.  Was not sure if standard practice is to name such params specific to module.  So 'so_priority', or just 'priority' would be OK?

>Should it be a bool if a non-zero value is sufficient?

Since various NIC vendors may use different actual so_priority values to trigger specific behavior thought is best to leave it adjustable.  True, in our case a Boolean could be made to work for now.  But future proofing may be a good thing.  If we leave as is, maybe would make sense to not allow a value that would exceed the possible range to be used/set.
_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme