[3/3] mlx5: Implement new counter infrastructure

Commit Message

Christoph Lameter (Ampere) March 15, 2016, 3:54 p.m. UTC

This provides the proper infrastructure to add counters to mlx5 and
shows some constant integer values in sysfs. It lacks the actual
access to the hardware counters. This is a vendor specific issue
so I hope that Mellanox can add that.

Signed-off-by: Mark Bloch <[e-mail]>
Signed-off-by: Christoph Lameter <[e-mail]>


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Doug Ledford May 13, 2016, 7:18 p.m. UTC | #1

On 03/15/2016 11:54 AM, Christoph Lameter wrote:

Please either get Mellanox to provide you what you need to actually set
the counters, or drop this patch from the series.

Christoph Lameter (Ampere) May 16, 2016, 1:52 p.m. UTC | #2

On Fri, 13 May 2016, Doug Ledford wrote:

> On 03/15/2016 11:54 AM, Christoph Lameter wrote:
>
> Please either get Mellanox to provide you what you need to actually set
> the counters, or drop this patch from the series.

Mellanox did a new patch that we are still internally revising.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Doug Ledford May 16, 2016, 3:44 p.m. UTC | #3

On 05/16/2016 09:52 AM, Christoph Lameter wrote:
> On Fri, 13 May 2016, Doug Ledford wrote:
> 
>> On 03/15/2016 11:54 AM, Christoph Lameter wrote:
>>
>> Please either get Mellanox to provide you what you need to actually set
>> the counters, or drop this patch from the series.
> 
> Mellanox did a new patch that we are still internally revising.
> 

Cool, I'll wait for the new series then.

Patch

Index: linux/drivers/infiniband/hw/mlx5/main.c
===================================================================
--- linux.orig/drivers/infiniband/hw/mlx5/main.c
+++ linux/drivers/infiniband/hw/mlx5/main.c
@@ -2122,6 +2122,76 @@  static int mlx5_port_immutable(struct ib
 	return 0;
 }
 
+#define MLX5_NUM_COUNTERS 28
+
+static char *mlx5_counter_names[MLX5_NUM_COUNTERS + 1] = {
+	"rq_num_lle",
+	"sq_num_lle",
+	"rq_num_lqpoe",
+	"sq_num_lqpoe",
+	"rq_num_lpe",
+	"sq_num_lpe",
+	"rq_num_wrfe",
+	"sq_num_wrfe",
+	"sq_num_mwbe",
+	"sq_num_bre",
+	"rq_num_lae",
+	"sq_num_rire",
+	"rq_num_rire",
+	"sq_num_rae",
+	"rq_num_rae",
+	"sq_num_roe",
+	"sq_num_tree",
+	"sq_num_rree",
+	"rq_num_rnr",
+	"sq_num_rnr",
+	"rq_num_oos",
+	"sq_num_oos",
+	"rq_num_mce",
+	"rq_num_udsdprd",
+	"rq_num_ucsdprd",
+	"num_cqovf",
+	"num_eqovf",
+	"num_baddb",
+	NULL };
+
+#define MLX5_NUM_PORT_COUNTERS 1
+static char *mlx5_port_counter_names[MLX5_NUM_PORT_COUNTERS + 1] = {
+	"p1",
+	NULL
+};
+
+static int mlx5_get_protocol_stats(struct ib_device *ibdev,
+				   struct rdma_protocol_stats *stats,
+				   u8 port)
+{
+	if (port != 0) {
+		/* Port Counters */
+		stats->name = mlx5_port_counter_names;
+		stats->dirname = "stats";
+		/*
+		 * Just one value for testing. Insert the right
+		 * way to retrieve all the counters here
+		 */
+		stats->value[0] = 1;
+		return 0;
+	}
+
+
+	/* Device Counters */
+
+	/* No idea how to retrieve the actual counters. Just some values */
+	stats->value[0] = 1;
+	stats->value[1] = 2;
+	stats->value[2] = 3;
+	stats->value[3] = 4;
+
+	stats->name = mlx5_counter_names;
+	stats->dirname = "ib_stats";
+
+	return 0;
+}
+
 static int mlx5_enable_roce(struct mlx5_ib_dev *dev)
 {
 	int err;
@@ -2266,6 +2336,7 @@  static void *mlx5_ib_add(struct mlx5_cor
 	dev->ib_dev.map_mr_sg		= mlx5_ib_map_mr_sg;
 	dev->ib_dev.check_mr_status	= mlx5_ib_check_mr_status;
 	dev->ib_dev.get_port_immutable  = mlx5_port_immutable;
+	dev->ib_dev.get_protocol_stats	= mlx5_get_protocol_stats;
 
 	mlx5_ib_internal_fill_odp_caps(dev);