[PATCH rdma-next v5 0/4] Use RDMA adaptive moderation library

[PATCH rdma-next v5 0/4] Use RDMA adaptive moderation library

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Hi,

This is RDMA part of previously sent DIM library improvements series
[1], which was pulled by Dave. It needs to be pulled to RDMA too as
a pre-requirements.

Changes since v4:
 * Separated mlx5 change from IB/core changes.

Changes since v3:
 * Renamed dim_owner to be priv
 * Added Sagi's ROBs
 * Removed casting of void pointer.

Changes since v2:
- renamed user-space knob from dim to adaptive-moderation (Sagi)
- some minor code clean ups (Sagi)
- Reordered patches to ensure that netlink expose is last in the series.
- Slightly cleaned commit messages
- Changed "bool use_cq_dim" flag to be bitwise to save space.

Changes since v1:
- added per ib device configuration knob for rdma-dim (Sagi)
- add NL directives for user-space / rdma tool to configure rdma dim
  (Sagi/Leon)
- use one header file for DIM implementations (Leon)
- various point changes in the rdma dim related code in the IB core
  (Leon)
- removed the RDMA specific patches form this pull request\

Thanks

[1] https://www.spinics.net/lists/netdev/msg581046.html

Leon Romanovsky (1):
  RDMA/mlx5: Set RDMA DIM to be enabled by default

Yamin Friedman (3):
  linux/dim: Implement RDMA adaptive moderation (DIM)
  RDMA/core: Provide RDMA DIM support for ULPs
  RDMA/nldev: Added configuration of RDMA dynamic interrupt moderation
    to netlink

 drivers/infiniband/Kconfig          |   1 +
 drivers/infiniband/core/core_priv.h |   1 +
 drivers/infiniband/core/cq.c        |  45 ++++++++++++
 drivers/infiniband/core/device.c    |   9 +++
 drivers/infiniband/core/nldev.c     |  14 ++++
 drivers/infiniband/hw/mlx5/main.c   |   2 +
 include/linux/dim.h                 |  36 ++++++++++
 include/rdma/ib_verbs.h             |   4 ++
 include/uapi/rdma/rdma_netlink.h    |   5 ++
 lib/dim/Makefile                    |   6 +-
 lib/dim/rdma_dim.c                  | 108 ++++++++++++++++++++++++++++
 11 files changed, 227 insertions(+), 4 deletions(-)
 create mode 100644 lib/dim/rdma_dim.c

--
2.20.1

[PATCH rdma-next v5 1/4] linux/dim: Implement RDMA adaptive moderation (DIM)

[Leon Romanovsky]

From: Yamin Friedman <yaminf@mellanox.com>

RDMA DIM implements a different algorithm from net DIM and is based on
completions which is how we can implement interrupt moderation in RDMA.

The algorithm optimizes for number of completions and ratio between
completions and events. In order to avoid long latencies, the
implementation performs fast reduction of moderation level when the
traffic changes.

Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 include/linux/dim.h |  36 +++++++++++++++
 lib/dim/Makefile    |   6 +--
 lib/dim/rdma_dim.c  | 108 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 146 insertions(+), 4 deletions(-)
 create mode 100644 lib/dim/rdma_dim.c

diff --git a/include/linux/dim.h b/include/linux/dim.h
index aa9bdd47a648..aa69730c3b8d 100644
--- a/include/linux/dim.h
+++ b/include/linux/dim.h
@@ -82,6 +82,7 @@ struct dim_stats {
  * @prev_stats: Measured rates from previous iteration (for comparison)
  * @start_sample: Sampled data at start of current iteration
  * @work: Work to perform on action required
+ * @priv: A pointer to the struct that points to dim
  * @profile_ix: Current moderation profile
  * @mode: CQ period count mode
  * @tune_state: Algorithm tuning state (see below)
@@ -95,6 +96,7 @@ struct dim {
 	struct dim_sample start_sample;
 	struct dim_sample measuring_sample;
 	struct work_struct work;
+	void *priv;
 	u8 profile_ix;
 	u8 mode;
 	u8 tune_state;
@@ -363,4 +365,38 @@ struct dim_cq_moder net_dim_get_def_tx_moderation(u8 cq_period_mode);
  */
 void net_dim(struct dim *dim, struct dim_sample end_sample);
 
+/* RDMA DIM */
+
+/*
+ * RDMA DIM profile:
+ * profile size must be of RDMA_DIM_PARAMS_NUM_PROFILES.
+ */
+#define RDMA_DIM_PARAMS_NUM_PROFILES 9
+#define RDMA_DIM_START_PROFILE 0
+
+static const struct dim_cq_moder
+rdma_dim_prof[RDMA_DIM_PARAMS_NUM_PROFILES] = {
+	{1,   0, 1,  0},
+	{1,   0, 4,  0},
+	{2,   0, 4,  0},
+	{2,   0, 8,  0},
+	{4,   0, 8,  0},
+	{16,  0, 8,  0},
+	{16,  0, 16, 0},
+	{32,  0, 16, 0},
+	{32,  0, 32, 0},
+};
+
+/**
+ * rdma_dim - Runs the adaptive moderation.
+ * @dim: The moderation struct.
+ * @completions: The number of completions collected in this round.
+ *
+ * Each call to rdma_dim takes the latest amount of completions that
+ * have been collected and counts them as a new event.
+ * Once enough events have been collected the algorithm decides a new
+ * moderation level.
+ */
+void rdma_dim(struct dim *dim, u64 completions);
+
 #endif /* DIM_H */
diff --git a/lib/dim/Makefile b/lib/dim/Makefile
index 160afe288df0..1d6858a108cb 100644
--- a/lib/dim/Makefile
+++ b/lib/dim/Makefile
@@ -2,8 +2,6 @@
 # DIM Dynamic Interrupt Moderation library
 #
 
-obj-$(CONFIG_DIMLIB) = net_dim.o
+obj-$(CONFIG_DIMLIB) += dim.o
 
-net_dim-y = \
-	dim.o		\
-	net_dim.o
+dim-y := dim.o net_dim.o rdma_dim.o
diff --git a/lib/dim/rdma_dim.c b/lib/dim/rdma_dim.c
new file mode 100644
index 000000000000..f7e26c7b4749
--- /dev/null
+++ b/lib/dim/rdma_dim.c
@@ -0,0 +1,108 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2019, Mellanox Technologies inc.  All rights reserved.
+ */
+
+#include <linux/dim.h>
+
+static int rdma_dim_step(struct dim *dim)
+{
+	if (dim->tune_state == DIM_GOING_RIGHT) {
+		if (dim->profile_ix == (RDMA_DIM_PARAMS_NUM_PROFILES - 1))
+			return DIM_ON_EDGE;
+		dim->profile_ix++;
+		dim->steps_right++;
+	}
+	if (dim->tune_state == DIM_GOING_LEFT) {
+		if (dim->profile_ix == 0)
+			return DIM_ON_EDGE;
+		dim->profile_ix--;
+		dim->steps_left++;
+	}
+
+	return DIM_STEPPED;
+}
+
+static int rdma_dim_stats_compare(struct dim_stats *curr,
+				  struct dim_stats *prev)
+{
+	/* first stat */
+	if (!prev->cpms)
+		return DIM_STATS_SAME;
+
+	if (IS_SIGNIFICANT_DIFF(curr->cpms, prev->cpms))
+		return (curr->cpms > prev->cpms) ? DIM_STATS_BETTER :
+						DIM_STATS_WORSE;
+
+	if (IS_SIGNIFICANT_DIFF(curr->cpe_ratio, prev->cpe_ratio))
+		return (curr->cpe_ratio > prev->cpe_ratio) ? DIM_STATS_BETTER :
+						DIM_STATS_WORSE;
+
+	return DIM_STATS_SAME;
+}
+
+static bool rdma_dim_decision(struct dim_stats *curr_stats, struct dim *dim)
+{
+	int prev_ix = dim->profile_ix;
+	u8 state = dim->tune_state;
+	int stats_res;
+	int step_res;
+
+	if (state != DIM_PARKING_ON_TOP && state != DIM_PARKING_TIRED) {
+		stats_res = rdma_dim_stats_compare(curr_stats,
+						   &dim->prev_stats);
+
+		switch (stats_res) {
+		case DIM_STATS_SAME:
+			if (curr_stats->cpe_ratio <= 50 * prev_ix)
+				dim->profile_ix = 0;
+			break;
+		case DIM_STATS_WORSE:
+			dim_turn(dim);
+			/* fall through */
+		case DIM_STATS_BETTER:
+			step_res = rdma_dim_step(dim);
+			if (step_res == DIM_ON_EDGE)
+				dim_turn(dim);
+			break;
+		}
+	}
+
+	dim->prev_stats = *curr_stats;
+
+	return dim->profile_ix != prev_ix;
+}
+
+void rdma_dim(struct dim *dim, u64 completions)
+{
+	struct dim_sample *curr_sample = &dim->measuring_sample;
+	struct dim_stats curr_stats;
+	u32 nevents;
+
+	dim_update_sample_with_comps(curr_sample->event_ctr + 1, 0, 0,
+				     curr_sample->comp_ctr + completions,
+				     &dim->measuring_sample);
+
+	switch (dim->state) {
+	case DIM_MEASURE_IN_PROGRESS:
+		nevents = curr_sample->event_ctr - dim->start_sample.event_ctr;
+		if (nevents < DIM_NEVENTS)
+			break;
+		dim_calc_stats(&dim->start_sample, curr_sample, &curr_stats);
+		if (rdma_dim_decision(&curr_stats, dim)) {
+			dim->state = DIM_APPLY_NEW_PROFILE;
+			schedule_work(&dim->work);
+			break;
+		}
+		/* fall through */
+	case DIM_START_MEASURE:
+		dim->state = DIM_MEASURE_IN_PROGRESS;
+		dim_update_sample_with_comps(curr_sample->event_ctr, 0, 0,
+					     curr_sample->comp_ctr,
+					     &dim->start_sample);
+		break;
+	case DIM_APPLY_NEW_PROFILE:
+		break;
+	}
+}
+EXPORT_SYMBOL(rdma_dim);
-- 
2.20.1

[PATCH rdma-next v5 2/4] RDMA/core: Provide RDMA DIM support for ULPs

[Leon Romanovsky]

From: Yamin Friedman <yaminf@mellanox.com>

Added the interface in the infiniband driver that applies the rdma_dim
adaptive moderation. There is now a special function for allocating an
ib_cq that uses rdma_dim.

Performance improvement (ConnectX-5 100GbE, x86) running FIO benchmark over
NVMf between two equal end-hosts with 56 cores across a Mellanox switch
using null_blk device:

READS without DIM:
blk size | BW       | IOPS | 99th percentile latency  | 99.99th latency
512B     | 3.8GiB/s | 7.7M | 1401  usec               | 2442  usec
4k       | 7.0GiB/s | 1.8M | 4817  usec               | 6587  usec
64k      | 10.7GiB/s| 175k | 9896  usec               | 10028 usec

IO WRITES without DIM:
blk size | BW       | IOPS | 99th percentile latency  | 99.99th latency
512B     | 3.6GiB/s | 7.5M | 1434  usec               | 2474  usec
4k       | 6.3GiB/s | 1.6M | 938   usec               | 1221  usec
64k      | 10.7GiB/s| 175k | 8979  usec               | 12780 usec

IO READS with DIM:
blk size | BW       | IOPS | 99th percentile latency  | 99.99th latency
512B     | 4GiB/s   | 8.2M | 816    usec              | 889   usec
4k       | 10.1GiB/s| 2.65M| 3359   usec              | 5080  usec
64k      | 10.7GiB/s| 175k | 9896   usec              | 10028 usec

IO WRITES with DIM:
blk size | BW       | IOPS  | 99th percentile latency | 99.99th latency
512B     | 3.9GiB/s | 8.1M  | 799   usec              | 922   usec
4k       | 9.6GiB/s | 2.5M  | 717   usec              | 1004  usec
64k      | 10.7GiB/s| 176k  | 8586  usec              | 12256 usec

The rdma_dim algorithm was designed to measure the effectiveness of
moderation on the flow in a general way and thus should be appropriate
for all RDMA storage protocols.

rdma_dim is configured to be the default option based on performance
improvement seen after extensive tests.

Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/core/cq.c | 45 ++++++++++++++++++++++++++++++++++++
 include/rdma/ib_verbs.h      |  4 ++++
 2 files changed, 49 insertions(+)

diff --git a/drivers/infiniband/core/cq.c b/drivers/infiniband/core/cq.c
index 00d70f166209..ffd6e24109d5 100644
--- a/drivers/infiniband/core/cq.c
+++ b/drivers/infiniband/core/cq.c
@@ -18,6 +18,40 @@
 #define IB_POLL_FLAGS \
 	(IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS)
 
+static void ib_cq_rdma_dim_work(struct work_struct *w)
+{
+	struct dim *dim = container_of(w, struct dim, work);
+	struct ib_cq *cq = dim->priv;
+
+	u16 usec = rdma_dim_prof[dim->profile_ix].usec;
+	u16 comps = rdma_dim_prof[dim->profile_ix].comps;
+
+	dim->state = DIM_START_MEASURE;
+
+	cq->device->ops.modify_cq(cq, comps, usec);
+}
+
+static void rdma_dim_init(struct ib_cq *cq)
+{
+	struct dim *dim;
+
+	if (!cq->device->ops.modify_cq || !cq->device->use_cq_dim ||
+	    cq->poll_ctx == IB_POLL_DIRECT)
+		return;
+
+	dim = kzalloc(sizeof(struct dim), GFP_KERNEL);
+	if (!dim)
+		return;
+
+	dim->state = DIM_START_MEASURE;
+	dim->tune_state = DIM_GOING_RIGHT;
+	dim->profile_ix = RDMA_DIM_START_PROFILE;
+	dim->priv = cq;
+	cq->dim = dim;
+
+	INIT_WORK(&dim->work, ib_cq_rdma_dim_work);
+}
+
 static int __ib_process_cq(struct ib_cq *cq, int budget, struct ib_wc *wcs,
 			   int batch)
 {
@@ -78,6 +112,7 @@ static void ib_cq_completion_direct(struct ib_cq *cq, void *private)
 static int ib_poll_handler(struct irq_poll *iop, int budget)
 {
 	struct ib_cq *cq = container_of(iop, struct ib_cq, iop);
+	struct dim *dim = cq->dim;
 	int completed;
 
 	completed = __ib_process_cq(cq, budget, cq->wc, IB_POLL_BATCH);
@@ -87,6 +122,9 @@ static int ib_poll_handler(struct irq_poll *iop, int budget)
 			irq_poll_sched(&cq->iop);
 	}
 
+	if (dim)
+		rdma_dim(dim, completed);
+
 	return completed;
 }
 
@@ -105,6 +143,8 @@ static void ib_cq_poll_work(struct work_struct *work)
 	if (completed >= IB_POLL_BUDGET_WORKQUEUE ||
 	    ib_req_notify_cq(cq, IB_POLL_FLAGS) > 0)
 		queue_work(cq->comp_wq, &cq->work);
+	else if (cq->dim)
+		rdma_dim(cq->dim, completed);
 }
 
 static void ib_cq_completion_workqueue(struct ib_cq *cq, void *private)
@@ -161,6 +201,8 @@ struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
 
 	rdma_restrack_kadd(&cq->res);
 
+	rdma_dim_init(cq);
+
 	switch (cq->poll_ctx) {
 	case IB_POLL_DIRECT:
 		cq->comp_handler = ib_cq_completion_direct;
@@ -223,6 +265,9 @@ void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata)
 
 	rdma_restrack_del(&cq->res);
 	cq->device->ops.destroy_cq(cq, udata);
+	if (cq->dim)
+		cancel_work_sync(&cq->dim->work);
+	kfree(cq->dim);
 	kfree(cq->wc);
 	kfree(cq);
 }
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 50806bef9f20..30eb68f36109 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -61,6 +61,7 @@
 #include <linux/cgroup_rdma.h>
 #include <linux/irqflags.h>
 #include <linux/preempt.h>
+#include <linux/dim.h>
 #include <uapi/rdma/ib_user_verbs.h>
 #include <rdma/rdma_counter.h>
 #include <rdma/restrack.h>
@@ -1509,6 +1510,7 @@ struct ib_cq {
 		struct work_struct	work;
 	};
 	struct workqueue_struct *comp_wq;
+	struct dim *dim;
 	/*
 	 * Implementation details of the RDMA core, don't use in drivers:
 	 */
@@ -2576,6 +2578,8 @@ struct ib_device {
 	u16                          is_switch:1;
 	/* Indicates kernel verbs support, should not be used in drivers */
 	u16                          kverbs_provider:1;
+	/* CQ adaptive moderation (RDMA DIM) */
+	u16                          use_cq_dim:1;
 	u8                           node_type;
 	u8                           phys_port_cnt;
 	struct ib_device_attr        attrs;
-- 
2.20.1

[PATCH rdma-next v5 3/4] RDMA/nldev: Added configuration of RDMA dynamic interrupt moderation to netlink

[Leon Romanovsky]

From: Yamin Friedman <yaminf@mellanox.com>

Added parameter in ib_device for enabling dynamic interrupt moderation so
that it can be configured in userspace using rdma tool.

In order to set adaptive-moderation for an ib device the command is:
rdma dev set [DEV] adaptive-moderation [on|off]
Please set on/off.

rdma dev show
0: mlx5_0: node_type ca fw 16.26.0055 node_guid 248a:0703:00a5:29d0
sys_image_guid 248a:0703:00a5:29d0 adaptive-moderation on

rdma resource show cq
dev mlx5_0 cqn 0 cqe 1023 users 4 poll-ctx UNBOUND_WORKQUEUE
adaptive-moderation off comm [ib_core]

Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/Kconfig          |  1 +
 drivers/infiniband/core/core_priv.h |  1 +
 drivers/infiniband/core/device.c    |  9 +++++++++
 drivers/infiniband/core/nldev.c     | 14 ++++++++++++++
 include/uapi/rdma/rdma_netlink.h    |  5 +++++
 5 files changed, 30 insertions(+)

diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index f277cb7aea29..85e103b147cc 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -7,6 +7,7 @@ menuconfig INFINIBAND
 	depends on m || IPV6 != m
 	depends on !ALPHA
 	select IRQ_POLL
+	select DIMLIB
 	---help---
 	  Core support for InfiniBand (IB).  Make sure to also select
 	  any protocols you wish to use as well as drivers for your
diff --git a/drivers/infiniband/core/core_priv.h b/drivers/infiniband/core/core_priv.h
index a953c2fa2e78..888d89ce81df 100644
--- a/drivers/infiniband/core/core_priv.h
+++ b/drivers/infiniband/core/core_priv.h
@@ -60,6 +60,7 @@ extern bool ib_devices_shared_netns;
 int ib_device_register_sysfs(struct ib_device *device);
 void ib_device_unregister_sysfs(struct ib_device *device);
 int ib_device_rename(struct ib_device *ibdev, const char *name);
+int ib_device_set_dim(struct ib_device *ibdev, u8 use_dim);
 
 typedef void (*roce_netdev_callback)(struct ib_device *device, u8 port,
 	      struct net_device *idev, void *cookie);
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index bdf61499e6d5..7f4affe8a10d 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -448,6 +448,15 @@ int ib_device_rename(struct ib_device *ibdev, const char *name)
 	return 0;
 }
 
+int ib_device_set_dim(struct ib_device *ibdev, u8 use_dim)
+{
+	if (use_dim > 1)
+		return -EINVAL;
+	ibdev->use_cq_dim = use_dim;
+
+	return 0;
+}
+
 static int alloc_name(struct ib_device *ibdev, const char *name)
 {
 	struct ib_device *device;
diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c
index a4431ed566b6..d9f2a30e6467 100644
--- a/drivers/infiniband/core/nldev.c
+++ b/drivers/infiniband/core/nldev.c
@@ -52,6 +52,7 @@ static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
 	[RDMA_NLDEV_ATTR_CHARDEV_TYPE]		= { .type = NLA_NUL_STRING,
 					.len = RDMA_NLDEV_ATTR_CHARDEV_TYPE_SIZE },
+	[RDMA_NLDEV_ATTR_DEV_DIM]               = { .type = NLA_U8 },
 	[RDMA_NLDEV_ATTR_DEV_INDEX]		= { .type = NLA_U32 },
 	[RDMA_NLDEV_ATTR_DEV_NAME]		= { .type = NLA_NUL_STRING,
 					.len = IB_DEVICE_NAME_MAX },
@@ -252,6 +253,8 @@ static int fill_dev_info(struct sk_buff *msg, struct ib_device *device)
 		return -EMSGSIZE;
 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_NODE_TYPE, device->node_type))
 		return -EMSGSIZE;
+	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, device->use_cq_dim))
+		return -EMSGSIZE;
 
 	/*
 	 * Link type is determined on first port and mlx4 device
@@ -552,6 +555,9 @@ static int fill_res_cq_entry(struct sk_buff *msg, bool has_cap_net_admin,
 	    nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_POLL_CTX, cq->poll_ctx))
 		goto err;
 
+	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, (cq->dim != NULL)))
+		goto err;
+
 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN, res->id))
 		goto err;
 	if (!rdma_is_kernel_res(res) &&
@@ -870,6 +876,14 @@ static int nldev_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 		goto put_done;
 	}
 
+	if (tb[RDMA_NLDEV_ATTR_DEV_DIM]) {
+		u8 use_dim;
+
+		use_dim = nla_get_u8(tb[RDMA_NLDEV_ATTR_DEV_DIM]);
+		err = ib_device_set_dim(device,  use_dim);
+		goto done;
+	}
+
 done:
 	ib_device_put(device);
 put_done:
diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h
index ce6fd66e7aa3..8e277783fa96 100644
--- a/include/uapi/rdma/rdma_netlink.h
+++ b/include/uapi/rdma/rdma_netlink.h
@@ -520,6 +520,11 @@ enum rdma_nldev_attr {
 	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,	/* string */
 	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,	/* u64 */
 
+	/*
+	 * CQ adaptive moderatio (DIM)
+	 */
+	RDMA_NLDEV_ATTR_DEV_DIM,                /* u8 */
+
 	/*
 	 * Always the end
 	 */
-- 
2.20.1

[PATCH rdma-next v5 4/4] RDMA/mlx5: Set RDMA DIM to be enabled by default

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Enable RDMA DIM by default for better user experience.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/main.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 7581571bd9cd..07a05b0b9e42 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -6424,6 +6424,8 @@ static int mlx5_ib_stage_caps_init(struct mlx5_ib_dev *dev)
 	     MLX5_CAP_GEN(dev->mdev, disable_local_lb_mc)))
 		mutex_init(&dev->lb.mutex);
 
+	dev->ib_dev.use_cq_dim = true;
+
 	return 0;
 }
 
-- 
2.20.1

Re: [PATCH rdma-next v5 0/4] Use RDMA adaptive moderation library

[Jason Gunthorpe]

On Mon, Jul 08, 2019 at 01:59:01PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@mellanox.com>
> 
> Hi,
> 
> This is RDMA part of previously sent DIM library improvements series
> [1], which was pulled by Dave. It needs to be pulled to RDMA too as
> a pre-requirements.
> 
> Changes since v4:
>  * Separated mlx5 change from IB/core changes.
> 
> Changes since v3:
>  * Renamed dim_owner to be priv
>  * Added Sagi's ROBs
>  * Removed casting of void pointer.
> 
> Changes since v2:
> - renamed user-space knob from dim to adaptive-moderation (Sagi)
> - some minor code clean ups (Sagi)
> - Reordered patches to ensure that netlink expose is last in the series.
> - Slightly cleaned commit messages
> - Changed "bool use_cq_dim" flag to be bitwise to save space.
> 
> Changes since v1:
> - added per ib device configuration knob for rdma-dim (Sagi)
> - add NL directives for user-space / rdma tool to configure rdma dim
>   (Sagi/Leon)
> - use one header file for DIM implementations (Leon)
> - various point changes in the rdma dim related code in the IB core
>   (Leon)
> - removed the RDMA specific patches form this pull request\
> 
> Thanks
> 
> [1] https://www.spinics.net/lists/netdev/msg581046.html
> 
> Leon Romanovsky (1):
>   RDMA/mlx5: Set RDMA DIM to be enabled by default
> 
> Yamin Friedman (3):
>   linux/dim: Implement RDMA adaptive moderation (DIM)
>   RDMA/core: Provide RDMA DIM support for ULPs
>   RDMA/nldev: Added configuration of RDMA dynamic interrupt moderation
>     to netlink

Applied to for-next

Thanks,
Jason