Merge tag 'blk-dim-v2' into rdma.git for-next

Generic DIM

From: Tal Gilboa and Yamin Fridman

Implement net DIM over a generic DIM library, add RDMA DIM

dim.h lib exposes an implementation of the DIM algorithm for
dynamically-tuned interrupt moderation for networking interfaces.

We want a similar functionality for other protocols, which might need to
optimize interrupts differently. Main motivation here is DIM for NVMf
storage protocol.

Current DIM implementation prioritizes reducing interrupt overhead over
latency. Also, in order to reduce DIM's own overhead, the algorithm might
take some time to identify it needs to change profiles. While this is
acceptable for networking, it might not work well on other scenarios.

Here we propose a new structure to DIM. The idea is to allow a slightly
modified functionality without the risk of breaking Net DIM behavior for
netdev. We verified there are no degradations in current DIM behavior with
the modified solution.

Suggested solution:
- Common logic is implemented in lib/dim/dim.c
- Net DIM (existing) logic is implemented in lib/dim/net_dim.c, which uses
  the common logic in dim.c
- Any new DIM logic will be implemented in "lib/dim/new_dim.c".
  This new implementation will expose modified versions of profiles,
  dim_step() and dim_decision().
- DIM API is declared in include/linux/dim.h for all implementations.

Pros for this solution are:
- Zero impact on existing net_dim implementation and usage
- Relatively more code reuse (compared to two separate solutions)
- Increased extensibility

Required for dependencies in the next series.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

5 files changed, 291 insertions, 0 deletions

diff --git a/lib/Kconfig b/lib/Kconfig
index 90623a0e1942b..78ddb9526b62f 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -562,6 +562,14 @@ config SIGNATURE
 	  Digital signature verification. Currently only RSA is supported.
 	  Implementation is done using GnuPG MPI library
 
+config DIMLIB
+	bool "DIM library"
+	default y
+	help
+	  Dynamic Interrupt Moderation library.
+	  Implements an algorithm for dynamically change CQ modertion values
+	  according to run time performance.
+
 #
 # libfdt files, only selected if needed.
 #
diff --git a/lib/Makefile b/lib/Makefile
index fb7697031a797..dcb558c7554d9 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -202,6 +202,7 @@ obj-$(CONFIG_GLOB) += glob.o
 obj-$(CONFIG_GLOB_SELFTEST) += globtest.o
 
 obj-$(CONFIG_MPILIB) += mpi/
+obj-$(CONFIG_DIMLIB) += dim/
 obj-$(CONFIG_SIGNATURE) += digsig.o
 
 lib-$(CONFIG_CLZ_TAB) += clz_tab.o
diff --git a/lib/dim/Makefile b/lib/dim/Makefile
new file mode 100644
index 0000000000000..160afe288df06
--- /dev/null
+++ b/lib/dim/Makefile
@@ -0,0 +1,9 @@
+#
+# DIM Dynamic Interrupt Moderation library
+#
+
+obj-$(CONFIG_DIMLIB) = net_dim.o
+
+net_dim-y = \
+	dim.o		\
+	net_dim.o
diff --git a/lib/dim/dim.c b/lib/dim/dim.c
new file mode 100644
index 0000000000000..439d641ec7964
--- /dev/null
+++ b/lib/dim/dim.c
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2019, Mellanox Technologies inc.  All rights reserved.
+ */
+
+#include <linux/dim.h>
+
+bool dim_on_top(struct dim *dim)
+{
+	switch (dim->tune_state) {
+	case DIM_PARKING_ON_TOP:
+	case DIM_PARKING_TIRED:
+		return true;
+	case DIM_GOING_RIGHT:
+		return (dim->steps_left > 1) && (dim->steps_right == 1);
+	default: /* DIM_GOING_LEFT */
+		return (dim->steps_right > 1) && (dim->steps_left == 1);
+	}
+}
+EXPORT_SYMBOL(dim_on_top);
+
+void dim_turn(struct dim *dim)
+{
+	switch (dim->tune_state) {
+	case DIM_PARKING_ON_TOP:
+	case DIM_PARKING_TIRED:
+		break;
+	case DIM_GOING_RIGHT:
+		dim->tune_state = DIM_GOING_LEFT;
+		dim->steps_left = 0;
+		break;
+	case DIM_GOING_LEFT:
+		dim->tune_state = DIM_GOING_RIGHT;
+		dim->steps_right = 0;
+		break;
+	}
+}
+EXPORT_SYMBOL(dim_turn);
+
+void dim_park_on_top(struct dim *dim)
+{
+	dim->steps_right  = 0;
+	dim->steps_left   = 0;
+	dim->tired        = 0;
+	dim->tune_state   = DIM_PARKING_ON_TOP;
+}
+EXPORT_SYMBOL(dim_park_on_top);
+
+void dim_park_tired(struct dim *dim)
+{
+	dim->steps_right  = 0;
+	dim->steps_left   = 0;
+	dim->tune_state   = DIM_PARKING_TIRED;
+}
+EXPORT_SYMBOL(dim_park_tired);
+
+void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
+		    struct dim_stats *curr_stats)
+{
+	/* u32 holds up to 71 minutes, should be enough */
+	u32 delta_us = ktime_us_delta(end->time, start->time);
+	u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
+	u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
+			     start->byte_ctr);
+	u32 ncomps = BIT_GAP(BITS_PER_TYPE(u32), end->comp_ctr,
+			     start->comp_ctr);
+
+	if (!delta_us)
+		return;
+
+	curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
+	curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
+	curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
+					delta_us);
+	curr_stats->cpms = DIV_ROUND_UP(ncomps * USEC_PER_MSEC, delta_us);
+	if (curr_stats->epms != 0)
+		curr_stats->cpe_ratio =
+				(curr_stats->cpms * 100) / curr_stats->epms;
+	else
+		curr_stats->cpe_ratio = 0;
+
+}
+EXPORT_SYMBOL(dim_calc_stats);
diff --git a/lib/dim/net_dim.c b/lib/dim/net_dim.c
new file mode 100644
index 0000000000000..5bcc902c53888
--- /dev/null
+++ b/lib/dim/net_dim.c
@@ -0,0 +1,190 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/*
+ * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
+ */
+
+#include <linux/dim.h>
+
+struct dim_cq_moder
+net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
+{
+	struct dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
+
+	cq_moder.cq_period_mode = cq_period_mode;
+	return cq_moder;
+}
+EXPORT_SYMBOL(net_dim_get_rx_moderation);
+
+struct dim_cq_moder
+net_dim_get_def_rx_moderation(u8 cq_period_mode)
+{
+	u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+			NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
+
+	return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
+}
+EXPORT_SYMBOL(net_dim_get_def_rx_moderation);
+
+struct dim_cq_moder
+net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
+{
+	struct dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
+
+	cq_moder.cq_period_mode = cq_period_mode;
+	return cq_moder;
+}
+EXPORT_SYMBOL(net_dim_get_tx_moderation);
+
+struct dim_cq_moder
+net_dim_get_def_tx_moderation(u8 cq_period_mode)
+{
+	u8 profile_ix = cq_period_mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+			NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
+
+	return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
+}
+EXPORT_SYMBOL(net_dim_get_def_tx_moderation);
+
+static int net_dim_step(struct dim *dim)
+{
+	if (dim->tired == (NET_DIM_PARAMS_NUM_PROFILES * 2))
+		return DIM_TOO_TIRED;
+
+	switch (dim->tune_state) {
+	case DIM_PARKING_ON_TOP:
+	case DIM_PARKING_TIRED:
+		break;
+	case DIM_GOING_RIGHT:
+		if (dim->profile_ix == (NET_DIM_PARAMS_NUM_PROFILES - 1))
+			return DIM_ON_EDGE;
+		dim->profile_ix++;
+		dim->steps_right++;
+		break;
+	case DIM_GOING_LEFT:
+		if (dim->profile_ix == 0)
+			return DIM_ON_EDGE;
+		dim->profile_ix--;
+		dim->steps_left++;
+		break;
+	}
+
+	dim->tired++;
+	return DIM_STEPPED;
+}
+
+static void net_dim_exit_parking(struct dim *dim)
+{
+	dim->tune_state = dim->profile_ix ? DIM_GOING_LEFT : DIM_GOING_RIGHT;
+	net_dim_step(dim);
+}
+
+static int net_dim_stats_compare(struct dim_stats *curr,
+				 struct dim_stats *prev)
+{
+	if (!prev->bpms)
+		return curr->bpms ? DIM_STATS_BETTER : DIM_STATS_SAME;
+
+	if (IS_SIGNIFICANT_DIFF(curr->bpms, prev->bpms))
+		return (curr->bpms > prev->bpms) ? DIM_STATS_BETTER :
+						   DIM_STATS_WORSE;
+
+	if (!prev->ppms)
+		return curr->ppms ? DIM_STATS_BETTER :
+				    DIM_STATS_SAME;
+
+	if (IS_SIGNIFICANT_DIFF(curr->ppms, prev->ppms))
+		return (curr->ppms > prev->ppms) ? DIM_STATS_BETTER :
+						   DIM_STATS_WORSE;
+
+	if (!prev->epms)
+		return DIM_STATS_SAME;
+
+	if (IS_SIGNIFICANT_DIFF(curr->epms, prev->epms))
+		return (curr->epms < prev->epms) ? DIM_STATS_BETTER :
+						   DIM_STATS_WORSE;
+
+	return DIM_STATS_SAME;
+}
+
+static bool net_dim_decision(struct dim_stats *curr_stats, struct dim *dim)
+{
+	int prev_state = dim->tune_state;
+	int prev_ix = dim->profile_ix;
+	int stats_res;
+	int step_res;
+
+	switch (dim->tune_state) {
+	case DIM_PARKING_ON_TOP:
+		stats_res = net_dim_stats_compare(curr_stats,
+						  &dim->prev_stats);
+		if (stats_res != DIM_STATS_SAME)
+			net_dim_exit_parking(dim);
+		break;
+
+	case DIM_PARKING_TIRED:
+		dim->tired--;
+		if (!dim->tired)
+			net_dim_exit_parking(dim);
+		break;
+
+	case DIM_GOING_RIGHT:
+	case DIM_GOING_LEFT:
+		stats_res = net_dim_stats_compare(curr_stats,
+						  &dim->prev_stats);
+		if (stats_res != DIM_STATS_BETTER)
+			dim_turn(dim);
+
+		if (dim_on_top(dim)) {
+			dim_park_on_top(dim);
+			break;
+		}
+
+		step_res = net_dim_step(dim);
+		switch (step_res) {
+		case DIM_ON_EDGE:
+			dim_park_on_top(dim);
+			break;
+		case DIM_TOO_TIRED:
+			dim_park_tired(dim);
+			break;
+		}
+
+		break;
+	}
+
+	if (prev_state != DIM_PARKING_ON_TOP ||
+	    dim->tune_state != DIM_PARKING_ON_TOP)
+		dim->prev_stats = *curr_stats;
+
+	return dim->profile_ix != prev_ix;
+}
+
+void net_dim(struct dim *dim, struct dim_sample end_sample)
+{
+	struct dim_stats curr_stats;
+	u16 nevents;
+
+	switch (dim->state) {
+	case DIM_MEASURE_IN_PROGRESS:
+		nevents = BIT_GAP(BITS_PER_TYPE(u16),
+				  end_sample.event_ctr,
+				  dim->start_sample.event_ctr);
+		if (nevents < DIM_NEVENTS)
+			break;
+		dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats);
+		if (net_dim_decision(&curr_stats, dim)) {
+			dim->state = DIM_APPLY_NEW_PROFILE;
+			schedule_work(&dim->work);
+			break;
+		}
+		/* fall through */
+	case DIM_START_MEASURE:
+		dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
+				  end_sample.byte_ctr, &dim->start_sample);
+		dim->state = DIM_MEASURE_IN_PROGRESS;
+		break;
+	case DIM_APPLY_NEW_PROFILE:
+		break;
+	}
+}
+EXPORT_SYMBOL(net_dim);