This patch series adds DDR performance monitor support
on Marvell cn10k series of processor.
First patch adds device tree binding changes.
Second patch add basic support (without overflow and event
ownership). Third and fourth patch adds overflow and event
ownership respectively.
Seems like 4th patch can be merged in second patch,
For easy review it is currently separate
Bharat Bhushan (4):
dt-bindings: perf: marvell: cn10k ddr performance monitor
perf/marvell: ADD cn10k DDR PMU basic support
perf/marvell: cn10k DDR perfmon event overflow handling
perf/marvell: cn10k DDR perf event core ownership
.../bindings/perf/marvell-cn10k-ddr.txt | 15 +
drivers/perf/Kconfig | 7 +
drivers/perf/Makefile | 1 +
drivers/perf/marvell_cn10k_ddr_pmu.c | 766 ++++++++++++++++++
include/linux/cpuhotplug.h | 1 +
5 files changed, 790 insertions(+)
create mode 100644 Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
create mode 100644 drivers/perf/marvell_cn10k_ddr_pmu.c
--
2.17.1
Add binding documentation for the Marvell CN10k DDR
performance monitor unit.
Signed-off-by: Bharat Bhushan <[email protected]>
---
.../bindings/perf/marvell-cn10k-ddr.txt | 15 +++++++++++++++
1 file changed, 15 insertions(+)
create mode 100644 Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
diff --git a/Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt b/Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
new file mode 100644
index 000000000000..a2a593e99963
--- /dev/null
+++ b/Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
@@ -0,0 +1,15 @@
+* Marvell CN10k DDR performance monitor
+
+Required properties:
+
+- compatible: should be one of:
+ "marvell,cn10k-ddr-pmu"
+
+- reg: physical address and size
+
+Example:
+
+ ddrcpmu {
+ compatible = "marvell,cn10k-ddr-pmu";
+ reg = <0x87e1 0xc0000000 0x0 0x10000>;
+ };
--
2.17.1
Marvell CN10k DRAM Subsystem (DSS) supports eight
event counters for monitoring performance and software
can program each counter to monitor any of the defined
performance event. Performance events are for interface
between the DDR controller and the PHY, interface between
the DDR Controller and the CHI interconnect, or within
the DDR Controller. Additionally DSS also supports two
fixed performance event counters, one for number of
ddr reads and other for ddr writes.
This patch add basic support for these performance
monitoring events on CN10k.
Signed-off-by: Bharat Bhushan <[email protected]>
---
drivers/perf/Kconfig | 7 +
drivers/perf/Makefile | 1 +
drivers/perf/marvell_cn10k_ddr_pmu.c | 609 +++++++++++++++++++++++++++
3 files changed, 617 insertions(+)
create mode 100644 drivers/perf/marvell_cn10k_ddr_pmu.c
diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
index 77522e5efe11..41a80a4b8d29 100644
--- a/drivers/perf/Kconfig
+++ b/drivers/perf/Kconfig
@@ -139,4 +139,11 @@ config ARM_DMC620_PMU
source "drivers/perf/hisilicon/Kconfig"
+config MARVELL_CN10K_DDR_PMU
+ tristate "Enable MARVELL CN10K DRAM Subsystem(DSS) PMU Support"
+ depends on ARM64
+ help
+ Enable perf support for Marvell DDR Performance monitoring
+ event on CN10K platform.
+
endmenu
diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
index 5260b116c7da..ee1126219d8d 100644
--- a/drivers/perf/Makefile
+++ b/drivers/perf/Makefile
@@ -14,3 +14,4 @@ obj-$(CONFIG_THUNDERX2_PMU) += thunderx2_pmu.o
obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o
obj-$(CONFIG_ARM_SPE_PMU) += arm_spe_pmu.o
obj-$(CONFIG_ARM_DMC620_PMU) += arm_dmc620_pmu.o
+obj-$(CONFIG_MARVELL_CN10K_DDR_PMU) += marvell_cn10k_ddr_pmu.o
diff --git a/drivers/perf/marvell_cn10k_ddr_pmu.c b/drivers/perf/marvell_cn10k_ddr_pmu.c
new file mode 100644
index 000000000000..b27df356e55c
--- /dev/null
+++ b/drivers/perf/marvell_cn10k_ddr_pmu.c
@@ -0,0 +1,609 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Marvell CN10K DRAM Subsystem (DSS) Performance Monitor Driver
+ *
+ * Copyright (C) 2021 Marvell.
+ */
+
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/perf_event.h>
+
+/* Performance Counters Operating Mode Control Registers */
+#define DDRC_PERF_CNT_OP_MODE_CTRL 0x8020
+#define OP_MODE_CTRL_VAL_MANNUAL 0x1
+
+/* Performance Counters Start Operation Control Registers */
+#define DDRC_PERF_CNT_START_OP_CTRL 0x8028
+#define START_OP_CTRL_VAL_START 0x1ULL
+#define START_OP_CTRL_VAL_ACTIVE 0x2
+
+/* Performance Counters End Operation Control Registers */
+#define DDRC_PERF_CNT_END_OP_CTRL 0x8030
+#define END_OP_CTRL_VAL_END 0x1ULL
+
+/* Performance Counters End Status Registers */
+#define DDRC_PERF_CNT_END_STATUS 0x8038
+#define END_STATUS_VAL_END_TIMER_MODE_END 0x1
+
+/* Performance Counters Configuration Registers */
+#define DDRC_PERF_CFG_BASE 0x8040
+
+/* 8 Generic event counter + 2 fixed event counters */
+#define DDRC_PERF_NUM_GEN_COUNTERS 8
+#define DDRC_PERF_NUM_FIX_COUNTERS 2
+#define DDRC_PERF_READ_COUNTER_IDX DDRC_PERF_NUM_GEN_COUNTERS
+#define DDRC_PERF_WRITE_COUNTER_IDX (DDRC_PERF_NUM_GEN_COUNTERS + 1)
+#define DDRC_PERF_NUM_COUNTERS (DDRC_PERF_NUM_GEN_COUNTERS + \
+ DDRC_PERF_NUM_FIX_COUNTERS)
+
+/* Generic event counter registers */
+#define DDRC_PERF_CFG(n) (DDRC_PERF_CFG_BASE + 8 * (n))
+#define EVENT_ENABLE BIT_ULL(63)
+
+/* Two dedicated event counters for DDR reads and writes */
+#define EVENT_DDR_READS 101
+#define EVENT_DDR_WRITES 100
+
+/*
+ * programmable events IDs in programmable event counters.
+ * DO NOT change these event-id numbers, they are used to
+ * program event bitmap in h/w.
+ */
+#define EVENT_OP_IS_ZQLATCH 55
+#define EVENT_OP_IS_ZQSTART 54
+#define EVENT_OP_IS_TCR_MRR 53
+#define EVENT_OP_IS_DQSOSC_MRR 52
+#define EVENT_OP_IS_DQSOSC_MPC 51
+#define EVENT_VISIBLE_WIN_LIMIT_REACHED_WR 50
+#define EVENT_VISIBLE_WIN_LIMIT_REACHED_RD 49
+#define EVENT_BSM_STARVATION 48
+#define EVENT_BSM_ALLOC 47
+#define EVENT_LPR_REQ_WITH_NOCREDIT 46
+#define EVENT_HPR_REQ_WITH_NOCREDIT 45
+#define EVENT_OP_IS_ZQCS 44
+#define EVENT_OP_IS_ZQCL 43
+#define EVENT_OP_IS_LOAD_MODE 42
+#define EVENT_OP_IS_SPEC_REF 41
+#define EVENT_OP_IS_CRIT_REF 40
+#define EVENT_OP_IS_REFRESH 39
+#define EVENT_OP_IS_ENTER_MPSM 35
+#define EVENT_OP_IS_ENTER_POWERDOWN 31
+#define EVENT_OP_IS_ENTER_SELFREF 27
+#define EVENT_WAW_HAZARD 26
+#define EVENT_RAW_HAZARD 25
+#define EVENT_WAR_HAZARD 24
+#define EVENT_WRITE_COMBINE 23
+#define EVENT_RDWR_TRANSITIONS 22
+#define EVENT_PRECHARGE_FOR_OTHER 21
+#define EVENT_PRECHARGE_FOR_RDWR 20
+#define EVENT_OP_IS_PRECHARGE 19
+#define EVENT_OP_IS_MWR 18
+#define EVENT_OP_IS_WR 17
+#define EVENT_OP_IS_RD 16
+#define EVENT_OP_IS_RD_ACTIVATE 15
+#define EVENT_OP_IS_RD_OR_WR 14
+#define EVENT_OP_IS_ACTIVATE 13
+#define EVENT_WR_XACT_WHEN_CRITICAL 12
+#define EVENT_LPR_XACT_WHEN_CRITICAL 11
+#define EVENT_HPR_XACT_WHEN_CRITICAL 10
+#define EVENT_DFI_RD_DATA_CYCLES 9
+#define EVENT_DFI_WR_DATA_CYCLES 8
+#define EVENT_ACT_BYPASS 7
+#define EVENT_READ_BYPASS 6
+#define EVENT_HIF_HI_PRI_RD 5
+#define EVENT_HIF_RMW 4
+#define EVENT_HIF_RD 3
+#define EVENT_HIF_WR 2
+#define EVENT_HIF_RD_OR_WR 1
+
+/* Event counter value registers */
+#define DDRC_PERF_CNT_VALUE_BASE 0x8080
+#define DDRC_PERF_CNT_VALUE(n) (DDRC_PERF_CNT_VALUE_BASE + 8 * (n))
+
+/* Fixed event counter enable/disable register */
+#define DDRC_PERF_CNT_FREERUN_EN 0x80C0
+#define DDRC_PERF_FREERUN_WRITE_EN 0x1
+#define DDRC_PERF_FREERUN_READ_EN 0x2
+
+/* Fixed event counter control register */
+#define DDRC_PERF_CNT_FREERUN_CTRL 0x80C8
+#define DDRC_FREERUN_WRITE_CNT_CLR 0x1
+#define DDRC_FREERUN_READ_CNT_CLR 0x2
+
+/* Fixed event counter value register */
+#define DDRC_PERF_CNT_VALUE_WR_OP 0x80D0
+#define DDRC_PERF_CNT_VALUE_RD_OP 0x80D8
+#define DDRC_PERF_CNT_VALUE_OVERFLOW BIT_ULL(48)
+#define DDRC_PERF_CNT_MAX_VALUE GENMASK_ULL(48, 0)
+
+struct cn10k_ddr_pmu {
+ struct pmu pmu;
+ int id;
+ void __iomem *base;
+ unsigned int cpu;
+ struct device *dev;
+ int active_events;
+ struct perf_event *events[DDRC_PERF_NUM_COUNTERS];
+};
+
+#define to_cn10k_ddr_pmu(p) container_of(p, struct cn10k_ddr_pmu, pmu)
+
+static ssize_t cn10k_ddr_pmu_event_show(struct device *dev,
+ struct device_attribute *attr,
+ char *page)
+{
+ struct perf_pmu_events_attr *pmu_attr;
+
+ pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+ return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define CN10K_DDR_PMU_EVENT_ATTR(_name, _id) \
+ (&((struct perf_pmu_events_attr[]) { \
+ { .attr = __ATTR(_name, 0444, cn10k_ddr_pmu_event_show, NULL), \
+ .id = _id, } \
+ })[0].attr.attr)
+
+static struct attribute *cn10k_ddr_perf_events_attrs[] = {
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hif_rd_or_wr_access, EVENT_HIF_RD_OR_WR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hif_wr_access, EVENT_HIF_WR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hif_rd_access, EVENT_HIF_RD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hif_rmw_access, EVENT_HIF_RMW),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hif_pri_rdaccess, EVENT_HIF_HI_PRI_RD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_rd_bypass_access, EVENT_READ_BYPASS),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_act_bypass_access, EVENT_ACT_BYPASS),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_dif_wr_data_access, EVENT_DFI_WR_DATA_CYCLES),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_dif_rd_data_access, EVENT_DFI_RD_DATA_CYCLES),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hpri_sched_rd_crit_access,
+ EVENT_HPR_XACT_WHEN_CRITICAL),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_lpri_sched_rd_crit_access,
+ EVENT_LPR_XACT_WHEN_CRITICAL),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_wr_trxn_crit_access,
+ EVENT_WR_XACT_WHEN_CRITICAL),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_active_access, EVENT_OP_IS_ACTIVATE),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_rd_or_wr_access, EVENT_OP_IS_RD_OR_WR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_rd_active_access, EVENT_OP_IS_RD_ACTIVATE),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_read, EVENT_OP_IS_RD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_write, EVENT_OP_IS_WR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_mwr, EVENT_OP_IS_MWR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_precharge, EVENT_OP_IS_PRECHARGE),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_precharge_for_rdwr, EVENT_PRECHARGE_FOR_RDWR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_precharge_for_other,
+ EVENT_PRECHARGE_FOR_OTHER),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_rdwr_transitions, EVENT_RDWR_TRANSITIONS),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_write_combine, EVENT_WRITE_COMBINE),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_war_hazard, EVENT_WAR_HAZARD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_raw_hazard, EVENT_RAW_HAZARD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_waw_hazard, EVENT_WAW_HAZARD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_enter_selfref, EVENT_OP_IS_ENTER_SELFREF),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_enter_powerdown, EVENT_OP_IS_ENTER_POWERDOWN),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_enter_mpsm, EVENT_OP_IS_ENTER_MPSM),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_refresh, EVENT_OP_IS_REFRESH),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_crit_ref, EVENT_OP_IS_CRIT_REF),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_spec_ref, EVENT_OP_IS_SPEC_REF),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_load_mode, EVENT_OP_IS_LOAD_MODE),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_zqcl, EVENT_OP_IS_ZQCL),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_cam_wr_access, EVENT_OP_IS_ZQCS),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_hpr_req_with_nocredit,
+ EVENT_HPR_REQ_WITH_NOCREDIT),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_lpr_req_with_nocredit,
+ EVENT_LPR_REQ_WITH_NOCREDIT),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_bsm_alloc, EVENT_BSM_ALLOC),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_bsm_starvation, EVENT_BSM_STARVATION),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_win_limit_reached_rd,
+ EVENT_VISIBLE_WIN_LIMIT_REACHED_RD),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_win_limit_reached_wr,
+ EVENT_VISIBLE_WIN_LIMIT_REACHED_WR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_dqsosc_mpc, EVENT_OP_IS_DQSOSC_MPC),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_dqsosc_mrr, EVENT_OP_IS_DQSOSC_MRR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_tcr_mrr, EVENT_OP_IS_TCR_MRR),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_zqstart, EVENT_OP_IS_ZQSTART),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_zqlatch, EVENT_OP_IS_ZQLATCH),
+ /* Free run event counters */
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_ddr_reads, EVENT_DDR_READS),
+ CN10K_DDR_PMU_EVENT_ATTR(ddr_ddr_writes, EVENT_DDR_WRITES),
+ NULL,
+};
+
+static struct attribute_group cn10k_ddr_perf_events_attr_group = {
+ .name = "events",
+ .attrs = cn10k_ddr_perf_events_attrs,
+};
+
+PMU_FORMAT_ATTR(event, "config:0-8");
+
+static struct attribute *cn10k_ddr_perf_format_attrs[] = {
+ &format_attr_event.attr,
+ NULL,
+};
+
+static struct attribute_group cn10k_ddr_perf_format_attr_group = {
+ .name = "format",
+ .attrs = cn10k_ddr_perf_format_attrs,
+};
+
+static ssize_t cn10k_ddr_perf_cpumask_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ struct cn10k_ddr_pmu *pmu = dev_get_drvdata(dev);
+
+ return cpumap_print_to_pagebuf(true, buf, cpumask_of(pmu->cpu));
+}
+
+static struct device_attribute cn10k_ddr_perf_cpumask_attr =
+ __ATTR(cpumask, 0444, cn10k_ddr_perf_cpumask_show, NULL);
+
+static struct attribute *cn10k_ddr_perf_cpumask_attrs[] = {
+ &cn10k_ddr_perf_cpumask_attr.attr,
+ NULL,
+};
+
+static struct attribute_group cn10k_ddr_perf_cpumask_attr_group = {
+ .attrs = cn10k_ddr_perf_cpumask_attrs,
+};
+
+static const struct attribute_group *cn10k_attr_groups[] = {
+ &cn10k_ddr_perf_events_attr_group,
+ &cn10k_ddr_perf_format_attr_group,
+ &cn10k_ddr_perf_cpumask_attr_group,
+ NULL,
+};
+
+static uint64_t ddr_perf_get_event_bitmap(int eventid)
+{
+ uint64_t event_bitmap = 0;
+
+ switch (eventid) {
+ case EVENT_HIF_RD_OR_WR ... EVENT_WAW_HAZARD:
+ case EVENT_OP_IS_REFRESH ... EVENT_OP_IS_ZQLATCH:
+ event_bitmap = (1ULL << (eventid - 1));
+ break;
+
+ case EVENT_OP_IS_ENTER_SELFREF:
+ case EVENT_OP_IS_ENTER_POWERDOWN:
+ case EVENT_OP_IS_ENTER_MPSM:
+ event_bitmap = (0xFULL << (eventid - 1));
+ break;
+ default:
+ pr_err("%s Invalid eventid %d\n", __func__, eventid);
+ break;
+ }
+
+ return event_bitmap;
+}
+
+static int cn10k_ddr_perf_alloc_counter(struct cn10k_ddr_pmu *pmu,
+ struct perf_event *event)
+{
+ uint8_t config = event->attr.config;
+ int i;
+
+ /* DDR read free-run counter index */
+ if (config == EVENT_DDR_READS) {
+ pmu->events[DDRC_PERF_READ_COUNTER_IDX] = event;
+ return DDRC_PERF_READ_COUNTER_IDX;
+ }
+
+ /* DDR write free-run counter index */
+ if (config == EVENT_DDR_WRITES) {
+ pmu->events[DDRC_PERF_WRITE_COUNTER_IDX] = event;
+ return DDRC_PERF_WRITE_COUNTER_IDX;
+ }
+
+ /* Allocate DDR generic counters */
+ for (i = 0; i < DDRC_PERF_NUM_GEN_COUNTERS; i++) {
+ if (pmu->events[i] == NULL) {
+ pmu->events[i] = event;
+ return i;
+ }
+ }
+
+ return -ENOENT;
+}
+
+static void cn10k_ddr_perf_free_counter(struct cn10k_ddr_pmu *pmu, int counter)
+{
+ pmu->events[counter] = NULL;
+}
+
+static int cn10k_ddr_perf_event_init(struct perf_event *event)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+
+ if (event->attr.type != event->pmu->type)
+ return -ENOENT;
+
+ if (is_sampling_event(event)) {
+ dev_info(pmu->dev, "Sampling not supported!\n");
+ return -EOPNOTSUPP;
+ }
+
+ if (event->cpu < 0) {
+ dev_warn(pmu->dev, "Can't provide per-task data!\n");
+ return -EOPNOTSUPP;
+ }
+
+ /* We must NOT create groups containing mixed PMUs */
+ if (event->group_leader->pmu != event->pmu &&
+ !is_software_event(event->group_leader))
+ return -EINVAL;
+
+ /* Set ownership of event to one CPU, same event can not be observed
+ * on multiple cpus at same time.
+ */
+ event->cpu = pmu->cpu;
+ hwc->idx = -1;
+ return 0;
+}
+
+static void cn10k_ddr_perf_counter_enable(struct cn10k_ddr_pmu *pmu,
+ int counter, bool enable)
+{
+ uint32_t reg;
+ uint64_t val;
+
+ if (counter > DDRC_PERF_NUM_COUNTERS) {
+ pr_err("Error: unsupported counter %d\n", counter);
+ return;
+ }
+
+ if (counter < DDRC_PERF_NUM_GEN_COUNTERS) {
+ reg = DDRC_PERF_CFG(counter);
+ val = readq(pmu->base + reg);
+
+ if (enable)
+ val |= EVENT_ENABLE;
+ else
+ val &= ~EVENT_ENABLE;
+
+ writeq(val, pmu->base + reg);
+ } else {
+ val = readq(pmu->base + DDRC_PERF_CNT_FREERUN_EN);
+ if (enable) {
+ if (counter == DDRC_PERF_READ_COUNTER_IDX)
+ val |= DDRC_PERF_FREERUN_READ_EN;
+ else
+ val |= DDRC_PERF_FREERUN_WRITE_EN;
+ } else {
+ if (counter == DDRC_PERF_READ_COUNTER_IDX)
+ val &= ~DDRC_PERF_FREERUN_READ_EN;
+ else
+ val &= ~DDRC_PERF_FREERUN_WRITE_EN;
+ }
+ writeq(val, pmu->base + DDRC_PERF_CNT_FREERUN_EN);
+ }
+}
+
+static uint64_t cn10k_ddr_perf_read_counter(struct cn10k_ddr_pmu *pmu,
+ int counter)
+{
+ uint64_t val;
+
+ if (counter == DDRC_PERF_READ_COUNTER_IDX)
+ return readq(pmu->base + DDRC_PERF_CNT_VALUE_RD_OP);
+
+ if (counter == DDRC_PERF_WRITE_COUNTER_IDX)
+ return readq(pmu->base + DDRC_PERF_CNT_VALUE_WR_OP);
+
+ val = readq(pmu->base + DDRC_PERF_CNT_VALUE(counter));
+ return val;
+}
+
+static void cn10k_ddr_perf_event_update(struct perf_event *event)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ uint64_t prev_count, new_count, mask;
+
+ do {
+ prev_count = local64_read(&hwc->prev_count);
+ new_count = cn10k_ddr_perf_read_counter(pmu, hwc->idx);
+ } while (local64_xchg(&hwc->prev_count, new_count) != prev_count);
+
+ mask = DDRC_PERF_CNT_MAX_VALUE;
+
+ local64_add((new_count - prev_count) & mask, &event->count);
+}
+
+static void cn10k_ddr_perf_event_start(struct perf_event *event, int flags)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ int counter = hwc->idx;
+
+ local64_set(&hwc->prev_count, 0);
+
+ cn10k_ddr_perf_counter_enable(pmu, counter, true);
+
+ hwc->state = 0;
+}
+
+static int cn10k_ddr_perf_event_add(struct perf_event *event, int flags)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ uint8_t config = event->attr.config;
+ uint32_t reg_offset;
+ uint64_t val;
+ int counter;
+
+ counter = cn10k_ddr_perf_alloc_counter(pmu, event);
+ if (counter < 0) {
+ dev_dbg(pmu->dev, "There are not enough counters\n");
+ return -EOPNOTSUPP;
+ }
+
+ pmu->active_events++;
+ hwc->idx = counter;
+
+ if (counter < DDRC_PERF_NUM_GEN_COUNTERS) {
+ /* Generic counters, configure event id */
+ reg_offset = DDRC_PERF_CFG(counter);
+ val = ddr_perf_get_event_bitmap(config);
+ writeq(val, pmu->base + reg_offset);
+ } else {
+ /* fixed event counter, clear counter value */
+ if (counter == DDRC_PERF_READ_COUNTER_IDX)
+ val = DDRC_FREERUN_READ_CNT_CLR;
+ else
+ val = DDRC_FREERUN_WRITE_CNT_CLR;
+
+ writeq(val, pmu->base + DDRC_PERF_CNT_FREERUN_CTRL);
+ }
+
+ hwc->state |= PERF_HES_STOPPED;
+
+ if (flags & PERF_EF_START)
+ cn10k_ddr_perf_event_start(event, flags);
+
+ return 0;
+}
+
+static void cn10k_ddr_perf_event_stop(struct perf_event *event, int flags)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ int counter = hwc->idx;
+
+ cn10k_ddr_perf_counter_enable(pmu, counter, false);
+
+ if (flags & PERF_EF_UPDATE)
+ cn10k_ddr_perf_event_update(event);
+
+ hwc->state |= PERF_HES_STOPPED;
+}
+
+static void cn10k_ddr_perf_event_del(struct perf_event *event, int flags)
+{
+ struct cn10k_ddr_pmu *pmu = to_cn10k_ddr_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ int counter = hwc->idx;
+
+ cn10k_ddr_perf_event_stop(event, PERF_EF_UPDATE);
+
+ cn10k_ddr_perf_free_counter(pmu, counter);
+ pmu->active_events--;
+ hwc->idx = -1;
+}
+
+static void cn10k_ddr_perf_pmu_enable(struct pmu *pmu)
+{
+ struct cn10k_ddr_pmu *ddr_pmu = to_cn10k_ddr_pmu(pmu);
+
+ writeq(START_OP_CTRL_VAL_START, ddr_pmu->base +
+ DDRC_PERF_CNT_START_OP_CTRL);
+}
+
+static void cn10k_ddr_perf_pmu_disable(struct pmu *pmu)
+{
+ struct cn10k_ddr_pmu *ddr_pmu = to_cn10k_ddr_pmu(pmu);
+
+ writeq(END_OP_CTRL_VAL_END, ddr_pmu->base +
+ DDRC_PERF_CNT_END_OP_CTRL);
+}
+
+static int cn10k_ddr_perf_probe(struct platform_device *pdev)
+{
+ struct cn10k_ddr_pmu *ddr_pmu;
+ struct resource *res;
+ void __iomem *base;
+ static int index;
+ char *name;
+ int ret;
+
+ ddr_pmu = devm_kzalloc(&pdev->dev, sizeof(*ddr_pmu), GFP_KERNEL);
+ if (!ddr_pmu)
+ return -ENOMEM;
+
+ ddr_pmu->dev = &pdev->dev;
+ platform_set_drvdata(pdev, ddr_pmu);
+
+ res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ base = devm_ioremap_resource(&pdev->dev, res);
+ if (IS_ERR(base))
+ return PTR_ERR(base);
+
+ ddr_pmu->base = base;
+
+ /* Setup the PMU counter to work in mannual mode */
+ writeq(OP_MODE_CTRL_VAL_MANNUAL, ddr_pmu->base +
+ DDRC_PERF_CNT_OP_MODE_CTRL);
+
+ ddr_pmu->pmu = (struct pmu) {
+ .module = THIS_MODULE,
+ .capabilities = PERF_PMU_CAP_NO_EXCLUDE,
+ .task_ctx_nr = perf_invalid_context,
+ .attr_groups = cn10k_attr_groups,
+ .event_init = cn10k_ddr_perf_event_init,
+ .add = cn10k_ddr_perf_event_add,
+ .del = cn10k_ddr_perf_event_del,
+ .start = cn10k_ddr_perf_event_start,
+ .stop = cn10k_ddr_perf_event_stop,
+ .read = cn10k_ddr_perf_event_update,
+ .pmu_enable = cn10k_ddr_perf_pmu_enable,
+ .pmu_disable = cn10k_ddr_perf_pmu_disable,
+ };
+
+ /* Choose this cpu to collect perf data */
+ ddr_pmu->cpu = raw_smp_processor_id();
+
+ name = devm_kasprintf(ddr_pmu->dev, GFP_KERNEL, "mrvl_ddr_pmu@%llx",
+ res->start);
+ if (!name)
+ return -ENOMEM;
+
+ ret = perf_pmu_register(&ddr_pmu->pmu, name, -1);
+ if (ret)
+ return ret;
+
+ ddr_pmu->id = index++;
+ pr_info("CN10K DDR PMU Driver for ddrc@%llx - id-%d\n",
+ res->start, ddr_pmu->id);
+ return 0;
+}
+
+static int cn10k_ddr_perf_remove(struct platform_device *pdev)
+{
+ struct cn10k_ddr_pmu *ddr_pmu = platform_get_drvdata(pdev);
+
+ perf_pmu_unregister(&ddr_pmu->pmu);
+ return 0;
+}
+
+static const struct of_device_id cn10k_ddr_pmu_of_match[] = {
+ { .compatible = "marvell,cn10k-ddr-pmu", },
+ { },
+};
+MODULE_DEVICE_TABLE(of, cn10k_ddr_pmu_of_match);
+
+static struct platform_driver cn10k_ddr_pmu_driver = {
+ .driver = {
+ .name = "cn10k-ddr-pmu",
+ .of_match_table = cn10k_ddr_pmu_of_match,
+ .suppress_bind_attrs = true,
+ },
+ .probe = cn10k_ddr_perf_probe,
+ .remove = cn10k_ddr_perf_remove,
+};
+
+static int __init cn10k_ddr_pmu_init(void)
+{
+ return platform_driver_register(&cn10k_ddr_pmu_driver);
+}
+
+static void __exit cn10k_ddr_pmu_exit(void)
+{
+ platform_driver_unregister(&cn10k_ddr_pmu_driver);
+}
+
+module_init(cn10k_ddr_pmu_init);
+module_exit(cn10k_ddr_pmu_exit);
+
+MODULE_AUTHOR("Bharat Bhushan <[email protected]>");
+MODULE_LICENSE("GPL v2");
--
2.17.1
CN10k DSS h/w perfmon does not support event overflow
interrupt, so periodic timer is being used. Each
event counter is 48bit, which in worst case scenario
can increment at maximum 5.6 GT/s. At this rate it may
take many hours to overflow these counters. Therefore
polling period for overflow is set to 100 sec, which
can be changed using sysfs parameter.
Two fixed event counters starts counting from zero on
overflow, so overflow condition is when new count less
than previous count. While eight programmable event
counters freezes at maximum value. Also individual
counter cannot be restarted, so need to restart all
eight counters.
Signed-off-by: Bharat Bhushan <[email protected]>
---
drivers/perf/marvell_cn10k_ddr_pmu.c | 111 +++++++++++++++++++++++++++
1 file changed, 111 insertions(+)
diff --git a/drivers/perf/marvell_cn10k_ddr_pmu.c b/drivers/perf/marvell_cn10k_ddr_pmu.c
index b27df356e55c..5e88fb7ffcb8 100644
--- a/drivers/perf/marvell_cn10k_ddr_pmu.c
+++ b/drivers/perf/marvell_cn10k_ddr_pmu.c
@@ -11,6 +11,7 @@
#include <linux/of_address.h>
#include <linux/of_device.h>
#include <linux/perf_event.h>
+#include <linux/hrtimer.h>
/* Performance Counters Operating Mode Control Registers */
#define DDRC_PERF_CNT_OP_MODE_CTRL 0x8020
@@ -128,6 +129,7 @@ struct cn10k_ddr_pmu {
struct device *dev;
int active_events;
struct perf_event *events[DDRC_PERF_NUM_COUNTERS];
+ struct hrtimer hrtimer;
};
#define to_cn10k_ddr_pmu(p) container_of(p, struct cn10k_ddr_pmu, pmu)
@@ -254,6 +256,18 @@ static const struct attribute_group *cn10k_attr_groups[] = {
NULL,
};
+/* Default poll timeout is 100 sec, which is very sufficient for
+ * 48 bit counter incremented max at 5.6 GT/s, which may take many
+ * hours to overflow.
+ */
+static unsigned long cn10k_ddr_pmu_poll_period_sec = 100;
+module_param_named(poll_period_sec, cn10k_ddr_pmu_poll_period_sec, ulong, 0644);
+
+static ktime_t cn10k_ddr_pmu_timer_period(void)
+{
+ return ms_to_ktime((u64)cn10k_ddr_pmu_poll_period_sec * USEC_PER_SEC);
+}
+
static uint64_t ddr_perf_get_event_bitmap(int eventid)
{
uint64_t event_bitmap = 0;
@@ -442,6 +456,10 @@ static int cn10k_ddr_perf_event_add(struct perf_event *event, int flags)
pmu->active_events++;
hwc->idx = counter;
+ if (pmu->active_events == 1)
+ hrtimer_start(&pmu->hrtimer, cn10k_ddr_pmu_timer_period(),
+ HRTIMER_MODE_REL_PINNED);
+
if (counter < DDRC_PERF_NUM_GEN_COUNTERS) {
/* Generic counters, configure event id */
reg_offset = DDRC_PERF_CFG(counter);
@@ -490,6 +508,10 @@ static void cn10k_ddr_perf_event_del(struct perf_event *event, int flags)
cn10k_ddr_perf_free_counter(pmu, counter);
pmu->active_events--;
hwc->idx = -1;
+
+ /* Cancel timer when no events to capture */
+ if (pmu->active_events == 0)
+ hrtimer_cancel(&pmu->hrtimer);
}
static void cn10k_ddr_perf_pmu_enable(struct pmu *pmu)
@@ -508,6 +530,92 @@ static void cn10k_ddr_perf_pmu_disable(struct pmu *pmu)
DDRC_PERF_CNT_END_OP_CTRL);
}
+static void cn10k_ddr_perf_event_update_all(struct cn10k_ddr_pmu *pmu)
+{
+ struct hw_perf_event *hwc;
+ int i;
+
+ for (i = 0; i < DDRC_PERF_NUM_GEN_COUNTERS; i++) {
+ if (pmu->events[i] == NULL)
+ continue;
+
+ cn10k_ddr_perf_event_update(pmu->events[i]);
+ }
+
+ /* Reset previous count as h/w counter are reset */
+ for (i = 0; i < DDRC_PERF_NUM_GEN_COUNTERS; i++) {
+ if (pmu->events[i] == NULL)
+ continue;
+
+ hwc = &pmu->events[i]->hw;
+ local64_set(&hwc->prev_count, 0);
+ }
+}
+
+static irqreturn_t cn10k_ddr_pmu_overflow_handler(struct cn10k_ddr_pmu *pmu)
+{
+ struct perf_event *event;
+ struct hw_perf_event *hwc;
+ uint64_t prev_count, new_count;
+ uint64_t value;
+ int i;
+
+ event = pmu->events[DDRC_PERF_READ_COUNTER_IDX];
+ if (event) {
+ hwc = &event->hw;
+ prev_count = local64_read(&hwc->prev_count);
+ new_count = cn10k_ddr_perf_read_counter(pmu, hwc->idx);
+
+ /* Overflow condition is when new count less than
+ * previous count
+ */
+ if (new_count < prev_count)
+ cn10k_ddr_perf_event_update(event);
+ }
+
+ event = pmu->events[DDRC_PERF_WRITE_COUNTER_IDX];
+ if (event) {
+ hwc = &event->hw;
+ prev_count = local64_read(&hwc->prev_count);
+ new_count = cn10k_ddr_perf_read_counter(pmu, hwc->idx);
+
+ /* Overflow condition is when new count less than
+ * previous count
+ */
+ if (new_count < prev_count)
+ cn10k_ddr_perf_event_update(event);
+ }
+
+ for (i = 0; i < DDRC_PERF_NUM_GEN_COUNTERS; i++) {
+ if (pmu->events[i] == NULL)
+ continue;
+
+ value = cn10k_ddr_perf_read_counter(pmu, i);
+ if (value == DDRC_PERF_CNT_MAX_VALUE) {
+ pr_info("Counter-(%d) reached max value\n", i);
+ cn10k_ddr_perf_event_update_all(pmu);
+ cn10k_ddr_perf_pmu_disable(&pmu->pmu);
+ cn10k_ddr_perf_pmu_enable(&pmu->pmu);
+ }
+ }
+
+ return IRQ_HANDLED;
+}
+
+static enum hrtimer_restart cn10k_ddr_pmu_timer_handler(struct hrtimer *hrtimer)
+{
+ struct cn10k_ddr_pmu *pmu = container_of(hrtimer, struct cn10k_ddr_pmu,
+ hrtimer);
+ unsigned long flags;
+
+ local_irq_save(flags);
+ cn10k_ddr_pmu_overflow_handler(pmu);
+ local_irq_restore(flags);
+
+ hrtimer_forward_now(hrtimer, cn10k_ddr_pmu_timer_period());
+ return HRTIMER_RESTART;
+}
+
static int cn10k_ddr_perf_probe(struct platform_device *pdev)
{
struct cn10k_ddr_pmu *ddr_pmu;
@@ -558,6 +666,9 @@ static int cn10k_ddr_perf_probe(struct platform_device *pdev)
if (!name)
return -ENOMEM;
+ hrtimer_init(&ddr_pmu->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+ ddr_pmu->hrtimer.function = cn10k_ddr_pmu_timer_handler;
+
ret = perf_pmu_register(&ddr_pmu->pmu, name, -1);
if (ret)
return ret;
--
2.17.1
As DDR perf event counters are not per core, so they
should be accessed only by one core at a time.
Select new core when previously owning core is going
offline.
Signed-off-by: Bharat Bhushan <[email protected]>
---
drivers/perf/marvell_cn10k_ddr_pmu.c | 50 ++++++++++++++++++++++++++--
include/linux/cpuhotplug.h | 1 +
2 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/drivers/perf/marvell_cn10k_ddr_pmu.c b/drivers/perf/marvell_cn10k_ddr_pmu.c
index 5e88fb7ffcb8..0b20c20c7f2f 100644
--- a/drivers/perf/marvell_cn10k_ddr_pmu.c
+++ b/drivers/perf/marvell_cn10k_ddr_pmu.c
@@ -130,6 +130,7 @@ struct cn10k_ddr_pmu {
int active_events;
struct perf_event *events[DDRC_PERF_NUM_COUNTERS];
struct hrtimer hrtimer;
+ struct hlist_node node;
};
#define to_cn10k_ddr_pmu(p) container_of(p, struct cn10k_ddr_pmu, pmu)
@@ -616,6 +617,24 @@ static enum hrtimer_restart cn10k_ddr_pmu_timer_handler(struct hrtimer *hrtimer)
return HRTIMER_RESTART;
}
+static int cn10k_ddr_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+ struct cn10k_ddr_pmu *pmu = hlist_entry_safe(node, struct cn10k_ddr_pmu,
+ node);
+ unsigned int target;
+
+ if (cpu != pmu->cpu)
+ return 0;
+
+ target = cpumask_any_but(cpu_online_mask, cpu);
+ if (target >= nr_cpu_ids)
+ return 0;
+
+ perf_pmu_migrate_context(&pmu->pmu, cpu, target);
+ pmu->cpu = target;
+ return 0;
+}
+
static int cn10k_ddr_perf_probe(struct platform_device *pdev)
{
struct cn10k_ddr_pmu *ddr_pmu;
@@ -669,20 +688,33 @@ static int cn10k_ddr_perf_probe(struct platform_device *pdev)
hrtimer_init(&ddr_pmu->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
ddr_pmu->hrtimer.function = cn10k_ddr_pmu_timer_handler;
+ cpuhp_state_add_instance_nocalls(
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE,
+ &ddr_pmu->node);
+
ret = perf_pmu_register(&ddr_pmu->pmu, name, -1);
if (ret)
- return ret;
+ goto error;
ddr_pmu->id = index++;
pr_info("CN10K DDR PMU Driver for ddrc@%llx - id-%d\n",
res->start, ddr_pmu->id);
return 0;
+error:
+ cpuhp_state_remove_instance_nocalls(
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE,
+ &ddr_pmu->node);
+ return ret;
}
static int cn10k_ddr_perf_remove(struct platform_device *pdev)
{
struct cn10k_ddr_pmu *ddr_pmu = platform_get_drvdata(pdev);
+ cpuhp_state_remove_instance_nocalls(
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE,
+ &ddr_pmu->node);
+
perf_pmu_unregister(&ddr_pmu->pmu);
return 0;
}
@@ -705,12 +737,26 @@ static struct platform_driver cn10k_ddr_pmu_driver = {
static int __init cn10k_ddr_pmu_init(void)
{
- return platform_driver_register(&cn10k_ddr_pmu_driver);
+ int ret;
+
+ ret = cpuhp_setup_state_multi(
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE,
+ "perf/marvell/cn10k/ddr:online", NULL,
+ cn10k_ddr_pmu_offline_cpu);
+ if (ret)
+ return ret;
+
+ ret = platform_driver_register(&cn10k_ddr_pmu_driver);
+ if (ret)
+ cpuhp_remove_multi_state(
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE);
+ return ret;
}
static void __exit cn10k_ddr_pmu_exit(void)
{
platform_driver_unregister(&cn10k_ddr_pmu_driver);
+ cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE);
}
module_init(cn10k_ddr_pmu_init);
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index 47e13582d9fc..ce7f2740d7db 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -185,6 +185,7 @@ enum cpuhp_state {
CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE,
CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE,
CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE,
+ CPUHP_AP_PERF_ARM_MARVELL_CN10K_DDR_ONLINE,
CPUHP_AP_PERF_POWERPC_NEST_IMC_ONLINE,
CPUHP_AP_PERF_POWERPC_CORE_IMC_ONLINE,
CPUHP_AP_PERF_POWERPC_THREAD_IMC_ONLINE,
--
2.17.1
On Mon, Jul 26, 2021 at 02:40:24PM +0530, Bharat Bhushan wrote:
> Add binding documentation for the Marvell CN10k DDR
> performance monitor unit.
>
> Signed-off-by: Bharat Bhushan <[email protected]>
> ---
> .../bindings/perf/marvell-cn10k-ddr.txt | 15 +++++++++++++++
> 1 file changed, 15 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
Bindings are in DT schema format now.
Rob
> -----Original Message-----
> From: Rob Herring <[email protected]>
> Sent: Friday, July 30, 2021 5:00 AM
> To: Bharat Bhushan <[email protected]>
> Cc: [email protected]; [email protected]; linux-arm-
> [email protected]; [email protected]; linux-
> [email protected]; Sunil Kovvuri Goutham <[email protected]>
> Subject: [EXT] Re: [PATCH 1/4] dt-bindings: perf: marvell: cn10k ddr performance
> monitor
>
> External Email
>
> ----------------------------------------------------------------------
> On Mon, Jul 26, 2021 at 02:40:24PM +0530, Bharat Bhushan wrote:
> > Add binding documentation for the Marvell CN10k DDR performance
> > monitor unit.
> >
> > Signed-off-by: Bharat Bhushan <[email protected]>
> > ---
> > .../bindings/perf/marvell-cn10k-ddr.txt | 15 +++++++++++++++
> > 1 file changed, 15 insertions(+)
> > create mode 100644
> > Documentation/devicetree/bindings/perf/marvell-cn10k-ddr.txt
>
> Bindings are in DT schema format now.
I did not paid attention it changed to DT schema. Will do required changed in next version.
Just waiting for comments in other patches.
Thanks
-Bharat
>
> Rob