PCI Express Interface PMU includes various performance counters
to monitor the data that is transmitted over the PCIe link. The
counters track various inbound and outbound transactions which
includes separate counters for posted/non-posted/completion TLPs.
Also, inbound and outbound memory read requests along with their
latencies can also be monitored. Address Translation Services(ATS)events
such as ATS Translation, ATS Page Request, ATS Invalidation along with
their corresponding latencies are also supported.
The performance counters are 64 bits wide.
For instance,
perf stat -e ib_tlp_pr <workload>
tracks the inbound posted TLPs for the workload.
Signed-off-by: Gowthami Thiagarajan <[email protected]>
---
MAINTAINERS | 6 +
drivers/perf/Kconfig | 7 +
drivers/perf/Makefile | 1 +
drivers/perf/marvell_pem_pmu.c | 425 +++++++++++++++++++++++++++++++++
include/linux/cpuhotplug.h | 1 +
5 files changed, 440 insertions(+)
create mode 100644 drivers/perf/marvell_pem_pmu.c
diff --git a/MAINTAINERS b/MAINTAINERS
index dd5de540ec0b..50c11e128a98 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -12759,6 +12759,12 @@ S: Supported
F: Documentation/networking/device_drivers/ethernet/marvell/octeontx2.rst
F: drivers/net/ethernet/marvell/octeontx2/af/
+MARVELL PEM PMU DRIVER
+M: Linu Cherian <[email protected]>
+M: Gowthami Thiagarajan <[email protected]>
+S: Supported
+F: drivers/perf/marvell_pem_pmu.c
+
MARVELL PRESTERA ETHERNET SWITCH DRIVER
M: Taras Chornyi <[email protected]>
S: Supported
diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
index 273d67ecf6d2..12ac8409263a 100644
--- a/drivers/perf/Kconfig
+++ b/drivers/perf/Kconfig
@@ -234,4 +234,11 @@ config CXL_PMU
If unsure say 'm'.
+config MARVELL_PEM_PMU
+ tristate "MARVELL PEM PMU Support"
+ depends on ARCH_THUNDER || (COMPILE_TEST && 64BIT)
+ help
+ Enable support for PCIe Interface performance monitoring
+ on Marvell platform.
+
endmenu
diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
index 16b3ec4db916..db3c473cc5c7 100644
--- a/drivers/perf/Makefile
+++ b/drivers/perf/Makefile
@@ -21,6 +21,7 @@ obj-$(CONFIG_ARM_SPE_PMU) += arm_spe_pmu.o
obj-$(CONFIG_ARM_DMC620_PMU) += arm_dmc620_pmu.o
obj-$(CONFIG_MARVELL_CN10K_TAD_PMU) += marvell_cn10k_tad_pmu.o
obj-$(CONFIG_MARVELL_CN10K_DDR_PMU) += marvell_cn10k_ddr_pmu.o
+obj-$(CONFIG_MARVELL_PEM_PMU) += marvell_pem_pmu.o
obj-$(CONFIG_APPLE_M1_CPU_PMU) += apple_m1_cpu_pmu.o
obj-$(CONFIG_ALIBABA_UNCORE_DRW_PMU) += alibaba_uncore_drw_pmu.o
obj-$(CONFIG_ARM_CORESIGHT_PMU_ARCH_SYSTEM_PMU) += arm_cspmu/
diff --git a/drivers/perf/marvell_pem_pmu.c b/drivers/perf/marvell_pem_pmu.c
new file mode 100644
index 000000000000..913175954a9c
--- /dev/null
+++ b/drivers/perf/marvell_pem_pmu.c
@@ -0,0 +1,425 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Marvell PEM(PCIe RC) Performance Monitor Driver
+ *
+ * Copyright (C) 2024 Marvell.
+ */
+
+#include <linux/acpi.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+
+/*
+ * Each of these events maps to a free running 64 bit counter
+ * with no event control, but can be reset.
+ *
+ */
+enum pem_events {
+ IB_TLP_NPR,
+ IB_TLP_PR,
+ IB_TLP_CPL,
+ IB_TLP_DWORDS_NPR,
+ IB_TLP_DWORDS_PR,
+ IB_TLP_DWORDS_CPL,
+ IB_INFLIGHT,
+ IB_READS,
+ IB_REQ_NO_RO_NCB,
+ IB_REQ_NO_RO_EBUS,
+ OB_TLP_NPR,
+ OB_TLP_PR,
+ OB_TLP_CPL,
+ OB_TLP_DWORDS_NPR,
+ OB_TLP_DWORDS_PR,
+ OB_TLP_DWORDS_CPL,
+ OB_INFLIGHT,
+ OB_READS,
+ OB_MERGES_NPR,
+ OB_MERGES_PR,
+ OB_MERGES_CPL,
+ ATS_TRANS,
+ ATS_TRANS_LATENCY,
+ ATS_PRI,
+ ATS_PRI_LATENCY,
+ ATS_INV,
+ ATS_INV_LATENCY,
+ PEM_EVENTIDS_MAX,
+};
+
+static u64 eventid_to_offset_table[] = {
+ [IB_TLP_NPR] = 0x0,
+ [IB_TLP_PR] = 0x8,
+ [IB_TLP_CPL] = 0x10,
+ [IB_TLP_DWORDS_NPR] = 0x100,
+ [IB_TLP_DWORDS_PR] = 0x108,
+ [IB_TLP_DWORDS_CPL] = 0x110,
+ [IB_INFLIGHT] = 0x200,
+ [IB_READS] = 0x300,
+ [IB_REQ_NO_RO_NCB] = 0x400,
+ [IB_REQ_NO_RO_EBUS] = 0x408,
+ [OB_TLP_NPR] = 0x500,
+ [OB_TLP_PR] = 0x508,
+ [OB_TLP_CPL] = 0x510,
+ [OB_TLP_DWORDS_NPR] = 0x600,
+ [OB_TLP_DWORDS_PR] = 0x608,
+ [OB_TLP_DWORDS_CPL] = 0x610,
+ [OB_INFLIGHT] = 0x700,
+ [OB_READS] = 0x800,
+ [OB_MERGES_NPR] = 0x900,
+ [OB_MERGES_PR] = 0x908,
+ [OB_MERGES_CPL] = 0x910,
+ [ATS_TRANS] = 0x2D18,
+ [ATS_TRANS_LATENCY] = 0x2D20,
+ [ATS_PRI] = 0x2D28,
+ [ATS_PRI_LATENCY] = 0x2D30,
+ [ATS_INV] = 0x2D38,
+ [ATS_INV_LATENCY] = 0x2D40,
+};
+
+struct pem_pmu {
+ struct pmu pmu;
+ void __iomem *base;
+ unsigned int cpu;
+ struct device *dev;
+ struct hlist_node node;
+};
+
+#define to_pem_pmu(p) container_of(p, struct pem_pmu, pmu)
+
+static int eventid_to_offset(int eventid)
+{
+ return eventid_to_offset_table[eventid];
+}
+
+/* Events */
+static ssize_t pem_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 sysfs_emit(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define PEM_EVENT_ATTR(_name, _id) \
+ (&((struct perf_pmu_events_attr[]) { \
+ { .attr = __ATTR(_name, 0444, pem_pmu_event_show, NULL), \
+ .id = _id, } \
+ })[0].attr.attr)
+
+static struct attribute *pem_perf_events_attrs[] = {
+ PEM_EVENT_ATTR(ib_tlp_npr, IB_TLP_NPR),
+ PEM_EVENT_ATTR(ib_tlp_pr, IB_TLP_PR),
+ PEM_EVENT_ATTR(ib_tlp_cpl_partid, IB_TLP_CPL),
+ PEM_EVENT_ATTR(ib_tlp_dwords_npr, IB_TLP_DWORDS_NPR),
+ PEM_EVENT_ATTR(ib_tlp_dwords_pr, IB_TLP_DWORDS_PR),
+ PEM_EVENT_ATTR(ib_tlp_dwords_cpl_partid, IB_TLP_DWORDS_CPL),
+ PEM_EVENT_ATTR(ib_inflight, IB_INFLIGHT),
+ PEM_EVENT_ATTR(ib_reads, IB_READS),
+ PEM_EVENT_ATTR(ib_req_no_ro_ncb, IB_REQ_NO_RO_NCB),
+ PEM_EVENT_ATTR(ib_req_no_ro_ebus, IB_REQ_NO_RO_EBUS),
+ PEM_EVENT_ATTR(ob_tlp_npr_partid, OB_TLP_NPR),
+ PEM_EVENT_ATTR(ob_tlp_pr_partid, OB_TLP_PR),
+ PEM_EVENT_ATTR(ob_tlp_cpl_partid, OB_TLP_CPL),
+ PEM_EVENT_ATTR(ob_tlp_dwords_npr_partid, OB_TLP_DWORDS_NPR),
+ PEM_EVENT_ATTR(ob_tlp_dwords_pr_partid, OB_TLP_DWORDS_PR),
+ PEM_EVENT_ATTR(ob_tlp_dwords_cpl_partid, OB_TLP_DWORDS_CPL),
+ PEM_EVENT_ATTR(ob_inflight_partid, OB_INFLIGHT),
+ PEM_EVENT_ATTR(ob_reads_partid, OB_READS),
+ PEM_EVENT_ATTR(ob_merges_npr_partid, OB_MERGES_NPR),
+ PEM_EVENT_ATTR(ob_merges_pr_partid, OB_MERGES_PR),
+ PEM_EVENT_ATTR(ob_merges_cpl_partid, OB_MERGES_CPL),
+ PEM_EVENT_ATTR(ats_trans, ATS_TRANS),
+ PEM_EVENT_ATTR(ats_trans_latency, ATS_TRANS_LATENCY),
+ PEM_EVENT_ATTR(ats_pri, ATS_PRI),
+ PEM_EVENT_ATTR(ats_pri_latency, ATS_PRI_LATENCY),
+ PEM_EVENT_ATTR(ats_inv, ATS_INV),
+ PEM_EVENT_ATTR(ats_inv_latency, ATS_INV_LATENCY),
+ NULL
+};
+
+static struct attribute_group pem_perf_events_attr_group = {
+ .name = "events",
+ .attrs = pem_perf_events_attrs,
+};
+
+PMU_FORMAT_ATTR(event, "config:0-5");
+
+static struct attribute *pem_perf_format_attrs[] = {
+ &format_attr_event.attr,
+ NULL
+};
+
+static struct attribute_group pem_perf_format_attr_group = {
+ .name = "format",
+ .attrs = pem_perf_format_attrs,
+};
+
+/* cpumask */
+static ssize_t pem_perf_cpumask_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ struct pem_pmu *pmu = dev_get_drvdata(dev);
+
+ return cpumap_print_to_pagebuf(true, buf, cpumask_of(pmu->cpu));
+}
+
+static struct device_attribute pem_perf_cpumask_attr =
+ __ATTR(cpumask, 0444, pem_perf_cpumask_show, NULL);
+
+static struct attribute *pem_perf_cpumask_attrs[] = {
+ &pem_perf_cpumask_attr.attr,
+ NULL
+};
+
+static struct attribute_group pem_perf_cpumask_attr_group = {
+ .attrs = pem_perf_cpumask_attrs,
+};
+
+static const struct attribute_group *pem_perf_attr_groups[] = {
+ &pem_perf_events_attr_group,
+ &pem_perf_cpumask_attr_group,
+ &pem_perf_format_attr_group,
+ NULL
+};
+
+static int pem_perf_event_init(struct perf_event *event)
+{
+ struct pem_pmu *pmu = to_pem_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+
+ if (event->attr.type != event->pmu->type)
+ return -ENOENT;
+
+ if (is_sampling_event(event) ||
+ event->attach_state & PERF_ATTACH_TASK) {
+ return -EOPNOTSUPP;
+ }
+
+ if (event->cpu < 0)
+ 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 pem_perf_counter_reset(struct pem_pmu *pmu,
+ struct perf_event *event, int eventid)
+{
+ writeq_relaxed(0x0, pmu->base + eventid_to_offset(eventid));
+}
+
+static u64 pem_perf_read_counter(struct pem_pmu *pmu,
+ struct perf_event *event, int eventid)
+{
+ return readq_relaxed(pmu->base + eventid_to_offset(eventid));
+}
+
+static void pem_perf_event_update(struct perf_event *event)
+{
+ struct pem_pmu *pmu = to_pem_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ u64 prev_count, new_count;
+
+ do {
+ prev_count = local64_read(&hwc->prev_count);
+ new_count = pem_perf_read_counter(pmu, event, hwc->idx);
+ } while (local64_xchg(&hwc->prev_count, new_count) != prev_count);
+
+ local64_add((new_count - prev_count), &event->count);
+}
+
+static void pem_perf_event_start(struct perf_event *event, int flags)
+{
+ struct pem_pmu *pmu = to_pem_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ int eventid = hwc->idx;
+
+ local64_set(&hwc->prev_count, 0);
+
+ pem_perf_counter_reset(pmu, event, eventid);
+
+ hwc->state = 0;
+}
+
+static int pem_perf_event_add(struct perf_event *event, int flags)
+{
+ struct hw_perf_event *hwc = &event->hw;
+
+ hwc->idx = event->attr.config;
+ if (hwc->idx >= PEM_EVENTIDS_MAX)
+ return -EINVAL;
+ hwc->state |= PERF_HES_STOPPED;
+
+ if (flags & PERF_EF_START)
+ pem_perf_event_start(event, flags);
+
+ return 0;
+}
+
+static void pem_perf_event_stop(struct perf_event *event, int flags)
+{
+ struct hw_perf_event *hwc = &event->hw;
+
+ if (flags & PERF_EF_UPDATE)
+ pem_perf_event_update(event);
+
+ hwc->state |= PERF_HES_STOPPED;
+}
+
+static void pem_perf_event_del(struct perf_event *event, int flags)
+{
+ struct hw_perf_event *hwc = &event->hw;
+
+ pem_perf_event_stop(event, PERF_EF_UPDATE);
+ hwc->idx = -1;
+}
+
+static int pem_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+ struct pem_pmu *pmu = hlist_entry_safe(node, struct pem_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 pem_perf_probe(struct platform_device *pdev)
+{
+ struct pem_pmu *pem_pmu;
+ struct resource *res;
+ void __iomem *base;
+ char *name;
+ int ret;
+
+ pem_pmu = devm_kzalloc(&pdev->dev, sizeof(*pem_pmu), GFP_KERNEL);
+ if (!pem_pmu)
+ return -ENOMEM;
+
+ pem_pmu->dev = &pdev->dev;
+ platform_set_drvdata(pdev, pem_pmu);
+
+ base = devm_platform_get_and_ioremap_resource(pdev, 0, &res);
+ if (IS_ERR(base))
+ return PTR_ERR(base);
+
+ pem_pmu->base = base;
+
+ pem_pmu->pmu = (struct pmu) {
+ .module = THIS_MODULE,
+ .capabilities = PERF_PMU_CAP_NO_EXCLUDE,
+ .task_ctx_nr = perf_invalid_context,
+ .attr_groups = pem_perf_attr_groups,
+ .event_init = pem_perf_event_init,
+ .add = pem_perf_event_add,
+ .del = pem_perf_event_del,
+ .start = pem_perf_event_start,
+ .stop = pem_perf_event_stop,
+ .read = pem_perf_event_update,
+ };
+
+ /* Choose this cpu to collect perf data */
+ pem_pmu->cpu = raw_smp_processor_id();
+
+ name = devm_kasprintf(pem_pmu->dev, GFP_KERNEL, "mrvl_pcie_rc_pmu_%llx",
+ res->start);
+ if (!name)
+ return -ENOMEM;
+
+ cpuhp_state_add_instance_nocalls
+ (CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE,
+ &pem_pmu->node);
+
+ ret = perf_pmu_register(&pem_pmu->pmu, name, -1);
+ if (ret)
+ goto error;
+
+ return 0;
+error:
+ cpuhp_state_remove_instance_nocalls
+ (CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE,
+ &pem_pmu->node);
+ return ret;
+}
+
+static int pem_perf_remove(struct platform_device *pdev)
+{
+ struct pem_pmu *pem_pmu = platform_get_drvdata(pdev);
+
+ cpuhp_state_remove_instance_nocalls
+ (CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE,
+ &pem_pmu->node);
+
+ perf_pmu_unregister(&pem_pmu->pmu);
+ return 0;
+}
+
+#ifdef CONFIG_ACPI
+static const struct acpi_device_id pem_pmu_acpi_match[] = {
+ {"MRVL000E", 0},
+ {},
+};
+MODULE_DEVICE_TABLE(acpi, pem_pmu_acpi_match);
+#endif
+
+static struct platform_driver pem_pmu_driver = {
+ .driver = {
+ .name = "pem-pmu",
+ .acpi_match_table = ACPI_PTR(pem_pmu_acpi_match),
+ .suppress_bind_attrs = true,
+ },
+ .probe = pem_perf_probe,
+ .remove = pem_perf_remove,
+};
+
+static int __init pem_pmu_init(void)
+{
+ int ret;
+
+ ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE,
+ "perf/marvell/pem:online", NULL,
+ pem_pmu_offline_cpu);
+ if (ret)
+ return ret;
+
+ ret = platform_driver_register(&pem_pmu_driver);
+ if (ret)
+ cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE);
+ return ret;
+}
+
+static void __exit pem_pmu_exit(void)
+{
+ platform_driver_unregister(&pem_pmu_driver);
+ cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_MARVELL_PEM_ONLINE);
+}
+
+module_init(pem_pmu_init);
+module_exit(pem_pmu_exit);
+
+MODULE_DESCRIPTION("Marvell PEM Perf driver");
+MODULE_AUTHOR("Linu Cherian <[email protected]>");
+MODULE_LICENSE("GPL");
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index 624d4a38c358..9bf34f66e5f5 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -238,6 +238,7 @@ enum cpuhp_state {
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_ARM_MARVELL_PEM_ONLINE,
CPUHP_AP_PERF_POWERPC_NEST_IMC_ONLINE,
CPUHP_AP_PERF_POWERPC_CORE_IMC_ONLINE,
CPUHP_AP_PERF_POWERPC_THREAD_IMC_ONLINE,
--
2.25.1
> +config MARVELL_PEM_PMU
> + tristate "MARVELL PEM PMU Support"
> + depends on ARCH_THUNDER || (COMPILE_TEST && 64BIT)
Why the && 64BIT?
I expect any kernel function/macro should work on 32bit systems as
well, but might internally need a different implementation. But
drivers should not need to care.
If you get warning / errors with 32bit systems, it might suggests your
driver is not as clean as it could be.
Andrew