Re: [PATCH v2 1/1] drivers: perf: Add LLC-TAD perf counter support

From: Will Deacon
Date: Mon Aug 02 2021 - 06:34:12 EST


On Mon, Jun 14, 2021 at 07:28:49PM +0530, Bhaskara Budiredla wrote:
> This driver adds support for Last-level cache tag-and-data unit
> (LLC-TAD) PMU that is featured in some of the Marvell's CN10K
> infrastructure silicons.
>
> The LLC is divided into 2N slices distributed across N Mesh tiles
> in a single-socket configuration. The driver always configures the
> same counter for all of the TADs. The user would end up effectively
> reserving one of eight counters in every TAD to look across all TADs.
> The occurrences of events are aggregated and presented to the user
> at the end of an application run. The driver does not provide a way
> for the user to partition TADs so that different TADs are used for
> different applications.
>
> The event counters are zeroed to start event counting to avoid any
> rollover issues. TAD perf counters are 64-bit, so it's not currently
> possible to overflow event counters at current mesh and core
> frequencies.

I couldn't spot where you disable sampling events, which rely heavily on
detecting overflow. Probably need PERF_PMU_CAP_NO_INTERRUPT.

> To measure tad pmu events use perf tool stat command. For instance:
>
> perf stat -e tad_dat_msh_in_dss,tad_req_msh_out_any <application>
> perf stat -e tad_alloc_any,tad_hit_any,tad_tag_rd <application>
>
> Signed-off-by: Bhaskara Budiredla <bbudiredla@xxxxxxxxxxx>
> ---
> .../bindings/perf/marvell-cn10k-tad-pmu.txt | 20 +

Adding the DT list for this ^^ as it looks a bit odd to me (also shouldn't
it be in YAML?)

> drivers/perf/Kconfig | 7 +
> drivers/perf/Makefile | 1 +
> drivers/perf/marvell_cn10k_tad_pmu.c | 428 ++++++++++++++++++

-ENODOCUMENTATION

> 4 files changed, 456 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/perf/marvell-cn10k-tad-pmu.txt
> create mode 100644 drivers/perf/marvell_cn10k_tad_pmu.c
>
> diff --git a/Documentation/devicetree/bindings/perf/marvell-cn10k-tad-pmu.txt b/Documentation/devicetree/bindings/perf/marvell-cn10k-tad-pmu.txt
> new file mode 100644
> index 000000000000..8b1f753303e2
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/perf/marvell-cn10k-tad-pmu.txt
> @@ -0,0 +1,20 @@
> +* Marvell CN10K LLC-TAD performace monitor unit
> +
> +Required properties:
> +- compatible: must be:
> + "marvell,cn10k-tad-pmu"
> +- tad-cnt: number of tad pmu regions
> +- tad-page-size: size of entire tad block
> +- tad-pmu-page-size: size of one tad pmu region
> +- reg: physical address and size
> +
> +Example:
> +
> +/* Actual values updated by firmware at boot time */
> +tad_pmu {
> + compatible = "marvell,cn10k-tad-pmu";
> + tad-cnt = <1>;
> + tad-page-size = <0x1000>;
> + tad-pmu-page-size = <0x1000>;
> + reg = <0x87e2 0x80000000 0x0 0x1000>;
> +};
> diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
> index 77522e5efe11..73dca11f8080 100644
> --- a/drivers/perf/Kconfig
> +++ b/drivers/perf/Kconfig
> @@ -137,6 +137,13 @@ config ARM_DMC620_PMU
> Support for PMU events monitoring on the ARM DMC-620 memory
> controller.
>
> +config MARVELL_CN10K_TAD_PMU
> + tristate "Marvell CN10K LLC-TAD PMU"
> + depends on ARM64
> + help
> + Provides support for Last-Level cache Tag-and-data Units (LLC-TAD)
> + performance monitors on CN10K family silicons.
> +
> source "drivers/perf/hisilicon/Kconfig"
>
> endmenu
> diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
> index 5260b116c7da..2db5418d5b0a 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_TAD_PMU) += marvell_cn10k_tad_pmu.o
> diff --git a/drivers/perf/marvell_cn10k_tad_pmu.c b/drivers/perf/marvell_cn10k_tad_pmu.c
> new file mode 100644
> index 000000000000..99878de481f0
> --- /dev/null
> +++ b/drivers/perf/marvell_cn10k_tad_pmu.c
> @@ -0,0 +1,428 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Marvell CN10K LLC-TAD perf driver
> + *
> + * Copyright (C) 2021 Marvell.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + */
> +
> +#define pr_fmt(fmt) "tad_pmu: " fmt
> +
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/of_address.h>
> +#include <linux/of_device.h>
> +#include <linux/cpuhotplug.h>
> +#include <linux/perf_event.h>
> +#include <linux/platform_device.h>
> +#include <linux/arm-smccc.h>
> +
> +#define TAD_PFC_OFFSET 0x0800
> +#define TAD_PFC(counter) (TAD_PFC_OFFSET | (counter << 3))
> +#define TAD_PRF_OFFSET 0x0900
> +#define TAD_PRF(counter) (TAD_PRF_OFFSET | (counter << 3))
> +#define TAD_PRF_CNTSEL_MASK 0xFF
> +#define TAD_MAX_COUNTERS 8
> +
> +#define to_tad_pmu(p) (container_of(p, struct tad_pmu, pmu))
> +
> +struct tad_region {
> + void __iomem *base;
> +};
> +
> +struct tad_pmu {
> + struct pmu pmu;
> + struct tad_region *regions;
> + u32 region_cnt;
> + unsigned int cpu;
> + struct hlist_node node;
> + struct perf_event *events[TAD_MAX_COUNTERS];
> + DECLARE_BITMAP(counters_map, TAD_MAX_COUNTERS);
> +};
> +
> +static int tad_pmu_cpuhp_state;
> +
> +static void tad_pmu_event_counter_read(struct perf_event *event)
> +{
> + struct tad_pmu *tad_pmu = to_tad_pmu(event->pmu);
> + struct hw_perf_event *hwc = &event->hw;
> + u32 counter_idx = hwc->idx;
> + u64 delta, prev, new;
> + int i;
> +
> + do {
> + prev = local64_read(&hwc->prev_count);
> + for (i = 0, new = 0; i < tad_pmu->region_cnt; i++)
> + new += readq(tad_pmu->regions[i].base +
> + TAD_PFC(counter_idx));
> + } while (local64_cmpxchg(&hwc->prev_count, prev, new) != prev);
> +
> + delta = (new - prev) & GENMASK_ULL(63, 0);
> + local64_add(delta, &event->count);
> +}
> +
> +static void tad_pmu_event_counter_stop(struct perf_event *event, int flags)
> +{
> + struct tad_pmu *tad_pmu = to_tad_pmu(event->pmu);
> + struct hw_perf_event *hwc = &event->hw;
> + u32 counter_idx = hwc->idx;
> + int i;
> +
> + /* TAD()_PFC() stop counting on the write
> + * which sets TAD()_PRF()[CNTSEL] == 0
> + */
> + for (i = 0; i < tad_pmu->region_cnt; i++)
> + writeq(0, tad_pmu->regions[i].base + TAD_PRF(counter_idx));

writeq_relaxed() should be sufficient for this (and the others in this
driver), no?

> +static ssize_t tad_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 TAD_PMU_EVENT_ATTR(_name, _id) \
> + (&((struct perf_pmu_events_attr[]) { \
> + { .attr = __ATTR(_name, 0444, tad_pmu_event_show, NULL),\
> + .id = _id, } \
> + })[0].attr.attr)

Use PMU_EVENT_ATTR_ID instead?

> +static int tad_pmu_probe(struct platform_device *pdev)
> +{
> + struct device_node *node = pdev->dev.of_node;
> + struct tad_region *regions;
> + struct tad_pmu *tad_pmu;
> + struct resource *res;
> + u32 tad_pmu_page_size;
> + u32 tad_page_size;
> + u32 tad_cnt;
> + int i, ret;
> + char *name;
> +
> + tad_pmu = devm_kzalloc(&pdev->dev, sizeof(*tad_pmu), GFP_KERNEL);
> + if (!tad_pmu)
> + return -ENOMEM;
> +
> + platform_set_drvdata(pdev, tad_pmu);
> +
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + if (!res) {
> + dev_err(&pdev->dev, "Mem resource not found\n");
> + return -ENODEV;
> + }
> +
> + ret = of_property_read_u32(node, "tad-page-size", &tad_page_size);
> + if (ret) {
> + dev_err(&pdev->dev, "Can't find tad-page-size property\n");
> + return ret;
> + }
> +
> + ret = of_property_read_u32(node, "tad-pmu-page-size",
> + &tad_pmu_page_size);
> + if (ret) {
> + dev_err(&pdev->dev, "Can't find tad-pmu-page-size property\n");
> + return ret;
> + }
> +
> + ret = of_property_read_u32(node, "tad-cnt", &tad_cnt);
> + if (ret) {
> + dev_err(&pdev->dev, "Can't find tad-cnt property\n");
> + return ret;
> + }
> +
> + regions = kcalloc(tad_cnt, sizeof(*regions), GFP_KERNEL);

devm_kcalloc() instead?

Will