On Thu, 16 Jan 2025 04:59:10 +0000 Yoshihiro Furudera <fj5100bi@xxxxxxxxxxx> wrote: > This adds a new dynamic PMU to the Perf Events framework to program and > control the Uncore MAC PMUs in Fujitsu chips. > > This driver was created with reference to drivers/perf/qcom_l3_pmu.c. > > This driver exports formatting and event information to sysfs so it can > be used by the perf user space tools with the syntaxes: > > perf stat -e mac_iod0_mac0_ch0/ea-mac/ ls > perf stat -e mac_iod0_mac0_ch0/event=0x80/ ls > > FUJITSU-MONAKA Specification URL: > https://github.com/fujitsu/FUJITSU-MONAKA > > Signed-off-by: Yoshihiro Furudera <fj5100bi@xxxxxxxxxxx> Hi, Other than the docs issue, just minor comments inline. With the docs adjusted, Reviewed-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx> It's been a while since I closely reviewed a PMU driver so I may have missed some stuff! Jonathan > --- > .../admin-guide/perf/fujitsu_mac_pmu.rst | 75 +++ > Documentation/admin-guide/perf/index.rst | 1 + > drivers/perf/Kconfig | 9 + > drivers/perf/Makefile | 1 + > drivers/perf/fujitsu_mac_pmu.c | 573 ++++++++++++++++++ > 5 files changed, 659 insertions(+) > create mode 100644 Documentation/admin-guide/perf/fujitsu_mac_pmu.rst > create mode 100644 drivers/perf/fujitsu_mac_pmu.c > > diff --git a/Documentation/admin-guide/perf/fujitsu_mac_pmu.rst b/Documentation/admin-guide/perf/fujitsu_mac_pmu.rst > new file mode 100644 > index 000000000000..b6676526bde2 > --- /dev/null > +++ b/Documentation/admin-guide/perf/fujitsu_mac_pmu.rst > @@ -0,0 +1,75 @@ > +==================================================== > +Fujitsu Uncore MAC Performance Monitoring Unit (PMU) > +==================================================== > + > +This driver supports the Uncore MAC PMUs found in Fujitsu chips. > +Each MAC PMU on these chips is exposed as a uncore perf PMU with device name > +mac_iod<iod>_mac<mac>_ch<ch>. > + > +The driver provides a description of its available events and configuration > +options in sysfs, see /sys/bus/event_sources/devices/mac_iod<iod>_mac<mac>_ch<ch>/. > +This driver exports: > +- formats, used by perf user space and other tools to configure events > +- events, used by perf user space and other tools to create events > + symbolically, e.g.: > + perf stat -a -e mac_iod0_mac0_ch0/event=0x21/ ls > +- cpumask, used by perf user space and other tools to know on which CPUs > + to open the events > + > +This driver supports the following events: > +- cycles > + This event counts MAC cycles at MAC frequency. > +- read-count > + This event counts the number of read requests to MAC. > +- read-count-request > + This event counts the number of read requests including retry to MAC. > +- read-count-return > + This event counts the number of read requests to MAC. > +- read-count-request-pftgt > + This event counts the number of read requests including retry with PFTGT > + flag. > +- read-count-request-normal > + This event counts the number of read requests including retry without PFTGT > + flag. > +- read-count-return-pftgt-hit > + This event counts the number of read requests which hit the PFTGT buffer. > +- read-count-return-pftgt-miss > + This event counts the number of read requests which miss the PFTGT buffer. > +- read-wait > + This event counts outstanding read requests issued by DDR memory controller > + per cycle. > +- write-count > + This event counts the number of write requests to MAC (including zero write, > + full write, partial write, write cancel). > +- write-count-write > + This event counts the number of full write requests to MAC (not including > + zero write). > +- write-count-pwrite > + This event counts the number of partial write requests to MAC. > +- memory-read-count > + This event counts the number of read requests from MAC to memory. > +- memory-write-count > + This event counts the number of full write requests from MAC to memory. > +- memory-pwrite-count > + This event counts the number of partial write requests from MAC to memory. > +- ea-mac > + This event counts energy consumption of the MAC. > +- ea-memory > + This event counts energy consumption of the memory. > +- ea-memory-mac-read > + This event counts the number of read requests from MAC to memory. > +- ea-memory-mac-write > + This event counts the number of write requests from MAC to memory. > +- ea-memory-mac-pwrite > + This event counts the number of partial write requests from MAC to memory Text identical to memory-pwrite-count which suggest two things. a) naming inconsistent. Why is mac mentioned here and not in the name earlier. b) This comment is perhaps wrong as I assume has something more tod owtih with energy estimation? > +- ea-ha > + This event counts energy consumption of the HA. > + > + 'ea' is the abbreviation for 'Energy Analyzer'. > diff --git a/drivers/perf/fujitsu_mac_pmu.c b/drivers/perf/fujitsu_mac_pmu.c > new file mode 100644 > index 000000000000..788bf05b05e7 > --- /dev/null > +++ b/drivers/perf/fujitsu_mac_pmu.c > @@ -0,0 +1,573 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* > + * Driver for the Uncore MAC PMUs in Fujitsu chips. > + * > + * See Documentation/admin-guide/perf/fujitsu_mac_pmu.rst for more details. > + * > + * This driver is based on drivers/perf/qcom_l3_pmu.c > + * Copyright (c) 2015-2017, The Linux Foundation. All rights reserved. > + * Copyright (c) 2024 Fujitsu. All rights reserved. > + */ > + > +#include <linux/acpi.h> > +#include <linux/bitops.h> > +#include <linux/interrupt.h> > +#include <linux/io.h> > +#include <linux/list.h> > +#include <linux/module.h> mod_devicetable.h for the acpi device table. > +#include <linux/perf_event.h> > +#include <linux/platform_device.h> > + > +/* Number of counters on each PMU */ > +#define MAC_NUM_COUNTERS 8 > +/* Mask for the event type field within perf_event_attr.config and EVTYPE reg */ > +#define MAC_EVTYPE_MASK 0xFF > + > +/* Perfmon registers */ > +#define MAC_PM_EVCNTR(__cntr) (0x000 + __cntr * 8) > +#define MAC_PM_CNTCTL(__cntr) (0x100 + __cntr * 8) > +#define MAC_PM_EVTYPE(__cntr) (0x200 + __cntr * 8) (0x200 + (__cntr) * 8) preferred as avoids any possibility of precedence issues if __cntr isn't simply a number. > +/* > + * We must NOT create groups containing events from multiple hardware PMUs, > + * although mixing different software and hardware PMUs is allowed. > + */ > +static bool fujitsu_mac__validate_event_group(struct perf_event *event) > +{ > + struct perf_event *leader = event->group_leader; > + struct perf_event *sibling; > + int counters = 0; > + > + if (leader->pmu != event->pmu && !is_software_event(leader)) > + return false; > + > + /* The sum of the counters used by the event and its leader event */ > + counters = 2; > + > + for_each_sibling_event(sibling, leader) { > + if (is_software_event(sibling)) > + continue; > + if (sibling->pmu != event->pmu) > + return false; > + counters += 1; counters++; ? > + } > + > + /* > + * If the group requires more counters than the HW has, it > + * cannot ever be scheduled. > + */ > + return counters <= MAC_NUM_COUNTERS; > +} > + > +#define MAC_PMU_FORMAT_ATTR(_name, _config) \ > + (&((struct dev_ext_attribute[]) { \ > + { .attr = __ATTR(_name, 0444, device_show_string, NULL), \ > + .var = (void *) _config, } \ > + })[0].attr.attr) > + > +static struct attribute *fujitsu_mac_pmu_formats[] = { > + MAC_PMU_FORMAT_ATTR(event, "config:0-7"), > + NULL > +}; > + > +static const struct attribute_group fujitsu_mac_pmu_format_group = { > + .name = "format", > + .attrs = fujitsu_mac_pmu_formats Add trailing comma. > +}; > + > +static const struct attribute_group fujitsu_mac_pmu_events_group = { > + .name = "events", > + .attrs = fujitsu_mac_pmu_events Add trailing comma. There might be other fields in future. > +}; > + > +static ssize_t cpumask_show(struct device *dev, > + struct device_attribute *attr, char *buf) > +{ > + struct mac_pmu *macpmu = to_mac_pmu(dev_get_drvdata(dev)); > + > + return cpumap_print_to_pagebuf(true, buf, &macpmu->cpumask); > +} > + I would drop this blank line to tightly associate the following declaration with the function above. > +static DEVICE_ATTR_RO(cpumask); > + > +static int fujitsu_mac_pmu_probe(struct platform_device *pdev) > +{ > + struct device *dev = &pdev->dev; > + struct acpi_device *acpi_dev; > + struct mac_pmu *macpmu; > + struct resource *memrc; > + char *name; > + int ret; > + u64 uid; > + > + acpi_dev = ACPI_COMPANION(dev); > + if (!acpi_dev) > + return -ENODEV; > + > + ret = acpi_dev_uid_to_integer(acpi_dev, &uid); > + if (ret) > + return dev_err_probe(dev, ret, "unable to read ACPI uid\n"); > + > + macpmu = devm_kzalloc(dev, sizeof(*macpmu), GFP_KERNEL); > + if (!macpmu) > + return -ENOMEM; > + > + name = devm_kasprintf(dev, GFP_KERNEL, "mac_iod%llu_mac%llu_ch%llu", > + (uid >> 8) & 0xF, (uid >> 4) & 0xF, uid & 0xF); Slightly novel to encode that much in a UID, but nothing stops you doing that so I don't really mind. We always went with additional properties for our platforms but this is fine I think. > + if (!name) > + return -ENOMEM; > + > + macpmu->pmu = (struct pmu) { > + .parent = dev, > + .task_ctx_nr = perf_invalid_context, > + > + .pmu_enable = fujitsu_mac__pmu_enable, > + .pmu_disable = fujitsu_mac__pmu_disable, > + .event_init = fujitsu_mac__event_init, > + .add = fujitsu_mac__event_add, > + .del = fujitsu_mac__event_del, > + .start = fujitsu_mac__event_start, > + .stop = fujitsu_mac__event_stop, > + .read = fujitsu_mac__event_read, > + > + .attr_groups = fujitsu_mac_pmu_attr_grps, > + .capabilities = PERF_PMU_CAP_NO_EXCLUDE > + }; > + > + macpmu->regs = devm_platform_get_and_ioremap_resource(pdev, 0, &memrc); > + if (IS_ERR(macpmu->regs)) > + return PTR_ERR(macpmu->regs); > + > + fujitsu_mac__init(macpmu); > + > + ret = platform_get_irq(pdev, 0); > + if (ret <= 0) > + return ret; If it were 0 you'd not want to return that as would look like your driver probed successfully and none of the devm cleanup would happen. if (ret < 0) is fine - see docs for platform_get_irq() that make it clear 0 is never returned. > + > + ret = devm_request_irq(dev, ret, fujitsu_mac__handle_irq, 0, > + name, macpmu); > + if (ret) > + return dev_err_probe(dev, ret, "Request for IRQ failed for slice @%pa\n", I would wrap this under d of dev. > + &memrc->start); Indent of this line is also unusual. > + > + /* Add this instance to the list used by the offline callback */ > + ret = cpuhp_state_add_instance(mac_pmu_cpuhp_state, &macpmu->node); > + if (ret) > + return dev_err_probe(dev, ret, "Error registering hotplug"); > + > + ret = perf_pmu_register(&macpmu->pmu, name, -1); > + if (ret < 0) > + return dev_err_probe(dev, ret, "Failed to register MAC PMU\n"); > + > + dev_dbg(dev, "Registered %s, type: %d\n", name, macpmu->pmu.type); > + > + return 0; > +} > + > +static const struct acpi_device_id fujitsu_mac_pmu_acpi_match[] = { > + { "FUJI200C", }, > + { } > +}; > +MODULE_DEVICE_TABLE(acpi, fujitsu_mac_pmu_acpi_match); > + > +static struct platform_driver fujitsu_mac_pmu_driver = { > + .driver = { > + .name = "fujitsu-mac-pmu", > + .acpi_match_table = fujitsu_mac_pmu_acpi_match, > + .suppress_bind_attrs = true Add trailing comma. Not impossible we will want to set another field in here one day. So we should not make that harder. > + }, > + .probe = fujitsu_mac_pmu_probe > +};