From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933319AbcCIQWK (ORCPT ); Wed, 9 Mar 2016 11:22:10 -0500 Received: from mail-wm0-f68.google.com ([74.125.82.68]:34962 "EHLO mail-wm0-f68.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932401AbcCIQVc (ORCPT ); Wed, 9 Mar 2016 11:21:32 -0500 From: Jan Glauber To: Mark Rutland , Will Deacon Cc: linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org, Jan Glauber Subject: [PATCH v2 1/5] arm64/perf: Basic uncore counter support for Cavium ThunderX Date: Wed, 9 Mar 2016 17:21:03 +0100 Message-Id: <2588abcb011df6e60cd8212a8583454999180748.1457539622.git.jglauber@cavium.com> X-Mailer: git-send-email 1.9.1 In-Reply-To: References: In-Reply-To: References: Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Provide "uncore" facilities for different non-CPU performance counter units. Based on Intel/AMD uncore pmu support. The uncore drivers cover quite different functionality including L2 Cache, memory controllers and interconnects. The uncore PMUs can be found under /sys/bus/event_source/devices. All counters are exported via sysfs in the corresponding events files under the PMU directory so the perf tool can list the event names. There are some points that are special in this implementation: 1) The PMU detection relies on PCI device detection. If a matching PCI device is found the PMU is created. The code can deal with multiple units of the same type, e.g. more than one memory controller. Note: There is also a CPUID check to determine the CPU variant, this is needed to support different hardware versions that use the same PCI IDs. 2) Counters are summarized across different units of the same type on one NUMA node. For instance L2C TAD 0..7 are presented as a single counter (adding the values from TAD 0 to 7). Although losing the ability to read a single value the merged values are easier to use. 3) NUMA support. The device node id is used to group devices by node so counters on one node can be merged. The NUMA node can be selected via a new sysfs node attribute. Without NUMA support all devices will be on node 0. 4) All counters are 64 bit wide without overflow interrupts. Signed-off-by: Jan Glauber --- drivers/perf/Makefile | 1 + drivers/perf/uncore/Makefile | 1 + drivers/perf/uncore/uncore_cavium.c | 301 ++++++++++++++++++++++++++++++++++++ drivers/perf/uncore/uncore_cavium.h | 78 ++++++++++ 4 files changed, 381 insertions(+) create mode 100644 drivers/perf/uncore/Makefile create mode 100644 drivers/perf/uncore/uncore_cavium.c create mode 100644 drivers/perf/uncore/uncore_cavium.h diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile index acd2397..61b6084 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -1 +1,2 @@ obj-$(CONFIG_ARM_PMU) += arm_pmu.o +obj-$(CONFIG_ARCH_THUNDER) += uncore/ diff --git a/drivers/perf/uncore/Makefile b/drivers/perf/uncore/Makefile new file mode 100644 index 0000000..b9c72c2 --- /dev/null +++ b/drivers/perf/uncore/Makefile @@ -0,0 +1 @@ +obj-$(CONFIG_ARCH_THUNDER) += uncore_cavium.o diff --git a/drivers/perf/uncore/uncore_cavium.c b/drivers/perf/uncore/uncore_cavium.c new file mode 100644 index 0000000..4fd5e45 --- /dev/null +++ b/drivers/perf/uncore/uncore_cavium.c @@ -0,0 +1,301 @@ +/* + * Cavium Thunder uncore PMU support. Derived from Intel and AMD uncore code. + * + * Copyright (C) 2015,2016 Cavium Inc. + * Author: Jan Glauber + */ + +#include +#include +#include + +#include "uncore_cavium.h" + +int thunder_uncore_version; + +struct thunder_uncore *event_to_thunder_uncore(struct perf_event *event) +{ + return NULL; +} + +void thunder_uncore_read(struct perf_event *event) +{ + struct thunder_uncore *uncore = event_to_thunder_uncore(event); + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + struct thunder_uncore_unit *unit; + u64 prev, new = 0; + s64 delta; + + node = get_node(hwc->config, uncore); + + /* + * No counter overflow interrupts so we do not + * have to worry about prev_count changing on us. + */ + prev = local64_read(&hwc->prev_count); + + /* read counter values from all units on the node */ + list_for_each_entry(unit, &node->unit_list, entry) + new += readq(hwc->event_base + unit->map); + + local64_set(&hwc->prev_count, new); + delta = new - prev; + local64_add(delta, &event->count); +} + +void thunder_uncore_del(struct perf_event *event, int flags) +{ + struct thunder_uncore *uncore = event_to_thunder_uncore(event); + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + int i; + + event->pmu->stop(event, PERF_EF_UPDATE); + + /* + * For programmable counters we need to check where we installed it. + * To keep this function generic always test the more complicated + * case (free running counters won't need the loop). + */ + node = get_node(hwc->config, uncore); + for (i = 0; i < node->num_counters; i++) { + if (cmpxchg(&node->events[i], event, NULL) == event) + break; + } + hwc->idx = -1; +} + +int thunder_uncore_event_init(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + struct thunder_uncore *uncore; + + if (event->attr.type != event->pmu->type) + return -ENOENT; + + /* we do not support sampling */ + if (is_sampling_event(event)) + return -EINVAL; + + /* counters do not have these bits */ + if (event->attr.exclude_user || + event->attr.exclude_kernel || + event->attr.exclude_host || + event->attr.exclude_guest || + event->attr.exclude_hv || + event->attr.exclude_idle) + return -EINVAL; + + /* counters are 64 bit wide and without overflow interrupts */ + + uncore = event_to_thunder_uncore(event); + if (!uncore) + return -ENODEV; + if (!uncore->event_valid(event->attr.config & UNCORE_EVENT_ID_MASK)) + return -EINVAL; + + /* check NUMA node */ + node = get_node(event->attr.config, uncore); + if (!node) { + pr_debug("Invalid numa node selected\n"); + return -EINVAL; + } + + hwc->config = event->attr.config; + hwc->idx = -1; + return 0; +} + +/* + * Thunder uncore events are independent from CPUs. Provide a cpumask + * nevertheless to prevent perf from adding the event per-cpu and just + * set the mask to one online CPU. Use the same cpumask for all uncore + * devices. + */ +static cpumask_t thunder_active_mask; + +static ssize_t thunder_uncore_attr_show_cpumask(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + return cpumap_print_to_pagebuf(true, buf, &thunder_active_mask); +} +static DEVICE_ATTR(cpumask, S_IRUGO, thunder_uncore_attr_show_cpumask, NULL); + +static struct attribute *thunder_uncore_attrs[] = { + &dev_attr_cpumask.attr, + NULL, +}; + +struct attribute_group thunder_uncore_attr_group = { + .attrs = thunder_uncore_attrs, +}; + +ssize_t thunder_events_sysfs_show(struct device *dev, + struct device_attribute *attr, + char *page) +{ + struct perf_pmu_events_attr *pmu_attr = + container_of(attr, struct perf_pmu_events_attr, attr); + + if (pmu_attr->event_str) + return sprintf(page, "%s", pmu_attr->event_str); + + return 0; +} + +/* node attribute depending on number of numa nodes */ +static ssize_t node_show(struct device *dev, struct device_attribute *attr, char *page) +{ + if (NODES_SHIFT) + return sprintf(page, "config:16-%d\n", 16 + NODES_SHIFT - 1); + else + return sprintf(page, "config:16\n"); +} + +struct device_attribute format_attr_node = __ATTR_RO(node); + +static int thunder_uncore_pmu_cpu_notifier(struct notifier_block *nb, + unsigned long action, void *data) +{ + struct thunder_uncore *uncore = container_of(nb, struct thunder_uncore, cpu_nb); + int new_cpu, old_cpu = (long) data; + + switch (action & ~CPU_TASKS_FROZEN) { + case CPU_DOWN_PREPARE: + if (!cpumask_test_and_clear_cpu(old_cpu, &thunder_active_mask)) + break; + new_cpu = cpumask_any_but(cpu_online_mask, old_cpu); + if (new_cpu >= nr_cpu_ids) + break; + perf_pmu_migrate_context(uncore->pmu, old_cpu, new_cpu); + cpumask_set_cpu(new_cpu, &thunder_active_mask); + break; + default: + break; + } + return NOTIFY_OK; +} + +static struct thunder_uncore_node *alloc_node(struct thunder_uncore *uncore, int node_id, int counters) +{ + struct thunder_uncore_node *node; + + node = kzalloc(sizeof(struct thunder_uncore_node), GFP_KERNEL); + if (!node) + return NULL; + node->num_counters = counters; + INIT_LIST_HEAD(&node->unit_list); + return node; +} + +int __init thunder_uncore_setup(struct thunder_uncore *uncore, int device_id, + unsigned long offset, unsigned long size, + struct pmu *pmu, int counters) +{ + struct thunder_uncore_unit *unit, *tmp; + struct thunder_uncore_node *node; + struct pci_dev *pdev = NULL; + int ret, node_id, found = 0; + + /* detect PCI devices */ + do { + pdev = pci_get_device(PCI_VENDOR_ID_CAVIUM, device_id, pdev); + if (!pdev) + break; + + node_id = dev_to_node(&pdev->dev); + /* + * -1 without NUMA, set to 0 because we always have at + * least node 0. + */ + if (node_id < 0) + node_id = 0; + + /* allocate node if necessary */ + if (!uncore->nodes[node_id]) + uncore->nodes[node_id] = alloc_node(uncore, node_id, counters); + + node = uncore->nodes[node_id]; + if (!node) { + ret = -ENOMEM; + goto fail; + } + + unit = kzalloc(sizeof(struct thunder_uncore_unit), GFP_KERNEL); + if (!unit) { + ret = -ENOMEM; + goto fail; + } + + unit->pdev = pdev; + unit->map = ioremap(pci_resource_start(pdev, 0) + offset, size); + list_add(&unit->entry, &node->unit_list); + node->nr_units++; + found++; + } while (1); + + if (!found) + return -ENODEV; + + /* + * perf PMU is CPU dependent in difference to our uncore devices. + * Just pick a CPU and migrate away if it goes offline. + */ + cpumask_set_cpu(smp_processor_id(), &thunder_active_mask); + + uncore->cpu_nb.notifier_call = thunder_uncore_pmu_cpu_notifier; + uncore->cpu_nb.priority = CPU_PRI_PERF + 1; + ret = register_cpu_notifier(&uncore->cpu_nb); + if (ret) + goto fail; + + ret = perf_pmu_register(pmu, pmu->name, -1); + if (ret) + goto fail_pmu; + + uncore->pmu = pmu; + return 0; + +fail_pmu: + unregister_cpu_notifier(&uncore->cpu_nb); +fail: + node_id = 0; + while (uncore->nodes[node_id]) { + node = uncore->nodes[node_id]; + + list_for_each_entry_safe(unit, tmp, &node->unit_list, entry) { + if (unit->pdev) { + if (unit->map) + iounmap(unit->map); + pci_dev_put(unit->pdev); + } + kfree(unit); + } + kfree(uncore->nodes[node_id]); + node_id++; + } + return ret; +} + +static int __init thunder_uncore_init(void) +{ + unsigned long implementor = read_cpuid_implementor(); + unsigned long part_number = read_cpuid_part_number(); + u32 variant; + + if (implementor != ARM_CPU_IMP_CAVIUM || + part_number != CAVIUM_CPU_PART_THUNDERX) + return -ENODEV; + + /* detect pass2 which contains different counters */ + variant = MIDR_VARIANT(read_cpuid_id()); + if (variant == 1) + thunder_uncore_version = 1; + pr_info("PMU version: %d\n", thunder_uncore_version); + + return 0; +} +late_initcall(thunder_uncore_init); diff --git a/drivers/perf/uncore/uncore_cavium.h b/drivers/perf/uncore/uncore_cavium.h new file mode 100644 index 0000000..c799709 --- /dev/null +++ b/drivers/perf/uncore/uncore_cavium.h @@ -0,0 +1,78 @@ +#include +#include +#include +#include + +#undef pr_fmt +#define pr_fmt(fmt) "thunderx_uncore: " fmt + +enum uncore_type { + NOP_TYPE, +}; + +extern int thunder_uncore_version; + +#define UNCORE_EVENT_ID_MASK 0xffff +#define UNCORE_EVENT_ID_SHIFT 16 + +/* maximum number of parallel hardware counters for all uncore parts */ +#define MAX_COUNTERS 64 + +struct thunder_uncore_unit { + struct list_head entry; + void __iomem *map; + struct pci_dev *pdev; +}; + +struct thunder_uncore_node { + int nr_units; + int num_counters; + struct list_head unit_list; + struct perf_event *events[MAX_COUNTERS]; +}; + +/* generic uncore struct for different pmu types */ +struct thunder_uncore { + int type; + struct pmu *pmu; + int (*event_valid)(u64); + struct notifier_block cpu_nb; + struct thunder_uncore_node *nodes[MAX_NUMNODES]; +}; + +#define EVENT_PTR(_id) (&event_attr_##_id.attr.attr) + +#define EVENT_ATTR(_name, _val) \ +static struct perf_pmu_events_attr event_attr_##_name = { \ + .attr = __ATTR(_name, 0444, thunder_events_sysfs_show, NULL), \ + .event_str = "event=" __stringify(_val), \ +} + +#define EVENT_ATTR_STR(_name, _str) \ +static struct perf_pmu_events_attr event_attr_##_name = { \ + .attr = __ATTR(_name, 0444, thunder_events_sysfs_show, NULL), \ + .event_str = _str, \ +} + +static inline struct thunder_uncore_node *get_node(u64 config, + struct thunder_uncore *uncore) +{ + return uncore->nodes[config >> UNCORE_EVENT_ID_SHIFT]; +} + +#define get_id(config) (config & UNCORE_EVENT_ID_MASK) + +extern struct attribute_group thunder_uncore_attr_group; +extern struct device_attribute format_attr_node; + +/* Prototypes */ +struct thunder_uncore *event_to_thunder_uncore(struct perf_event *event); +void thunder_uncore_del(struct perf_event *event, int flags); +int thunder_uncore_event_init(struct perf_event *event); +void thunder_uncore_read(struct perf_event *event); +int thunder_uncore_setup(struct thunder_uncore *uncore, int id, + unsigned long offset, unsigned long size, + struct pmu *pmu, int counters); +ssize_t thunder_events_sysfs_show(struct device *dev, + struct device_attribute *attr, + char *page); -- 1.9.1 From mboxrd@z Thu Jan 1 00:00:00 1970 From: jglauber@cavium.com (Jan Glauber) Date: Wed, 9 Mar 2016 17:21:03 +0100 Subject: [PATCH v2 1/5] arm64/perf: Basic uncore counter support for Cavium ThunderX In-Reply-To: References: Message-ID: <2588abcb011df6e60cd8212a8583454999180748.1457539622.git.jglauber@cavium.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Provide "uncore" facilities for different non-CPU performance counter units. Based on Intel/AMD uncore pmu support. The uncore drivers cover quite different functionality including L2 Cache, memory controllers and interconnects. The uncore PMUs can be found under /sys/bus/event_source/devices. All counters are exported via sysfs in the corresponding events files under the PMU directory so the perf tool can list the event names. There are some points that are special in this implementation: 1) The PMU detection relies on PCI device detection. If a matching PCI device is found the PMU is created. The code can deal with multiple units of the same type, e.g. more than one memory controller. Note: There is also a CPUID check to determine the CPU variant, this is needed to support different hardware versions that use the same PCI IDs. 2) Counters are summarized across different units of the same type on one NUMA node. For instance L2C TAD 0..7 are presented as a single counter (adding the values from TAD 0 to 7). Although losing the ability to read a single value the merged values are easier to use. 3) NUMA support. The device node id is used to group devices by node so counters on one node can be merged. The NUMA node can be selected via a new sysfs node attribute. Without NUMA support all devices will be on node 0. 4) All counters are 64 bit wide without overflow interrupts. Signed-off-by: Jan Glauber --- drivers/perf/Makefile | 1 + drivers/perf/uncore/Makefile | 1 + drivers/perf/uncore/uncore_cavium.c | 301 ++++++++++++++++++++++++++++++++++++ drivers/perf/uncore/uncore_cavium.h | 78 ++++++++++ 4 files changed, 381 insertions(+) create mode 100644 drivers/perf/uncore/Makefile create mode 100644 drivers/perf/uncore/uncore_cavium.c create mode 100644 drivers/perf/uncore/uncore_cavium.h diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile index acd2397..61b6084 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -1 +1,2 @@ obj-$(CONFIG_ARM_PMU) += arm_pmu.o +obj-$(CONFIG_ARCH_THUNDER) += uncore/ diff --git a/drivers/perf/uncore/Makefile b/drivers/perf/uncore/Makefile new file mode 100644 index 0000000..b9c72c2 --- /dev/null +++ b/drivers/perf/uncore/Makefile @@ -0,0 +1 @@ +obj-$(CONFIG_ARCH_THUNDER) += uncore_cavium.o diff --git a/drivers/perf/uncore/uncore_cavium.c b/drivers/perf/uncore/uncore_cavium.c new file mode 100644 index 0000000..4fd5e45 --- /dev/null +++ b/drivers/perf/uncore/uncore_cavium.c @@ -0,0 +1,301 @@ +/* + * Cavium Thunder uncore PMU support. Derived from Intel and AMD uncore code. + * + * Copyright (C) 2015,2016 Cavium Inc. + * Author: Jan Glauber + */ + +#include +#include +#include + +#include "uncore_cavium.h" + +int thunder_uncore_version; + +struct thunder_uncore *event_to_thunder_uncore(struct perf_event *event) +{ + return NULL; +} + +void thunder_uncore_read(struct perf_event *event) +{ + struct thunder_uncore *uncore = event_to_thunder_uncore(event); + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + struct thunder_uncore_unit *unit; + u64 prev, new = 0; + s64 delta; + + node = get_node(hwc->config, uncore); + + /* + * No counter overflow interrupts so we do not + * have to worry about prev_count changing on us. + */ + prev = local64_read(&hwc->prev_count); + + /* read counter values from all units on the node */ + list_for_each_entry(unit, &node->unit_list, entry) + new += readq(hwc->event_base + unit->map); + + local64_set(&hwc->prev_count, new); + delta = new - prev; + local64_add(delta, &event->count); +} + +void thunder_uncore_del(struct perf_event *event, int flags) +{ + struct thunder_uncore *uncore = event_to_thunder_uncore(event); + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + int i; + + event->pmu->stop(event, PERF_EF_UPDATE); + + /* + * For programmable counters we need to check where we installed it. + * To keep this function generic always test the more complicated + * case (free running counters won't need the loop). + */ + node = get_node(hwc->config, uncore); + for (i = 0; i < node->num_counters; i++) { + if (cmpxchg(&node->events[i], event, NULL) == event) + break; + } + hwc->idx = -1; +} + +int thunder_uncore_event_init(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct thunder_uncore_node *node; + struct thunder_uncore *uncore; + + if (event->attr.type != event->pmu->type) + return -ENOENT; + + /* we do not support sampling */ + if (is_sampling_event(event)) + return -EINVAL; + + /* counters do not have these bits */ + if (event->attr.exclude_user || + event->attr.exclude_kernel || + event->attr.exclude_host || + event->attr.exclude_guest || + event->attr.exclude_hv || + event->attr.exclude_idle) + return -EINVAL; + + /* counters are 64 bit wide and without overflow interrupts */ + + uncore = event_to_thunder_uncore(event); + if (!uncore) + return -ENODEV; + if (!uncore->event_valid(event->attr.config & UNCORE_EVENT_ID_MASK)) + return -EINVAL; + + /* check NUMA node */ + node = get_node(event->attr.config, uncore); + if (!node) { + pr_debug("Invalid numa node selected\n"); + return -EINVAL; + } + + hwc->config = event->attr.config; + hwc->idx = -1; + return 0; +} + +/* + * Thunder uncore events are independent from CPUs. Provide a cpumask + * nevertheless to prevent perf from adding the event per-cpu and just + * set the mask to one online CPU. Use the same cpumask for all uncore + * devices. + */ +static cpumask_t thunder_active_mask; + +static ssize_t thunder_uncore_attr_show_cpumask(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + return cpumap_print_to_pagebuf(true, buf, &thunder_active_mask); +} +static DEVICE_ATTR(cpumask, S_IRUGO, thunder_uncore_attr_show_cpumask, NULL); + +static struct attribute *thunder_uncore_attrs[] = { + &dev_attr_cpumask.attr, + NULL, +}; + +struct attribute_group thunder_uncore_attr_group = { + .attrs = thunder_uncore_attrs, +}; + +ssize_t thunder_events_sysfs_show(struct device *dev, + struct device_attribute *attr, + char *page) +{ + struct perf_pmu_events_attr *pmu_attr = + container_of(attr, struct perf_pmu_events_attr, attr); + + if (pmu_attr->event_str) + return sprintf(page, "%s", pmu_attr->event_str); + + return 0; +} + +/* node attribute depending on number of numa nodes */ +static ssize_t node_show(struct device *dev, struct device_attribute *attr, char *page) +{ + if (NODES_SHIFT) + return sprintf(page, "config:16-%d\n", 16 + NODES_SHIFT - 1); + else + return sprintf(page, "config:16\n"); +} + +struct device_attribute format_attr_node = __ATTR_RO(node); + +static int thunder_uncore_pmu_cpu_notifier(struct notifier_block *nb, + unsigned long action, void *data) +{ + struct thunder_uncore *uncore = container_of(nb, struct thunder_uncore, cpu_nb); + int new_cpu, old_cpu = (long) data; + + switch (action & ~CPU_TASKS_FROZEN) { + case CPU_DOWN_PREPARE: + if (!cpumask_test_and_clear_cpu(old_cpu, &thunder_active_mask)) + break; + new_cpu = cpumask_any_but(cpu_online_mask, old_cpu); + if (new_cpu >= nr_cpu_ids) + break; + perf_pmu_migrate_context(uncore->pmu, old_cpu, new_cpu); + cpumask_set_cpu(new_cpu, &thunder_active_mask); + break; + default: + break; + } + return NOTIFY_OK; +} + +static struct thunder_uncore_node *alloc_node(struct thunder_uncore *uncore, int node_id, int counters) +{ + struct thunder_uncore_node *node; + + node = kzalloc(sizeof(struct thunder_uncore_node), GFP_KERNEL); + if (!node) + return NULL; + node->num_counters = counters; + INIT_LIST_HEAD(&node->unit_list); + return node; +} + +int __init thunder_uncore_setup(struct thunder_uncore *uncore, int device_id, + unsigned long offset, unsigned long size, + struct pmu *pmu, int counters) +{ + struct thunder_uncore_unit *unit, *tmp; + struct thunder_uncore_node *node; + struct pci_dev *pdev = NULL; + int ret, node_id, found = 0; + + /* detect PCI devices */ + do { + pdev = pci_get_device(PCI_VENDOR_ID_CAVIUM, device_id, pdev); + if (!pdev) + break; + + node_id = dev_to_node(&pdev->dev); + /* + * -1 without NUMA, set to 0 because we always have at + * least node 0. + */ + if (node_id < 0) + node_id = 0; + + /* allocate node if necessary */ + if (!uncore->nodes[node_id]) + uncore->nodes[node_id] = alloc_node(uncore, node_id, counters); + + node = uncore->nodes[node_id]; + if (!node) { + ret = -ENOMEM; + goto fail; + } + + unit = kzalloc(sizeof(struct thunder_uncore_unit), GFP_KERNEL); + if (!unit) { + ret = -ENOMEM; + goto fail; + } + + unit->pdev = pdev; + unit->map = ioremap(pci_resource_start(pdev, 0) + offset, size); + list_add(&unit->entry, &node->unit_list); + node->nr_units++; + found++; + } while (1); + + if (!found) + return -ENODEV; + + /* + * perf PMU is CPU dependent in difference to our uncore devices. + * Just pick a CPU and migrate away if it goes offline. + */ + cpumask_set_cpu(smp_processor_id(), &thunder_active_mask); + + uncore->cpu_nb.notifier_call = thunder_uncore_pmu_cpu_notifier; + uncore->cpu_nb.priority = CPU_PRI_PERF + 1; + ret = register_cpu_notifier(&uncore->cpu_nb); + if (ret) + goto fail; + + ret = perf_pmu_register(pmu, pmu->name, -1); + if (ret) + goto fail_pmu; + + uncore->pmu = pmu; + return 0; + +fail_pmu: + unregister_cpu_notifier(&uncore->cpu_nb); +fail: + node_id = 0; + while (uncore->nodes[node_id]) { + node = uncore->nodes[node_id]; + + list_for_each_entry_safe(unit, tmp, &node->unit_list, entry) { + if (unit->pdev) { + if (unit->map) + iounmap(unit->map); + pci_dev_put(unit->pdev); + } + kfree(unit); + } + kfree(uncore->nodes[node_id]); + node_id++; + } + return ret; +} + +static int __init thunder_uncore_init(void) +{ + unsigned long implementor = read_cpuid_implementor(); + unsigned long part_number = read_cpuid_part_number(); + u32 variant; + + if (implementor != ARM_CPU_IMP_CAVIUM || + part_number != CAVIUM_CPU_PART_THUNDERX) + return -ENODEV; + + /* detect pass2 which contains different counters */ + variant = MIDR_VARIANT(read_cpuid_id()); + if (variant == 1) + thunder_uncore_version = 1; + pr_info("PMU version: %d\n", thunder_uncore_version); + + return 0; +} +late_initcall(thunder_uncore_init); diff --git a/drivers/perf/uncore/uncore_cavium.h b/drivers/perf/uncore/uncore_cavium.h new file mode 100644 index 0000000..c799709 --- /dev/null +++ b/drivers/perf/uncore/uncore_cavium.h @@ -0,0 +1,78 @@ +#include +#include +#include +#include + +#undef pr_fmt +#define pr_fmt(fmt) "thunderx_uncore: " fmt + +enum uncore_type { + NOP_TYPE, +}; + +extern int thunder_uncore_version; + +#define UNCORE_EVENT_ID_MASK 0xffff +#define UNCORE_EVENT_ID_SHIFT 16 + +/* maximum number of parallel hardware counters for all uncore parts */ +#define MAX_COUNTERS 64 + +struct thunder_uncore_unit { + struct list_head entry; + void __iomem *map; + struct pci_dev *pdev; +}; + +struct thunder_uncore_node { + int nr_units; + int num_counters; + struct list_head unit_list; + struct perf_event *events[MAX_COUNTERS]; +}; + +/* generic uncore struct for different pmu types */ +struct thunder_uncore { + int type; + struct pmu *pmu; + int (*event_valid)(u64); + struct notifier_block cpu_nb; + struct thunder_uncore_node *nodes[MAX_NUMNODES]; +}; + +#define EVENT_PTR(_id) (&event_attr_##_id.attr.attr) + +#define EVENT_ATTR(_name, _val) \ +static struct perf_pmu_events_attr event_attr_##_name = { \ + .attr = __ATTR(_name, 0444, thunder_events_sysfs_show, NULL), \ + .event_str = "event=" __stringify(_val), \ +} + +#define EVENT_ATTR_STR(_name, _str) \ +static struct perf_pmu_events_attr event_attr_##_name = { \ + .attr = __ATTR(_name, 0444, thunder_events_sysfs_show, NULL), \ + .event_str = _str, \ +} + +static inline struct thunder_uncore_node *get_node(u64 config, + struct thunder_uncore *uncore) +{ + return uncore->nodes[config >> UNCORE_EVENT_ID_SHIFT]; +} + +#define get_id(config) (config & UNCORE_EVENT_ID_MASK) + +extern struct attribute_group thunder_uncore_attr_group; +extern struct device_attribute format_attr_node; + +/* Prototypes */ +struct thunder_uncore *event_to_thunder_uncore(struct perf_event *event); +void thunder_uncore_del(struct perf_event *event, int flags); +int thunder_uncore_event_init(struct perf_event *event); +void thunder_uncore_read(struct perf_event *event); +int thunder_uncore_setup(struct thunder_uncore *uncore, int id, + unsigned long offset, unsigned long size, + struct pmu *pmu, int counters); +ssize_t thunder_events_sysfs_show(struct device *dev, + struct device_attribute *attr, + char *page); -- 1.9.1