Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754145AbdDDK4z (ORCPT ); Tue, 4 Apr 2017 06:56:55 -0400 Received: from mail-pg0-f67.google.com ([74.125.83.67]:33421 "EHLO mail-pg0-f67.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753408AbdDDK4w (ORCPT ); Tue, 4 Apr 2017 06:56:52 -0400 From: Anurup M X-Google-Original-From: Anurup M To: mark.rutland@arm.com, will.deacon@arm.com Cc: linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org, anurup.m@huawei.com, zhangshaokun@hisilicon.com, tanxiaojun@huawei.com, xuwei5@hisilicon.com, sanil.kumar@hisilicon.com, john.garry@huawei.com, gabriele.paoloni@huawei.com, shiju.jose@huawei.com, huangdaode@hisilicon.com, linuxarm@huawei.com, dikshit.n@huawei.com, shyju.pv@huawei.com, anurupvasu@gmail.com Subject: [PATCH v7 8/9] drivers: perf: hisi: Miscellanous node(MN) event counting in perf Date: Tue, 4 Apr 2017 06:56:31 -0400 Message-Id: <1491303391-140554-1-git-send-email-anurup.m@huawei.com> X-Mailer: git-send-email 2.1.4 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 14836 Lines: 512 From: Shaokun Zhang 1. Add support to count MN hardware events. 2. Mn events are listed in sysfs at /sys/devices/hisi_mn_2/events/ The events can be selected as shown in perf list e.g.: For MN_READ_REQUEST event for Super CPU cluster 2 the event format is -e "hisi_mn_2/read_req/" 3. MN PMU doesnot support counter overflow IRQ in HiP05/06/07, So use hrtimer to poll and avoid counter overflow. Signed-off-by: Shaokun Zhang Signed-off-by: Dikshit N Signed-off-by: Anurup M --- drivers/perf/hisilicon/Makefile | 2 +- drivers/perf/hisilicon/hisi_uncore_mn.c | 478 ++++++++++++++++++++++++++++++++ 2 files changed, 479 insertions(+), 1 deletion(-) create mode 100644 drivers/perf/hisilicon/hisi_uncore_mn.c diff --git a/drivers/perf/hisilicon/Makefile b/drivers/perf/hisilicon/Makefile index 0887b56..26b2507 100644 --- a/drivers/perf/hisilicon/Makefile +++ b/drivers/perf/hisilicon/Makefile @@ -1 +1 @@ -obj-$(CONFIG_HISI_PMU) += djtag.o hisi_uncore_pmu.o hisi_uncore_l3c.o +obj-$(CONFIG_HISI_PMU) += djtag.o hisi_uncore_pmu.o hisi_uncore_l3c.o hisi_uncore_mn.o diff --git a/drivers/perf/hisilicon/hisi_uncore_mn.c b/drivers/perf/hisilicon/hisi_uncore_mn.c new file mode 100644 index 0000000..63a4060 --- /dev/null +++ b/drivers/perf/hisilicon/hisi_uncore_mn.c @@ -0,0 +1,478 @@ +/* + * HiSilicon SoC MN Hardware event counters support + * + * Copyright (C) 2017 Hisilicon Limited + * Author: Shaokun Zhang + * + * This code is based on the uncore PMUs like arm-cci and + * arm-ccn. + * + * 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. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + */ +#include +#include +#include +#include +#include +#include +#include "hisi_uncore_pmu.h" + +/* + * ARMv8 HiSilicon MN event types. + */ +enum armv8_hisi_mn_event_types { + HISI_HWEVENT_MN_EO_BARR_REQ = 0x0, + HISI_HWEVENT_MN_EC_BARR_REQ = 0x01, + HISI_HWEVENT_MN_DVM_OP_REQ = 0x02, + HISI_HWEVENT_MN_DVM_SYNC_REQ = 0x03, + HISI_HWEVENT_MN_READ_REQ = 0x04, + HISI_HWEVENT_MN_WRITE_REQ = 0x05, + HISI_HWEVENT_MN_EVENT_MAX = 0x08, +}; + +/* + * ARMv8 HiSilicon Hardware counter Index. + */ +enum armv8_hisi_mn_counters { + HISI_IDX_MN_COUNTER0 = 0x0, + HISI_IDX_MN_COUNTER_MAX = 0x4, +}; + +#define MN1_EVTYPE_REG_OFF 0x48 +#define MN1_EVCTRL_REG_OFF 0x40 +#define MN1_CNT0_REG_OFF 0x30 +#define MN1_EVENT_EN 0x01 +#define MN1_BANK_SELECT 0x01 + +#define GET_MODULE_ID(hwmod_data) hwmod_data->module_id + +/* + * Default timer frequency to poll and avoid counter overflow. + * CPU speed = 2.4Ghz and number of CPU cores in a SCCL is 16. + * For a single MN event on a CPU core consumes 200 cycles. + * So overflow time = (2^31 * 200) / (16 * 2.4G) which is about 21 seconds + * So on a safe side we use a timer interval of 8sec + */ +#define MN1_HRTIMER_INTERVAL (8LL * MSEC_PER_SEC) + +struct hisi_mn_data { + struct hisi_djtag_client *client; + u32 module_id; +}; + +static inline int hisi_mn_counter_valid(int idx) +{ + return (idx >= HISI_IDX_MN_COUNTER0 && + idx < HISI_IDX_MN_COUNTER_MAX); +} + +/* Select the counter register offset from the index */ +static inline u32 get_counter_reg_off(int cntr_idx) +{ + return (MN1_CNT0_REG_OFF + (cntr_idx * 4)); +} + +static u32 hisi_mn_read_counter(struct hisi_mn_data *mn_data, int cntr_idx) +{ + struct hisi_djtag_client *client = mn_data->client; + u32 module_id = GET_MODULE_ID(mn_data); + u32 reg_off, value; + + reg_off = get_counter_reg_off(cntr_idx); + hisi_djtag_readreg(module_id, MN1_BANK_SELECT, reg_off, + client, &value); + + return value; +} + +static u64 hisi_mn_event_update(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + struct hisi_pmu *mn_pmu = to_hisi_pmu(event->pmu); + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + u64 delta, prev_raw_count, new_raw_count = 0; + + if (!hisi_mn_counter_valid(idx)) { + dev_err(mn_pmu->dev, + "Unsupported event index:%d!\n", idx); + return 0; + } + + do { + /* Get count from the MN */ + prev_raw_count = local64_read(&hwc->prev_count); + new_raw_count = hisi_mn_read_counter(mn_data, idx); + } while (local64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count); + + /* + * compute the delta and add to event->count + */ + delta = (new_raw_count - prev_raw_count) & HISI_MAX_PERIOD; + local64_add(delta, &event->count); + + return new_raw_count; +} + +static void hisi_mn_set_evtype(struct hisi_pmu *mn_pmu, int idx, u32 val) +{ + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + struct hisi_djtag_client *client = mn_data->client; + u32 module_id = GET_MODULE_ID(mn_data); + u32 event_value, value = 0; + + event_value = (val - HISI_HWEVENT_MN_EO_BARR_REQ); + + /* + * Write the event code in event select register. + * Each byte in the 32 bit event select register is used + * to configure the event code. Each byte correspond to a + * counter register to use. + */ + val = event_value << (8 * idx); + + hisi_djtag_readreg(module_id, MN1_BANK_SELECT, MN1_EVTYPE_REG_OFF, + client, &value); + value &= ~(0xff << (8 * idx)); + value |= val; + hisi_djtag_writereg(module_id, MN1_BANK_SELECT, MN1_EVTYPE_REG_OFF, + value, client); +} + +static void hisi_mn_clear_evtype(struct hisi_pmu *mn_pmu, int idx) +{ + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + struct hisi_djtag_client *client = mn_data->client; + u32 module_id = GET_MODULE_ID(mn_data); + u32 value; + + if (!hisi_mn_counter_valid(idx)) { + dev_err(mn_pmu->dev, + "Unsupported event index:%d!\n", idx); + return; + } + + /* + * Clear the event code in event select register by writing value 0xff. + * Each byte in the 32 bit event select register is used to configure + * the event code. Each byte correspond to a counter register to use. + */ + hisi_djtag_readreg(module_id, MN1_BANK_SELECT, MN1_EVTYPE_REG_OFF, + client, &value); + value &= ~(0xff << (8 * idx)); + value |= (0xff << (8 * idx)); + hisi_djtag_writereg(module_id, MN1_BANK_SELECT, MN1_EVTYPE_REG_OFF, + value, client); +} + +static void hisi_mn_write_counter(struct hisi_pmu *mn_pmu, + struct hw_perf_event *hwc, u32 value) +{ + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + struct hisi_djtag_client *client = mn_data->client; + u32 module_id = GET_MODULE_ID(mn_data); + u32 reg_off; + int idx = GET_CNTR_IDX(hwc); + + reg_off = get_counter_reg_off(idx); + hisi_djtag_writereg(module_id, MN1_BANK_SELECT, reg_off, value, client); +} + +static void hisi_mn_start_counters(struct hisi_pmu *mn_pmu) +{ + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + struct hisi_djtag_client *client = mn_data->client; + unsigned long *used_mask = mn_pmu->pmu_events.used_mask; + u32 module_id = GET_MODULE_ID(mn_data); + u32 num_counters = mn_pmu->num_counters; + u32 value; + int enabled = bitmap_weight(used_mask, num_counters); + + if (!enabled) + return; + + /* Set the event_bus_en bit in MN_EVENT_CTRL to start counting */ + hisi_djtag_readreg(module_id, MN1_BANK_SELECT, MN1_EVCTRL_REG_OFF, + client, &value); + value |= MN1_EVENT_EN; + hisi_djtag_writereg(module_id, MN1_BANK_SELECT, MN1_EVCTRL_REG_OFF, + value, client); +} + +static void hisi_mn_stop_counters(struct hisi_pmu *mn_pmu) +{ + struct hisi_mn_data *mn_data = mn_pmu->hwmod_data; + struct hisi_djtag_client *client = mn_data->client; + u32 module_id = GET_MODULE_ID(mn_data); + u32 value; + + /* + * Clear the event_bus_en bit in MN_EVENT_CTRL + */ + hisi_djtag_readreg(module_id, MN1_BANK_SELECT, MN1_EVCTRL_REG_OFF, + client, &value); + value &= ~(MN1_EVENT_EN); + hisi_djtag_writereg(module_id, MN1_BANK_SELECT, MN1_EVCTRL_REG_OFF, + value, client); +} + +static void hisi_mn_clear_event_idx(struct hisi_pmu *mn_pmu, int idx) +{ + if (!hisi_mn_counter_valid(idx)) { + dev_err(mn_pmu->dev, "Unsupported event index:%d!\n", idx); + return; + } + clear_bit(idx, mn_pmu->pmu_events.used_mask); +} + +static int hisi_mn_get_event_idx(struct hisi_pmu *mn_pmu) +{ + const unsigned long *used_mask = mn_pmu->pmu_events.used_mask; + u32 num_counters = mn_pmu->num_counters; + int event_idx; + + event_idx = find_first_zero_bit(used_mask, num_counters); + if (event_idx == num_counters) + return -EAGAIN; + + set_bit(event_idx, mn_pmu->pmu_events.used_mask); + + return event_idx; +} + +static const struct of_device_id mn_of_match[] = { + { .compatible = "hisilicon,hip05-pmu-mn-v1", }, + { .compatible = "hisilicon,hip06-pmu-mn-v1", }, + { .compatible = "hisilicon,hip07-pmu-mn-v2", }, + {}, +}; +MODULE_DEVICE_TABLE(of, mn_of_match); + +/* Use hrtimer when no IRQ, to poll for avoiding counter overflow */ +static void hisi_mn_hrtimer_init(struct hisi_pmu *mn_pmu) +{ + INIT_LIST_HEAD(&mn_pmu->active_list); + mn_pmu->ops->start_hrtimer = hisi_hrtimer_start; + mn_pmu->ops->stop_hrtimer = hisi_hrtimer_stop; + hisi_hrtimer_init(mn_pmu, MN1_HRTIMER_INTERVAL); +} + +static int hisi_mn_init_data(struct hisi_pmu *mn_pmu, + struct hisi_djtag_client *client) +{ + struct hisi_mn_data *mn_data; + struct device *dev = &client->dev; + const struct of_device_id *of_id; + int ret; + + mn_data = devm_kzalloc(dev, sizeof(*mn_data), GFP_KERNEL); + if (!mn_data) + return -ENOMEM; + + /* Set the djtag Identifier */ + mn_data->client = client; + mn_pmu->hwmod_data = mn_data; + + of_id = of_match_device(mn_of_match, dev); + if (!of_id) { + dev_err(dev, "DT: Match device fail!\n"); + return -EINVAL; + } + + ret = device_property_read_u32(dev, "hisilicon,module-id", + &mn_data->module_id); + if (ret < 0) { + dev_err(dev, "DT: Could not read module-id!\n"); + return -EINVAL; + } + + /* + * Use poll method to avoid counter overflow when + * overflow IRQ is not supported in hardware. + */ + hisi_mn_hrtimer_init(mn_pmu); + + return 0; +} + +static struct attribute *hisi_mn_format_attr[] = { + HISI_PMU_FORMAT_ATTR(event, "config:0-7"), + NULL, +}; + +static const struct attribute_group hisi_mn_format_group = { + .name = "format", + .attrs = hisi_mn_format_attr, +}; + +static struct attribute *hisi_mn_events_attr[] = { + HISI_PMU_EVENT_ATTR_STR(eo_barrier_req, "event=0x0"), + HISI_PMU_EVENT_ATTR_STR(ec_barrier_req, "event=0x01"), + HISI_PMU_EVENT_ATTR_STR(dvm_op_req, "event=0x02"), + HISI_PMU_EVENT_ATTR_STR(dvm_sync_req, "event=0x03"), + HISI_PMU_EVENT_ATTR_STR(read_req, "event=0x04"), + HISI_PMU_EVENT_ATTR_STR(write_req, "event=0x05"), + NULL, +}; + +static const struct attribute_group hisi_mn_events_group = { + .name = "events", + .attrs = hisi_mn_events_attr, +}; + +static struct attribute *hisi_mn_attrs[] = { + NULL, +}; + +static const struct attribute_group hisi_mn_attr_group = { + .attrs = hisi_mn_attrs, +}; + +static DEVICE_ATTR(cpumask, 0444, hisi_cpumask_sysfs_show, NULL); + +static struct attribute *hisi_mn_cpumask_attrs[] = { + &dev_attr_cpumask.attr, + NULL, +}; + +static const struct attribute_group hisi_mn_cpumask_attr_group = { + .attrs = hisi_mn_cpumask_attrs, +}; + +static const struct attribute_group *hisi_mn_pmu_attr_groups[] = { + &hisi_mn_attr_group, + &hisi_mn_format_group, + &hisi_mn_events_group, + &hisi_mn_cpumask_attr_group, + NULL, +}; + +static struct hisi_uncore_ops hisi_uncore_mn_ops = { + .set_evtype = hisi_mn_set_evtype, + .clear_evtype = hisi_mn_clear_evtype, + .set_event_period = hisi_pmu_set_event_period, + .get_event_idx = hisi_mn_get_event_idx, + .clear_event_idx = hisi_mn_clear_event_idx, + .event_update = hisi_mn_event_update, + .start_counters = hisi_mn_start_counters, + .stop_counters = hisi_mn_stop_counters, + .write_counter = hisi_mn_write_counter, +}; + +static void hisi_mn_pmu_init(struct hisi_pmu *mn_pmu, + struct hisi_djtag_client *client) +{ + struct device *dev = &client->dev; + + mn_pmu->num_events = HISI_HWEVENT_MN_EVENT_MAX; + mn_pmu->num_counters = HISI_IDX_MN_COUNTER_MAX; + mn_pmu->scl_id = hisi_djtag_get_sclid(client); + + mn_pmu->name = devm_kasprintf(dev, GFP_KERNEL, "hisi_mn_%d", + mn_pmu->scl_id); + mn_pmu->ops = &hisi_uncore_mn_ops; + mn_pmu->dev = dev; + + /* Pick one core to use for cpumask attributes */ + cpumask_set_cpu(smp_processor_id(), &mn_pmu->cpus); +} + +static int hisi_pmu_mn_dev_probe(struct hisi_djtag_client *client) +{ + struct hisi_pmu *mn_pmu; + struct device *dev = &client->dev; + int ret; + + mn_pmu = hisi_pmu_alloc(dev, HISI_IDX_MN_COUNTER_MAX); + if (!mn_pmu) + return -ENOMEM; + + hisi_mn_pmu_init(mn_pmu, client); + + ret = hisi_mn_init_data(mn_pmu, client); + if (ret) + goto fail; + + mn_pmu->pmu = (struct pmu) { + .name = mn_pmu->name, + .task_ctx_nr = perf_invalid_context, + .event_init = hisi_uncore_pmu_event_init, + .pmu_enable = hisi_uncore_pmu_enable, + .pmu_disable = hisi_uncore_pmu_disable, + .add = hisi_uncore_pmu_add, + .del = hisi_uncore_pmu_del, + .start = hisi_uncore_pmu_start, + .stop = hisi_uncore_pmu_stop, + .read = hisi_uncore_pmu_read, + .attr_groups = hisi_mn_pmu_attr_groups, + }; + + ret = hisi_uncore_pmu_setup(mn_pmu, mn_pmu->name); + if (ret) { + dev_err(mn_pmu->dev, "hisi_uncore_pmu_init FAILED!!\n"); + goto fail; + } + + /* Set the drv data to MN pmu */ + dev_set_drvdata(dev, mn_pmu); + + return 0; + +fail: + return ret; +} + +static int hisi_pmu_mn_dev_remove(struct hisi_djtag_client *client) +{ + struct hisi_pmu *mn_pmu; + struct device *dev = &client->dev; + + mn_pmu = dev_get_drvdata(dev); + perf_pmu_unregister(&mn_pmu->pmu); + kfree(mn_pmu->name); + + return 0; +} + +static struct hisi_djtag_driver hisi_pmu_mn_driver = { + .driver = { + .name = "hisi-pmu-mn", + .of_match_table = mn_of_match, + }, + .probe = hisi_pmu_mn_dev_probe, + .remove = hisi_pmu_mn_dev_remove, +}; + +static int __init hisi_pmu_mn_init(void) +{ + int ret; + + ret = hisi_djtag_register_driver(THIS_MODULE, &hisi_pmu_mn_driver); + if (ret < 0) { + pr_err("hisi pmu MN init failed, ret=%d\n", ret); + return ret; + } + + return 0; +} +module_init(hisi_pmu_mn_init); + +static void __exit hisi_pmu_mn_exit(void) +{ + hisi_djtag_unregister_driver(&hisi_pmu_mn_driver); +} +module_exit(hisi_pmu_mn_exit); + +MODULE_DESCRIPTION("HiSilicon SoC HIP0x MN PMU driver"); +MODULE_LICENSE("GPL v2"); +MODULE_AUTHOR("Shaokun Zhang"); -- 2.1.4