diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2017-11-15 10:56:56 -0800 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2017-11-15 10:56:56 -0800 |
commit | c9b012e5f4a1d01dfa8abc6318211a67ba7d5db2 (patch) | |
tree | 97b2f1c654fc4333e9e3111f76a26ec5503ee5b9 /drivers/perf | |
parent | b293fca43be544483b6488d33ad4b3ed55881064 (diff) | |
parent | 6cfa7cc46b1a7a15d81d5389c99cfca633c12b8e (diff) |
Merge tag 'arm64-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Pull arm64 updates from Will Deacon:
"The big highlight is support for the Scalable Vector Extension (SVE)
which required extensive ABI work to ensure we don't break existing
applications by blowing away their signal stack with the rather large
new vector context (<= 2 kbit per vector register). There's further
work to be done optimising things like exception return, but the ABI
is solid now.
Much of the line count comes from some new PMU drivers we have, but
they're pretty self-contained and I suspect we'll have more of them in
future.
Plenty of acronym soup here:
- initial support for the Scalable Vector Extension (SVE)
- improved handling for SError interrupts (required to handle RAS
events)
- enable GCC support for 128-bit integer types
- remove kernel text addresses from backtraces and register dumps
- use of WFE to implement long delay()s
- ACPI IORT updates from Lorenzo Pieralisi
- perf PMU driver for the Statistical Profiling Extension (SPE)
- perf PMU driver for Hisilicon's system PMUs
- misc cleanups and non-critical fixes"
* tag 'arm64-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux: (97 commits)
arm64: Make ARMV8_DEPRECATED depend on SYSCTL
arm64: Implement __lshrti3 library function
arm64: support __int128 on gcc 5+
arm64/sve: Add documentation
arm64/sve: Detect SVE and activate runtime support
arm64/sve: KVM: Hide SVE from CPU features exposed to guests
arm64/sve: KVM: Treat guest SVE use as undefined instruction execution
arm64/sve: KVM: Prevent guests from using SVE
arm64/sve: Add sysctl to set the default vector length for new processes
arm64/sve: Add prctl controls for userspace vector length management
arm64/sve: ptrace and ELF coredump support
arm64/sve: Preserve SVE registers around EFI runtime service calls
arm64/sve: Preserve SVE registers around kernel-mode NEON use
arm64/sve: Probe SVE capabilities and usable vector lengths
arm64: cpufeature: Move sys_caps_initialised declarations
arm64/sve: Backend logic for setting the vector length
arm64/sve: Signal handling support
arm64/sve: Support vector length resetting for new processes
arm64/sve: Core task context handling
arm64/sve: Low-level CPU setup
...
Diffstat (limited to 'drivers/perf')
-rw-r--r-- | drivers/perf/Kconfig | 15 | ||||
-rw-r--r-- | drivers/perf/Makefile | 2 | ||||
-rw-r--r-- | drivers/perf/arm_pmu.c | 10 | ||||
-rw-r--r-- | drivers/perf/arm_pmu_acpi.c | 3 | ||||
-rw-r--r-- | drivers/perf/arm_pmu_platform.c | 4 | ||||
-rw-r--r-- | drivers/perf/arm_spe_pmu.c | 1249 | ||||
-rw-r--r-- | drivers/perf/hisilicon/Makefile | 1 | ||||
-rw-r--r-- | drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 463 | ||||
-rw-r--r-- | drivers/perf/hisilicon/hisi_uncore_hha_pmu.c | 473 | ||||
-rw-r--r-- | drivers/perf/hisilicon/hisi_uncore_l3c_pmu.c | 463 | ||||
-rw-r--r-- | drivers/perf/hisilicon/hisi_uncore_pmu.c | 447 | ||||
-rw-r--r-- | drivers/perf/hisilicon/hisi_uncore_pmu.h | 102 | ||||
-rw-r--r-- | drivers/perf/qcom_l2_pmu.c | 54 |
13 files changed, 3276 insertions, 10 deletions
diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig index e5197ffb7422..b8f44b068fc6 100644 --- a/drivers/perf/Kconfig +++ b/drivers/perf/Kconfig @@ -17,6 +17,13 @@ config ARM_PMU_ACPI depends on ARM_PMU && ACPI def_bool y +config HISI_PMU + bool "HiSilicon SoC PMU" + depends on ARM64 && ACPI + help + Support for HiSilicon SoC uncore performance monitoring + unit (PMU), such as: L3C, HHA and DDRC. + config QCOM_L2_PMU bool "Qualcomm Technologies L2-cache PMU" depends on ARCH_QCOM && ARM64 && ACPI @@ -43,4 +50,12 @@ config XGENE_PMU help Say y if you want to use APM X-Gene SoC performance monitors. +config ARM_SPE_PMU + tristate "Enable support for the ARMv8.2 Statistical Profiling Extension" + depends on PERF_EVENTS && ARM64 + help + Enable perf support for the ARMv8.2 Statistical Profiling + Extension, which provides periodic sampling of operations in + the CPU pipeline and reports this via the perf AUX interface. + endmenu diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile index 9402dc8ff22c..710a0135bd61 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -1,6 +1,8 @@ # SPDX-License-Identifier: GPL-2.0 obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o +obj-$(CONFIG_HISI_PMU) += hisilicon/ obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o +obj-$(CONFIG_ARM_SPE_PMU) += arm_spe_pmu.o diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c index d14fc2e67f93..7bc5eee96b31 100644 --- a/drivers/perf/arm_pmu.c +++ b/drivers/perf/arm_pmu.c @@ -539,7 +539,7 @@ void armpmu_free_irq(struct arm_pmu *armpmu, int cpu) if (!cpumask_test_and_clear_cpu(cpu, &armpmu->active_irqs)) return; - if (irq_is_percpu(irq)) { + if (irq_is_percpu_devid(irq)) { free_percpu_irq(irq, &hw_events->percpu_pmu); cpumask_clear(&armpmu->active_irqs); return; @@ -565,10 +565,10 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) if (!irq) return 0; - if (irq_is_percpu(irq) && cpumask_empty(&armpmu->active_irqs)) { + if (irq_is_percpu_devid(irq) && cpumask_empty(&armpmu->active_irqs)) { err = request_percpu_irq(irq, handler, "arm-pmu", &hw_events->percpu_pmu); - } else if (irq_is_percpu(irq)) { + } else if (irq_is_percpu_devid(irq)) { int other_cpu = cpumask_first(&armpmu->active_irqs); int other_irq = per_cpu(hw_events->irq, other_cpu); @@ -649,7 +649,7 @@ static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node) irq = armpmu_get_cpu_irq(pmu, cpu); if (irq) { - if (irq_is_percpu(irq)) { + if (irq_is_percpu_devid(irq)) { enable_percpu_irq(irq, IRQ_TYPE_NONE); return 0; } @@ -667,7 +667,7 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node) return 0; irq = armpmu_get_cpu_irq(pmu, cpu); - if (irq && irq_is_percpu(irq)) + if (irq && irq_is_percpu_devid(irq)) disable_percpu_irq(irq); return 0; diff --git a/drivers/perf/arm_pmu_acpi.c b/drivers/perf/arm_pmu_acpi.c index 3303dd8d8eb5..705f1a390e31 100644 --- a/drivers/perf/arm_pmu_acpi.c +++ b/drivers/perf/arm_pmu_acpi.c @@ -193,9 +193,6 @@ int arm_pmu_acpi_probe(armpmu_init_fn init_fn) int pmu_idx = 0; int cpu, ret; - if (acpi_disabled) - return 0; - /* * Initialise and register the set of PMUs which we know about right * now. Ideally we'd do this in arm_pmu_acpi_cpu_starting() so that we diff --git a/drivers/perf/arm_pmu_platform.c b/drivers/perf/arm_pmu_platform.c index 4428852e1da1..91b224eced18 100644 --- a/drivers/perf/arm_pmu_platform.c +++ b/drivers/perf/arm_pmu_platform.c @@ -127,7 +127,7 @@ static int pmu_parse_irqs(struct arm_pmu *pmu) if (num_irqs == 1) { int irq = platform_get_irq(pdev, 0); - if (irq && irq_is_percpu(irq)) + if (irq && irq_is_percpu_devid(irq)) return pmu_parse_percpu_irq(pmu, irq); } @@ -150,7 +150,7 @@ static int pmu_parse_irqs(struct arm_pmu *pmu) if (WARN_ON(irq <= 0)) continue; - if (irq_is_percpu(irq)) { + if (irq_is_percpu_devid(irq)) { pr_warn("multiple PPIs or mismatched SPI/PPI detected\n"); return -EINVAL; } diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c new file mode 100644 index 000000000000..8ce262fc2561 --- /dev/null +++ b/drivers/perf/arm_spe_pmu.c @@ -0,0 +1,1249 @@ +/* + * Perf support for the Statistical Profiling Extension, introduced as + * part of ARMv8.2. + * + * 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 <http://www.gnu.org/licenses/>. + * + * Copyright (C) 2016 ARM Limited + * + * Author: Will Deacon <will.deacon@arm.com> + */ + +#define PMUNAME "arm_spe" +#define DRVNAME PMUNAME "_pmu" +#define pr_fmt(fmt) DRVNAME ": " fmt + +#include <linux/cpuhotplug.h> +#include <linux/interrupt.h> +#include <linux/irq.h> +#include <linux/module.h> +#include <linux/of_address.h> +#include <linux/of_device.h> +#include <linux/perf_event.h> +#include <linux/platform_device.h> +#include <linux/slab.h> + +#include <asm/sysreg.h> + +#define ARM_SPE_BUF_PAD_BYTE 0 + +struct arm_spe_pmu_buf { + int nr_pages; + bool snapshot; + void *base; +}; + +struct arm_spe_pmu { + struct pmu pmu; + struct platform_device *pdev; + cpumask_t supported_cpus; + struct hlist_node hotplug_node; + + int irq; /* PPI */ + + u16 min_period; + u16 counter_sz; + +#define SPE_PMU_FEAT_FILT_EVT (1UL << 0) +#define SPE_PMU_FEAT_FILT_TYP (1UL << 1) +#define SPE_PMU_FEAT_FILT_LAT (1UL << 2) +#define SPE_PMU_FEAT_ARCH_INST (1UL << 3) +#define SPE_PMU_FEAT_LDS (1UL << 4) +#define SPE_PMU_FEAT_ERND (1UL << 5) +#define SPE_PMU_FEAT_DEV_PROBED (1UL << 63) + u64 features; + + u16 max_record_sz; + u16 align; + struct perf_output_handle __percpu *handle; +}; + +#define to_spe_pmu(p) (container_of(p, struct arm_spe_pmu, pmu)) + +/* Convert a free-running index from perf into an SPE buffer offset */ +#define PERF_IDX2OFF(idx, buf) ((idx) % ((buf)->nr_pages << PAGE_SHIFT)) + +/* Keep track of our dynamic hotplug state */ +static enum cpuhp_state arm_spe_pmu_online; + +enum arm_spe_pmu_buf_fault_action { + SPE_PMU_BUF_FAULT_ACT_SPURIOUS, + SPE_PMU_BUF_FAULT_ACT_FATAL, + SPE_PMU_BUF_FAULT_ACT_OK, +}; + +/* This sysfs gunk was really good fun to write. */ +enum arm_spe_pmu_capabilities { + SPE_PMU_CAP_ARCH_INST = 0, + SPE_PMU_CAP_ERND, + SPE_PMU_CAP_FEAT_MAX, + SPE_PMU_CAP_CNT_SZ = SPE_PMU_CAP_FEAT_MAX, + SPE_PMU_CAP_MIN_IVAL, +}; + +static int arm_spe_pmu_feat_caps[SPE_PMU_CAP_FEAT_MAX] = { + [SPE_PMU_CAP_ARCH_INST] = SPE_PMU_FEAT_ARCH_INST, + [SPE_PMU_CAP_ERND] = SPE_PMU_FEAT_ERND, +}; + +static u32 arm_spe_pmu_cap_get(struct arm_spe_pmu *spe_pmu, int cap) +{ + if (cap < SPE_PMU_CAP_FEAT_MAX) + return !!(spe_pmu->features & arm_spe_pmu_feat_caps[cap]); + + switch (cap) { + case SPE_PMU_CAP_CNT_SZ: + return spe_pmu->counter_sz; + case SPE_PMU_CAP_MIN_IVAL: + return spe_pmu->min_period; + default: + WARN(1, "unknown cap %d\n", cap); + } + + return 0; +} + +static ssize_t arm_spe_pmu_cap_show(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct platform_device *pdev = to_platform_device(dev); + struct arm_spe_pmu *spe_pmu = platform_get_drvdata(pdev); + struct dev_ext_attribute *ea = + container_of(attr, struct dev_ext_attribute, attr); + int cap = (long)ea->var; + + return snprintf(buf, PAGE_SIZE, "%u\n", + arm_spe_pmu_cap_get(spe_pmu, cap)); +} + +#define SPE_EXT_ATTR_ENTRY(_name, _func, _var) \ + &((struct dev_ext_attribute[]) { \ + { __ATTR(_name, S_IRUGO, _func, NULL), (void *)_var } \ + })[0].attr.attr + +#define SPE_CAP_EXT_ATTR_ENTRY(_name, _var) \ + SPE_EXT_ATTR_ENTRY(_name, arm_spe_pmu_cap_show, _var) + +static struct attribute *arm_spe_pmu_cap_attr[] = { + SPE_CAP_EXT_ATTR_ENTRY(arch_inst, SPE_PMU_CAP_ARCH_INST), + SPE_CAP_EXT_ATTR_ENTRY(ernd, SPE_PMU_CAP_ERND), + SPE_CAP_EXT_ATTR_ENTRY(count_size, SPE_PMU_CAP_CNT_SZ), + SPE_CAP_EXT_ATTR_ENTRY(min_interval, SPE_PMU_CAP_MIN_IVAL), + NULL, +}; + +static struct attribute_group arm_spe_pmu_cap_group = { + .name = "caps", + .attrs = arm_spe_pmu_cap_attr, +}; + +/* User ABI */ +#define ATTR_CFG_FLD_ts_enable_CFG config /* PMSCR_EL1.TS */ +#define ATTR_CFG_FLD_ts_enable_LO 0 +#define ATTR_CFG_FLD_ts_enable_HI 0 +#define ATTR_CFG_FLD_pa_enable_CFG config /* PMSCR_EL1.PA */ +#define ATTR_CFG_FLD_pa_enable_LO 1 +#define ATTR_CFG_FLD_pa_enable_HI 1 +#define ATTR_CFG_FLD_pct_enable_CFG config /* PMSCR_EL1.PCT */ +#define ATTR_CFG_FLD_pct_enable_LO 2 +#define ATTR_CFG_FLD_pct_enable_HI 2 +#define ATTR_CFG_FLD_jitter_CFG config /* PMSIRR_EL1.RND */ +#define ATTR_CFG_FLD_jitter_LO 16 +#define ATTR_CFG_FLD_jitter_HI 16 +#define ATTR_CFG_FLD_branch_filter_CFG config /* PMSFCR_EL1.B */ +#define ATTR_CFG_FLD_branch_filter_LO 32 +#define ATTR_CFG_FLD_branch_filter_HI 32 +#define ATTR_CFG_FLD_load_filter_CFG config /* PMSFCR_EL1.LD */ +#define ATTR_CFG_FLD_load_filter_LO 33 +#define ATTR_CFG_FLD_load_filter_HI 33 +#define ATTR_CFG_FLD_store_filter_CFG config /* PMSFCR_EL1.ST */ +#define ATTR_CFG_FLD_store_filter_LO 34 +#define ATTR_CFG_FLD_store_filter_HI 34 + +#define ATTR_CFG_FLD_event_filter_CFG config1 /* PMSEVFR_EL1 */ +#define ATTR_CFG_FLD_event_filter_LO 0 +#define ATTR_CFG_FLD_event_filter_HI 63 + +#define ATTR_CFG_FLD_min_latency_CFG config2 /* PMSLATFR_EL1.MINLAT */ +#define ATTR_CFG_FLD_min_latency_LO 0 +#define ATTR_CFG_FLD_min_latency_HI 11 + +/* Why does everything I do descend into this? */ +#define __GEN_PMU_FORMAT_ATTR(cfg, lo, hi) \ + (lo) == (hi) ? #cfg ":" #lo "\n" : #cfg ":" #lo "-" #hi + +#define _GEN_PMU_FORMAT_ATTR(cfg, lo, hi) \ + __GEN_PMU_FORMAT_ATTR(cfg, lo, hi) + +#define GEN_PMU_FORMAT_ATTR(name) \ + PMU_FORMAT_ATTR(name, \ + _GEN_PMU_FORMAT_ATTR(ATTR_CFG_FLD_##name##_CFG, \ + ATTR_CFG_FLD_##name##_LO, \ + ATTR_CFG_FLD_##name##_HI)) + +#define _ATTR_CFG_GET_FLD(attr, cfg, lo, hi) \ + ((((attr)->cfg) >> lo) & GENMASK(hi - lo, 0)) + +#define ATTR_CFG_GET_FLD(attr, name) \ + _ATTR_CFG_GET_FLD(attr, \ + ATTR_CFG_FLD_##name##_CFG, \ + ATTR_CFG_FLD_##name##_LO, \ + ATTR_CFG_FLD_##name##_HI) + +GEN_PMU_FORMAT_ATTR(ts_enable); +GEN_PMU_FORMAT_ATTR(pa_enable); +GEN_PMU_FORMAT_ATTR(pct_enable); +GEN_PMU_FORMAT_ATTR(jitter); +GEN_PMU_FORMAT_ATTR(branch_filter); +GEN_PMU_FORMAT_ATTR(load_filter); +GEN_PMU_FORMAT_ATTR(store_filter); +GEN_PMU_FORMAT_ATTR(event_filter); +GEN_PMU_FORMAT_ATTR(min_latency); + +static struct attribute *arm_spe_pmu_formats_attr[] = { + &format_attr_ts_enable.attr, + &format_attr_pa_enable.attr, + &format_attr_pct_enable.attr, + &format_attr_jitter.attr, + &format_attr_branch_filter.attr, + &format_attr_load_filter.attr, + &format_attr_store_filter.attr, + &format_attr_event_filter.attr, + &format_attr_min_latency.attr, + NULL, +}; + +static struct attribute_group arm_spe_pmu_format_group = { + .name = "format", + .attrs = arm_spe_pmu_formats_attr, +}; + +static ssize_t arm_spe_pmu_get_attr_cpumask(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct platform_device *pdev = to_platform_device(dev); + struct arm_spe_pmu *spe_pmu = platform_get_drvdata(pdev); + + return cpumap_print_to_pagebuf(true, buf, &spe_pmu->supported_cpus); +} +static DEVICE_ATTR(cpumask, S_IRUGO, arm_spe_pmu_get_attr_cpumask, NULL); + +static struct attribute *arm_spe_pmu_attrs[] = { + &dev_attr_cpumask.attr, + NULL, +}; + +static struct attribute_group arm_spe_pmu_group = { + .attrs = arm_spe_pmu_attrs, +}; + +static const struct attribute_group *arm_spe_pmu_attr_groups[] = { + &arm_spe_pmu_group, + &arm_spe_pmu_cap_group, + &arm_spe_pmu_format_group, + NULL, +}; + +/* Convert between user ABI and register values */ +static u64 arm_spe_event_to_pmscr(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + u64 reg = 0; + + reg |= ATTR_CFG_GET_FLD(attr, ts_enable) << SYS_PMSCR_EL1_TS_SHIFT; + reg |= ATTR_CFG_GET_FLD(attr, pa_enable) << SYS_PMSCR_EL1_PA_SHIFT; + reg |= ATTR_CFG_GET_FLD(attr, pct_enable) << SYS_PMSCR_EL1_PCT_SHIFT; + + if (!attr->exclude_user) + reg |= BIT(SYS_PMSCR_EL1_E0SPE_SHIFT); + + if (!attr->exclude_kernel) + reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT); + + if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && capable(CAP_SYS_ADMIN)) + reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT); + + return reg; +} + +static void arm_spe_event_sanitise_period(struct perf_event *event) +{ + struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + u64 period = event->hw.sample_period; + u64 max_period = SYS_PMSIRR_EL1_INTERVAL_MASK + << SYS_PMSIRR_EL1_INTERVAL_SHIFT; + + if (period < spe_pmu->min_period) + period = spe_pmu->min_period; + else if (period > max_period) + period = max_period; + else + period &= max_period; + + event->hw.sample_period = period; +} + +static u64 arm_spe_event_to_pmsirr(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + u64 reg = 0; + + arm_spe_event_sanitise_period(event); + + reg |= ATTR_CFG_GET_FLD(attr, jitter) << SYS_PMSIRR_EL1_RND_SHIFT; + reg |= event->hw.sample_period; + + return reg; +} + +static u64 arm_spe_event_to_pmsfcr(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + u64 reg = 0; + + reg |= ATTR_CFG_GET_FLD(attr, load_filter) << SYS_PMSFCR_EL1_LD_SHIFT; + reg |= ATTR_CFG_GET_FLD(attr, store_filter) << SYS_PMSFCR_EL1_ST_SHIFT; + reg |= ATTR_CFG_GET_FLD(attr, branch_filter) << SYS_PMSFCR_EL1_B_SHIFT; + + if (reg) + reg |= BIT(SYS_PMSFCR_EL1_FT_SHIFT); + + if (ATTR_CFG_GET_FLD(attr, event_filter)) + reg |= BIT(SYS_PMSFCR_EL1_FE_SHIFT); + + if (ATTR_CFG_GET_FLD(attr, min_latency)) + reg |= BIT(SYS_PMSFCR_EL1_FL_SHIFT); + + return reg; +} + +static u64 arm_spe_event_to_pmsevfr(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + return ATTR_CFG_GET_FLD(attr, event_filter); +} + +static u64 arm_spe_event_to_pmslatfr(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + return ATTR_CFG_GET_FLD(attr, min_latency) + << SYS_PMSLATFR_EL1_MINLAT_SHIFT; +} + +static void arm_spe_pmu_pad_buf(struct perf_output_handle *handle, int len) +{ + struct arm_spe_pmu_buf *buf = perf_get_aux(handle); + u64 head = PERF_IDX2OFF(handle->head, buf); + + memset(buf->base + head, ARM_SPE_BUF_PAD_BYTE, len); + if (!buf->snapshot) + perf_aux_output_skip(handle, len); +} + +static u64 arm_spe_pmu_next_snapshot_off(struct perf_output_handle *handle) +{ + struct arm_spe_pmu_buf *buf = perf_get_aux(handle); + struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu); + u64 head = PERF_IDX2OFF(handle->head, buf); + u64 limit = buf->nr_pages * PAGE_SIZE; + + /* + * The trace format isn't parseable in reverse, so clamp + * the limit to half of the buffer size in snapshot mode + * so that the worst case is half a buffer of records, as + * opposed to a single record. + */ + if (head < limit >> 1) + limit >>= 1; + + /* + * If we're within max_record_sz of the limit, we must + * pad, move the head index and recompute the limit. + */ + if (limit - head < spe_pmu->max_record_sz) { + arm_spe_pmu_pad_buf(handle, limit - head); + handle->head = PERF_IDX2OFF(limit, buf); + limit = ((buf->nr_pages * PAGE_SIZE) >> 1) + handle->head; + } + + return limit; +} + +static u64 __arm_spe_pmu_next_off(struct perf_output_handle *handle) +{ + struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu); + struct arm_spe_pmu_buf *buf = perf_get_aux(handle); + const u64 bufsize = buf->nr_pages * PAGE_SIZE; + u64 limit = bufsize; + u64 head, tail, wakeup; + + /* + * The head can be misaligned for two reasons: + * + * 1. The hardware left PMBPTR pointing to the first byte after + * a record when generating a buffer management event. + * + * 2. We used perf_aux_output_skip to consume handle->size bytes + * and CIRC_SPACE was used to compute the size, which always + * leaves one entry free. + * + * Deal with this by padding to the next alignment boundary and + * moving the head index. If we run out of buffer space, we'll + * reduce handle->size to zero and end up reporting truncation. + */ + head = PERF_IDX2OFF(handle->head, buf); + if (!IS_ALIGNED(head, spe_pmu->align)) { + unsigned long delta = roundup(head, spe_pmu->align) - head; + + delta = min(delta, handle->size); + arm_spe_pmu_pad_buf(handle, delta); + head = PERF_IDX2OFF(handle->head, buf); + } + + /* If we've run out of free space, then nothing more to do */ + if (!handle->size) + goto no_space; + + /* Compute the tail and wakeup indices now that we've aligned head */ + tail = PERF_IDX2OFF(handle->head + handle->size, buf); + wakeup = PERF_IDX2OFF(handle->wakeup, buf); + + /* + * Avoid clobbering unconsumed data. We know we have space, so + * if we see head == tail we know that the buffer is empty. If + * head > tail, then there's nothing to clobber prior to + * wrapping. + */ + if (head < tail) + limit = round_down(tail, PAGE_SIZE); + + /* + * Wakeup may be arbitrarily far into the future. If it's not in + * the current generation, either we'll wrap before hitting it, + * or it's in the past and has been handled already. + * + * If there's a wakeup before we wrap, arrange to be woken up by + * the page boundary following it. Keep the tail boundary if + * that's lower. + */ + if (handle->wakeup < (handle->head + handle->size) && head <= wakeup) + limit = min(limit, round_up(wakeup, PAGE_SIZE)); + + if (limit > head) + return limit; + + arm_spe_pmu_pad_buf(handle, handle->size); +no_space: + perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED); + perf_aux_output_end(handle, 0); + return 0; +} + +static u64 arm_spe_pmu_next_off(struct perf_output_handle *handle) +{ + struct arm_spe_pmu_buf *buf = perf_get_aux(handle); + struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu); + u64 limit = __arm_spe_pmu_next_off(handle); + u64 head = PERF_IDX2OFF(handle->head, buf); + + /* + * If the head has come too close to the end of the buffer, + * then pad to the end and recompute the limit. + */ + if (limit && (limit - head < spe_pmu->max_record_sz)) { + arm_spe_pmu_pad_buf(handle, limit - head); + limit = __arm_spe_pmu_next_off(handle); + } + + return limit; +} + +static void arm_spe_perf_aux_output_begin(struct perf_output_handle *handle, + struct perf_event *event) +{ + u64 base, limit; + struct arm_spe_pmu_buf *buf; + + /* Start a new aux session */ + buf = perf_aux_output_begin(handle, event); + if (!buf) { + event->hw.state |= PERF_HES_STOPPED; + /* + * We still need to clear the limit pointer, since the + * profiler might only be disabled by virtue of a fault. + */ + limit = 0; + goto out_write_limit; + } + + limit = buf->snapshot ? arm_spe_pmu_next_snapshot_off(handle) + : arm_spe_pmu_next_off(handle); + if (limit) + limit |= BIT(SYS_PMBLIMITR_EL1_E_SHIFT); + + limit += (u64)buf->base; + base = (u64)buf->base + PERF_IDX2OFF(handle->head, buf); + write_sysreg_s(base, SYS_PMBPTR_EL1); + +out_write_limit: + write_sysreg_s(limit, SYS_PMBLIMITR_EL1); +} + +static void arm_spe_perf_aux_output_end(struct perf_output_handle *handle) +{ + struct arm_spe_pmu_buf *buf = perf_get_aux(handle); + u64 offset, size; + + offset = read_sysreg_s(SYS_PMBPTR_EL1) - (u64)buf->base; + size = offset - PERF_IDX2OFF(handle->head, buf); + + if (buf->snapshot) + handle->head = offset; + + perf_aux_output_end(handle, size); +} + +static void arm_spe_pmu_disable_and_drain_local(void) +{ + /* Disable profiling at EL0 and EL1 */ + write_sysreg_s(0, SYS_PMSCR_EL1); + isb(); + + /* Drain any buffered data */ + psb_csync(); + dsb(nsh); + + /* Disable the profiling buffer */ + write_sysreg_s(0, SYS_PMBLIMITR_EL1); + isb(); +} + +/* IRQ handling */ +static enum arm_spe_pmu_buf_fault_action +arm_spe_pmu_buf_get_fault_act(struct perf_output_handle *handle) +{ + const char *err_str; + u64 pmbsr; + enum arm_spe_pmu_buf_fault_action ret; + + /* + * Ensure new profiling data is visible to the CPU and any external + * aborts have been resolved. + */ + psb_csync(); + dsb(nsh); + + /* Ensure hardware updates to PMBPTR_EL1 are visible */ + isb(); + + /* Service required? */ + pmbsr = read_sysreg_s(SYS_PMBSR_EL1); + if (!(pmbsr & BIT(SYS_PMBSR_EL1_S_SHIFT))) + return SPE_PMU_BUF_FAULT_ACT_SPURIOUS; + + /* + * If we've lost data, disable profiling and also set the PARTIAL + * flag to indicate that the last record is corrupted. + */ + if (pmbsr & BIT(SYS_PMBSR_EL1_DL_SHIFT)) + perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED | + PERF_AUX_FLAG_PARTIAL); + + /* Report collisions to userspace so that it can up the period */ + if (pmbsr & BIT(SYS_PMBSR_EL1_COLL_SHIFT)) + perf_aux_output_flag(handle, PERF_AUX_FLAG_COLLISION); + + /* We only expect buffer management events */ + switch (pmbsr & (SYS_PMBSR_EL1_EC_MASK << SYS_PMBSR_EL1_EC_SHIFT)) { + case SYS_PMBSR_EL1_EC_BUF: + /* Handled below */ + break; + case SYS_PMBSR_EL1_EC_FAULT_S1: + case SYS_PMBSR_EL1_EC_FAULT_S2: + err_str = "Unexpected buffer fault"; + goto out_err; + default: + err_str = "Unknown error code"; + goto out_err; + } + + /* Buffer management event */ + switch (pmbsr & + (SYS_PMBSR_EL1_BUF_BSC_MASK << SYS_PMBSR_EL1_BUF_BSC_SHIFT)) { + case SYS_PMBSR_EL1_BUF_BSC_FULL: + ret = SPE_PMU_BUF_FAULT_ACT_OK; + goto out_stop; + default: + err_str = "Unknown buffer status code"; + } + +out_err: + pr_err_ratelimited("%s on CPU %d [PMBSR=0x%016llx, PMBPTR=0x%016llx, PMBLIMITR=0x%016llx]\n", + err_str, smp_processor_id(), pmbsr, + read_sysreg_s(SYS_PMBPTR_EL1), + read_sysreg_s(SYS_PMBLIMITR_EL1)); + ret = SPE_PMU_BUF_FAULT_ACT_FATAL; + +out_stop: + arm_spe_perf_aux_output_end(handle); + return ret; +} + +static irqreturn_t arm_spe_pmu_irq_handler(int irq, void *dev) +{ + struct perf_output_handle *handle = dev; + struct perf_event *event = handle->event; + enum arm_spe_pmu_buf_fault_action act; + + if (!perf_get_aux(handle)) + return IRQ_NONE; + + act = arm_spe_pmu_buf_get_fault_act(handle); + if (act == SPE_PMU_BUF_FAULT_ACT_SPURIOUS) + return IRQ_NONE; + + /* + * Ensure perf callbacks have completed, which may disable the + * profiling buffer in response to a TRUNCATION flag. + */ + irq_work_run(); + + switch (act) { + case SPE_PMU_BUF_FAULT_ACT_FATAL: + /* + * If a fatal exception occurred then leaving the profiling + * buffer enabled is a recipe waiting to happen. Since + * fatal faults don't always imply truncation, make sure + * that the profiling buffer is disabled explicitly before + * clearing the syndrome register. + */ + arm_spe_pmu_disable_and_drain_local(); + break; + case SPE_PMU_BUF_FAULT_ACT_OK: + /* + * We handled the fault (the buffer was full), so resume + * profiling as long as we didn't detect truncation. + * PMBPTR might be misaligned, but we'll burn that bridge + * when we get to it. + */ + if (!(handle->aux_flags & PERF_AUX_FLAG_TRUNCATED)) { + arm_spe_perf_aux_output_begin(handle, event); + isb(); + } + break; + case SPE_PMU_BUF_FAULT_ACT_SPURIOUS: + /* We've seen you before, but GCC has the memory of a sieve. */ + break; + } + + /* The buffer pointers are now sane, so resume profiling. */ + write_sysreg_s(0, SYS_PMBSR_EL1); + return IRQ_HANDLED; +} + +/* Perf callbacks */ +static int arm_spe_pmu_event_init(struct perf_event *event) +{ + u64 reg; + struct perf_event_attr *attr = &event->attr; + struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + + /* This is, of course, deeply driver-specific */ + if (attr->type != event->pmu->type) + return -ENOENT; + + if (event->cpu >= 0 && + !cpumask_test_cpu(event->cpu, &spe_pmu->supported_cpus)) + return -ENOENT; + + if (arm_spe_event_to_pmsevfr(event) & SYS_PMSEVFR_EL1_RES0) + return -EOPNOTSUPP; + + if (attr->exclude_idle) + return -EOPNOTSUPP; + + /* + * Feedback-directed frequency throttling doesn't work when we + * have a buffer of samples. We'd need to manually count the + * samples in the buffer when it fills up and adjust the event + * count to reflect that. Instead, just force the user to specify + * a sample period. + */ + if (attr->freq) + return -EINVAL; + + reg = arm_spe_event_to_pmsfcr(event); + if ((reg & BIT(SYS_PMSFCR_EL1_FE_SHIFT)) && + !(spe_pmu->features & SPE_PMU_FEAT_FILT_EVT)) + return -EOPNOTSUPP; + + if ((reg & BIT(SYS_PMSFCR_EL1_FT_SHIFT)) && + !(spe_pmu->features & SPE_PMU_FEAT_FILT_TYP)) + return -EOPNOTSUPP; + + if ((reg & BIT(SYS_PMSFCR_EL1_FL_SHIFT)) && + !(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT)) + return -EOPNOTSUPP; + + reg = arm_spe_event_to_pmscr(event); + if (!capable(CAP_SYS_ADMIN) && + (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) | + BIT(SYS_PMSCR_EL1_CX_SHIFT) | + BIT(SYS_PMSCR_EL1_PCT_SHIFT)))) + return -EACCES; + + return 0; +} + +static void arm_spe_pmu_start(struct perf_event *event, int flags) +{ + u64 reg; + struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + struct hw_perf_event *hwc = &event->hw; + struct perf_output_handle *handle = this_cpu_ptr(spe_pmu->handle); + + hwc->state = 0; + arm_spe_perf_aux_output_begin(handle, event); + if (hwc->state) + return; + + reg = arm_spe_event_to_pmsfcr(event); + write_sysreg_s(reg, SYS_PMSFCR_EL1); + + reg = arm_spe_event_to_pmsevfr(event); + write_sysreg_s(reg, SYS_PMSEVFR_EL1); + + reg = arm_spe_event_to_pmslatfr(event); + write_sysreg_s(reg, SYS_PMSLATFR_EL1); + + if (flags & PERF_EF_RELOAD) { + reg = arm_spe_event_to_pmsirr(event); + write_sysreg_s(reg, SYS_PMSIRR_EL1); + isb(); + reg = local64_read(&hwc->period_left); + write_sysreg_s(reg, SYS_PMSICR_EL1); + } + + reg = arm_spe_event_to_pmscr(event); + isb(); + write_sysreg_s(reg, SYS_PMSCR_EL1); +} + +static void arm_spe_pmu_stop(struct perf_event *event, int flags) +{ + struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + struct hw_perf_event *hwc = &event->hw; + struct perf_output_handle *handle = this_cpu_ptr(spe_pmu->handle); + + /* If we're already stopped, then nothing to do */ + if (hwc->state & PERF_HES_STOPPED) + return; + + /* Stop all trace generation */ + arm_spe_pmu_disable_and_drain_local(); + + if (flags & PERF_EF_UPDATE) { + /* + * If there's a fault pending then ensure we contain it + * to this buffer, since we might be on the context-switch + * path. + */ + if (perf_get_aux(handle)) { + enum arm_spe_pmu_buf_fault_action act; + + act = arm_spe_pmu_buf_get_fault_act(handle); + if (act == SPE_PMU_BUF_FAULT_ACT_SPURIOUS) + arm_spe_perf_aux_output_end(handle); + else + write_sysreg_s(0, SYS_PMBSR_EL1); + } + + /* + * This may also contain ECOUNT, but nobody else should + * be looking at period_left, since we forbid frequency + * based sampling. + */ + local64_set(&hwc->period_left, read_sysreg_s(SYS_PMSICR_EL1)); + hwc->state |= PERF_HES_UPTODATE; + } + + hwc->state |= PERF_HES_STOPPED; +} + +static int arm_spe_pmu_add(struct perf_event *event, int flags) +{ + int ret = 0; + struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + struct hw_perf_event *hwc = &event->hw; + int cpu = event->cpu == -1 ? smp_processor_id() : event->cpu; + + if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus)) + return -ENOENT; + + hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED; + + if (flags & PERF_EF_START) { + arm_spe_pmu_start(event, PERF_EF_RELOAD); + if (hwc->state & PERF_HES_STOPPED) + ret = -EINVAL; + } + + return ret; +} + +static void arm_spe_pmu_del(struct perf_event *event, int flags) +{ + arm_spe_pmu_stop(event, PERF_EF_UPDATE); |