From patchwork Thu Sep 28 14:09:50 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Will Deacon X-Patchwork-Id: 9976219 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork.web.codeaurora.org (Postfix) with ESMTP id 6CFE260365 for ; Thu, 28 Sep 2017 14:13:45 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 5A388295DA for ; Thu, 28 Sep 2017 14:13:45 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 4EB7029602; Thu, 28 Sep 2017 14:13:45 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-4.2 required=2.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,RCVD_IN_DNSWL_MED autolearn=unavailable version=3.3.1 Received: from bombadil.infradead.org (bombadil.infradead.org [65.50.211.133]) (using TLSv1.2 with cipher AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id 7E1C7295DA for ; Thu, 28 Sep 2017 14:13:43 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.infradead.org; s=bombadil.20170209; h=Sender: Content-Transfer-Encoding:Content-Type:MIME-Version:Cc:List-Subscribe: List-Help:List-Post:List-Archive:List-Unsubscribe:List-Id:References: In-Reply-To:Message-Id:Date:Subject:To:From:Reply-To:Content-ID: Content-Description:Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc :Resent-Message-ID:List-Owner; bh=i3EgfvSlfD6+wXyZWOtyE0Q7BcWZSe0K1ivme08mHuY=; b=QMJvS8HiyeVbLFCP0NqotP1GMJ WLecylYNmXVMWnu1rAiNEDfX+wakCslc2XWiC129mTl2u240uZx30xxMZkQjZEQU5Jv44HZmiTK6e 5hrmEcsOQ/EhHIqThsvoY3uIYqLuDDrKkh1soalNxely/EtWJ+IEHPqI5NGxNIvwb/w2u5hybqe6H YHt0HWnWaYJFcRGZcoxlGwLpkFnTpCpLUOtDVO/lrfER4UzUyLtNGoKUIgCLb3FelYloYhsEqP73b Ga1E2gd4TyO9YowbxadbTG4gPhYuzl1UxZN858xi8pyNm2GWlafsz8TS2SRPU6WdSCjdWPlV54cMr 7MJYn75g==; Received: from localhost ([127.0.0.1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.87 #1 (Red Hat Linux)) id 1dxZZH-0006tE-JE; Thu, 28 Sep 2017 14:13:35 +0000 Received: from foss.arm.com ([217.140.101.70]) by bombadil.infradead.org with esmtp (Exim 4.87 #1 (Red Hat Linux)) id 1dxZW8-0003Re-QW for linux-arm-kernel@lists.infradead.org; Thu, 28 Sep 2017 14:10:48 +0000 Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.72.51.249]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 57759169E; Thu, 28 Sep 2017 07:09:39 -0700 (PDT) Received: from edgewater-inn.cambridge.arm.com (usa-sjc-imap-foss1.foss.arm.com [10.72.51.249]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id D69853F5C1; Thu, 28 Sep 2017 07:09:38 -0700 (PDT) Received: by edgewater-inn.cambridge.arm.com (Postfix, from userid 1000) id C87061AE3096; Thu, 28 Sep 2017 15:09:52 +0100 (BST) From: Will Deacon To: linux-arm-kernel@lists.infradead.org Subject: [PATCH v5 6/7] drivers/perf: Add support for ARMv8.2 Statistical Profiling Extension Date: Thu, 28 Sep 2017 15:09:50 +0100 Message-Id: <1506607791-8621-7-git-send-email-will.deacon@arm.com> X-Mailer: git-send-email 2.1.4 In-Reply-To: <1506607791-8621-1-git-send-email-will.deacon@arm.com> References: <1506607791-8621-1-git-send-email-will.deacon@arm.com> X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20170928_071021_615881_2C4E9EF8 X-CRM114-Status: GOOD ( 25.02 ) X-BeenThere: linux-arm-kernel@lists.infradead.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: mark.rutland@arm.com, robh@kernel.org, kim.phillips@arm.com, mathieu.poirier@linaro.org, pawel.moll@arm.com, suzuki.poulose@arm.com, marc.zyngier@arm.com, Will Deacon , linux-kernel@vger.kernel.org, alexander.shishkin@linux.intel.com, peterz@infradead.org, mingo@redhat.com, tglx@linutronix.de MIME-Version: 1.0 Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+patchwork-linux-arm=patchwork.kernel.org@lists.infradead.org X-Virus-Scanned: ClamAV using ClamSMTP The ARMv8.2 architecture introduces the optional Statistical Profiling Extension (SPE). SPE can be used to profile a population of operations in the CPU pipeline after instruction decode. These are either architected instructions (i.e. a dynamic instruction trace) or CPU-specific uops and the choice is fixed statically in the hardware and advertised to userspace via caps/. Sampling is controlled using a sampling interval, similar to a regular PMU counter, but also with an optional random perturbation to avoid falling into patterns where you continuously profile the same instruction in a hot loop. After each operation is decoded, the interval counter is decremented. When it hits zero, an operation is chosen for profiling and tracked within the pipeline until it retires. Along the way, information such as TLB lookups, cache misses, time spent to issue etc is captured in the form of a sample. The sample is then filtered according to certain criteria (e.g. load latency) that can be specified in the event config (described under format/) and, if the sample satisfies the filter, it is written out to memory as a record, otherwise it is discarded. Only one operation can be sampled at a time. The in-memory buffer is linear and virtually addressed, raising an interrupt when it fills up. The PMU driver handles these interrupts to give the appearance of a ring buffer, as expected by the AUX code. The in-memory trace-like format is self-describing (though not parseable in reverse) and written as a series of records, with each record corresponding to a sample and consisting of a sequence of packets. These packets are defined by the architecture, although some have CPU-specific fields for recording information specific to the microarchitecture. As a simple example, a record generated for a branch instruction may consist of the following packets: 0 (Address) : Virtual PC of the branch instruction 1 (Type) : Conditional direct branch 2 (Counter) : Number of cycles taken from Dispatch to Issue 3 (Address) : Virtual branch target + condition flags 4 (Counter) : Number of cycles taken from Dispatch to Complete 5 (Events) : Mispredicted as not-taken 6 (END) : End of record It is also possible to toggle properties such as timestamp packets in each record. This patch adds support for SPE in the form of a new perf driver. Cc: Mark Rutland Cc: Alexander Shishkin Signed-off-by: Will Deacon --- drivers/perf/Kconfig | 8 + drivers/perf/Makefile | 1 + drivers/perf/arm_spe_pmu.c | 1246 ++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 1255 insertions(+) create mode 100644 drivers/perf/arm_spe_pmu.c diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig index e5197ffb7422..43525a9dc947 100644 --- a/drivers/perf/Kconfig +++ b/drivers/perf/Kconfig @@ -43,4 +43,12 @@ 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 6420bd4394d5..eaee60cf4b1b 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -3,3 +3,4 @@ obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o obj-$(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_spe_pmu.c b/drivers/perf/arm_spe_pmu.c new file mode 100644 index 000000000000..a9cc7b25c116 --- /dev/null +++ b/drivers/perf/arm_spe_pmu.c @@ -0,0 +1,1246 @@ +/* + * 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 . + * + * Copyright (C) 2016 ARM Limited + * + * Author: Will Deacon + */ + +#define PMUNAME "arm_spe" +#define DRVNAME PMUNAME "_pmu" +#define pr_fmt(fmt) DRVNAME ": " fmt + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#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(jitter); +GEN_PMU_FORMAT_ATTR(load_filter); +GEN_PMU_FORMAT_ATTR(store_filter); +GEN_PMU_FORMAT_ATTR(branch_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_jitter.attr, + &format_attr_load_filter.attr, + &format_attr_store_filter.attr, + &format_attr_branch_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); +} + +static void arm_spe_pmu_read(struct perf_event *event) +{ +} + +static void *arm_spe_pmu_setup_aux(int cpu, void **pages, int nr_pages, + bool snapshot) +{ + int i; + struct page **pglist; + struct arm_spe_pmu_buf *buf; + + /* We need at least two pages for this to work. */ + if (nr_pages < 2) + return NULL; + + /* + * We require an even number of pages for snapshot mode, so that + * we can effectively treat the buffer as consisting of two equal + * parts and give userspace a fighting chance of getting some + * useful data out of it. + */ + if (!nr_pages || (snapshot && (nr_pages & 1))) + return NULL; + + if (cpu == -1) + cpu = raw_smp_processor_id(); + + buf = kzalloc_node(sizeof(*buf), GFP_KERNEL, cpu_to_node(cpu)); + if (!buf) + return NULL; + + pglist = kcalloc(nr_pages, sizeof(*pglist), GFP_KERNEL); + if (!pglist) + goto out_free_buf; + + for (i = 0; i < nr_pages; ++i) { + struct page *page = virt_to_page(pages[i]); + + if (PagePrivate(page)) { + pr_warn("unexpected high-order page for auxbuf!"); + goto out_free_pglist; + } + + pglist[i] = virt_to_page(pages[i]); + } + + buf->base = vmap(pglist, nr_pages, VM_MAP, PAGE_KERNEL); + if (!buf->base) + goto out_free_pglist; + + buf->nr_pages = nr_pages; + buf->snapshot = snapshot; + + kfree(pglist); + return buf; + +out_free_pglist: + kfree(pglist); +out_free_buf: + kfree(buf); + return NULL; +} + +static void arm_spe_pmu_free_aux(void *aux) +{ + struct arm_spe_pmu_buf *buf = aux; + + vunmap(buf->base); + kfree(buf); +} + +/* Initialisation and teardown functions */ +static int arm_spe_pmu_perf_init(struct arm_spe_pmu *spe_pmu) +{ + static atomic_t pmu_idx = ATOMIC_INIT(-1); + + int idx; + char *name; + struct device *dev = &spe_pmu->pdev->dev; + + spe_pmu->pmu = (struct pmu) { + .capabilities = PERF_PMU_CAP_EXCLUSIVE | PERF_PMU_CAP_ITRACE, + .attr_groups = arm_spe_pmu_attr_groups, + /* + * We hitch a ride on the software context here, so that + * we can support per-task profiling (which is not possible + * with the invalid context as it doesn't get sched callbacks). + * This requires that userspace either uses a dummy event for + * perf_event_open, since the aux buffer is not setup until + * a subsequent mmap, or creates the profiling event in a + * disabled state and explicitly PERF_EVENT_IOC_ENABLEs it + * once the buffer has been created. + */ + .task_ctx_nr = perf_sw_context, + .event_init = arm_spe_pmu_event_init, + .add = arm_spe_pmu_add, + .del = arm_spe_pmu_del, + .start = arm_spe_pmu_start, + .stop = arm_spe_pmu_stop, + .read = arm_spe_pmu_read, + .setup_aux = arm_spe_pmu_setup_aux, + .free_aux = arm_spe_pmu_free_aux, + }; + + idx = atomic_inc_return(&pmu_idx); + name = devm_kasprintf(dev, GFP_KERNEL, "%s_%d", PMUNAME, idx); + return perf_pmu_register(&spe_pmu->pmu, name, -1); +} + +static void arm_spe_pmu_perf_destroy(struct arm_spe_pmu *spe_pmu) +{ + perf_pmu_unregister(&spe_pmu->pmu); +} + +static void __arm_spe_pmu_dev_probe(void *info) +{ + int fld; + u64 reg; + struct arm_spe_pmu *spe_pmu = info; + struct device *dev = &spe_pmu->pdev->dev; + + fld = cpuid_feature_extract_unsigned_field(read_cpuid(ID_AA64DFR0_EL1), + ID_AA64DFR0_PMSVER_SHIFT); + if (!fld) { + dev_err(dev, + "unsupported ID_AA64DFR0_EL1.PMSVer [%d] on CPU %d\n", + fld, smp_processor_id()); + return; + } + + /* Read PMBIDR first to determine whether or not we have access */ + reg = read_sysreg_s(SYS_PMBIDR_EL1); + if (reg & BIT(SYS_PMBIDR_EL1_P_SHIFT)) { + dev_err(dev, + "profiling buffer owned by higher exception level\n"); + return; + } + + /* Minimum alignment. If it's out-of-range, then fail the probe */ + fld = reg >> SYS_PMBIDR_EL1_ALIGN_SHIFT & SYS_PMBIDR_EL1_ALIGN_MASK; + spe_pmu->align = 1 << fld; + if (spe_pmu->align > SZ_2K) { + dev_err(dev, "unsupported PMBIDR.Align [%d] on CPU %d\n", + fld, smp_processor_id()); + return; + } + + /* It's now safe to read PMSIDR and figure out what we've got */ + reg = read_sysreg_s(SYS_PMSIDR_EL1); + if (reg & BIT(SYS_PMSIDR_EL1_FE_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_FILT_EVT; + + if (reg & BIT(SYS_PMSIDR_EL1_FT_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_FILT_TYP; + + if (reg & BIT(SYS_PMSIDR_EL1_FL_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_FILT_LAT; + + if (reg & BIT(SYS_PMSIDR_EL1_ARCHINST_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_ARCH_INST; + + if (reg & BIT(SYS_PMSIDR_EL1_LDS_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_LDS; + + if (reg & BIT(SYS_PMSIDR_EL1_ERND_SHIFT)) + spe_pmu->features |= SPE_PMU_FEAT_ERND; + + /* This field has a spaced out encoding, so just use a look-up */ + fld = reg >> SYS_PMSIDR_EL1_INTERVAL_SHIFT & SYS_PMSIDR_EL1_INTERVAL_MASK; + switch (fld) { + case 0: + spe_pmu->min_period = 256; + break; + case 2: + spe_pmu->min_period = 512; + break; + case 3: + spe_pmu->min_period = 768; + break; + case 4: + spe_pmu->min_period = 1024; + break; + case 5: + spe_pmu->min_period = 1536; + break; + case 6: + spe_pmu->min_period = 2048; + break; + case 7: + spe_pmu->min_period = 3072; + break; + default: + dev_warn(dev, "unknown PMSIDR_EL1.Interval [%d]; assuming 8\n", + fld); + /* Fallthrough */ + case 8: + spe_pmu->min_period = 4096; + } + + /* Maximum record size. If it's out-of-range, then fail the probe */ + fld = reg >> SYS_PMSIDR_EL1_MAXSIZE_SHIFT & SYS_PMSIDR_EL1_MAXSIZE_MASK; + spe_pmu->max_record_sz = 1 << fld; + if (spe_pmu->max_record_sz > SZ_2K || spe_pmu->max_record_sz < 16) { + dev_err(dev, "unsupported PMSIDR_EL1.MaxSize [%d] on CPU %d\n", + fld, smp_processor_id()); + return; + } + + fld = reg >> SYS_PMSIDR_EL1_COUNTSIZE_SHIFT & SYS_PMSIDR_EL1_COUNTSIZE_MASK; + switch (fld) { + default: + dev_warn(dev, "unknown PMSIDR_EL1.CountSize [%d]; assuming 2\n", + fld); + /* Fallthrough */ + case 2: + spe_pmu->counter_sz = 12; + } + + dev_info(dev, + "probed for CPUs %*pbl [max_record_sz %u, align %u, features 0x%llx]\n", + cpumask_pr_args(&spe_pmu->supported_cpus), + spe_pmu->max_record_sz, spe_pmu->align, spe_pmu->features); + + spe_pmu->features |= SPE_PMU_FEAT_DEV_PROBED; + return; +} + +static void __arm_spe_pmu_reset_local(void) +{ + /* + * This is probably overkill, as we have no idea where we're + * draining any buffered data to... + */ + arm_spe_pmu_disable_and_drain_local(); + + /* Reset the buffer base pointer */ + write_sysreg_s(0, SYS_PMBPTR_EL1); + isb(); + + /* Clear any pending management interrupts */ + write_sysreg_s(0, SYS_PMBSR_EL1); + isb(); +} + +static void __arm_spe_pmu_setup_one(void *info) +{ + struct arm_spe_pmu *spe_pmu = info; + + __arm_spe_pmu_reset_local(); + enable_percpu_irq(spe_pmu->irq, IRQ_TYPE_NONE); +} + +static void __arm_spe_pmu_stop_one(void *info) +{ + struct arm_spe_pmu *spe_pmu = info; + + disable_percpu_irq(spe_pmu->irq); + __arm_spe_pmu_reset_local(); +} + +static int arm_spe_pmu_cpu_startup(unsigned int cpu, struct hlist_node *node) +{ + struct arm_spe_pmu *spe_pmu; + + spe_pmu = hlist_entry_safe(node, struct arm_spe_pmu, hotplug_node); + if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus)) + return 0; + + __arm_spe_pmu_setup_one(spe_pmu); + return 0; +} + +static int arm_spe_pmu_cpu_teardown(unsigned int cpu, struct hlist_node *node) +{ + struct arm_spe_pmu *spe_pmu; + + spe_pmu = hlist_entry_safe(node, struct arm_spe_pmu, hotplug_node); + if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus)) + return 0; + + __arm_spe_pmu_stop_one(spe_pmu); + return 0; +} + +static int arm_spe_pmu_dev_init(struct arm_spe_pmu *spe_pmu) +{ + int ret; + cpumask_t *mask = &spe_pmu->supported_cpus; + + /* Make sure we probe the hardware on a relevant CPU */ + ret = smp_call_function_any(mask, __arm_spe_pmu_dev_probe, spe_pmu, 1); + if (ret || !(spe_pmu->features & SPE_PMU_FEAT_DEV_PROBED)) + return -ENXIO; + + /* Request our PPIs (note that the IRQ is still disabled) */ + ret = request_percpu_irq(spe_pmu->irq, arm_spe_pmu_irq_handler, DRVNAME, + spe_pmu->handle); + if (ret) + return ret; + + /* + * Register our hotplug notifier now so we don't miss any events. + * This will enable the IRQ for any supported CPUs that are already + * up. + */ + ret = cpuhp_state_add_instance(arm_spe_pmu_online, + &spe_pmu->hotplug_node); + if (ret) + free_percpu_irq(spe_pmu->irq, spe_pmu->handle); + + return ret; +} + +static void arm_spe_pmu_dev_teardown(struct arm_spe_pmu *spe_pmu) +{ + cpuhp_state_remove_instance(arm_spe_pmu_online, &spe_pmu->hotplug_node); + free_percpu_irq(spe_pmu->irq, spe_pmu->handle); +} + +/* Driver and device probing */ +static int arm_spe_pmu_irq_probe(struct arm_spe_pmu *spe_pmu) +{ + struct platform_device *pdev = spe_pmu->pdev; + int irq = platform_get_irq(pdev, 0); + + if (irq < 0) { + dev_err(&pdev->dev, "failed to get IRQ (%d)\n", irq); + return -ENXIO; + } + + if (!irq_is_percpu(irq)) { + dev_err(&pdev->dev, "expected PPI but got SPI (%d)\n", irq); + return -EINVAL; + } + + if (irq_get_percpu_devid_partition(irq, &spe_pmu->supported_cpus)) { + dev_err(&pdev->dev, "failed to get PPI partition (%d)\n", irq); + return -EINVAL; + } + + spe_pmu->irq = irq; + return 0; +} + +static const struct of_device_id arm_spe_pmu_of_match[] = { + { .compatible = "arm,statistical-profiling-extension-v1", .data = (void *)1 }, + { /* Sentinel */ }, +}; + +static int arm_spe_pmu_device_dt_probe(struct platform_device *pdev) +{ + int ret; + struct arm_spe_pmu *spe_pmu; + struct device *dev = &pdev->dev; + + spe_pmu = devm_kzalloc(dev, sizeof(*spe_pmu), GFP_KERNEL); + if (!spe_pmu) { + dev_err(dev, "failed to allocate spe_pmu\n"); + return -ENOMEM; + } + + spe_pmu->handle = alloc_percpu(typeof(*spe_pmu->handle)); + if (!spe_pmu->handle) + return -ENOMEM; + + spe_pmu->pdev = pdev; + platform_set_drvdata(pdev, spe_pmu); + + ret = arm_spe_pmu_irq_probe(spe_pmu); + if (ret) + goto out_free_handle; + + ret = arm_spe_pmu_dev_init(spe_pmu); + if (ret) + goto out_free_handle; + + ret = arm_spe_pmu_perf_init(spe_pmu); + if (ret) + goto out_teardown_dev; + + return 0; + +out_teardown_dev: + arm_spe_pmu_dev_teardown(spe_pmu); +out_free_handle: + free_percpu(spe_pmu->handle); + return ret; +} + +static int arm_spe_pmu_device_remove(struct platform_device *pdev) +{ + struct arm_spe_pmu *spe_pmu = platform_get_drvdata(pdev); + + arm_spe_pmu_perf_destroy(spe_pmu); + arm_spe_pmu_dev_teardown(spe_pmu); + free_percpu(spe_pmu->handle); + return 0; +} + +static struct platform_driver arm_spe_pmu_driver = { + .driver = { + .name = DRVNAME, + .of_match_table = of_match_ptr(arm_spe_pmu_of_match), + }, + .probe = arm_spe_pmu_device_dt_probe, + .remove = arm_spe_pmu_device_remove, +}; + +static int __init arm_spe_pmu_init(void) +{ + int ret; + + ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, DRVNAME, + arm_spe_pmu_cpu_startup, + arm_spe_pmu_cpu_teardown); + if (ret < 0) + return ret; + arm_spe_pmu_online = ret; + + ret = platform_driver_register(&arm_spe_pmu_driver); + if (ret) + cpuhp_remove_multi_state(arm_spe_pmu_online); + + return ret; +} + +static void __exit arm_spe_pmu_exit(void) +{ + platform_driver_unregister(&arm_spe_pmu_driver); + cpuhp_remove_multi_state(arm_spe_pmu_online); +} + +module_init(arm_spe_pmu_init); +module_exit(arm_spe_pmu_exit); + +MODULE_DESCRIPTION("Perf driver for the ARMv8.2 Statistical Profiling Extension"); +MODULE_AUTHOR("Will Deacon "); +MODULE_LICENSE("GPL v2");