From mboxrd@z Thu Jan 1 00:00:00 1970 From: Shameerali Kolothum Thodi Subject: RE: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver Date: Wed, 23 Jan 2019 11:02:48 +0000 Message-ID: <5FC3163CFD30C246ABAA99954A238FA83929855B@lhreml524-mbs.china.huawei.com> References: <20181130154751.28580-1-shameerali.kolothum.thodi@huawei.com> <20181130154751.28580-3-shameerali.kolothum.thodi@huawei.com> <20190122162351.GE8120@e119886-lin.cambridge.arm.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT Return-path: In-Reply-To: <20190122162351.GE8120@e119886-lin.cambridge.arm.com> Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org To: Andrew Murray Cc: "lorenzo.pieralisi@arm.com" , "robin.murphy@arm.com" , "mark.rutland@arm.com" , "vkilari@codeaurora.org" , "neil.m.leeder@gmail.com" , "jean-philippe.brucker@arm.com" , "pabba@codeaurora.org" , John Garry , "will.deacon@arm.com" , "rruigrok@codeaurora.org" , Linuxarm , "linux-kernel@vger.kernel.org" , "linux-acpi@vger.kernel.org" , "Guohanjun (Hanjun Guo)" , "linux-arm-kernel@lists.infradead.org" List-Id: linux-acpi@vger.kernel.org Hi Andrew, Thanks for taking a look at this. > -----Original Message----- > From: Andrew Murray [mailto:andrew.murray@arm.com] > Sent: 22 January 2019 16:24 > To: Shameerali Kolothum Thodi > Cc: lorenzo.pieralisi@arm.com; robin.murphy@arm.com; > mark.rutland@arm.com; vkilari@codeaurora.org; neil.m.leeder@gmail.com; > jean-philippe.brucker@arm.com; pabba@codeaurora.org; John Garry > ; will.deacon@arm.com; rruigrok@codeaurora.org; > Linuxarm ; linux-kernel@vger.kernel.org; > linux-acpi@vger.kernel.org; Guohanjun (Hanjun Guo) > ; linux-arm-kernel@lists.infradead.org > Subject: Re: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver > > On Fri, Nov 30, 2018 at 03:47:49PM +0000, Shameer Kolothum wrote: > > From: Neil Leeder > > > > Adds a new driver to support the SMMUv3 PMU and add it into the > > perf events framework. > > > > Each SMMU node may have multiple PMUs associated with it, each of > > which may support different events. > > > > SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > is the physical page address of the SMMU PMCG > > wrapped to 4K boundary. For example, the PMCG at 0xff88840000 is > > named smmuv3_pmcg_ff88840 > > > > Filtering by stream id is done by specifying filtering parameters > > with the event. options are: > > filter_enable - 0 = no filtering, 1 = filtering enabled > > filter_span - 0 = exact match, 1 = pattern match > > filter_stream_id - pattern to filter against > > > > Example: perf stat -e smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > filter_span=1,filter_stream_id=0x42/ -a netperf > > > > Applies filter pattern 0x42 to transaction events, which means events > > matching stream ids 0x42 & 0x43 are counted as only upper StreamID > > bits are required to match the given filter. Further filtering > > information is available in the SMMU documentation. > > > > SMMU events are not attributable to a CPU, so task mode and sampling > > are not supported. > > > > Signed-off-by: Neil Leeder > > Signed-off-by: Shameer Kolothum > > --- > > drivers/perf/Kconfig | 9 + > > drivers/perf/Makefile | 1 + > > drivers/perf/arm_smmuv3_pmu.c | 778 > ++++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 788 insertions(+) > > create mode 100644 drivers/perf/arm_smmuv3_pmu.c > > > > diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig > > index 08ebaf7..92be6a3 100644 > > --- a/drivers/perf/Kconfig > > +++ b/drivers/perf/Kconfig > > @@ -52,6 +52,15 @@ config ARM_PMU_ACPI > > depends on ARM_PMU && ACPI > > def_bool y > > > > +config ARM_SMMU_V3_PMU > > + bool "ARM SMMUv3 Performance Monitors Extension" > > + depends on ARM64 && ACPI && ARM_SMMU_V3 > > + help > > + Provides support for the SMMU version 3 performance monitor unit > (PMU) > > + on ARM-based systems. > > + Adds the SMMU PMU into the perf events subsystem for > > + monitoring SMMU performance events. > > + > > config ARM_DSU_PMU > > tristate "ARM DynamIQ Shared Unit (DSU) PMU" > > depends on ARM64 > > diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile > > index b3902bd..f10a932 100644 > > --- a/drivers/perf/Makefile > > +++ b/drivers/perf/Makefile > > @@ -4,6 +4,7 @@ obj-$(CONFIG_ARM_CCN) += arm-ccn.o > > obj-$(CONFIG_ARM_DSU_PMU) += arm_dsu_pmu.o > > obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o > > obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o > > +obj-$(CONFIG_ARM_SMMU_V3_PMU) += arm_smmuv3_pmu.o > > obj-$(CONFIG_HISI_PMU) += hisilicon/ > > obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o > > obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o > > diff --git a/drivers/perf/arm_smmuv3_pmu.c > b/drivers/perf/arm_smmuv3_pmu.c > > new file mode 100644 > > index 0000000..fb9dcd8 > > --- /dev/null > > +++ b/drivers/perf/arm_smmuv3_pmu.c > > @@ -0,0 +1,778 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > + > > +/* > > + * This driver adds support for perf events to use the Performance > > + * Monitor Counter Groups (PMCG) associated with an SMMUv3 node > > + * to monitor that node. > > + * > > + * SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > + * is the physical page address of the SMMU PMCG > wrapped > > + * to 4K boundary. For example, the PMCG at 0xff88840000 is named > > + * smmuv3_pmcg_ff88840 > > + * > > + * Filtering by stream id is done by specifying filtering parameters > > + * with the event. options are: > > + * filter_enable - 0 = no filtering, 1 = filtering enabled > > + * filter_span - 0 = exact match, 1 = pattern match > > + * filter_stream_id - pattern to filter against > > + * > > + * To match a partial StreamID where the X most-significant bits must match > > + * but the Y least-significant bits might differ, STREAMID is programmed > > + * with a value that contains: > > + * STREAMID[Y - 1] == 0. > > + * STREAMID[Y - 2:0] == 1 (where Y > 1). > > + * The remainder of implemented bits of STREAMID (X bits, from bit Y > upwards) > > + * contain a value to match from the corresponding bits of event StreamID. > > + * > > + * Example: perf stat -e > smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > + * filter_span=1,filter_stream_id=0x42/ -a netperf > > + * Applies filter pattern 0x42 to transaction events, which means events > > + * matching stream ids 0x42 and 0x43 are counted. Further filtering > > + * information is available in the SMMU documentation. > > + * > > + * SMMU events are not attributable to a CPU, so task mode and sampling > > + * are not supported. > > + */ > > + > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > + > > +#define SMMU_PMCG_EVCNTR0 0x0 > > +#define SMMU_PMCG_EVCNTR(n, stride) (SMMU_PMCG_EVCNTR0 + > (n) * (stride)) > > +#define SMMU_PMCG_EVTYPER0 0x400 > > +#define SMMU_PMCG_EVTYPER(n) > (SMMU_PMCG_EVTYPER0 + (n) * 4) > > +#define SMMU_PMCG_SID_SPAN_SHIFT 29 > > +#define SMMU_PMCG_SID_SPAN_MASK GENMASK(29, 29) > > +#define SMMU_PMCG_SMR0 0xA00 > > +#define SMMU_PMCG_SMR(n) (SMMU_PMCG_SMR0 + > (n) * 4) > > +#define SMMU_PMCG_CNTENSET0 0xC00 > > +#define SMMU_PMCG_CNTENCLR0 0xC20 > > +#define SMMU_PMCG_INTENSET0 0xC40 > > +#define SMMU_PMCG_INTENCLR0 0xC60 > > +#define SMMU_PMCG_OVSCLR0 0xC80 > > +#define SMMU_PMCG_OVSSET0 0xCC0 > > +#define SMMU_PMCG_CFGR 0xE00 > > +#define SMMU_PMCG_CFGR_RELOC_CTRS BIT(20) > > +#define SMMU_PMCG_CFGR_SID_FILTER_TYPE BIT(23) > > +#define SMMU_PMCG_CFGR_SIZE_MASK GENMASK(13, 8) > > +#define SMMU_PMCG_CFGR_NCTR_MASK GENMASK(5, 0) > > +#define SMMU_PMCG_CR 0xE04 > > +#define SMMU_PMCG_CR_ENABLE BIT(0) > > +#define SMMU_PMCG_CEID0 0xE20 > > +#define SMMU_PMCG_CEID1 0xE28 > > +#define SMMU_PMCG_IRQ_CTRL 0xE50 > > +#define SMMU_PMCG_IRQ_CTRL_IRQEN BIT(0) > > +#define SMMU_PMCG_IRQ_CFG0 0xE58 > > + > > +#define SMMU_DEFAULT_FILTER_SPAN 1 > > +#define SMMU_DEFAULT_FILTER_STREAM_ID GENMASK(31, 0) > > + > > +#define SMMU_MAX_COUNTERS 64 > > +#define SMMU_ARCH_MAX_EVENT_ID 0x7F > > +#define SMMU_IMPDEF_MAX_EVENT_ID 0xFFFF > > +#define SMMU_ARCH_MAX_EVENTS > (SMMU_ARCH_MAX_EVENT_ID + 1) > > + > > +#define SMMU_PA_SHIFT 12 > > + > > +static int cpuhp_state_num; > > + > > +struct smmu_pmu { > > + bool sid_filter_global; > > + struct hlist_node node; > > + struct perf_event *events[SMMU_MAX_COUNTERS]; > > + DECLARE_BITMAP(used_counters, SMMU_MAX_COUNTERS); > > + DECLARE_BITMAP(supported_events, SMMU_ARCH_MAX_EVENTS); > > + unsigned int irq; > > + unsigned int on_cpu; > > + struct pmu pmu; > > + unsigned int num_counters; > > + struct device *dev; > > + void __iomem *reg_base; > > + void __iomem *reloc_base; > > + u64 counter_present_mask; > > + u64 counter_mask; > > +}; > > + > > +#define to_smmu_pmu(p) (container_of(p, struct smmu_pmu, pmu)) > > + > > +#define SMMU_PMU_EVENT_ATTR_EXTRACTOR(_name, _config, _start, > _end) \ > > + static inline u32 get_##_name(struct perf_event *event) \ > > + > { > \ > > + return FIELD_GET(GENMASK_ULL(_end, _start), \ > > + event->attr._config); \ > > + } > \ > > + > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(event, config, 0, 15); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_stream_id, config1, 0, 31); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_span, config1, 32, 32); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_enable, config1, 33, 33); > > + > > +static inline void smmu_pmu_enable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(SMMU_PMCG_CR_ENABLE, smmu_pmu->reg_base + > SMMU_PMCG_CR); > > + writel(SMMU_PMCG_IRQ_CTRL_IRQEN, > > + smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > Should the interrupts be enabled before enabling the PMU? Yes, it make sense to do so. > > +} > > + > > +static inline void smmu_pmu_disable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_CR); > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > +} > > + > > +static inline void smmu_pmu_counter_set_value(struct smmu_pmu > *smmu_pmu, > > + u32 idx, u64 value) > > +{ > > + if (smmu_pmu->counter_mask & BIT(32)) > > + writeq(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + writel(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > The arm64 IO macros use __force u32 and so it's probably OK to provide a 64 > bit > value to writel. But you could use something like lower_32_bits for clarity. Yes, macro uses __force u32. I will change it to make it more clear though. > > +} > > + > > +static inline u64 smmu_pmu_counter_get_value(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + u64 value; > > + > > + if (smmu_pmu->counter_mask & BIT(32)) > > + value = readq(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + value = readl(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > + > > + return value; > > +} > > + > > +static inline void smmu_pmu_counter_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENSET0); > > +} > > + > > +static inline void smmu_pmu_counter_disable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > +} > > + > > +static inline void smmu_pmu_interrupt_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENSET0); > > +} > > + > > +static inline void smmu_pmu_interrupt_disable(struct smmu_pmu > *smmu_pmu, > > + u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > +} > > + > > +static inline void smmu_pmu_set_evtyper(struct smmu_pmu *smmu_pmu, > u32 idx, > > + u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx)); > > +} > > + > > +static inline void smmu_pmu_set_smr(struct smmu_pmu *smmu_pmu, u32 > idx, u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_SMR(idx)); > > +} > > + > > +static void smmu_pmu_event_update(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + u64 delta, prev, now; > > + u32 idx = hwc->idx; > > + > > + do { > > + prev = local64_read(&hwc->prev_count); > > + now = smmu_pmu_counter_get_value(smmu_pmu, idx); > > + } while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev); > > + > > + /* handle overflow. */ > > + delta = now - prev; > > + delta &= smmu_pmu->counter_mask; > > + > > + local64_add(delta, &event->count); > > +} > > + > > +static void smmu_pmu_set_period(struct smmu_pmu *smmu_pmu, > > + struct hw_perf_event *hwc) > > +{ > > + u32 idx = hwc->idx; > > + u64 new; > > + > > + /* > > + * We limit the max period to half the max counter value of the counter > > + * size, so that even in the case of extreme interrupt latency the > > + * counter will (hopefully) not wrap past its initial value. > > + */ > > + new = smmu_pmu->counter_mask >> 1; > > + > > + local64_set(&hwc->prev_count, new); > > + smmu_pmu_counter_set_value(smmu_pmu, idx, new); > > +} > > + > > +static void smmu_pmu_get_event_filter(struct perf_event *event, u32 > *span, > > + u32 *stream_id) > > +{ > > + bool filter_en = !!get_filter_enable(event); > > + > > + *span = filter_en ? get_filter_span(event) : > SMMU_DEFAULT_FILTER_SPAN; > > + *stream_id = filter_en ? get_filter_stream_id(event) : > > + SMMU_DEFAULT_FILTER_STREAM_ID; > > +} > > + > > +static bool smmu_pmu_event_filter_valid(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event, int idx) > > +{ > > + u32 filter_span, filter_sid; > > + u32 curr_span, curr_sid; > > + > > + if (!idx || !smmu_pmu->sid_filter_global) > > + return true; > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + /* Read the current global filter setting */ > > + curr_span = FIELD_GET(SMMU_PMCG_SID_SPAN_MASK, > > + readl(smmu_pmu->reg_base + > SMMU_PMCG_EVTYPER0)); > > + curr_sid = readl(smmu_pmu->reg_base + SMMU_PMCG_SMR0); > > + > > + if (filter_span != curr_span || filter_sid != curr_sid) > > + return false; > > + > > + return true; > > +} > > + > > +static int smmu_pmu_get_event_idx(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event) > > +{ > > + unsigned int idx; > > + unsigned int num_ctrs = smmu_pmu->num_counters; > > + > > + idx = find_first_zero_bit(smmu_pmu->used_counters, num_ctrs); > > + if (idx == num_ctrs) > > + /* The counters are all in use. */ > > + return -EAGAIN; > > + > > + if (!smmu_pmu_event_filter_valid(smmu_pmu, event, idx)) > > + return -EAGAIN; > > + > > + set_bit(idx, smmu_pmu->used_counters); > > + > > + return idx; > > +} > > + > > +/* > > + * Implementation of abstract pmu functionality required by > > + * the core perf events code. > > + */ > > + > > +static int smmu_pmu_event_init(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct device *dev = smmu_pmu->dev; > > + struct perf_event *sibling; > > + u32 event_id; > > + > > + if (event->attr.type != event->pmu->type) > > + return -ENOENT; > > + > > + if (hwc->sample_period) { > > + dev_dbg(dev, "Sampling not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + if (event->cpu < 0) { > > + dev_dbg(dev, "Per-task mode not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + /* We cannot filter accurately so we just don't allow it. */ > > + if (event->attr.exclude_user || event->attr.exclude_kernel || > > + event->attr.exclude_hv || event->attr.exclude_idle) { > > + dev_dbg(dev, "Can't exclude execution levels\n"); > > + return -EOPNOTSUPP; > > + } > > You should also test for attr.exclude_host and attr.exclude_guest, unless > it makes sense to stop counting these events when entering guest/host? TBH, I haven't considered that scenario yet and don't think we have to stop the counters. I will check that. > > Also returning -EINVAL may be more desirable than -EOPNOTSUPP. If a user > makes use of the perf utility and attempts to filter one of the above > exclusion flags - then perf will retry opening the event but without some > of the unsupported flags, but it will only do this if you return -EINVAL. > > (See line 1927 of tools/perf/util/evsel.c and run the perf stat tool with > the verbose flag to see this in action.) Ok. I will take a look. > > + > > + /* Verify specified event is supported on this PMU */ > > + event_id = get_event(event); > > + if (((event_id <= SMMU_ARCH_MAX_EVENT_ID) && > > + (!test_bit(event_id, smmu_pmu->supported_events))) || > > + (event_id > SMMU_IMPDEF_MAX_EVENT_ID)) { > > + dev_dbg(dev, "Invalid event %d for this PMU\n", event_id); > > + return -EINVAL; > > + } > > + > > + /* Don't allow groups with mixed PMUs, except for s/w events */ > > + if (event->group_leader->pmu != event->pmu && > > + !is_software_event(event->group_leader)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + list_for_each_entry(sibling, &event->group_leader->sibling_list, > > + sibling_list) > > + if (sibling->pmu != event->pmu && > > + !is_software_event(sibling)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + /* Ensure all events in a group are on the same cpu */ > > + if ((event->group_leader != event) && > > + (event->cpu != event->group_leader->cpu)) { > > + dev_dbg(dev, "Can't create group on CPUs %d and %d\n", > > + event->cpu, event->group_leader->cpu); > > + return -EINVAL; > > + } > > + > > + hwc->idx = -1; > > + > > + /* > > + * Ensure all events are on the same cpu so all events are in the > > + * same cpu context, to avoid races on pmu_enable etc. > > + */ > > + event->cpu = smmu_pmu->on_cpu; > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_start(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + u32 filter_span, filter_sid; > > + u32 evtyper; > > + > > + hwc->state = 0; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + evtyper = get_event(event) | > > + filter_span << SMMU_PMCG_SID_SPAN_SHIFT; > > + > > + smmu_pmu_set_evtyper(smmu_pmu, idx, evtyper); > > + smmu_pmu_set_smr(smmu_pmu, idx, filter_sid); > > + smmu_pmu_interrupt_enable(smmu_pmu, idx); > > + smmu_pmu_counter_enable(smmu_pmu, idx); > > +} > > + > > +static void smmu_pmu_event_stop(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + > > + if (hwc->state & PERF_HES_STOPPED) > > + return; > > + > > + smmu_pmu_counter_disable(smmu_pmu, idx); > > Is it intentional not to call smmu_pmu_interrupt_disable here? Yes, it is. Earlier patch had the interrupt toggling and Robin pointed out that it is not really needed as counters are anyway stopped and explicitly disabling may not solve the spurious interrupt case as well. > > + > > + if (flags & PERF_EF_UPDATE) > > + smmu_pmu_event_update(event); > > + hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE; > > Shouldn't we only set PERF_HES_UPTODATE if we've updated the count (e.g. > when PERF_EF_UPDATE was set?) Hmm..I had a look at couple of other drivers(arm-cci.c/arm_dsu_pmu.c etc) and those ones always call update ignoring PERF_EF_UPDATE and sets PERF_HES_UPTODATE. I think the reason being the counter is always reprogrammed on start. Do you think changing to that logic is a better way? > This may be a nit, but shouldn't smmu_pmu_event_update come before the > test for PERF_HES_STOPPED? I guess in the odd case where stop is called > without the PERF_EF_UPDATE flag (to stop the counter) and then called > a second time with PERF_EF_UPDATE to update the count. Not sure such a scenario exists. I think most of the ones I checked has the logic -its already stopped, nothing to do. > > +} > > + > > +static int smmu_pmu_event_add(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + int idx; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + > > + idx = smmu_pmu_get_event_idx(smmu_pmu, event); > > + if (idx < 0) > > + return idx; > > + > > + hwc->idx = idx; > > + hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE; > > + smmu_pmu->events[idx] = event; > > + local64_set(&hwc->prev_count, 0); > > + > > + if (flags & PERF_EF_START) > > + smmu_pmu_event_start(event, flags); > > + > > + /* Propagate changes to the userspace mapping. */ > > + perf_event_update_userpage(event); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_del(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + int idx = hwc->idx; > > + > > + smmu_pmu_event_stop(event, flags | PERF_EF_UPDATE); > > + smmu_pmu->events[idx] = NULL; > > + clear_bit(idx, smmu_pmu->used_counters); > > + > > + perf_event_update_userpage(event); > > +} > > + > > +static void smmu_pmu_event_read(struct perf_event *event) > > +{ > > + smmu_pmu_event_update(event); > > +} > > + > > +/* cpumask */ > > + > > +static ssize_t smmu_pmu_cpumask_show(struct device *dev, > > + struct device_attribute *attr, > > + char *buf) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + > > + return cpumap_print_to_pagebuf(true, buf, > cpumask_of(smmu_pmu->on_cpu)); > > +} > > + > > +static struct device_attribute smmu_pmu_cpumask_attr = > > + __ATTR(cpumask, 0444, smmu_pmu_cpumask_show, NULL); > > + > > +static struct attribute *smmu_pmu_cpumask_attrs[] = { > > + &smmu_pmu_cpumask_attr.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_cpumask_group = { > > + .attrs = smmu_pmu_cpumask_attrs, > > +}; > > + > > +/* Events */ > > + > > +ssize_t smmu_pmu_event_show(struct device *dev, > > + struct device_attribute *attr, char *page) > > +{ > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr); > > + > > + return sprintf(page, "event=0x%02llx\n", pmu_attr->id); > > +} > > + > > +#define SMMU_EVENT_ATTR(_name, _id) \ > > + (&((struct perf_pmu_events_attr[]) { \ > > + { .attr = __ATTR(_name, 0444, smmu_pmu_event_show, NULL), \ > > + .id = _id, } \ > > + })[0].attr.attr) > > This is very similar to PMU_EVENT_ATTR, have you considered taking an > approach > similar to the ARMV8_EVENT_ATTR macro that wraps up PMU_EVENT_ATTR. > It may > be more readable if we define these attributes consistently across PMUs. Ok. I will take a look at that. Thanks, Shameer > Thanks, > > Andrew Murray > > > + > > +static struct attribute *smmu_pmu_events[] = { > > + SMMU_EVENT_ATTR(cycles, 0), > > + SMMU_EVENT_ATTR(transaction, 1), > > + SMMU_EVENT_ATTR(tlb_miss, 2), > > + SMMU_EVENT_ATTR(config_cache_miss, 3), > > + SMMU_EVENT_ATTR(trans_table_walk_access, 4), > > + SMMU_EVENT_ATTR(config_struct_access, 5), > > + SMMU_EVENT_ATTR(pcie_ats_trans_rq, 6), > > + SMMU_EVENT_ATTR(pcie_ats_trans_passed, 7), > > + NULL > > +}; > > + > > +static umode_t smmu_pmu_event_is_visible(struct kobject *kobj, > > + struct attribute *attr, int unused) > > +{ > > + struct device *dev = kobj_to_dev(kobj); > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr); > > + > > + if (test_bit(pmu_attr->id, smmu_pmu->supported_events)) > > + return attr->mode; > > + > > + return 0; > > +} > > +static struct attribute_group smmu_pmu_events_group = { > > + .name = "events", > > + .attrs = smmu_pmu_events, > > + .is_visible = smmu_pmu_event_is_visible, > > +}; > > + > > +/* Formats */ > > +PMU_FORMAT_ATTR(event, "config:0-15"); > > +PMU_FORMAT_ATTR(filter_stream_id, "config1:0-31"); > > +PMU_FORMAT_ATTR(filter_span, "config1:32"); > > +PMU_FORMAT_ATTR(filter_enable, "config1:33"); > > + > > +static struct attribute *smmu_pmu_formats[] = { > > + &format_attr_event.attr, > > + &format_attr_filter_stream_id.attr, > > + &format_attr_filter_span.attr, > > + &format_attr_filter_enable.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_format_group = { > > + .name = "format", > > + .attrs = smmu_pmu_formats, > > +}; > > + > > +static const struct attribute_group *smmu_pmu_attr_grps[] = { > > + &smmu_pmu_cpumask_group, > > + &smmu_pmu_events_group, > > + &smmu_pmu_format_group, > > + NULL > > +}; > > + > > +/* > > + * Generic device handlers > > + */ > > + > > +static int smmu_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + unsigned int target; > > + > > + smmu_pmu = hlist_entry_safe(node, struct smmu_pmu, node); > > + if (cpu != smmu_pmu->on_cpu) > > + return 0; > > + > > + target = cpumask_any_but(cpu_online_mask, cpu); > > + if (target >= nr_cpu_ids) > > + return 0; > > + > > + perf_pmu_migrate_context(&smmu_pmu->pmu, cpu, target); > > + smmu_pmu->on_cpu = target; > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, cpumask_of(target))); > > + > > + return 0; > > +} > > + > > +static irqreturn_t smmu_pmu_handle_irq(int irq_num, void *data) > > +{ > > + struct smmu_pmu *smmu_pmu = data; > > + u64 ovsr; > > + unsigned int idx; > > + > > + ovsr = readq(smmu_pmu->reloc_base + SMMU_PMCG_OVSSET0); > > + if (!ovsr) > > + return IRQ_NONE; > > + > > + writeq(ovsr, smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > + > > + for_each_set_bit(idx, (unsigned long *)&ovsr, smmu_pmu->num_counters) > { > > + struct perf_event *event = smmu_pmu->events[idx]; > > + struct hw_perf_event *hwc; > > + > > + if (WARN_ON_ONCE(!event)) > > + continue; > > + > > + smmu_pmu_event_update(event); > > + hwc = &event->hw; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + } > > + > > + return IRQ_HANDLED; > > +} > > + > > +static int smmu_pmu_setup_irq(struct smmu_pmu *pmu) > > +{ > > + unsigned long flags = IRQF_NOBALANCING | IRQF_SHARED | > IRQF_NO_THREAD; > > + int irq, ret = -ENXIO; > > + > > + irq = pmu->irq; > > + if (irq) > > + ret = devm_request_irq(pmu->dev, irq, smmu_pmu_handle_irq, > > + flags, "smmuv3-pmu", pmu); > > + return ret; > > +} > > + > > +static void smmu_pmu_reset(struct smmu_pmu *smmu_pmu) > > +{ > > + smmu_pmu_disable(&smmu_pmu->pmu); > > + > > + /* Disable counter and interrupt */ > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > +} > > + > > +static int smmu_pmu_probe(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + struct resource *res_0, *res_1; > > + u32 cfgr, reg_size; > > + u64 ceid_64[2]; > > + int irq, err; > > + char *name; > > + struct device *dev = &pdev->dev; > > + > > + smmu_pmu = devm_kzalloc(dev, sizeof(*smmu_pmu), GFP_KERNEL); > > + if (!smmu_pmu) > > + return -ENOMEM; > > + > > + smmu_pmu->dev = dev; > > + platform_set_drvdata(pdev, smmu_pmu); > > + > > + smmu_pmu->pmu = (struct pmu) { > > + .task_ctx_nr = perf_invalid_context, > > + .pmu_enable = smmu_pmu_enable, > > + .pmu_disable = smmu_pmu_disable, > > + .event_init = smmu_pmu_event_init, > > + .add = smmu_pmu_event_add, > > + .del = smmu_pmu_event_del, > > + .start = smmu_pmu_event_start, > > + .stop = smmu_pmu_event_stop, > > + .read = smmu_pmu_event_read, > > + .attr_groups = smmu_pmu_attr_grps, > > + }; > > + > > + res_0 = platform_get_resource(pdev, IORESOURCE_MEM, 0); > > + smmu_pmu->reg_base = devm_ioremap_resource(dev, res_0); > > + if (IS_ERR(smmu_pmu->reg_base)) > > + return PTR_ERR(smmu_pmu->reg_base); > > + > > + cfgr = readl_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CFGR); > > + > > + /* Determine if page 1 is present */ > > + if (cfgr & SMMU_PMCG_CFGR_RELOC_CTRS) { > > + res_1 = platform_get_resource(pdev, IORESOURCE_MEM, 1); > > + smmu_pmu->reloc_base = devm_ioremap_resource(dev, res_1); > > + if (IS_ERR(smmu_pmu->reloc_base)) > > + return PTR_ERR(smmu_pmu->reloc_base); > > + } else { > > + smmu_pmu->reloc_base = smmu_pmu->reg_base; > > + } > > + > > + irq = platform_get_irq(pdev, 0); > > + if (irq > 0) > > + smmu_pmu->irq = irq; > > + > > + ceid_64[0] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID0); > > + ceid_64[1] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID1); > > + bitmap_from_arr32(smmu_pmu->supported_events, (u32 *)ceid_64, > > + SMMU_ARCH_MAX_EVENTS); > > + > > + smmu_pmu->num_counters = > FIELD_GET(SMMU_PMCG_CFGR_NCTR_MASK, cfgr) + 1; > > + smmu_pmu->counter_present_mask = > GENMASK(smmu_pmu->num_counters - 1, 0); > > + > > + smmu_pmu->sid_filter_global = !!(cfgr & > SMMU_PMCG_CFGR_SID_FILTER_TYPE); > > + > > + reg_size = FIELD_GET(SMMU_PMCG_CFGR_SIZE_MASK, cfgr); > > + smmu_pmu->counter_mask = GENMASK_ULL(reg_size, 0); > > + > > + smmu_pmu_reset(smmu_pmu); > > + > > + err = smmu_pmu_setup_irq(smmu_pmu); > > + if (err) { > > + dev_err(dev, "Setup irq failed, PMU @%pa\n", &res_0->start); > > + return err; > > + } > > + > > + name = devm_kasprintf(&pdev->dev, GFP_KERNEL, > "smmuv3_pmcg_%llx", > > + (res_0->start) >> SMMU_PA_SHIFT); > > + if (!name) { > > + dev_err(dev, "Create name failed, PMU @%pa\n", &res_0->start); > > + return -EINVAL; > > + } > > + > > + /* Pick one CPU to be the preferred one to use */ > > + smmu_pmu->on_cpu = get_cpu(); > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, > cpumask_of(smmu_pmu->on_cpu))); > > + > > + err = cpuhp_state_add_instance_nocalls(cpuhp_state_num, > > + &smmu_pmu->node); > > + if (err) { > > + dev_err(dev, "Error %d registering hotplug, PMU @%pa\n", > > + err, &res_0->start); > > + goto out_cpuhp_err; > > + } > > + > > + err = perf_pmu_register(&smmu_pmu->pmu, name, -1); > > + if (err) { > > + dev_err(dev, "Error %d registering PMU @%pa\n", > > + err, &res_0->start); > > + goto out_unregister; > > + } > > + > > + put_cpu(); > > + dev_info(dev, "Registered PMU @ %pa using %d counters with %s filter > settings\n", > > + &res_0->start, smmu_pmu->num_counters, > > + smmu_pmu->sid_filter_global ? "Global(Counter0)" : > > + "Individual"); > > + > > + return 0; > > + > > +out_unregister: > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > +out_cpuhp_err: > > + put_cpu(); > > + return err; > > +} > > + > > +static int smmu_pmu_remove(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + perf_pmu_unregister(&smmu_pmu->pmu); > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_shutdown(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + smmu_pmu_disable(&smmu_pmu->pmu); > > +} > > + > > +static struct platform_driver smmu_pmu_driver = { > > + .driver = { > > + .name = "arm-smmu-v3-pmu", > > + }, > > + .probe = smmu_pmu_probe, > > + .remove = smmu_pmu_remove, > > + .shutdown = smmu_pmu_shutdown, > > +}; > > + > > +static int __init arm_smmu_pmu_init(void) > > +{ > > + cpuhp_state_num = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, > > + "perf/arm/pmcg:online", > > + NULL, > > + smmu_pmu_offline_cpu); > > + if (cpuhp_state_num < 0) > > + return cpuhp_state_num; > > + > > + return platform_driver_register(&smmu_pmu_driver); > > +} > > +module_init(arm_smmu_pmu_init); > > + > > +static void __exit arm_smmu_pmu_exit(void) > > +{ > > + platform_driver_unregister(&smmu_pmu_driver); > > + cpuhp_remove_multi_state(cpuhp_state_num); > > +} > > + > > +module_exit(arm_smmu_pmu_exit); > > + > > +MODULE_DESCRIPTION("PMU driver for ARM SMMUv3 Performance > Monitors Extension"); > > +MODULE_AUTHOR("Neil Leeder "); > > +MODULE_AUTHOR("Shameer Kolothum > "); > > +MODULE_LICENSE("GPL v2"); > > -- > > 2.7.4 > > > > > > > > _______________________________________________ > > linux-arm-kernel mailing list > > linux-arm-kernel@lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.0 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_PASS,URIBL_BLOCKED autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 9FC59C282C2 for ; Wed, 23 Jan 2019 11:03:05 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 5666C21019 for ; Wed, 23 Jan 2019 11:03:05 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1727593AbfAWLDD convert rfc822-to-8bit (ORCPT ); Wed, 23 Jan 2019 06:03:03 -0500 Received: from lhrrgout.huawei.com ([185.176.76.210]:32834 "EHLO huawei.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1726322AbfAWLDC (ORCPT ); Wed, 23 Jan 2019 06:03:02 -0500 Received: from LHREML710-CAH.china.huawei.com (unknown [172.18.7.107]) by Forcepoint Email with ESMTP id 110CA88E94A8DD5A75A6; Wed, 23 Jan 2019 11:03:00 +0000 (GMT) Received: from LHREML524-MBS.china.huawei.com ([169.254.2.138]) by LHREML710-CAH.china.huawei.com ([10.201.108.33]) with mapi id 14.03.0415.000; Wed, 23 Jan 2019 11:02:49 +0000 From: Shameerali Kolothum Thodi To: Andrew Murray CC: "lorenzo.pieralisi@arm.com" , "robin.murphy@arm.com" , "mark.rutland@arm.com" , "vkilari@codeaurora.org" , "neil.m.leeder@gmail.com" , "jean-philippe.brucker@arm.com" , "pabba@codeaurora.org" , John Garry , "will.deacon@arm.com" , "rruigrok@codeaurora.org" , Linuxarm , "linux-kernel@vger.kernel.org" , "linux-acpi@vger.kernel.org" , "Guohanjun (Hanjun Guo)" , "linux-arm-kernel@lists.infradead.org" Subject: RE: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver Thread-Topic: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver Thread-Index: AQHUiMReCYaliznTZUOBIOWTrrVwNqW7zC+AgAEf2sA= Date: Wed, 23 Jan 2019 11:02:48 +0000 Message-ID: <5FC3163CFD30C246ABAA99954A238FA83929855B@lhreml524-mbs.china.huawei.com> References: <20181130154751.28580-1-shameerali.kolothum.thodi@huawei.com> <20181130154751.28580-3-shameerali.kolothum.thodi@huawei.com> <20190122162351.GE8120@e119886-lin.cambridge.arm.com> In-Reply-To: <20190122162351.GE8120@e119886-lin.cambridge.arm.com> Accept-Language: en-GB, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.202.227.237] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT MIME-Version: 1.0 X-CFilter-Loop: Reflected Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Andrew, Thanks for taking a look at this. > -----Original Message----- > From: Andrew Murray [mailto:andrew.murray@arm.com] > Sent: 22 January 2019 16:24 > To: Shameerali Kolothum Thodi > Cc: lorenzo.pieralisi@arm.com; robin.murphy@arm.com; > mark.rutland@arm.com; vkilari@codeaurora.org; neil.m.leeder@gmail.com; > jean-philippe.brucker@arm.com; pabba@codeaurora.org; John Garry > ; will.deacon@arm.com; rruigrok@codeaurora.org; > Linuxarm ; linux-kernel@vger.kernel.org; > linux-acpi@vger.kernel.org; Guohanjun (Hanjun Guo) > ; linux-arm-kernel@lists.infradead.org > Subject: Re: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver > > On Fri, Nov 30, 2018 at 03:47:49PM +0000, Shameer Kolothum wrote: > > From: Neil Leeder > > > > Adds a new driver to support the SMMUv3 PMU and add it into the > > perf events framework. > > > > Each SMMU node may have multiple PMUs associated with it, each of > > which may support different events. > > > > SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > is the physical page address of the SMMU PMCG > > wrapped to 4K boundary. For example, the PMCG at 0xff88840000 is > > named smmuv3_pmcg_ff88840 > > > > Filtering by stream id is done by specifying filtering parameters > > with the event. options are: > > filter_enable - 0 = no filtering, 1 = filtering enabled > > filter_span - 0 = exact match, 1 = pattern match > > filter_stream_id - pattern to filter against > > > > Example: perf stat -e smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > filter_span=1,filter_stream_id=0x42/ -a netperf > > > > Applies filter pattern 0x42 to transaction events, which means events > > matching stream ids 0x42 & 0x43 are counted as only upper StreamID > > bits are required to match the given filter. Further filtering > > information is available in the SMMU documentation. > > > > SMMU events are not attributable to a CPU, so task mode and sampling > > are not supported. > > > > Signed-off-by: Neil Leeder > > Signed-off-by: Shameer Kolothum > > --- > > drivers/perf/Kconfig | 9 + > > drivers/perf/Makefile | 1 + > > drivers/perf/arm_smmuv3_pmu.c | 778 > ++++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 788 insertions(+) > > create mode 100644 drivers/perf/arm_smmuv3_pmu.c > > > > diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig > > index 08ebaf7..92be6a3 100644 > > --- a/drivers/perf/Kconfig > > +++ b/drivers/perf/Kconfig > > @@ -52,6 +52,15 @@ config ARM_PMU_ACPI > > depends on ARM_PMU && ACPI > > def_bool y > > > > +config ARM_SMMU_V3_PMU > > + bool "ARM SMMUv3 Performance Monitors Extension" > > + depends on ARM64 && ACPI && ARM_SMMU_V3 > > + help > > + Provides support for the SMMU version 3 performance monitor unit > (PMU) > > + on ARM-based systems. > > + Adds the SMMU PMU into the perf events subsystem for > > + monitoring SMMU performance events. > > + > > config ARM_DSU_PMU > > tristate "ARM DynamIQ Shared Unit (DSU) PMU" > > depends on ARM64 > > diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile > > index b3902bd..f10a932 100644 > > --- a/drivers/perf/Makefile > > +++ b/drivers/perf/Makefile > > @@ -4,6 +4,7 @@ obj-$(CONFIG_ARM_CCN) += arm-ccn.o > > obj-$(CONFIG_ARM_DSU_PMU) += arm_dsu_pmu.o > > obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o > > obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o > > +obj-$(CONFIG_ARM_SMMU_V3_PMU) += arm_smmuv3_pmu.o > > obj-$(CONFIG_HISI_PMU) += hisilicon/ > > obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o > > obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o > > diff --git a/drivers/perf/arm_smmuv3_pmu.c > b/drivers/perf/arm_smmuv3_pmu.c > > new file mode 100644 > > index 0000000..fb9dcd8 > > --- /dev/null > > +++ b/drivers/perf/arm_smmuv3_pmu.c > > @@ -0,0 +1,778 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > + > > +/* > > + * This driver adds support for perf events to use the Performance > > + * Monitor Counter Groups (PMCG) associated with an SMMUv3 node > > + * to monitor that node. > > + * > > + * SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > + * is the physical page address of the SMMU PMCG > wrapped > > + * to 4K boundary. For example, the PMCG at 0xff88840000 is named > > + * smmuv3_pmcg_ff88840 > > + * > > + * Filtering by stream id is done by specifying filtering parameters > > + * with the event. options are: > > + * filter_enable - 0 = no filtering, 1 = filtering enabled > > + * filter_span - 0 = exact match, 1 = pattern match > > + * filter_stream_id - pattern to filter against > > + * > > + * To match a partial StreamID where the X most-significant bits must match > > + * but the Y least-significant bits might differ, STREAMID is programmed > > + * with a value that contains: > > + * STREAMID[Y - 1] == 0. > > + * STREAMID[Y - 2:0] == 1 (where Y > 1). > > + * The remainder of implemented bits of STREAMID (X bits, from bit Y > upwards) > > + * contain a value to match from the corresponding bits of event StreamID. > > + * > > + * Example: perf stat -e > smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > + * filter_span=1,filter_stream_id=0x42/ -a netperf > > + * Applies filter pattern 0x42 to transaction events, which means events > > + * matching stream ids 0x42 and 0x43 are counted. Further filtering > > + * information is available in the SMMU documentation. > > + * > > + * SMMU events are not attributable to a CPU, so task mode and sampling > > + * are not supported. > > + */ > > + > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > + > > +#define SMMU_PMCG_EVCNTR0 0x0 > > +#define SMMU_PMCG_EVCNTR(n, stride) (SMMU_PMCG_EVCNTR0 + > (n) * (stride)) > > +#define SMMU_PMCG_EVTYPER0 0x400 > > +#define SMMU_PMCG_EVTYPER(n) > (SMMU_PMCG_EVTYPER0 + (n) * 4) > > +#define SMMU_PMCG_SID_SPAN_SHIFT 29 > > +#define SMMU_PMCG_SID_SPAN_MASK GENMASK(29, 29) > > +#define SMMU_PMCG_SMR0 0xA00 > > +#define SMMU_PMCG_SMR(n) (SMMU_PMCG_SMR0 + > (n) * 4) > > +#define SMMU_PMCG_CNTENSET0 0xC00 > > +#define SMMU_PMCG_CNTENCLR0 0xC20 > > +#define SMMU_PMCG_INTENSET0 0xC40 > > +#define SMMU_PMCG_INTENCLR0 0xC60 > > +#define SMMU_PMCG_OVSCLR0 0xC80 > > +#define SMMU_PMCG_OVSSET0 0xCC0 > > +#define SMMU_PMCG_CFGR 0xE00 > > +#define SMMU_PMCG_CFGR_RELOC_CTRS BIT(20) > > +#define SMMU_PMCG_CFGR_SID_FILTER_TYPE BIT(23) > > +#define SMMU_PMCG_CFGR_SIZE_MASK GENMASK(13, 8) > > +#define SMMU_PMCG_CFGR_NCTR_MASK GENMASK(5, 0) > > +#define SMMU_PMCG_CR 0xE04 > > +#define SMMU_PMCG_CR_ENABLE BIT(0) > > +#define SMMU_PMCG_CEID0 0xE20 > > +#define SMMU_PMCG_CEID1 0xE28 > > +#define SMMU_PMCG_IRQ_CTRL 0xE50 > > +#define SMMU_PMCG_IRQ_CTRL_IRQEN BIT(0) > > +#define SMMU_PMCG_IRQ_CFG0 0xE58 > > + > > +#define SMMU_DEFAULT_FILTER_SPAN 1 > > +#define SMMU_DEFAULT_FILTER_STREAM_ID GENMASK(31, 0) > > + > > +#define SMMU_MAX_COUNTERS 64 > > +#define SMMU_ARCH_MAX_EVENT_ID 0x7F > > +#define SMMU_IMPDEF_MAX_EVENT_ID 0xFFFF > > +#define SMMU_ARCH_MAX_EVENTS > (SMMU_ARCH_MAX_EVENT_ID + 1) > > + > > +#define SMMU_PA_SHIFT 12 > > + > > +static int cpuhp_state_num; > > + > > +struct smmu_pmu { > > + bool sid_filter_global; > > + struct hlist_node node; > > + struct perf_event *events[SMMU_MAX_COUNTERS]; > > + DECLARE_BITMAP(used_counters, SMMU_MAX_COUNTERS); > > + DECLARE_BITMAP(supported_events, SMMU_ARCH_MAX_EVENTS); > > + unsigned int irq; > > + unsigned int on_cpu; > > + struct pmu pmu; > > + unsigned int num_counters; > > + struct device *dev; > > + void __iomem *reg_base; > > + void __iomem *reloc_base; > > + u64 counter_present_mask; > > + u64 counter_mask; > > +}; > > + > > +#define to_smmu_pmu(p) (container_of(p, struct smmu_pmu, pmu)) > > + > > +#define SMMU_PMU_EVENT_ATTR_EXTRACTOR(_name, _config, _start, > _end) \ > > + static inline u32 get_##_name(struct perf_event *event) \ > > + > { > \ > > + return FIELD_GET(GENMASK_ULL(_end, _start), \ > > + event->attr._config); \ > > + } > \ > > + > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(event, config, 0, 15); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_stream_id, config1, 0, 31); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_span, config1, 32, 32); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_enable, config1, 33, 33); > > + > > +static inline void smmu_pmu_enable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(SMMU_PMCG_CR_ENABLE, smmu_pmu->reg_base + > SMMU_PMCG_CR); > > + writel(SMMU_PMCG_IRQ_CTRL_IRQEN, > > + smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > Should the interrupts be enabled before enabling the PMU? Yes, it make sense to do so. > > +} > > + > > +static inline void smmu_pmu_disable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_CR); > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > +} > > + > > +static inline void smmu_pmu_counter_set_value(struct smmu_pmu > *smmu_pmu, > > + u32 idx, u64 value) > > +{ > > + if (smmu_pmu->counter_mask & BIT(32)) > > + writeq(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + writel(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > The arm64 IO macros use __force u32 and so it's probably OK to provide a 64 > bit > value to writel. But you could use something like lower_32_bits for clarity. Yes, macro uses __force u32. I will change it to make it more clear though. > > +} > > + > > +static inline u64 smmu_pmu_counter_get_value(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + u64 value; > > + > > + if (smmu_pmu->counter_mask & BIT(32)) > > + value = readq(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + value = readl(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > + > > + return value; > > +} > > + > > +static inline void smmu_pmu_counter_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENSET0); > > +} > > + > > +static inline void smmu_pmu_counter_disable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > +} > > + > > +static inline void smmu_pmu_interrupt_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENSET0); > > +} > > + > > +static inline void smmu_pmu_interrupt_disable(struct smmu_pmu > *smmu_pmu, > > + u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > +} > > + > > +static inline void smmu_pmu_set_evtyper(struct smmu_pmu *smmu_pmu, > u32 idx, > > + u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx)); > > +} > > + > > +static inline void smmu_pmu_set_smr(struct smmu_pmu *smmu_pmu, u32 > idx, u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_SMR(idx)); > > +} > > + > > +static void smmu_pmu_event_update(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + u64 delta, prev, now; > > + u32 idx = hwc->idx; > > + > > + do { > > + prev = local64_read(&hwc->prev_count); > > + now = smmu_pmu_counter_get_value(smmu_pmu, idx); > > + } while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev); > > + > > + /* handle overflow. */ > > + delta = now - prev; > > + delta &= smmu_pmu->counter_mask; > > + > > + local64_add(delta, &event->count); > > +} > > + > > +static void smmu_pmu_set_period(struct smmu_pmu *smmu_pmu, > > + struct hw_perf_event *hwc) > > +{ > > + u32 idx = hwc->idx; > > + u64 new; > > + > > + /* > > + * We limit the max period to half the max counter value of the counter > > + * size, so that even in the case of extreme interrupt latency the > > + * counter will (hopefully) not wrap past its initial value. > > + */ > > + new = smmu_pmu->counter_mask >> 1; > > + > > + local64_set(&hwc->prev_count, new); > > + smmu_pmu_counter_set_value(smmu_pmu, idx, new); > > +} > > + > > +static void smmu_pmu_get_event_filter(struct perf_event *event, u32 > *span, > > + u32 *stream_id) > > +{ > > + bool filter_en = !!get_filter_enable(event); > > + > > + *span = filter_en ? get_filter_span(event) : > SMMU_DEFAULT_FILTER_SPAN; > > + *stream_id = filter_en ? get_filter_stream_id(event) : > > + SMMU_DEFAULT_FILTER_STREAM_ID; > > +} > > + > > +static bool smmu_pmu_event_filter_valid(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event, int idx) > > +{ > > + u32 filter_span, filter_sid; > > + u32 curr_span, curr_sid; > > + > > + if (!idx || !smmu_pmu->sid_filter_global) > > + return true; > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + /* Read the current global filter setting */ > > + curr_span = FIELD_GET(SMMU_PMCG_SID_SPAN_MASK, > > + readl(smmu_pmu->reg_base + > SMMU_PMCG_EVTYPER0)); > > + curr_sid = readl(smmu_pmu->reg_base + SMMU_PMCG_SMR0); > > + > > + if (filter_span != curr_span || filter_sid != curr_sid) > > + return false; > > + > > + return true; > > +} > > + > > +static int smmu_pmu_get_event_idx(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event) > > +{ > > + unsigned int idx; > > + unsigned int num_ctrs = smmu_pmu->num_counters; > > + > > + idx = find_first_zero_bit(smmu_pmu->used_counters, num_ctrs); > > + if (idx == num_ctrs) > > + /* The counters are all in use. */ > > + return -EAGAIN; > > + > > + if (!smmu_pmu_event_filter_valid(smmu_pmu, event, idx)) > > + return -EAGAIN; > > + > > + set_bit(idx, smmu_pmu->used_counters); > > + > > + return idx; > > +} > > + > > +/* > > + * Implementation of abstract pmu functionality required by > > + * the core perf events code. > > + */ > > + > > +static int smmu_pmu_event_init(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct device *dev = smmu_pmu->dev; > > + struct perf_event *sibling; > > + u32 event_id; > > + > > + if (event->attr.type != event->pmu->type) > > + return -ENOENT; > > + > > + if (hwc->sample_period) { > > + dev_dbg(dev, "Sampling not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + if (event->cpu < 0) { > > + dev_dbg(dev, "Per-task mode not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + /* We cannot filter accurately so we just don't allow it. */ > > + if (event->attr.exclude_user || event->attr.exclude_kernel || > > + event->attr.exclude_hv || event->attr.exclude_idle) { > > + dev_dbg(dev, "Can't exclude execution levels\n"); > > + return -EOPNOTSUPP; > > + } > > You should also test for attr.exclude_host and attr.exclude_guest, unless > it makes sense to stop counting these events when entering guest/host? TBH, I haven't considered that scenario yet and don't think we have to stop the counters. I will check that. > > Also returning -EINVAL may be more desirable than -EOPNOTSUPP. If a user > makes use of the perf utility and attempts to filter one of the above > exclusion flags - then perf will retry opening the event but without some > of the unsupported flags, but it will only do this if you return -EINVAL. > > (See line 1927 of tools/perf/util/evsel.c and run the perf stat tool with > the verbose flag to see this in action.) Ok. I will take a look. > > + > > + /* Verify specified event is supported on this PMU */ > > + event_id = get_event(event); > > + if (((event_id <= SMMU_ARCH_MAX_EVENT_ID) && > > + (!test_bit(event_id, smmu_pmu->supported_events))) || > > + (event_id > SMMU_IMPDEF_MAX_EVENT_ID)) { > > + dev_dbg(dev, "Invalid event %d for this PMU\n", event_id); > > + return -EINVAL; > > + } > > + > > + /* Don't allow groups with mixed PMUs, except for s/w events */ > > + if (event->group_leader->pmu != event->pmu && > > + !is_software_event(event->group_leader)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + list_for_each_entry(sibling, &event->group_leader->sibling_list, > > + sibling_list) > > + if (sibling->pmu != event->pmu && > > + !is_software_event(sibling)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + /* Ensure all events in a group are on the same cpu */ > > + if ((event->group_leader != event) && > > + (event->cpu != event->group_leader->cpu)) { > > + dev_dbg(dev, "Can't create group on CPUs %d and %d\n", > > + event->cpu, event->group_leader->cpu); > > + return -EINVAL; > > + } > > + > > + hwc->idx = -1; > > + > > + /* > > + * Ensure all events are on the same cpu so all events are in the > > + * same cpu context, to avoid races on pmu_enable etc. > > + */ > > + event->cpu = smmu_pmu->on_cpu; > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_start(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + u32 filter_span, filter_sid; > > + u32 evtyper; > > + > > + hwc->state = 0; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + evtyper = get_event(event) | > > + filter_span << SMMU_PMCG_SID_SPAN_SHIFT; > > + > > + smmu_pmu_set_evtyper(smmu_pmu, idx, evtyper); > > + smmu_pmu_set_smr(smmu_pmu, idx, filter_sid); > > + smmu_pmu_interrupt_enable(smmu_pmu, idx); > > + smmu_pmu_counter_enable(smmu_pmu, idx); > > +} > > + > > +static void smmu_pmu_event_stop(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + > > + if (hwc->state & PERF_HES_STOPPED) > > + return; > > + > > + smmu_pmu_counter_disable(smmu_pmu, idx); > > Is it intentional not to call smmu_pmu_interrupt_disable here? Yes, it is. Earlier patch had the interrupt toggling and Robin pointed out that it is not really needed as counters are anyway stopped and explicitly disabling may not solve the spurious interrupt case as well. > > + > > + if (flags & PERF_EF_UPDATE) > > + smmu_pmu_event_update(event); > > + hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE; > > Shouldn't we only set PERF_HES_UPTODATE if we've updated the count (e.g. > when PERF_EF_UPDATE was set?) Hmm..I had a look at couple of other drivers(arm-cci.c/arm_dsu_pmu.c etc) and those ones always call update ignoring PERF_EF_UPDATE and sets PERF_HES_UPTODATE. I think the reason being the counter is always reprogrammed on start. Do you think changing to that logic is a better way? > This may be a nit, but shouldn't smmu_pmu_event_update come before the > test for PERF_HES_STOPPED? I guess in the odd case where stop is called > without the PERF_EF_UPDATE flag (to stop the counter) and then called > a second time with PERF_EF_UPDATE to update the count. Not sure such a scenario exists. I think most of the ones I checked has the logic -its already stopped, nothing to do. > > +} > > + > > +static int smmu_pmu_event_add(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + int idx; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + > > + idx = smmu_pmu_get_event_idx(smmu_pmu, event); > > + if (idx < 0) > > + return idx; > > + > > + hwc->idx = idx; > > + hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE; > > + smmu_pmu->events[idx] = event; > > + local64_set(&hwc->prev_count, 0); > > + > > + if (flags & PERF_EF_START) > > + smmu_pmu_event_start(event, flags); > > + > > + /* Propagate changes to the userspace mapping. */ > > + perf_event_update_userpage(event); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_del(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + int idx = hwc->idx; > > + > > + smmu_pmu_event_stop(event, flags | PERF_EF_UPDATE); > > + smmu_pmu->events[idx] = NULL; > > + clear_bit(idx, smmu_pmu->used_counters); > > + > > + perf_event_update_userpage(event); > > +} > > + > > +static void smmu_pmu_event_read(struct perf_event *event) > > +{ > > + smmu_pmu_event_update(event); > > +} > > + > > +/* cpumask */ > > + > > +static ssize_t smmu_pmu_cpumask_show(struct device *dev, > > + struct device_attribute *attr, > > + char *buf) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + > > + return cpumap_print_to_pagebuf(true, buf, > cpumask_of(smmu_pmu->on_cpu)); > > +} > > + > > +static struct device_attribute smmu_pmu_cpumask_attr = > > + __ATTR(cpumask, 0444, smmu_pmu_cpumask_show, NULL); > > + > > +static struct attribute *smmu_pmu_cpumask_attrs[] = { > > + &smmu_pmu_cpumask_attr.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_cpumask_group = { > > + .attrs = smmu_pmu_cpumask_attrs, > > +}; > > + > > +/* Events */ > > + > > +ssize_t smmu_pmu_event_show(struct device *dev, > > + struct device_attribute *attr, char *page) > > +{ > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr); > > + > > + return sprintf(page, "event=0x%02llx\n", pmu_attr->id); > > +} > > + > > +#define SMMU_EVENT_ATTR(_name, _id) \ > > + (&((struct perf_pmu_events_attr[]) { \ > > + { .attr = __ATTR(_name, 0444, smmu_pmu_event_show, NULL), \ > > + .id = _id, } \ > > + })[0].attr.attr) > > This is very similar to PMU_EVENT_ATTR, have you considered taking an > approach > similar to the ARMV8_EVENT_ATTR macro that wraps up PMU_EVENT_ATTR. > It may > be more readable if we define these attributes consistently across PMUs. Ok. I will take a look at that. Thanks, Shameer > Thanks, > > Andrew Murray > > > + > > +static struct attribute *smmu_pmu_events[] = { > > + SMMU_EVENT_ATTR(cycles, 0), > > + SMMU_EVENT_ATTR(transaction, 1), > > + SMMU_EVENT_ATTR(tlb_miss, 2), > > + SMMU_EVENT_ATTR(config_cache_miss, 3), > > + SMMU_EVENT_ATTR(trans_table_walk_access, 4), > > + SMMU_EVENT_ATTR(config_struct_access, 5), > > + SMMU_EVENT_ATTR(pcie_ats_trans_rq, 6), > > + SMMU_EVENT_ATTR(pcie_ats_trans_passed, 7), > > + NULL > > +}; > > + > > +static umode_t smmu_pmu_event_is_visible(struct kobject *kobj, > > + struct attribute *attr, int unused) > > +{ > > + struct device *dev = kobj_to_dev(kobj); > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr); > > + > > + if (test_bit(pmu_attr->id, smmu_pmu->supported_events)) > > + return attr->mode; > > + > > + return 0; > > +} > > +static struct attribute_group smmu_pmu_events_group = { > > + .name = "events", > > + .attrs = smmu_pmu_events, > > + .is_visible = smmu_pmu_event_is_visible, > > +}; > > + > > +/* Formats */ > > +PMU_FORMAT_ATTR(event, "config:0-15"); > > +PMU_FORMAT_ATTR(filter_stream_id, "config1:0-31"); > > +PMU_FORMAT_ATTR(filter_span, "config1:32"); > > +PMU_FORMAT_ATTR(filter_enable, "config1:33"); > > + > > +static struct attribute *smmu_pmu_formats[] = { > > + &format_attr_event.attr, > > + &format_attr_filter_stream_id.attr, > > + &format_attr_filter_span.attr, > > + &format_attr_filter_enable.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_format_group = { > > + .name = "format", > > + .attrs = smmu_pmu_formats, > > +}; > > + > > +static const struct attribute_group *smmu_pmu_attr_grps[] = { > > + &smmu_pmu_cpumask_group, > > + &smmu_pmu_events_group, > > + &smmu_pmu_format_group, > > + NULL > > +}; > > + > > +/* > > + * Generic device handlers > > + */ > > + > > +static int smmu_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + unsigned int target; > > + > > + smmu_pmu = hlist_entry_safe(node, struct smmu_pmu, node); > > + if (cpu != smmu_pmu->on_cpu) > > + return 0; > > + > > + target = cpumask_any_but(cpu_online_mask, cpu); > > + if (target >= nr_cpu_ids) > > + return 0; > > + > > + perf_pmu_migrate_context(&smmu_pmu->pmu, cpu, target); > > + smmu_pmu->on_cpu = target; > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, cpumask_of(target))); > > + > > + return 0; > > +} > > + > > +static irqreturn_t smmu_pmu_handle_irq(int irq_num, void *data) > > +{ > > + struct smmu_pmu *smmu_pmu = data; > > + u64 ovsr; > > + unsigned int idx; > > + > > + ovsr = readq(smmu_pmu->reloc_base + SMMU_PMCG_OVSSET0); > > + if (!ovsr) > > + return IRQ_NONE; > > + > > + writeq(ovsr, smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > + > > + for_each_set_bit(idx, (unsigned long *)&ovsr, smmu_pmu->num_counters) > { > > + struct perf_event *event = smmu_pmu->events[idx]; > > + struct hw_perf_event *hwc; > > + > > + if (WARN_ON_ONCE(!event)) > > + continue; > > + > > + smmu_pmu_event_update(event); > > + hwc = &event->hw; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + } > > + > > + return IRQ_HANDLED; > > +} > > + > > +static int smmu_pmu_setup_irq(struct smmu_pmu *pmu) > > +{ > > + unsigned long flags = IRQF_NOBALANCING | IRQF_SHARED | > IRQF_NO_THREAD; > > + int irq, ret = -ENXIO; > > + > > + irq = pmu->irq; > > + if (irq) > > + ret = devm_request_irq(pmu->dev, irq, smmu_pmu_handle_irq, > > + flags, "smmuv3-pmu", pmu); > > + return ret; > > +} > > + > > +static void smmu_pmu_reset(struct smmu_pmu *smmu_pmu) > > +{ > > + smmu_pmu_disable(&smmu_pmu->pmu); > > + > > + /* Disable counter and interrupt */ > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > +} > > + > > +static int smmu_pmu_probe(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + struct resource *res_0, *res_1; > > + u32 cfgr, reg_size; > > + u64 ceid_64[2]; > > + int irq, err; > > + char *name; > > + struct device *dev = &pdev->dev; > > + > > + smmu_pmu = devm_kzalloc(dev, sizeof(*smmu_pmu), GFP_KERNEL); > > + if (!smmu_pmu) > > + return -ENOMEM; > > + > > + smmu_pmu->dev = dev; > > + platform_set_drvdata(pdev, smmu_pmu); > > + > > + smmu_pmu->pmu = (struct pmu) { > > + .task_ctx_nr = perf_invalid_context, > > + .pmu_enable = smmu_pmu_enable, > > + .pmu_disable = smmu_pmu_disable, > > + .event_init = smmu_pmu_event_init, > > + .add = smmu_pmu_event_add, > > + .del = smmu_pmu_event_del, > > + .start = smmu_pmu_event_start, > > + .stop = smmu_pmu_event_stop, > > + .read = smmu_pmu_event_read, > > + .attr_groups = smmu_pmu_attr_grps, > > + }; > > + > > + res_0 = platform_get_resource(pdev, IORESOURCE_MEM, 0); > > + smmu_pmu->reg_base = devm_ioremap_resource(dev, res_0); > > + if (IS_ERR(smmu_pmu->reg_base)) > > + return PTR_ERR(smmu_pmu->reg_base); > > + > > + cfgr = readl_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CFGR); > > + > > + /* Determine if page 1 is present */ > > + if (cfgr & SMMU_PMCG_CFGR_RELOC_CTRS) { > > + res_1 = platform_get_resource(pdev, IORESOURCE_MEM, 1); > > + smmu_pmu->reloc_base = devm_ioremap_resource(dev, res_1); > > + if (IS_ERR(smmu_pmu->reloc_base)) > > + return PTR_ERR(smmu_pmu->reloc_base); > > + } else { > > + smmu_pmu->reloc_base = smmu_pmu->reg_base; > > + } > > + > > + irq = platform_get_irq(pdev, 0); > > + if (irq > 0) > > + smmu_pmu->irq = irq; > > + > > + ceid_64[0] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID0); > > + ceid_64[1] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID1); > > + bitmap_from_arr32(smmu_pmu->supported_events, (u32 *)ceid_64, > > + SMMU_ARCH_MAX_EVENTS); > > + > > + smmu_pmu->num_counters = > FIELD_GET(SMMU_PMCG_CFGR_NCTR_MASK, cfgr) + 1; > > + smmu_pmu->counter_present_mask = > GENMASK(smmu_pmu->num_counters - 1, 0); > > + > > + smmu_pmu->sid_filter_global = !!(cfgr & > SMMU_PMCG_CFGR_SID_FILTER_TYPE); > > + > > + reg_size = FIELD_GET(SMMU_PMCG_CFGR_SIZE_MASK, cfgr); > > + smmu_pmu->counter_mask = GENMASK_ULL(reg_size, 0); > > + > > + smmu_pmu_reset(smmu_pmu); > > + > > + err = smmu_pmu_setup_irq(smmu_pmu); > > + if (err) { > > + dev_err(dev, "Setup irq failed, PMU @%pa\n", &res_0->start); > > + return err; > > + } > > + > > + name = devm_kasprintf(&pdev->dev, GFP_KERNEL, > "smmuv3_pmcg_%llx", > > + (res_0->start) >> SMMU_PA_SHIFT); > > + if (!name) { > > + dev_err(dev, "Create name failed, PMU @%pa\n", &res_0->start); > > + return -EINVAL; > > + } > > + > > + /* Pick one CPU to be the preferred one to use */ > > + smmu_pmu->on_cpu = get_cpu(); > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, > cpumask_of(smmu_pmu->on_cpu))); > > + > > + err = cpuhp_state_add_instance_nocalls(cpuhp_state_num, > > + &smmu_pmu->node); > > + if (err) { > > + dev_err(dev, "Error %d registering hotplug, PMU @%pa\n", > > + err, &res_0->start); > > + goto out_cpuhp_err; > > + } > > + > > + err = perf_pmu_register(&smmu_pmu->pmu, name, -1); > > + if (err) { > > + dev_err(dev, "Error %d registering PMU @%pa\n", > > + err, &res_0->start); > > + goto out_unregister; > > + } > > + > > + put_cpu(); > > + dev_info(dev, "Registered PMU @ %pa using %d counters with %s filter > settings\n", > > + &res_0->start, smmu_pmu->num_counters, > > + smmu_pmu->sid_filter_global ? "Global(Counter0)" : > > + "Individual"); > > + > > + return 0; > > + > > +out_unregister: > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > +out_cpuhp_err: > > + put_cpu(); > > + return err; > > +} > > + > > +static int smmu_pmu_remove(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + perf_pmu_unregister(&smmu_pmu->pmu); > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_shutdown(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + smmu_pmu_disable(&smmu_pmu->pmu); > > +} > > + > > +static struct platform_driver smmu_pmu_driver = { > > + .driver = { > > + .name = "arm-smmu-v3-pmu", > > + }, > > + .probe = smmu_pmu_probe, > > + .remove = smmu_pmu_remove, > > + .shutdown = smmu_pmu_shutdown, > > +}; > > + > > +static int __init arm_smmu_pmu_init(void) > > +{ > > + cpuhp_state_num = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, > > + "perf/arm/pmcg:online", > > + NULL, > > + smmu_pmu_offline_cpu); > > + if (cpuhp_state_num < 0) > > + return cpuhp_state_num; > > + > > + return platform_driver_register(&smmu_pmu_driver); > > +} > > +module_init(arm_smmu_pmu_init); > > + > > +static void __exit arm_smmu_pmu_exit(void) > > +{ > > + platform_driver_unregister(&smmu_pmu_driver); > > + cpuhp_remove_multi_state(cpuhp_state_num); > > +} > > + > > +module_exit(arm_smmu_pmu_exit); > > + > > +MODULE_DESCRIPTION("PMU driver for ARM SMMUv3 Performance > Monitors Extension"); > > +MODULE_AUTHOR("Neil Leeder "); > > +MODULE_AUTHOR("Shameer Kolothum > "); > > +MODULE_LICENSE("GPL v2"); > > -- > > 2.7.4 > > > > > > > > _______________________________________________ > > linux-arm-kernel mailing list > > linux-arm-kernel@lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-8.1 required=3.0 tests=DKIMWL_WL_HIGH,DKIM_SIGNED, DKIM_VALID,HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_PATCH,MAILING_LIST_MULTI, SIGNED_OFF_BY,SPF_PASS,URIBL_BLOCKED autolearn=unavailable autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id DAE9EC282C0 for ; Wed, 23 Jan 2019 11:03:21 +0000 (UTC) Received: from bombadil.infradead.org (bombadil.infradead.org [198.137.202.133]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 806AE21019 for ; Wed, 23 Jan 2019 11:03:21 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=lists.infradead.org header.i=@lists.infradead.org header.b="cM1oknGB" DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 806AE21019 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=huawei.com Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=linux-arm-kernel-bounces+infradead-linux-arm-kernel=archiver.kernel.org@lists.infradead.org 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:Cc:List-Subscribe:List-Help:List-Post: List-Archive:List-Unsubscribe:List-Id:MIME-Version:In-Reply-To:References: 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=h2lPDqR0aPCYfQhurbiB628UJFEEy7DfuY5y2vLplyA=; b=cM1oknGBKOPqgX RZ6xe9pm5M46W4yGVif1Rwds9kvDXQCJDT5N1lOi60no/Z7kPgTB7TWWbkrFGoT9tVl08zvNaKfZd CfZhr4bPJrR5nLlJ/78yDdf4B1359c4WQepStrvs/jXd0TDlOCAeV14VcTEKgmOm5jks0CDco0xSL Cy9QJh3ZKRee64hPuAQnTKigJojMTbVIWuXjJJZm7vbW0etinyNSnCPdkXbs1zrY81+UQSNx3FCzF A0plvePrmRmLzc4adawFbL1bOfiEVux7s4SyIoSgxN74Ux7UONaK6ZUkMESOST4H+KkdLd021/nkY W/+CTUACVX9KWQElOgNQ==; Received: from localhost ([127.0.0.1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.90_1 #2 (Red Hat Linux)) id 1gmGJL-0005gH-Ah; Wed, 23 Jan 2019 11:03:11 +0000 Received: from lhrrgout.huawei.com ([185.176.76.210] helo=huawei.com) by bombadil.infradead.org with esmtps (Exim 4.90_1 #2 (Red Hat Linux)) id 1gmGJD-0005f2-Rf for linux-arm-kernel@lists.infradead.org; Wed, 23 Jan 2019 11:03:08 +0000 Received: from LHREML710-CAH.china.huawei.com (unknown [172.18.7.107]) by Forcepoint Email with ESMTP id 110CA88E94A8DD5A75A6; Wed, 23 Jan 2019 11:03:00 +0000 (GMT) Received: from LHREML524-MBS.china.huawei.com ([169.254.2.138]) by LHREML710-CAH.china.huawei.com ([10.201.108.33]) with mapi id 14.03.0415.000; Wed, 23 Jan 2019 11:02:49 +0000 From: Shameerali Kolothum Thodi To: Andrew Murray Subject: RE: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver Thread-Topic: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver Thread-Index: AQHUiMReCYaliznTZUOBIOWTrrVwNqW7zC+AgAEf2sA= Date: Wed, 23 Jan 2019 11:02:48 +0000 Message-ID: <5FC3163CFD30C246ABAA99954A238FA83929855B@lhreml524-mbs.china.huawei.com> References: <20181130154751.28580-1-shameerali.kolothum.thodi@huawei.com> <20181130154751.28580-3-shameerali.kolothum.thodi@huawei.com> <20190122162351.GE8120@e119886-lin.cambridge.arm.com> In-Reply-To: <20190122162351.GE8120@e119886-lin.cambridge.arm.com> Accept-Language: en-GB, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.202.227.237] MIME-Version: 1.0 X-CFilter-Loop: Reflected X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20190123_030304_411436_FE551126 X-CRM114-Status: GOOD ( 32.99 ) 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" , "vkilari@codeaurora.org" , "lorenzo.pieralisi@arm.com" , "neil.m.leeder@gmail.com" , "jean-philippe.brucker@arm.com" , "pabba@codeaurora.org" , John Garry , "will.deacon@arm.com" , "rruigrok@codeaurora.org" , Linuxarm , "linux-acpi@vger.kernel.org" , "linux-arm-kernel@lists.infradead.org" , "Guohanjun \(Hanjun Guo\)" , "robin.murphy@arm.com" , "linux-kernel@vger.kernel.org" Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+infradead-linux-arm-kernel=archiver.kernel.org@lists.infradead.org Hi Andrew, Thanks for taking a look at this. > -----Original Message----- > From: Andrew Murray [mailto:andrew.murray@arm.com] > Sent: 22 January 2019 16:24 > To: Shameerali Kolothum Thodi > Cc: lorenzo.pieralisi@arm.com; robin.murphy@arm.com; > mark.rutland@arm.com; vkilari@codeaurora.org; neil.m.leeder@gmail.com; > jean-philippe.brucker@arm.com; pabba@codeaurora.org; John Garry > ; will.deacon@arm.com; rruigrok@codeaurora.org; > Linuxarm ; linux-kernel@vger.kernel.org; > linux-acpi@vger.kernel.org; Guohanjun (Hanjun Guo) > ; linux-arm-kernel@lists.infradead.org > Subject: Re: [PATCH v5 2/4] perf: add arm64 smmuv3 pmu driver > > On Fri, Nov 30, 2018 at 03:47:49PM +0000, Shameer Kolothum wrote: > > From: Neil Leeder > > > > Adds a new driver to support the SMMUv3 PMU and add it into the > > perf events framework. > > > > Each SMMU node may have multiple PMUs associated with it, each of > > which may support different events. > > > > SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > is the physical page address of the SMMU PMCG > > wrapped to 4K boundary. For example, the PMCG at 0xff88840000 is > > named smmuv3_pmcg_ff88840 > > > > Filtering by stream id is done by specifying filtering parameters > > with the event. options are: > > filter_enable - 0 = no filtering, 1 = filtering enabled > > filter_span - 0 = exact match, 1 = pattern match > > filter_stream_id - pattern to filter against > > > > Example: perf stat -e smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > filter_span=1,filter_stream_id=0x42/ -a netperf > > > > Applies filter pattern 0x42 to transaction events, which means events > > matching stream ids 0x42 & 0x43 are counted as only upper StreamID > > bits are required to match the given filter. Further filtering > > information is available in the SMMU documentation. > > > > SMMU events are not attributable to a CPU, so task mode and sampling > > are not supported. > > > > Signed-off-by: Neil Leeder > > Signed-off-by: Shameer Kolothum > > --- > > drivers/perf/Kconfig | 9 + > > drivers/perf/Makefile | 1 + > > drivers/perf/arm_smmuv3_pmu.c | 778 > ++++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 788 insertions(+) > > create mode 100644 drivers/perf/arm_smmuv3_pmu.c > > > > diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig > > index 08ebaf7..92be6a3 100644 > > --- a/drivers/perf/Kconfig > > +++ b/drivers/perf/Kconfig > > @@ -52,6 +52,15 @@ config ARM_PMU_ACPI > > depends on ARM_PMU && ACPI > > def_bool y > > > > +config ARM_SMMU_V3_PMU > > + bool "ARM SMMUv3 Performance Monitors Extension" > > + depends on ARM64 && ACPI && ARM_SMMU_V3 > > + help > > + Provides support for the SMMU version 3 performance monitor unit > (PMU) > > + on ARM-based systems. > > + Adds the SMMU PMU into the perf events subsystem for > > + monitoring SMMU performance events. > > + > > config ARM_DSU_PMU > > tristate "ARM DynamIQ Shared Unit (DSU) PMU" > > depends on ARM64 > > diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile > > index b3902bd..f10a932 100644 > > --- a/drivers/perf/Makefile > > +++ b/drivers/perf/Makefile > > @@ -4,6 +4,7 @@ obj-$(CONFIG_ARM_CCN) += arm-ccn.o > > obj-$(CONFIG_ARM_DSU_PMU) += arm_dsu_pmu.o > > obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o > > obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o > > +obj-$(CONFIG_ARM_SMMU_V3_PMU) += arm_smmuv3_pmu.o > > obj-$(CONFIG_HISI_PMU) += hisilicon/ > > obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o > > obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o > > diff --git a/drivers/perf/arm_smmuv3_pmu.c > b/drivers/perf/arm_smmuv3_pmu.c > > new file mode 100644 > > index 0000000..fb9dcd8 > > --- /dev/null > > +++ b/drivers/perf/arm_smmuv3_pmu.c > > @@ -0,0 +1,778 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > + > > +/* > > + * This driver adds support for perf events to use the Performance > > + * Monitor Counter Groups (PMCG) associated with an SMMUv3 node > > + * to monitor that node. > > + * > > + * SMMUv3 PMCG devices are named as smmuv3_pmcg_ > where > > + * is the physical page address of the SMMU PMCG > wrapped > > + * to 4K boundary. For example, the PMCG at 0xff88840000 is named > > + * smmuv3_pmcg_ff88840 > > + * > > + * Filtering by stream id is done by specifying filtering parameters > > + * with the event. options are: > > + * filter_enable - 0 = no filtering, 1 = filtering enabled > > + * filter_span - 0 = exact match, 1 = pattern match > > + * filter_stream_id - pattern to filter against > > + * > > + * To match a partial StreamID where the X most-significant bits must match > > + * but the Y least-significant bits might differ, STREAMID is programmed > > + * with a value that contains: > > + * STREAMID[Y - 1] == 0. > > + * STREAMID[Y - 2:0] == 1 (where Y > 1). > > + * The remainder of implemented bits of STREAMID (X bits, from bit Y > upwards) > > + * contain a value to match from the corresponding bits of event StreamID. > > + * > > + * Example: perf stat -e > smmuv3_pmcg_ff88840/transaction,filter_enable=1, > > + * filter_span=1,filter_stream_id=0x42/ -a netperf > > + * Applies filter pattern 0x42 to transaction events, which means events > > + * matching stream ids 0x42 and 0x43 are counted. Further filtering > > + * information is available in the SMMU documentation. > > + * > > + * SMMU events are not attributable to a CPU, so task mode and sampling > > + * are not supported. > > + */ > > + > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > +#include > > + > > +#define SMMU_PMCG_EVCNTR0 0x0 > > +#define SMMU_PMCG_EVCNTR(n, stride) (SMMU_PMCG_EVCNTR0 + > (n) * (stride)) > > +#define SMMU_PMCG_EVTYPER0 0x400 > > +#define SMMU_PMCG_EVTYPER(n) > (SMMU_PMCG_EVTYPER0 + (n) * 4) > > +#define SMMU_PMCG_SID_SPAN_SHIFT 29 > > +#define SMMU_PMCG_SID_SPAN_MASK GENMASK(29, 29) > > +#define SMMU_PMCG_SMR0 0xA00 > > +#define SMMU_PMCG_SMR(n) (SMMU_PMCG_SMR0 + > (n) * 4) > > +#define SMMU_PMCG_CNTENSET0 0xC00 > > +#define SMMU_PMCG_CNTENCLR0 0xC20 > > +#define SMMU_PMCG_INTENSET0 0xC40 > > +#define SMMU_PMCG_INTENCLR0 0xC60 > > +#define SMMU_PMCG_OVSCLR0 0xC80 > > +#define SMMU_PMCG_OVSSET0 0xCC0 > > +#define SMMU_PMCG_CFGR 0xE00 > > +#define SMMU_PMCG_CFGR_RELOC_CTRS BIT(20) > > +#define SMMU_PMCG_CFGR_SID_FILTER_TYPE BIT(23) > > +#define SMMU_PMCG_CFGR_SIZE_MASK GENMASK(13, 8) > > +#define SMMU_PMCG_CFGR_NCTR_MASK GENMASK(5, 0) > > +#define SMMU_PMCG_CR 0xE04 > > +#define SMMU_PMCG_CR_ENABLE BIT(0) > > +#define SMMU_PMCG_CEID0 0xE20 > > +#define SMMU_PMCG_CEID1 0xE28 > > +#define SMMU_PMCG_IRQ_CTRL 0xE50 > > +#define SMMU_PMCG_IRQ_CTRL_IRQEN BIT(0) > > +#define SMMU_PMCG_IRQ_CFG0 0xE58 > > + > > +#define SMMU_DEFAULT_FILTER_SPAN 1 > > +#define SMMU_DEFAULT_FILTER_STREAM_ID GENMASK(31, 0) > > + > > +#define SMMU_MAX_COUNTERS 64 > > +#define SMMU_ARCH_MAX_EVENT_ID 0x7F > > +#define SMMU_IMPDEF_MAX_EVENT_ID 0xFFFF > > +#define SMMU_ARCH_MAX_EVENTS > (SMMU_ARCH_MAX_EVENT_ID + 1) > > + > > +#define SMMU_PA_SHIFT 12 > > + > > +static int cpuhp_state_num; > > + > > +struct smmu_pmu { > > + bool sid_filter_global; > > + struct hlist_node node; > > + struct perf_event *events[SMMU_MAX_COUNTERS]; > > + DECLARE_BITMAP(used_counters, SMMU_MAX_COUNTERS); > > + DECLARE_BITMAP(supported_events, SMMU_ARCH_MAX_EVENTS); > > + unsigned int irq; > > + unsigned int on_cpu; > > + struct pmu pmu; > > + unsigned int num_counters; > > + struct device *dev; > > + void __iomem *reg_base; > > + void __iomem *reloc_base; > > + u64 counter_present_mask; > > + u64 counter_mask; > > +}; > > + > > +#define to_smmu_pmu(p) (container_of(p, struct smmu_pmu, pmu)) > > + > > +#define SMMU_PMU_EVENT_ATTR_EXTRACTOR(_name, _config, _start, > _end) \ > > + static inline u32 get_##_name(struct perf_event *event) \ > > + > { > \ > > + return FIELD_GET(GENMASK_ULL(_end, _start), \ > > + event->attr._config); \ > > + } > \ > > + > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(event, config, 0, 15); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_stream_id, config1, 0, 31); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_span, config1, 32, 32); > > +SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_enable, config1, 33, 33); > > + > > +static inline void smmu_pmu_enable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(SMMU_PMCG_CR_ENABLE, smmu_pmu->reg_base + > SMMU_PMCG_CR); > > + writel(SMMU_PMCG_IRQ_CTRL_IRQEN, > > + smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > Should the interrupts be enabled before enabling the PMU? Yes, it make sense to do so. > > +} > > + > > +static inline void smmu_pmu_disable(struct pmu *pmu) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu); > > + > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_CR); > > + writel(0, smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL); > > +} > > + > > +static inline void smmu_pmu_counter_set_value(struct smmu_pmu > *smmu_pmu, > > + u32 idx, u64 value) > > +{ > > + if (smmu_pmu->counter_mask & BIT(32)) > > + writeq(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + writel(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > The arm64 IO macros use __force u32 and so it's probably OK to provide a 64 > bit > value to writel. But you could use something like lower_32_bits for clarity. Yes, macro uses __force u32. I will change it to make it more clear though. > > +} > > + > > +static inline u64 smmu_pmu_counter_get_value(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + u64 value; > > + > > + if (smmu_pmu->counter_mask & BIT(32)) > > + value = readq(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 8)); > > + else > > + value = readl(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, > 4)); > > + > > + return value; > > +} > > + > > +static inline void smmu_pmu_counter_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENSET0); > > +} > > + > > +static inline void smmu_pmu_counter_disable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > +} > > + > > +static inline void smmu_pmu_interrupt_enable(struct smmu_pmu > *smmu_pmu, u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENSET0); > > +} > > + > > +static inline void smmu_pmu_interrupt_disable(struct smmu_pmu > *smmu_pmu, > > + u32 idx) > > +{ > > + writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > +} > > + > > +static inline void smmu_pmu_set_evtyper(struct smmu_pmu *smmu_pmu, > u32 idx, > > + u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx)); > > +} > > + > > +static inline void smmu_pmu_set_smr(struct smmu_pmu *smmu_pmu, u32 > idx, u32 val) > > +{ > > + writel(val, smmu_pmu->reg_base + SMMU_PMCG_SMR(idx)); > > +} > > + > > +static void smmu_pmu_event_update(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + u64 delta, prev, now; > > + u32 idx = hwc->idx; > > + > > + do { > > + prev = local64_read(&hwc->prev_count); > > + now = smmu_pmu_counter_get_value(smmu_pmu, idx); > > + } while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev); > > + > > + /* handle overflow. */ > > + delta = now - prev; > > + delta &= smmu_pmu->counter_mask; > > + > > + local64_add(delta, &event->count); > > +} > > + > > +static void smmu_pmu_set_period(struct smmu_pmu *smmu_pmu, > > + struct hw_perf_event *hwc) > > +{ > > + u32 idx = hwc->idx; > > + u64 new; > > + > > + /* > > + * We limit the max period to half the max counter value of the counter > > + * size, so that even in the case of extreme interrupt latency the > > + * counter will (hopefully) not wrap past its initial value. > > + */ > > + new = smmu_pmu->counter_mask >> 1; > > + > > + local64_set(&hwc->prev_count, new); > > + smmu_pmu_counter_set_value(smmu_pmu, idx, new); > > +} > > + > > +static void smmu_pmu_get_event_filter(struct perf_event *event, u32 > *span, > > + u32 *stream_id) > > +{ > > + bool filter_en = !!get_filter_enable(event); > > + > > + *span = filter_en ? get_filter_span(event) : > SMMU_DEFAULT_FILTER_SPAN; > > + *stream_id = filter_en ? get_filter_stream_id(event) : > > + SMMU_DEFAULT_FILTER_STREAM_ID; > > +} > > + > > +static bool smmu_pmu_event_filter_valid(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event, int idx) > > +{ > > + u32 filter_span, filter_sid; > > + u32 curr_span, curr_sid; > > + > > + if (!idx || !smmu_pmu->sid_filter_global) > > + return true; > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + /* Read the current global filter setting */ > > + curr_span = FIELD_GET(SMMU_PMCG_SID_SPAN_MASK, > > + readl(smmu_pmu->reg_base + > SMMU_PMCG_EVTYPER0)); > > + curr_sid = readl(smmu_pmu->reg_base + SMMU_PMCG_SMR0); > > + > > + if (filter_span != curr_span || filter_sid != curr_sid) > > + return false; > > + > > + return true; > > +} > > + > > +static int smmu_pmu_get_event_idx(struct smmu_pmu *smmu_pmu, > > + struct perf_event *event) > > +{ > > + unsigned int idx; > > + unsigned int num_ctrs = smmu_pmu->num_counters; > > + > > + idx = find_first_zero_bit(smmu_pmu->used_counters, num_ctrs); > > + if (idx == num_ctrs) > > + /* The counters are all in use. */ > > + return -EAGAIN; > > + > > + if (!smmu_pmu_event_filter_valid(smmu_pmu, event, idx)) > > + return -EAGAIN; > > + > > + set_bit(idx, smmu_pmu->used_counters); > > + > > + return idx; > > +} > > + > > +/* > > + * Implementation of abstract pmu functionality required by > > + * the core perf events code. > > + */ > > + > > +static int smmu_pmu_event_init(struct perf_event *event) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct device *dev = smmu_pmu->dev; > > + struct perf_event *sibling; > > + u32 event_id; > > + > > + if (event->attr.type != event->pmu->type) > > + return -ENOENT; > > + > > + if (hwc->sample_period) { > > + dev_dbg(dev, "Sampling not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + if (event->cpu < 0) { > > + dev_dbg(dev, "Per-task mode not supported\n"); > > + return -EOPNOTSUPP; > > + } > > + > > + /* We cannot filter accurately so we just don't allow it. */ > > + if (event->attr.exclude_user || event->attr.exclude_kernel || > > + event->attr.exclude_hv || event->attr.exclude_idle) { > > + dev_dbg(dev, "Can't exclude execution levels\n"); > > + return -EOPNOTSUPP; > > + } > > You should also test for attr.exclude_host and attr.exclude_guest, unless > it makes sense to stop counting these events when entering guest/host? TBH, I haven't considered that scenario yet and don't think we have to stop the counters. I will check that. > > Also returning -EINVAL may be more desirable than -EOPNOTSUPP. If a user > makes use of the perf utility and attempts to filter one of the above > exclusion flags - then perf will retry opening the event but without some > of the unsupported flags, but it will only do this if you return -EINVAL. > > (See line 1927 of tools/perf/util/evsel.c and run the perf stat tool with > the verbose flag to see this in action.) Ok. I will take a look. > > + > > + /* Verify specified event is supported on this PMU */ > > + event_id = get_event(event); > > + if (((event_id <= SMMU_ARCH_MAX_EVENT_ID) && > > + (!test_bit(event_id, smmu_pmu->supported_events))) || > > + (event_id > SMMU_IMPDEF_MAX_EVENT_ID)) { > > + dev_dbg(dev, "Invalid event %d for this PMU\n", event_id); > > + return -EINVAL; > > + } > > + > > + /* Don't allow groups with mixed PMUs, except for s/w events */ > > + if (event->group_leader->pmu != event->pmu && > > + !is_software_event(event->group_leader)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + list_for_each_entry(sibling, &event->group_leader->sibling_list, > > + sibling_list) > > + if (sibling->pmu != event->pmu && > > + !is_software_event(sibling)) { > > + dev_dbg(dev, "Can't create mixed PMU group\n"); > > + return -EINVAL; > > + } > > + > > + /* Ensure all events in a group are on the same cpu */ > > + if ((event->group_leader != event) && > > + (event->cpu != event->group_leader->cpu)) { > > + dev_dbg(dev, "Can't create group on CPUs %d and %d\n", > > + event->cpu, event->group_leader->cpu); > > + return -EINVAL; > > + } > > + > > + hwc->idx = -1; > > + > > + /* > > + * Ensure all events are on the same cpu so all events are in the > > + * same cpu context, to avoid races on pmu_enable etc. > > + */ > > + event->cpu = smmu_pmu->on_cpu; > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_start(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + u32 filter_span, filter_sid; > > + u32 evtyper; > > + > > + hwc->state = 0; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + > > + smmu_pmu_get_event_filter(event, &filter_span, &filter_sid); > > + > > + evtyper = get_event(event) | > > + filter_span << SMMU_PMCG_SID_SPAN_SHIFT; > > + > > + smmu_pmu_set_evtyper(smmu_pmu, idx, evtyper); > > + smmu_pmu_set_smr(smmu_pmu, idx, filter_sid); > > + smmu_pmu_interrupt_enable(smmu_pmu, idx); > > + smmu_pmu_counter_enable(smmu_pmu, idx); > > +} > > + > > +static void smmu_pmu_event_stop(struct perf_event *event, int flags) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + struct hw_perf_event *hwc = &event->hw; > > + int idx = hwc->idx; > > + > > + if (hwc->state & PERF_HES_STOPPED) > > + return; > > + > > + smmu_pmu_counter_disable(smmu_pmu, idx); > > Is it intentional not to call smmu_pmu_interrupt_disable here? Yes, it is. Earlier patch had the interrupt toggling and Robin pointed out that it is not really needed as counters are anyway stopped and explicitly disabling may not solve the spurious interrupt case as well. > > + > > + if (flags & PERF_EF_UPDATE) > > + smmu_pmu_event_update(event); > > + hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE; > > Shouldn't we only set PERF_HES_UPTODATE if we've updated the count (e.g. > when PERF_EF_UPDATE was set?) Hmm..I had a look at couple of other drivers(arm-cci.c/arm_dsu_pmu.c etc) and those ones always call update ignoring PERF_EF_UPDATE and sets PERF_HES_UPTODATE. I think the reason being the counter is always reprogrammed on start. Do you think changing to that logic is a better way? > This may be a nit, but shouldn't smmu_pmu_event_update come before the > test for PERF_HES_STOPPED? I guess in the odd case where stop is called > without the PERF_EF_UPDATE flag (to stop the counter) and then called > a second time with PERF_EF_UPDATE to update the count. Not sure such a scenario exists. I think most of the ones I checked has the logic -its already stopped, nothing to do. > > +} > > + > > +static int smmu_pmu_event_add(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + int idx; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + > > + idx = smmu_pmu_get_event_idx(smmu_pmu, event); > > + if (idx < 0) > > + return idx; > > + > > + hwc->idx = idx; > > + hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE; > > + smmu_pmu->events[idx] = event; > > + local64_set(&hwc->prev_count, 0); > > + > > + if (flags & PERF_EF_START) > > + smmu_pmu_event_start(event, flags); > > + > > + /* Propagate changes to the userspace mapping. */ > > + perf_event_update_userpage(event); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_event_del(struct perf_event *event, int flags) > > +{ > > + struct hw_perf_event *hwc = &event->hw; > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu); > > + int idx = hwc->idx; > > + > > + smmu_pmu_event_stop(event, flags | PERF_EF_UPDATE); > > + smmu_pmu->events[idx] = NULL; > > + clear_bit(idx, smmu_pmu->used_counters); > > + > > + perf_event_update_userpage(event); > > +} > > + > > +static void smmu_pmu_event_read(struct perf_event *event) > > +{ > > + smmu_pmu_event_update(event); > > +} > > + > > +/* cpumask */ > > + > > +static ssize_t smmu_pmu_cpumask_show(struct device *dev, > > + struct device_attribute *attr, > > + char *buf) > > +{ > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + > > + return cpumap_print_to_pagebuf(true, buf, > cpumask_of(smmu_pmu->on_cpu)); > > +} > > + > > +static struct device_attribute smmu_pmu_cpumask_attr = > > + __ATTR(cpumask, 0444, smmu_pmu_cpumask_show, NULL); > > + > > +static struct attribute *smmu_pmu_cpumask_attrs[] = { > > + &smmu_pmu_cpumask_attr.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_cpumask_group = { > > + .attrs = smmu_pmu_cpumask_attrs, > > +}; > > + > > +/* Events */ > > + > > +ssize_t smmu_pmu_event_show(struct device *dev, > > + struct device_attribute *attr, char *page) > > +{ > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr); > > + > > + return sprintf(page, "event=0x%02llx\n", pmu_attr->id); > > +} > > + > > +#define SMMU_EVENT_ATTR(_name, _id) \ > > + (&((struct perf_pmu_events_attr[]) { \ > > + { .attr = __ATTR(_name, 0444, smmu_pmu_event_show, NULL), \ > > + .id = _id, } \ > > + })[0].attr.attr) > > This is very similar to PMU_EVENT_ATTR, have you considered taking an > approach > similar to the ARMV8_EVENT_ATTR macro that wraps up PMU_EVENT_ATTR. > It may > be more readable if we define these attributes consistently across PMUs. Ok. I will take a look at that. Thanks, Shameer > Thanks, > > Andrew Murray > > > + > > +static struct attribute *smmu_pmu_events[] = { > > + SMMU_EVENT_ATTR(cycles, 0), > > + SMMU_EVENT_ATTR(transaction, 1), > > + SMMU_EVENT_ATTR(tlb_miss, 2), > > + SMMU_EVENT_ATTR(config_cache_miss, 3), > > + SMMU_EVENT_ATTR(trans_table_walk_access, 4), > > + SMMU_EVENT_ATTR(config_struct_access, 5), > > + SMMU_EVENT_ATTR(pcie_ats_trans_rq, 6), > > + SMMU_EVENT_ATTR(pcie_ats_trans_passed, 7), > > + NULL > > +}; > > + > > +static umode_t smmu_pmu_event_is_visible(struct kobject *kobj, > > + struct attribute *attr, int unused) > > +{ > > + struct device *dev = kobj_to_dev(kobj); > > + struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev)); > > + struct perf_pmu_events_attr *pmu_attr; > > + > > + pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr); > > + > > + if (test_bit(pmu_attr->id, smmu_pmu->supported_events)) > > + return attr->mode; > > + > > + return 0; > > +} > > +static struct attribute_group smmu_pmu_events_group = { > > + .name = "events", > > + .attrs = smmu_pmu_events, > > + .is_visible = smmu_pmu_event_is_visible, > > +}; > > + > > +/* Formats */ > > +PMU_FORMAT_ATTR(event, "config:0-15"); > > +PMU_FORMAT_ATTR(filter_stream_id, "config1:0-31"); > > +PMU_FORMAT_ATTR(filter_span, "config1:32"); > > +PMU_FORMAT_ATTR(filter_enable, "config1:33"); > > + > > +static struct attribute *smmu_pmu_formats[] = { > > + &format_attr_event.attr, > > + &format_attr_filter_stream_id.attr, > > + &format_attr_filter_span.attr, > > + &format_attr_filter_enable.attr, > > + NULL > > +}; > > + > > +static struct attribute_group smmu_pmu_format_group = { > > + .name = "format", > > + .attrs = smmu_pmu_formats, > > +}; > > + > > +static const struct attribute_group *smmu_pmu_attr_grps[] = { > > + &smmu_pmu_cpumask_group, > > + &smmu_pmu_events_group, > > + &smmu_pmu_format_group, > > + NULL > > +}; > > + > > +/* > > + * Generic device handlers > > + */ > > + > > +static int smmu_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + unsigned int target; > > + > > + smmu_pmu = hlist_entry_safe(node, struct smmu_pmu, node); > > + if (cpu != smmu_pmu->on_cpu) > > + return 0; > > + > > + target = cpumask_any_but(cpu_online_mask, cpu); > > + if (target >= nr_cpu_ids) > > + return 0; > > + > > + perf_pmu_migrate_context(&smmu_pmu->pmu, cpu, target); > > + smmu_pmu->on_cpu = target; > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, cpumask_of(target))); > > + > > + return 0; > > +} > > + > > +static irqreturn_t smmu_pmu_handle_irq(int irq_num, void *data) > > +{ > > + struct smmu_pmu *smmu_pmu = data; > > + u64 ovsr; > > + unsigned int idx; > > + > > + ovsr = readq(smmu_pmu->reloc_base + SMMU_PMCG_OVSSET0); > > + if (!ovsr) > > + return IRQ_NONE; > > + > > + writeq(ovsr, smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > + > > + for_each_set_bit(idx, (unsigned long *)&ovsr, smmu_pmu->num_counters) > { > > + struct perf_event *event = smmu_pmu->events[idx]; > > + struct hw_perf_event *hwc; > > + > > + if (WARN_ON_ONCE(!event)) > > + continue; > > + > > + smmu_pmu_event_update(event); > > + hwc = &event->hw; > > + > > + smmu_pmu_set_period(smmu_pmu, hwc); > > + } > > + > > + return IRQ_HANDLED; > > +} > > + > > +static int smmu_pmu_setup_irq(struct smmu_pmu *pmu) > > +{ > > + unsigned long flags = IRQF_NOBALANCING | IRQF_SHARED | > IRQF_NO_THREAD; > > + int irq, ret = -ENXIO; > > + > > + irq = pmu->irq; > > + if (irq) > > + ret = devm_request_irq(pmu->dev, irq, smmu_pmu_handle_irq, > > + flags, "smmuv3-pmu", pmu); > > + return ret; > > +} > > + > > +static void smmu_pmu_reset(struct smmu_pmu *smmu_pmu) > > +{ > > + smmu_pmu_disable(&smmu_pmu->pmu); > > + > > + /* Disable counter and interrupt */ > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0); > > + writeq_relaxed(smmu_pmu->counter_present_mask, > > + smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0); > > +} > > + > > +static int smmu_pmu_probe(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu; > > + struct resource *res_0, *res_1; > > + u32 cfgr, reg_size; > > + u64 ceid_64[2]; > > + int irq, err; > > + char *name; > > + struct device *dev = &pdev->dev; > > + > > + smmu_pmu = devm_kzalloc(dev, sizeof(*smmu_pmu), GFP_KERNEL); > > + if (!smmu_pmu) > > + return -ENOMEM; > > + > > + smmu_pmu->dev = dev; > > + platform_set_drvdata(pdev, smmu_pmu); > > + > > + smmu_pmu->pmu = (struct pmu) { > > + .task_ctx_nr = perf_invalid_context, > > + .pmu_enable = smmu_pmu_enable, > > + .pmu_disable = smmu_pmu_disable, > > + .event_init = smmu_pmu_event_init, > > + .add = smmu_pmu_event_add, > > + .del = smmu_pmu_event_del, > > + .start = smmu_pmu_event_start, > > + .stop = smmu_pmu_event_stop, > > + .read = smmu_pmu_event_read, > > + .attr_groups = smmu_pmu_attr_grps, > > + }; > > + > > + res_0 = platform_get_resource(pdev, IORESOURCE_MEM, 0); > > + smmu_pmu->reg_base = devm_ioremap_resource(dev, res_0); > > + if (IS_ERR(smmu_pmu->reg_base)) > > + return PTR_ERR(smmu_pmu->reg_base); > > + > > + cfgr = readl_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CFGR); > > + > > + /* Determine if page 1 is present */ > > + if (cfgr & SMMU_PMCG_CFGR_RELOC_CTRS) { > > + res_1 = platform_get_resource(pdev, IORESOURCE_MEM, 1); > > + smmu_pmu->reloc_base = devm_ioremap_resource(dev, res_1); > > + if (IS_ERR(smmu_pmu->reloc_base)) > > + return PTR_ERR(smmu_pmu->reloc_base); > > + } else { > > + smmu_pmu->reloc_base = smmu_pmu->reg_base; > > + } > > + > > + irq = platform_get_irq(pdev, 0); > > + if (irq > 0) > > + smmu_pmu->irq = irq; > > + > > + ceid_64[0] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID0); > > + ceid_64[1] = readq_relaxed(smmu_pmu->reg_base + > SMMU_PMCG_CEID1); > > + bitmap_from_arr32(smmu_pmu->supported_events, (u32 *)ceid_64, > > + SMMU_ARCH_MAX_EVENTS); > > + > > + smmu_pmu->num_counters = > FIELD_GET(SMMU_PMCG_CFGR_NCTR_MASK, cfgr) + 1; > > + smmu_pmu->counter_present_mask = > GENMASK(smmu_pmu->num_counters - 1, 0); > > + > > + smmu_pmu->sid_filter_global = !!(cfgr & > SMMU_PMCG_CFGR_SID_FILTER_TYPE); > > + > > + reg_size = FIELD_GET(SMMU_PMCG_CFGR_SIZE_MASK, cfgr); > > + smmu_pmu->counter_mask = GENMASK_ULL(reg_size, 0); > > + > > + smmu_pmu_reset(smmu_pmu); > > + > > + err = smmu_pmu_setup_irq(smmu_pmu); > > + if (err) { > > + dev_err(dev, "Setup irq failed, PMU @%pa\n", &res_0->start); > > + return err; > > + } > > + > > + name = devm_kasprintf(&pdev->dev, GFP_KERNEL, > "smmuv3_pmcg_%llx", > > + (res_0->start) >> SMMU_PA_SHIFT); > > + if (!name) { > > + dev_err(dev, "Create name failed, PMU @%pa\n", &res_0->start); > > + return -EINVAL; > > + } > > + > > + /* Pick one CPU to be the preferred one to use */ > > + smmu_pmu->on_cpu = get_cpu(); > > + WARN_ON(irq_set_affinity(smmu_pmu->irq, > cpumask_of(smmu_pmu->on_cpu))); > > + > > + err = cpuhp_state_add_instance_nocalls(cpuhp_state_num, > > + &smmu_pmu->node); > > + if (err) { > > + dev_err(dev, "Error %d registering hotplug, PMU @%pa\n", > > + err, &res_0->start); > > + goto out_cpuhp_err; > > + } > > + > > + err = perf_pmu_register(&smmu_pmu->pmu, name, -1); > > + if (err) { > > + dev_err(dev, "Error %d registering PMU @%pa\n", > > + err, &res_0->start); > > + goto out_unregister; > > + } > > + > > + put_cpu(); > > + dev_info(dev, "Registered PMU @ %pa using %d counters with %s filter > settings\n", > > + &res_0->start, smmu_pmu->num_counters, > > + smmu_pmu->sid_filter_global ? "Global(Counter0)" : > > + "Individual"); > > + > > + return 0; > > + > > +out_unregister: > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > +out_cpuhp_err: > > + put_cpu(); > > + return err; > > +} > > + > > +static int smmu_pmu_remove(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + perf_pmu_unregister(&smmu_pmu->pmu); > > + cpuhp_state_remove_instance_nocalls(cpuhp_state_num, > &smmu_pmu->node); > > + > > + return 0; > > +} > > + > > +static void smmu_pmu_shutdown(struct platform_device *pdev) > > +{ > > + struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev); > > + > > + smmu_pmu_disable(&smmu_pmu->pmu); > > +} > > + > > +static struct platform_driver smmu_pmu_driver = { > > + .driver = { > > + .name = "arm-smmu-v3-pmu", > > + }, > > + .probe = smmu_pmu_probe, > > + .remove = smmu_pmu_remove, > > + .shutdown = smmu_pmu_shutdown, > > +}; > > + > > +static int __init arm_smmu_pmu_init(void) > > +{ > > + cpuhp_state_num = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, > > + "perf/arm/pmcg:online", > > + NULL, > > + smmu_pmu_offline_cpu); > > + if (cpuhp_state_num < 0) > > + return cpuhp_state_num; > > + > > + return platform_driver_register(&smmu_pmu_driver); > > +} > > +module_init(arm_smmu_pmu_init); > > + > > +static void __exit arm_smmu_pmu_exit(void) > > +{ > > + platform_driver_unregister(&smmu_pmu_driver); > > + cpuhp_remove_multi_state(cpuhp_state_num); > > +} > > + > > +module_exit(arm_smmu_pmu_exit); > > + > > +MODULE_DESCRIPTION("PMU driver for ARM SMMUv3 Performance > Monitors Extension"); > > +MODULE_AUTHOR("Neil Leeder "); > > +MODULE_AUTHOR("Shameer Kolothum > "); > > +MODULE_LICENSE("GPL v2"); > > -- > > 2.7.4 > > > > > > > > _______________________________________________ > > linux-arm-kernel mailing list > > linux-arm-kernel@lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel