From 7f35d62698160b3384276a04a8eb0e31fdab979f Mon Sep 17 00:00:00 2001 From: Kan Liang Date: Thu, 16 Nov 2023 06:22:42 -0800 Subject: [PATCH 1/4] perf/x86/intel/cstate: Cleanup duplicate attr_groups ANBZ: #8054 commit 243218ca93037631f0224fdbefea045912cb761a upstream. The events of the cstate_core and cstate_pkg PMU have the same format. They both need to create a "events" group (with empty attrs). The attr_groups can be shared. Remove the dedicated attr_groups for each cstate PMU. Use the shared cstate_attr_groups to replace. Intel-SIG: commit 243218ca9303 perf/x86/intel/cstate: Cleanup duplicate attr_groups Backport SRF/GRR perf cstate support to kernel 5.10. Signed-off-by: Kan Liang Signed-off-by: Peter Zijlstra (Intel) Link: https://lore.kernel.org/r/20231116142245.1233485-1-kan.liang@linux.intel.com [ Yunying Sun: amend commit log ] Signed-off-by: Yunying Sun --- arch/x86/events/intel/cstate.c | 44 +++++++++------------------------- 1 file changed, 11 insertions(+), 33 deletions(-) diff --git a/arch/x86/events/intel/cstate.c b/arch/x86/events/intel/cstate.c index 218d862d6d3f..b4dd66867978 100644 --- a/arch/x86/events/intel/cstate.c +++ b/arch/x86/events/intel/cstate.c @@ -188,20 +188,20 @@ static struct attribute *attrs_empty[] = { * "events" group (with empty attrs) before updating * it with detected events. */ -static struct attribute_group core_events_attr_group = { +static struct attribute_group cstate_events_attr_group = { .name = "events", .attrs = attrs_empty, }; -DEFINE_CSTATE_FORMAT_ATTR(core_event, event, "config:0-63"); -static struct attribute *core_format_attrs[] = { - &format_attr_core_event.attr, +DEFINE_CSTATE_FORMAT_ATTR(cstate_event, event, "config:0-63"); +static struct attribute *cstate_format_attrs[] = { + &format_attr_cstate_event.attr, NULL, }; -static struct attribute_group core_format_attr_group = { +static struct attribute_group cstate_format_attr_group = { .name = "format", - .attrs = core_format_attrs, + .attrs = cstate_format_attrs, }; static cpumask_t cstate_core_cpu_mask; @@ -216,9 +216,9 @@ static struct attribute_group cpumask_attr_group = { .attrs = cstate_cpumask_attrs, }; -static const struct attribute_group *core_attr_groups[] = { - &core_events_attr_group, - &core_format_attr_group, +static const struct attribute_group *cstate_attr_groups[] = { + &cstate_events_attr_group, + &cstate_format_attr_group, &cpumask_attr_group, NULL, }; @@ -267,30 +267,8 @@ static struct perf_msr pkg_msr[] = { [PERF_CSTATE_PKG_C10_RES] = { MSR_PKG_C10_RESIDENCY, &group_cstate_pkg_c10, test_msr }, }; -static struct attribute_group pkg_events_attr_group = { - .name = "events", - .attrs = attrs_empty, -}; - -DEFINE_CSTATE_FORMAT_ATTR(pkg_event, event, "config:0-63"); -static struct attribute *pkg_format_attrs[] = { - &format_attr_pkg_event.attr, - NULL, -}; -static struct attribute_group pkg_format_attr_group = { - .name = "format", - .attrs = pkg_format_attrs, -}; - static cpumask_t cstate_pkg_cpu_mask; -static const struct attribute_group *pkg_attr_groups[] = { - &pkg_events_attr_group, - &pkg_format_attr_group, - &cpumask_attr_group, - NULL, -}; - static ssize_t cstate_get_attr_cpumask(struct device *dev, struct device_attribute *attr, char *buf) @@ -476,7 +454,7 @@ static const struct attribute_group *pkg_attr_update[] = { }; static struct pmu cstate_core_pmu = { - .attr_groups = core_attr_groups, + .attr_groups = cstate_attr_groups, .attr_update = core_attr_update, .name = "cstate_core", .task_ctx_nr = perf_invalid_context, @@ -491,7 +469,7 @@ static struct pmu cstate_core_pmu = { }; static struct pmu cstate_pkg_pmu = { - .attr_groups = pkg_attr_groups, + .attr_groups = cstate_attr_groups, .attr_update = pkg_attr_update, .name = "cstate_pkg", .task_ctx_nr = perf_invalid_context, -- Gitee From e160cce2b5463c0cd046332811b094b6f7f47ad8 Mon Sep 17 00:00:00 2001 From: Kan Liang Date: Thu, 16 Nov 2023 06:22:43 -0800 Subject: [PATCH 2/4] x86/smp: Export symbol cpu_clustergroup_mask() ANBZ: #8054 commit c3dd1995620cdcd65cf4944c4164b0dbc16e557c upstream. Intel cstate PMU driver will invoke the topology_cluster_cpumask() to retrieve the CPU mask of a cluster. A modpost error is triggered since the symbol cpu_clustergroup_mask is not exported. Intel-SIG: commit c3dd1995620c x86/smp: Export symbol cpu_clustergroup_mask() Backport SRF/GRR perf cstate support to kernel 5.10. Signed-off-by: Kan Liang Signed-off-by: Peter Zijlstra (Intel) Link: https://lore.kernel.org/r/20231116142245.1233485-2-kan.liang@linux.intel.com [ Yunying Sun: amend commit log ] Signed-off-by: Yunying Sun --- arch/x86/kernel/smpboot.c | 1 + 1 file changed, 1 insertion(+) diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index fda3e4eaec73..dfb11ce17ad4 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -696,6 +696,7 @@ const struct cpumask *cpu_clustergroup_mask(int cpu) { return cpu_l2c_shared_mask(cpu); } +EXPORT_SYMBOL_GPL(cpu_clustergroup_mask); static void impress_friends(void) { -- Gitee From e7e135c57969eb2f593d84e1c0769b46096545e2 Mon Sep 17 00:00:00 2001 From: Kan Liang Date: Thu, 16 Nov 2023 06:22:44 -0800 Subject: [PATCH 3/4] perf/x86/intel/cstate: Add Sierra Forest support ANBZ: #8054 commit 3877d55a0db2688c2e4ab8a319614a0c81f8e2d2 upstream. A new module C6 Residency Counter is introduced in the Sierra Forest. The scope of the new counter is module (A cluster of cores shared L2 cache). Create a brand new cstate_module PMU to profile the new counter. The only differences between the new cstate_module PMU and the existing cstate PMU are the scope and events. Regarding the choice of the new cstate_module PMU name, the current naming rule of a cstate PMU is "cstate_" + the scope of the PMU. The scope of the PMU is the cores shared L2. On SRF, Intel calls it "module", while the internal Linux sched code calls it "cluster". The "cstate_module" is used as the new PMU name, because - The Cstate PMU driver is a Intel specific driver. It doesn't impact other ARCHs. The name makes it consistent with the documentation. - The "cluster" mainly be used by the scheduler developer, while the user of cstate PMU is more likely a researcher reading HW docs and optimizing power. - In the Intel's SDM, the "cluster" has a different meaning/scope for topology. Using it will mislead the end users. Besides the module C6, the core C1/C6 and pkg C6 residency counters are supported in the Sierra Forest as well. Intel-SIG: commit 3877d55a0db2 perf/x86/intel/cstate: Add Sierra Forest support Backport SRF/GRR perf cstate support to kernel 5.10. Suggested-by: Artem Bityutskiy Signed-off-by: Kan Liang Signed-off-by: Peter Zijlstra (Intel) Link: https://lore.kernel.org/r/20231116142245.1233485-3-kan.liang@linux.intel.com [ Yunying Sun: amend commit log ] Signed-off-by: Yunying Sun --- arch/x86/events/intel/cstate.c | 113 +++++++++++++++++++++++++++++++-- 1 file changed, 109 insertions(+), 4 deletions(-) diff --git a/arch/x86/events/intel/cstate.c b/arch/x86/events/intel/cstate.c index b4dd66867978..63481f828507 100644 --- a/arch/x86/events/intel/cstate.c +++ b/arch/x86/events/intel/cstate.c @@ -41,7 +41,7 @@ * MSR_CORE_C1_RES: CORE C1 Residency Counter * perf code: 0x00 * Available model: SLM,AMT,GLM,CNL,ICX,TNT,ADL,RPL - * MTL + * MTL,SRF * Scope: Core (each processor core has a MSR) * MSR_CORE_C3_RESIDENCY: CORE C3 Residency Counter * perf code: 0x01 @@ -52,7 +52,7 @@ * perf code: 0x02 * Available model: SLM,AMT,NHM,WSM,SNB,IVB,HSW,BDW, * SKL,KNL,GLM,CNL,KBL,CML,ICL,ICX, - * TGL,TNT,SPR,RPL,MTL + * TGL,TNT,SPR,RPL,MTL,SRF * Scope: Core * MSR_CORE_C7_RESIDENCY: CORE C7 Residency Counter * perf code: 0x03 @@ -74,7 +74,7 @@ * perf code: 0x02 * Available model: SLM,AMT,NHM,WSM,SNB,IVB,HSW,BDW, * SKL,KNL,GLM,CNL,KBL,CML,ICL,ICX, - * TGL,TNT,SPR,ADL,RPL,MTL + * TGL,TNT,SPR,ADL,RPL,MTL,SRF * Scope: Package (physical package) * MSR_PKG_C7_RESIDENCY: Package C7 Residency Counter. * perf code: 0x03 @@ -96,6 +96,10 @@ * Available model: HSW ULT,KBL,GLM,CNL,CML,ICL,TGL, * TNT,ADL,RPL,MTL * Scope: Package (physical package) + * MSR_MODULE_C6_RES_MS: Module C6 Residency Counter. + * perf code: 0x00 + * Available model: SRF + * Scope: A cluster of cores shared L2 cache * */ @@ -129,6 +133,7 @@ static ssize_t cstate_get_attr_cpumask(struct device *dev, struct cstate_model { unsigned long core_events; unsigned long pkg_events; + unsigned long module_events; unsigned long quirks; }; @@ -269,6 +274,28 @@ static struct perf_msr pkg_msr[] = { static cpumask_t cstate_pkg_cpu_mask; +/* cstate_module PMU */ +static struct pmu cstate_module_pmu; +static bool has_cstate_module; + +enum perf_cstate_module_events { + PERF_CSTATE_MODULE_C6_RES = 0, + + PERF_CSTATE_MODULE_EVENT_MAX, +}; + +PMU_EVENT_ATTR_STRING(c6-residency, attr_cstate_module_c6, "event=0x00"); + +static unsigned long module_msr_mask; + +PMU_EVENT_GROUP(events, cstate_module_c6); + +static struct perf_msr module_msr[] = { + [PERF_CSTATE_MODULE_C6_RES] = { MSR_MODULE_C6_RES_MS, &group_cstate_module_c6, test_msr }, +}; + +static cpumask_t cstate_module_cpu_mask; + static ssize_t cstate_get_attr_cpumask(struct device *dev, struct device_attribute *attr, char *buf) @@ -279,6 +306,8 @@ static ssize_t cstate_get_attr_cpumask(struct device *dev, return cpumap_print_to_pagebuf(true, buf, &cstate_core_cpu_mask); else if (pmu == &cstate_pkg_pmu) return cpumap_print_to_pagebuf(true, buf, &cstate_pkg_cpu_mask); + else if (pmu == &cstate_module_pmu) + return cpumap_print_to_pagebuf(true, buf, &cstate_module_cpu_mask); else return 0; } @@ -316,6 +345,15 @@ static int cstate_pmu_event_init(struct perf_event *event) event->hw.event_base = pkg_msr[cfg].msr; cpu = cpumask_any_and(&cstate_pkg_cpu_mask, topology_die_cpumask(event->cpu)); + } else if (event->pmu == &cstate_module_pmu) { + if (cfg >= PERF_CSTATE_MODULE_EVENT_MAX) + return -EINVAL; + cfg = array_index_nospec((unsigned long)cfg, PERF_CSTATE_MODULE_EVENT_MAX); + if (!(module_msr_mask & (1 << cfg))) + return -EINVAL; + event->hw.event_base = module_msr[cfg].msr; + cpu = cpumask_any_and(&cstate_module_cpu_mask, + topology_cluster_cpumask(event->cpu)); } else { return -ENOENT; } @@ -405,6 +443,17 @@ static int cstate_cpu_exit(unsigned int cpu) perf_pmu_migrate_context(&cstate_pkg_pmu, cpu, target); } } + + if (has_cstate_module && + cpumask_test_and_clear_cpu(cpu, &cstate_module_cpu_mask)) { + + target = cpumask_any_but(topology_cluster_cpumask(cpu), cpu); + /* Migrate events if there is a valid target */ + if (target < nr_cpu_ids) { + cpumask_set_cpu(target, &cstate_module_cpu_mask); + perf_pmu_migrate_context(&cstate_module_pmu, cpu, target); + } + } return 0; } @@ -431,6 +480,15 @@ static int cstate_cpu_init(unsigned int cpu) if (has_cstate_pkg && target >= nr_cpu_ids) cpumask_set_cpu(cpu, &cstate_pkg_cpu_mask); + /* + * If this is the first online thread of that cluster, set it + * in the cluster cpu mask as the designated reader. + */ + target = cpumask_any_and(&cstate_module_cpu_mask, + topology_cluster_cpumask(cpu)); + if (has_cstate_module && target >= nr_cpu_ids) + cpumask_set_cpu(cpu, &cstate_module_cpu_mask); + return 0; } @@ -453,6 +511,11 @@ static const struct attribute_group *pkg_attr_update[] = { NULL, }; +static const struct attribute_group *module_attr_update[] = { + &group_cstate_module_c6, + NULL +}; + static struct pmu cstate_core_pmu = { .attr_groups = cstate_attr_groups, .attr_update = core_attr_update, @@ -483,6 +546,21 @@ static struct pmu cstate_pkg_pmu = { .module = THIS_MODULE, }; +static struct pmu cstate_module_pmu = { + .attr_groups = cstate_attr_groups, + .attr_update = module_attr_update, + .name = "cstate_module", + .task_ctx_nr = perf_invalid_context, + .event_init = cstate_pmu_event_init, + .add = cstate_pmu_event_add, + .del = cstate_pmu_event_del, + .start = cstate_pmu_event_start, + .stop = cstate_pmu_event_stop, + .read = cstate_pmu_event_update, + .capabilities = PERF_PMU_CAP_NO_INTERRUPT | PERF_PMU_CAP_NO_EXCLUDE, + .module = THIS_MODULE, +}; + static const struct cstate_model nhm_cstates __initconst = { .core_events = BIT(PERF_CSTATE_CORE_C3_RES) | BIT(PERF_CSTATE_CORE_C6_RES), @@ -597,6 +675,15 @@ static const struct cstate_model glm_cstates __initconst = { BIT(PERF_CSTATE_PKG_C10_RES), }; +static const struct cstate_model srf_cstates __initconst = { + .core_events = BIT(PERF_CSTATE_CORE_C1_RES) | + BIT(PERF_CSTATE_CORE_C6_RES), + + .pkg_events = BIT(PERF_CSTATE_PKG_C6_RES), + + .module_events = BIT(PERF_CSTATE_MODULE_C6_RES), +}; + static const struct x86_cpu_id intel_cstates_match[] __initconst = { X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &nhm_cstates), @@ -649,6 +736,7 @@ static const struct x86_cpu_id intel_cstates_match[] __initconst = { X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT, &glm_cstates), X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_L, &glm_cstates), X86_MATCH_INTEL_FAM6_MODEL(ATOM_GRACEMONT, &adl_cstates), + X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT_X, &srf_cstates), X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_L, &icl_cstates), X86_MATCH_INTEL_FAM6_MODEL(ICELAKE, &icl_cstates), @@ -688,10 +776,14 @@ static int __init cstate_probe(const struct cstate_model *cm) pkg_msr_mask = perf_msr_probe(pkg_msr, PERF_CSTATE_PKG_EVENT_MAX, true, (void *) &cm->pkg_events); + module_msr_mask = perf_msr_probe(module_msr, PERF_CSTATE_MODULE_EVENT_MAX, + true, (void *) &cm->module_events); + has_cstate_core = !!core_msr_mask; has_cstate_pkg = !!pkg_msr_mask; + has_cstate_module = !!module_msr_mask; - return (has_cstate_core || has_cstate_pkg) ? 0 : -ENODEV; + return (has_cstate_core || has_cstate_pkg || has_cstate_module) ? 0 : -ENODEV; } static inline void cstate_cleanup(void) @@ -704,6 +796,9 @@ static inline void cstate_cleanup(void) if (has_cstate_pkg) perf_pmu_unregister(&cstate_pkg_pmu); + + if (has_cstate_module) + perf_pmu_unregister(&cstate_module_pmu); } static int __init cstate_init(void) @@ -740,6 +835,16 @@ static int __init cstate_init(void) return err; } } + + if (has_cstate_module) { + err = perf_pmu_register(&cstate_module_pmu, cstate_module_pmu.name, -1); + if (err) { + has_cstate_module = false; + pr_info("Failed to register cstate cluster pmu\n"); + cstate_cleanup(); + return err; + } + } return 0; } -- Gitee From 5beb4bdcfbd941129390f999b05d0a0a6d4c8306 Mon Sep 17 00:00:00 2001 From: Kan Liang Date: Thu, 16 Nov 2023 06:22:45 -0800 Subject: [PATCH 4/4] perf/x86/intel/cstate: Add Grand Ridge support ANBZ: #8054 commit bbb968696d0f3442ab823598def3b756cf4735c6 upstream. The same as the Sierra Forest, the Grand Ridge supports core C1/C6 and module C6. But it doesn't support pkg C6 residency counter. Intel-SIG: commit bbb968696d0f perf/x86/intel/cstate: Add Grand Ridge support Backport SRF/GRR perf cstate support to kernel 5.10. Signed-off-by: Kan Liang Signed-off-by: Peter Zijlstra (Intel) Link: https://lore.kernel.org/r/20231116142245.1233485-4-kan.liang@linux.intel.com [ Yunying Sun: amend commit log ] Signed-off-by: Yunying Sun --- arch/x86/events/intel/cstate.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/arch/x86/events/intel/cstate.c b/arch/x86/events/intel/cstate.c index 63481f828507..14eb6351657d 100644 --- a/arch/x86/events/intel/cstate.c +++ b/arch/x86/events/intel/cstate.c @@ -41,7 +41,7 @@ * MSR_CORE_C1_RES: CORE C1 Residency Counter * perf code: 0x00 * Available model: SLM,AMT,GLM,CNL,ICX,TNT,ADL,RPL - * MTL,SRF + * MTL,SRF,GRR * Scope: Core (each processor core has a MSR) * MSR_CORE_C3_RESIDENCY: CORE C3 Residency Counter * perf code: 0x01 @@ -52,7 +52,8 @@ * perf code: 0x02 * Available model: SLM,AMT,NHM,WSM,SNB,IVB,HSW,BDW, * SKL,KNL,GLM,CNL,KBL,CML,ICL,ICX, - * TGL,TNT,SPR,RPL,MTL,SRF + * TGL,TNT,SPR,RPL,MTL,SRF, + * GRR * Scope: Core * MSR_CORE_C7_RESIDENCY: CORE C7 Residency Counter * perf code: 0x03 @@ -98,7 +99,7 @@ * Scope: Package (physical package) * MSR_MODULE_C6_RES_MS: Module C6 Residency Counter. * perf code: 0x00 - * Available model: SRF + * Available model: SRF,GRR * Scope: A cluster of cores shared L2 cache * */ @@ -675,6 +676,13 @@ static const struct cstate_model glm_cstates __initconst = { BIT(PERF_CSTATE_PKG_C10_RES), }; +static const struct cstate_model grr_cstates __initconst = { + .core_events = BIT(PERF_CSTATE_CORE_C1_RES) | + BIT(PERF_CSTATE_CORE_C6_RES), + + .module_events = BIT(PERF_CSTATE_MODULE_C6_RES), +}; + static const struct cstate_model srf_cstates __initconst = { .core_events = BIT(PERF_CSTATE_CORE_C1_RES) | BIT(PERF_CSTATE_CORE_C6_RES), @@ -737,6 +745,7 @@ static const struct x86_cpu_id intel_cstates_match[] __initconst = { X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_L, &glm_cstates), X86_MATCH_INTEL_FAM6_MODEL(ATOM_GRACEMONT, &adl_cstates), X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT_X, &srf_cstates), + X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT, &grr_cstates), X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_L, &icl_cstates), X86_MATCH_INTEL_FAM6_MODEL(ICELAKE, &icl_cstates), -- Gitee