From a63eaf34ae60bdb067a354cc8def2e8f4a01f5f4 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 22 May 2009 14:17:31 +1000 Subject: perf_counter: Dynamically allocate tasks' perf_counter_context struct This replaces the struct perf_counter_context in the task_struct with a pointer to a dynamically allocated perf_counter_context struct. The main reason for doing is this is to allow us to transfer a perf_counter_context from one task to another when we do lazy PMU switching in a later patch. This has a few side-benefits: the task_struct becomes a little smaller, we save some memory because only tasks that have perf_counters attached get a perf_counter_context allocated for them, and we can remove the inclusion of in sched.h, meaning that we don't end up recompiling nearly everything whenever perf_counter.h changes. The perf_counter_context structures are reference-counted and freed when the last reference is dropped. A context can have references from its task and the counters on its task. Counters can outlive the task so it is possible that a context will be freed well after its task has exited. Contexts are allocated on fork if the parent had a context, or otherwise the first time that a per-task counter is created on a task. In the latter case, we set the context pointer in the task struct locklessly using an atomic compare-and-exchange operation in case we raced with some other task in creating a context for the subject task. This also removes the task pointer from the perf_counter struct. The task pointer was not used anywhere and would make it harder to move a context from one task to another. Anything that needed to know which task a counter was attached to was already using counter->ctx->task. The __perf_counter_init_context function moves up in perf_counter.c so that it can be called from find_get_context, and now initializes the refcount, but is otherwise unchanged. We were potentially calling list_del_counter twice: once from __perf_counter_exit_task when the task exits and once from __perf_counter_remove_from_context when the counter's fd gets closed. This adds a check in list_del_counter so it doesn't do anything if the counter has already been removed from the lists. Since perf_counter_task_sched_in doesn't do anything if the task doesn't have a context, and leaves cpuctx->task_ctx = NULL, this adds code to __perf_install_in_context to set cpuctx->task_ctx if necessary, i.e. in the case where the current task adds the first counter to itself and thus creates a context for itself. This also adds similar code to __perf_counter_enable to handle a similar situation which can arise when the counters have been disabled using prctl; that also leaves cpuctx->task_ctx = NULL. [ Impact: refactor counter context management to prepare for new feature ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: <18966.10075.781053.231153@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic/apic.c | 1 + 1 file changed, 1 insertion(+) (limited to 'arch/x86/kernel/apic/apic.c') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index e9021a90802..b4f64402a82 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -14,6 +14,7 @@ * Mikael Pettersson : PM converted to driver model. */ +#include #include #include #include -- cgit v1.2.3-70-g09d2 From 48e22d56ecdeddd1ffb42a02fccba5c6ef42b133 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 17:39:04 +0200 Subject: perf_counter: x86: Remove interrupt throttle remove the x86 specific interrupt throttle Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525153931.616671838@chello.nl> Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic/apic.c | 2 -- arch/x86/kernel/cpu/perf_counter.c | 47 ++++---------------------------------- include/linux/perf_counter.h | 2 -- 3 files changed, 5 insertions(+), 46 deletions(-) (limited to 'arch/x86/kernel/apic/apic.c') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index b4f64402a82..89b63b5fad3 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -763,8 +763,6 @@ static void local_apic_timer_interrupt(void) inc_irq_stat(apic_timer_irqs); evt->event_handler(evt); - - perf_counter_unthrottle(); } /* diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index c14437faf5d..8c8177f859f 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -718,11 +718,6 @@ static void intel_pmu_save_and_restart(struct perf_counter *counter) intel_pmu_enable_counter(hwc, idx); } -/* - * Maximum interrupt frequency of 100KHz per CPU - */ -#define PERFMON_MAX_INTERRUPTS (100000/HZ) - /* * This handler is triggered by the local APIC, so the APIC IRQ handling * rules apply: @@ -775,15 +770,14 @@ again: if (status) goto again; - if (++cpuc->interrupts != PERFMON_MAX_INTERRUPTS) - perf_enable(); + perf_enable(); return 1; } static int amd_pmu_handle_irq(struct pt_regs *regs, int nmi) { - int cpu, idx, throttle = 0, handled = 0; + int cpu, idx, handled = 0; struct cpu_hw_counters *cpuc; struct perf_counter *counter; struct hw_perf_counter *hwc; @@ -792,16 +786,7 @@ static int amd_pmu_handle_irq(struct pt_regs *regs, int nmi) cpu = smp_processor_id(); cpuc = &per_cpu(cpu_hw_counters, cpu); - if (++cpuc->interrupts == PERFMON_MAX_INTERRUPTS) { - throttle = 1; - __perf_disable(); - cpuc->enabled = 0; - barrier(); - } - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - int disable = 0; - if (!test_bit(idx, cpuc->active_mask)) continue; @@ -809,45 +794,23 @@ static int amd_pmu_handle_irq(struct pt_regs *regs, int nmi) hwc = &counter->hw; if (counter->hw_event.nmi != nmi) - goto next; + continue; val = x86_perf_counter_update(counter, hwc, idx); if (val & (1ULL << (x86_pmu.counter_bits - 1))) - goto next; + continue; /* counter overflow */ x86_perf_counter_set_period(counter, hwc, idx); handled = 1; inc_irq_stat(apic_perf_irqs); - disable = perf_counter_overflow(counter, nmi, regs, 0); - -next: - if (disable || throttle) + if (perf_counter_overflow(counter, nmi, regs, 0)) amd_pmu_disable_counter(hwc, idx); } return handled; } -void perf_counter_unthrottle(void) -{ - struct cpu_hw_counters *cpuc; - - if (!x86_pmu_initialized()) - return; - - cpuc = &__get_cpu_var(cpu_hw_counters); - if (cpuc->interrupts >= PERFMON_MAX_INTERRUPTS) { - /* - * Clear them before re-enabling irqs/NMIs again: - */ - cpuc->interrupts = 0; - perf_enable(); - } else { - cpuc->interrupts = 0; - } -} - void smp_perf_counter_interrupt(struct pt_regs *regs) { irq_enter(); diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index d3e85de9bf1..0c160be2078 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -570,7 +570,6 @@ extern int perf_counter_init_task(struct task_struct *child); extern void perf_counter_exit_task(struct task_struct *child); extern void perf_counter_do_pending(void); extern void perf_counter_print_debug(void); -extern void perf_counter_unthrottle(void); extern void __perf_disable(void); extern bool __perf_enable(void); extern void perf_disable(void); @@ -635,7 +634,6 @@ static inline int perf_counter_init_task(struct task_struct *child) { } static inline void perf_counter_exit_task(struct task_struct *child) { } static inline void perf_counter_do_pending(void) { } static inline void perf_counter_print_debug(void) { } -static inline void perf_counter_unthrottle(void) { } static inline void perf_disable(void) { } static inline void perf_enable(void) { } static inline int perf_counter_task_disable(void) { return -EINVAL; } -- cgit v1.2.3-70-g09d2 From c323d95fa4dbe0b6bf6d59e24a0b7db067dd08a7 Mon Sep 17 00:00:00 2001 From: Yong Wang Date: Fri, 29 May 2009 13:28:35 +0800 Subject: perf_counter/x86: Always use NMI for performance-monitoring interrupt Always use NMI for performance-monitoring interrupt as there could be racy situations if we switch between irq and nmi mode frequently. Signed-off-by: Yong Wang LKML-Reference: <20090529052835.GA13657@ywang-moblin2.bj.intel.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/perf_counter.h | 4 ++-- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/cpu/perf_counter.c | 19 +++++-------------- 3 files changed, 8 insertions(+), 17 deletions(-) (limited to 'arch/x86/kernel/apic/apic.c') diff --git a/arch/x86/include/asm/perf_counter.h b/arch/x86/include/asm/perf_counter.h index d08dd52cb8f..876ed97147b 100644 --- a/arch/x86/include/asm/perf_counter.h +++ b/arch/x86/include/asm/perf_counter.h @@ -91,10 +91,10 @@ extern void set_perf_counter_pending(void); #ifdef CONFIG_PERF_COUNTERS extern void init_hw_perf_counters(void); -extern void perf_counters_lapic_init(int nmi); +extern void perf_counters_lapic_init(void); #else static inline void init_hw_perf_counters(void) { } -static inline void perf_counters_lapic_init(int nmi) { } +static inline void perf_counters_lapic_init(void) { } #endif #endif /* _ASM_X86_PERF_COUNTER_H */ diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index 89b63b5fad3..60df2efd7c8 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -1135,7 +1135,7 @@ void __cpuinit setup_local_APIC(void) apic_write(APIC_ESR, 0); } #endif - perf_counters_lapic_init(0); + perf_counters_lapic_init(); preempt_disable(); diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index 2eeaa99add1..316b0c995f3 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -604,7 +604,7 @@ try_generic: hwc->counter_base = x86_pmu.perfctr; } - perf_counters_lapic_init(hwc->nmi); + perf_counters_lapic_init(); x86_pmu.disable(hwc, idx); @@ -863,24 +863,15 @@ void set_perf_counter_pending(void) apic->send_IPI_self(LOCAL_PENDING_VECTOR); } -void perf_counters_lapic_init(int nmi) +void perf_counters_lapic_init(void) { - u32 apic_val; - if (!x86_pmu_initialized()) return; /* - * Enable the performance counter vector in the APIC LVT: + * Always use NMI for PMU */ - apic_val = apic_read(APIC_LVTERR); - - apic_write(APIC_LVTERR, apic_val | APIC_LVT_MASKED); - if (nmi) - apic_write(APIC_LVTPC, APIC_DM_NMI); - else - apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR); - apic_write(APIC_LVTERR, apic_val); + apic_write(APIC_LVTPC, APIC_DM_NMI); } static int __kprobes @@ -1054,7 +1045,7 @@ void __init init_hw_perf_counters(void) pr_info("... counter mask: %016Lx\n", perf_counter_mask); - perf_counters_lapic_init(0); + perf_counters_lapic_init(); register_die_notifier(&perf_counter_nmi_notifier); } -- cgit v1.2.3-70-g09d2 From 3d58829b0510244596079c1d2f1762c53aef2e97 Mon Sep 17 00:00:00 2001 From: Jiri Slaby Date: Thu, 28 May 2009 09:54:47 +0200 Subject: x86, apic: Restore irqs on fail paths lapic_resume forgets to restore interrupts on fail paths. Fix that. Signed-off-by: Jiri Slaby Acked-by: Cyrill Gorcunov LKML-Reference: <1243497289-18591-1-git-send-email-jirislaby@gmail.com> Signed-off-by: Ingo Molnar Cc: H. Peter Anvin --- arch/x86/kernel/apic/apic.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) (limited to 'arch/x86/kernel/apic/apic.c') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index b0fd26442c4..e82488d3f0b 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -2027,7 +2027,7 @@ static int lapic_resume(struct sys_device *dev) unsigned int l, h; unsigned long flags; int maxlvt; - int ret; + int ret = 0; struct IO_APIC_route_entry **ioapic_entries = NULL; if (!apic_pm_state.active) @@ -2038,14 +2038,15 @@ static int lapic_resume(struct sys_device *dev) ioapic_entries = alloc_ioapic_entries(); if (!ioapic_entries) { WARN(1, "Alloc ioapic_entries in lapic resume failed."); - return -ENOMEM; + ret = -ENOMEM; + goto restore; } ret = save_IO_APIC_setup(ioapic_entries); if (ret) { WARN(1, "Saving IO-APIC state failed: %d\n", ret); free_ioapic_entries(ioapic_entries); - return ret; + goto restore; } mask_IO_APIC_setup(ioapic_entries); @@ -2097,10 +2098,10 @@ static int lapic_resume(struct sys_device *dev) restore_IO_APIC_setup(ioapic_entries); free_ioapic_entries(ioapic_entries); } - +restore: local_irq_restore(flags); - return 0; + return ret; } /* -- cgit v1.2.3-70-g09d2 From 103428e57be323c3c5545db8ad12667099bc6005 Mon Sep 17 00:00:00 2001 From: Cyrill Gorcunov Date: Sun, 7 Jun 2009 16:48:40 +0400 Subject: x86, apic: Fix dummy apic read operation together with broken MP handling Ingo Molnar reported that read_apic is buggy novadays: [ 0.000000] Using APIC driver default [ 0.000000] SMP: Allowing 1 CPUs, 0 hotplug CPUs [ 0.000000] Local APIC disabled by BIOS -- you can enable it with "lapic" [ 0.000000] APIC: disable apic facility [ 0.000000] ------------[ cut here ]------------ [ 0.000000] WARNING: at arch/x86/kernel/apic/apic.c:254 native_apic_read_dummy+0x2d/0x3b() [ 0.000000] Hardware name: HP OmniBook PC Indeed we still rely on apic->read operation for SMP compiled kernel. And instead of disfigure the SMP code with #ifdef we allow to call apic->read. To capture any unexpected results we check for apic->read being called for sane reason via WARN_ON_ONCE but(!) instead of OR we should use AND logical operation (thanks Yinghai for spotting the root of the problem). Along with that we could be have bad MP table and we are to fix it that way no SMP started and no complains about BIOS bug if apic was just disabled via command line. Signed-off-by: Cyrill Gorcunov Cc: Yinghai Lu LKML-Reference: <20090607124840.GD4547@lenovo> Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic/apic.c | 9 ++++++++- arch/x86/kernel/smpboot.c | 8 +++++--- 2 files changed, 13 insertions(+), 4 deletions(-) (limited to 'arch/x86/kernel/apic/apic.c') diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index e82488d3f0b..a4c9cf0bf70 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -249,7 +249,7 @@ static void native_apic_write_dummy(u32 reg, u32 v) static u32 native_apic_read_dummy(u32 reg) { - WARN_ON_ONCE((cpu_has_apic || !disable_apic)); + WARN_ON_ONCE((cpu_has_apic && !disable_apic)); return 0; } @@ -1609,6 +1609,13 @@ void __init init_apic_mappings(void) new_apicid = read_apic_id(); if (boot_cpu_physical_apicid != new_apicid) { boot_cpu_physical_apicid = new_apicid; + /* + * yeah -- we lie about apic_version + * in case if apic was disabled via boot option + * but it's not a problem for SMP compiled kernel + * since smp_sanity_check is prepared for such a case + * and disable smp mode + */ apic_version[new_apicid] = GET_APIC_VERSION(apic_read(APIC_LVR)); } diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index d2e8de95815..7c80007ea5f 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -992,10 +992,12 @@ static int __init smp_sanity_check(unsigned max_cpus) */ if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) && !cpu_has_apic) { - printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n", - boot_cpu_physical_apicid); - printk(KERN_ERR "... forcing use of dummy APIC emulation." + if (!disable_apic) { + pr_err("BIOS bug, local APIC #%d not detected!...\n", + boot_cpu_physical_apicid); + pr_err("... forcing use of dummy APIC emulation." "(tell your hw vendor)\n"); + } smpboot_clear_io_apic(); arch_disable_smp_support(); return -1; -- cgit v1.2.3-70-g09d2