diff options
Diffstat (limited to 'arch/i386/kernel/cpu/perfctr-watchdog.c')
| -rw-r--r-- | arch/i386/kernel/cpu/perfctr-watchdog.c | 658 |
1 files changed, 658 insertions, 0 deletions
diff --git a/arch/i386/kernel/cpu/perfctr-watchdog.c b/arch/i386/kernel/cpu/perfctr-watchdog.c new file mode 100644 index 00000000000..2b04c8f1db6 --- /dev/null +++ b/arch/i386/kernel/cpu/perfctr-watchdog.c @@ -0,0 +1,658 @@ +/* local apic based NMI watchdog for various CPUs. + This file also handles reservation of performance counters for coordination + with other users (like oprofile). + + Note that these events normally don't tick when the CPU idles. This means + the frequency varies with CPU load. + + Original code for K7/P6 written by Keith Owens */ + +#include <linux/percpu.h> +#include <linux/module.h> +#include <linux/kernel.h> +#include <linux/bitops.h> +#include <linux/smp.h> +#include <linux/nmi.h> +#include <asm/apic.h> +#include <asm/intel_arch_perfmon.h> + +struct nmi_watchdog_ctlblk { + unsigned int cccr_msr; + unsigned int perfctr_msr; /* the MSR to reset in NMI handler */ + unsigned int evntsel_msr; /* the MSR to select the events to handle */ +}; + +/* Interface defining a CPU specific perfctr watchdog */ +struct wd_ops { + int (*reserve)(void); + void (*unreserve)(void); + int (*setup)(unsigned nmi_hz); + void (*rearm)(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz); + void (*stop)(void *); + unsigned perfctr; + unsigned evntsel; + u64 checkbit; +}; + +static struct wd_ops *wd_ops; + +/* this number is calculated from Intel's MSR_P4_CRU_ESCR5 register and it's + * offset from MSR_P4_BSU_ESCR0. It will be the max for all platforms (for now) + */ +#define NMI_MAX_COUNTER_BITS 66 + +/* perfctr_nmi_owner tracks the ownership of the perfctr registers: + * evtsel_nmi_owner tracks the ownership of the event selection + * - different performance counters/ event selection may be reserved for + * different subsystems this reservation system just tries to coordinate + * things a little + */ +static DECLARE_BITMAP(perfctr_nmi_owner, NMI_MAX_COUNTER_BITS); +static DECLARE_BITMAP(evntsel_nmi_owner, NMI_MAX_COUNTER_BITS); + +static DEFINE_PER_CPU(struct nmi_watchdog_ctlblk, nmi_watchdog_ctlblk); + +/* converts an msr to an appropriate reservation bit */ +static inline unsigned int nmi_perfctr_msr_to_bit(unsigned int msr) +{ + return wd_ops ? msr - wd_ops->perfctr : 0; +} + +/* converts an msr to an appropriate reservation bit */ +/* returns the bit offset of the event selection register */ +static inline unsigned int nmi_evntsel_msr_to_bit(unsigned int msr) +{ + return wd_ops ? msr - wd_ops->evntsel : 0; +} + +/* checks for a bit availability (hack for oprofile) */ +int avail_to_resrv_perfctr_nmi_bit(unsigned int counter) +{ + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + return (!test_bit(counter, perfctr_nmi_owner)); +} + +/* checks the an msr for availability */ +int avail_to_resrv_perfctr_nmi(unsigned int msr) +{ + unsigned int counter; + + counter = nmi_perfctr_msr_to_bit(msr); + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + return (!test_bit(counter, perfctr_nmi_owner)); +} + +int reserve_perfctr_nmi(unsigned int msr) +{ + unsigned int counter; + + counter = nmi_perfctr_msr_to_bit(msr); + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + if (!test_and_set_bit(counter, perfctr_nmi_owner)) + return 1; + return 0; +} + +void release_perfctr_nmi(unsigned int msr) +{ + unsigned int counter; + + counter = nmi_perfctr_msr_to_bit(msr); + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + clear_bit(counter, perfctr_nmi_owner); +} + +int reserve_evntsel_nmi(unsigned int msr) +{ + unsigned int counter; + + counter = nmi_evntsel_msr_to_bit(msr); + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + if (!test_and_set_bit(counter, evntsel_nmi_owner)) + return 1; + return 0; +} + +void release_evntsel_nmi(unsigned int msr) +{ + unsigned int counter; + + counter = nmi_evntsel_msr_to_bit(msr); + BUG_ON(counter > NMI_MAX_COUNTER_BITS); + + clear_bit(counter, evntsel_nmi_owner); +} + +EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi); +EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi_bit); +EXPORT_SYMBOL(reserve_perfctr_nmi); +EXPORT_SYMBOL(release_perfctr_nmi); +EXPORT_SYMBOL(reserve_evntsel_nmi); +EXPORT_SYMBOL(release_evntsel_nmi); + +void disable_lapic_nmi_watchdog(void) +{ + BUG_ON(nmi_watchdog != NMI_LOCAL_APIC); + + if (atomic_read(&nmi_active) <= 0) + return; + + on_each_cpu(wd_ops->stop, NULL, 0, 1); + wd_ops->unreserve(); + + BUG_ON(atomic_read(&nmi_active) != 0); +} + +void enable_lapic_nmi_watchdog(void) +{ + BUG_ON(nmi_watchdog != NMI_LOCAL_APIC); + + /* are we already enabled */ + if (atomic_read(&nmi_active) != 0) + return; + + /* are we lapic aware */ + if (!wd_ops) + return; + if (!wd_ops->reserve()) { + printk(KERN_ERR "NMI watchdog: cannot reserve perfctrs\n"); + return; + } + + on_each_cpu(setup_apic_nmi_watchdog, NULL, 0, 1); + touch_nmi_watchdog(); +} + +/* + * Activate the NMI watchdog via the local APIC. + */ + +static unsigned int adjust_for_32bit_ctr(unsigned int hz) +{ + u64 counter_val; + unsigned int retval = hz; + + /* + * On Intel CPUs with P6/ARCH_PERFMON only 32 bits in the counter + * are writable, with higher bits sign extending from bit 31. + * So, we can only program the counter with 31 bit values and + * 32nd bit should be 1, for 33.. to be 1. + * Find the appropriate nmi_hz + */ + counter_val = (u64)cpu_khz * 1000; + do_div(counter_val, retval); + if (counter_val > 0x7fffffffULL) { + u64 count = (u64)cpu_khz * 1000; + do_div(count, 0x7fffffffUL); + retval = count + 1; + } + return retval; +} + +static void +write_watchdog_counter(unsigned int perfctr_msr, const char *descr, unsigned nmi_hz) +{ + u64 count = (u64)cpu_khz * 1000; + + do_div(count, nmi_hz); + if(descr) + Dprintk("setting %s to -0x%08Lx\n", descr, count); + wrmsrl(perfctr_msr, 0 - count); +} + +static void write_watchdog_counter32(unsigned int perfctr_msr, + const char *descr, unsigned nmi_hz) +{ + u64 count = (u64)cpu_khz * 1000; + + do_div(count, nmi_hz); + if(descr) + Dprintk("setting %s to -0x%08Lx\n", descr, count); + wrmsr(perfctr_msr, (u32)(-count), 0); +} + +/* AMD K7/K8/Family10h/Family11h support. AMD keeps this interface + nicely stable so there is not much variety */ + +#define K7_EVNTSEL_ENABLE (1 << 22) +#define K7_EVNTSEL_INT (1 << 20) +#define K7_EVNTSEL_OS (1 << 17) +#define K7_EVNTSEL_USR (1 << 16) +#define K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING 0x76 +#define K7_NMI_EVENT K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING + +static int setup_k7_watchdog(unsigned nmi_hz) +{ + unsigned int perfctr_msr, evntsel_msr; + unsigned int evntsel; + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + + perfctr_msr = MSR_K7_PERFCTR0; + evntsel_msr = MSR_K7_EVNTSEL0; + + wrmsrl(perfctr_msr, 0UL); + + evntsel = K7_EVNTSEL_INT + | K7_EVNTSEL_OS + | K7_EVNTSEL_USR + | K7_NMI_EVENT; + + /* setup the timer */ + wrmsr(evntsel_msr, evntsel, 0); + write_watchdog_counter(perfctr_msr, "K7_PERFCTR0",nmi_hz); + apic_write(APIC_LVTPC, APIC_DM_NMI); + evntsel |= K7_EVNTSEL_ENABLE; + wrmsr(evntsel_msr, evntsel, 0); + + wd->perfctr_msr = perfctr_msr; + wd->evntsel_msr = evntsel_msr; + wd->cccr_msr = 0; //unused + return 1; +} + +static void single_msr_stop_watchdog(void *arg) +{ + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + + wrmsr(wd->evntsel_msr, 0, 0); +} + +static int single_msr_reserve(void) +{ + if (!reserve_perfctr_nmi(wd_ops->perfctr)) + return 0; + + if (!reserve_evntsel_nmi(wd_ops->evntsel)) { + release_perfctr_nmi(wd_ops->perfctr); + return 0; + } + return 1; +} + +static void single_msr_unreserve(void) +{ + release_evntsel_nmi(wd_ops->perfctr); + release_perfctr_nmi(wd_ops->evntsel); +} + +static void single_msr_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz) +{ + /* start the cycle over again */ + write_watchdog_counter(wd->perfctr_msr, NULL, nmi_hz); +} + +static struct wd_ops k7_wd_ops = { + .reserve = single_msr_reserve, + .unreserve = single_msr_unreserve, + .setup = setup_k7_watchdog, + .rearm = single_msr_rearm, + .stop = single_msr_stop_watchdog, + .perfctr = MSR_K7_PERFCTR0, + .evntsel = MSR_K7_EVNTSEL0, + .checkbit = 1ULL<<63, +}; + +/* Intel Model 6 (PPro+,P2,P3,P-M,Core1) */ + +#define P6_EVNTSEL0_ENABLE (1 << 22) +#define P6_EVNTSEL_INT (1 << 20) +#define P6_EVNTSEL_OS (1 << 17) +#define P6_EVNTSEL_USR (1 << 16) +#define P6_EVENT_CPU_CLOCKS_NOT_HALTED 0x79 +#define P6_NMI_EVENT P6_EVENT_CPU_CLOCKS_NOT_HALTED + +static int setup_p6_watchdog(unsigned nmi_hz) +{ + unsigned int perfctr_msr, evntsel_msr; + unsigned int evntsel; + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + + perfctr_msr = MSR_P6_PERFCTR0; + evntsel_msr = MSR_P6_EVNTSEL0; + + wrmsrl(perfctr_msr, 0UL); + + evntsel = P6_EVNTSEL_INT + | P6_EVNTSEL_OS + | P6_EVNTSEL_USR + | P6_NMI_EVENT; + + /* setup the timer */ + wrmsr(evntsel_msr, evntsel, 0); + nmi_hz = adjust_for_32bit_ctr(nmi_hz); + write_watchdog_counter32(perfctr_msr, "P6_PERFCTR0",nmi_hz); + apic_write(APIC_LVTPC, APIC_DM_NMI); + evntsel |= P6_EVNTSEL0_ENABLE; + wrmsr(evntsel_msr, evntsel, 0); + + wd->perfctr_msr = perfctr_msr; + wd->evntsel_msr = evntsel_msr; + wd->cccr_msr = 0; //unused + return 1; +} + +static void p6_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz) +{ + /* P6 based Pentium M need to re-unmask + * the apic vector but it doesn't hurt + * other P6 variant. + * ArchPerfom/Core Duo also needs this */ + apic_write(APIC_LVTPC, APIC_DM_NMI); + /* P6/ARCH_PERFMON has 32 bit counter write */ + write_watchdog_counter32(wd->perfctr_msr, NULL,nmi_hz); +} + +static struct wd_ops p6_wd_ops = { + .reserve = single_msr_reserve, + .unreserve = single_msr_unreserve, + .setup = setup_p6_watchdog, + .rearm = p6_rearm, + .stop = single_msr_stop_watchdog, + .perfctr = MSR_P6_PERFCTR0, + .evntsel = MSR_P6_EVNTSEL0, + .checkbit = 1ULL<<39, +}; + +/* Intel P4 performance counters. By far the most complicated of all. */ + +#define MSR_P4_MISC_ENABLE_PERF_AVAIL (1<<7) +#define P4_ESCR_EVENT_SELECT(N) ((N)<<25) +#define P4_ESCR_OS (1<<3) +#define P4_ESCR_USR (1<<2) +#define P4_CCCR_OVF_PMI0 (1<<26) +#define P4_CCCR_OVF_PMI1 (1<<27) +#define P4_CCCR_THRESHOLD(N) ((N)<<20) +#define P4_CCCR_COMPLEMENT (1<<19) +#define P4_CCCR_COMPARE (1<<18) +#define P4_CCCR_REQUIRED (3<<16) +#define P4_CCCR_ESCR_SELECT(N) ((N)<<13) +#define P4_CCCR_ENABLE (1<<12) +#define P4_CCCR_OVF (1<<31) + +/* Set up IQ_COUNTER0 to behave like a clock, by having IQ_CCCR0 filter + CRU_ESCR0 (with any non-null event selector) through a complemented + max threshold. [IA32-Vol3, Section 14.9.9] */ + +static int setup_p4_watchdog(unsigned nmi_hz) +{ + unsigned int perfctr_msr, evntsel_msr, cccr_msr; + unsigned int evntsel, cccr_val; + unsigned int misc_enable, dummy; + unsigned int ht_num; + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + + rdmsr(MSR_IA32_MISC_ENABLE, misc_enable, dummy); + if (!(misc_enable & MSR_P4_MISC_ENABLE_PERF_AVAIL)) + return 0; + +#ifdef CONFIG_SMP + /* detect which hyperthread we are on */ + if (smp_num_siblings == 2) { + unsigned int ebx, apicid; + + ebx = cpuid_ebx(1); + apicid = (ebx >> 24) & 0xff; + ht_num = apicid & 1; + } else +#endif + ht_num = 0; + + /* performance counters are shared resources + * assign each hyperthread its own set + * (re-use the ESCR0 register, seems safe + * and keeps the cccr_val the same) + */ + if (!ht_num) { + /* logical cpu 0 */ + perfctr_msr = MSR_P4_IQ_PERFCTR0; + evntsel_msr = MSR_P4_CRU_ESCR0; + cccr_msr = MSR_P4_IQ_CCCR0; + cccr_val = P4_CCCR_OVF_PMI0 | P4_CCCR_ESCR_SELECT(4); + } else { + /* logical cpu 1 */ + perfctr_msr = MSR_P4_IQ_PERFCTR1; + evntsel_msr = MSR_P4_CRU_ESCR0; + cccr_msr = MSR_P4_IQ_CCCR1; + cccr_val = P4_CCCR_OVF_PMI1 | P4_CCCR_ESCR_SELECT(4); + } + + evntsel = P4_ESCR_EVENT_SELECT(0x3F) + | P4_ESCR_OS + | P4_ESCR_USR; + + cccr_val |= P4_CCCR_THRESHOLD(15) + | P4_CCCR_COMPLEMENT + | P4_CCCR_COMPARE + | P4_CCCR_REQUIRED; + + wrmsr(evntsel_msr, evntsel, 0); + wrmsr(cccr_msr, cccr_val, 0); + write_watchdog_counter(perfctr_msr, "P4_IQ_COUNTER0", nmi_hz); + apic_write(APIC_LVTPC, APIC_DM_NMI); + cccr_val |= P4_CCCR_ENABLE; + wrmsr(cccr_msr, cccr_val, 0); + wd->perfctr_msr = perfctr_msr; + wd->evntsel_msr = evntsel_msr; + wd->cccr_msr = cccr_msr; + return 1; +} + +static void stop_p4_watchdog(void *arg) +{ + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + wrmsr(wd->cccr_msr, 0, 0); + wrmsr(wd->evntsel_msr, 0, 0); +} + +static int p4_reserve(void) +{ + if (!reserve_perfctr_nmi(MSR_P4_IQ_PERFCTR0)) + return 0; +#ifdef CONFIG_SMP + if (smp_num_siblings > 1 && !reserve_perfctr_nmi(MSR_P4_IQ_PERFCTR1)) + goto fail1; +#endif + if (!reserve_evntsel_nmi(MSR_P4_CRU_ESCR0)) + goto fail2; + /* RED-PEN why is ESCR1 not reserved here? */ + return 1; + fail2: +#ifdef CONFIG_SMP + if (smp_num_siblings > 1) + release_perfctr_nmi(MSR_P4_IQ_PERFCTR1); + fail1: +#endif + release_perfctr_nmi(MSR_P4_IQ_PERFCTR0); + return 0; +} + +static void p4_unreserve(void) +{ +#ifdef CONFIG_SMP + if (smp_num_siblings > 1) + release_evntsel_nmi(MSR_P4_IQ_PERFCTR1); +#endif + release_evntsel_nmi(MSR_P4_IQ_PERFCTR0); + release_perfctr_nmi(MSR_P4_CRU_ESCR0); +} + +static void p4_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz) +{ + unsigned dummy; + /* + * P4 quirks: + * - An overflown perfctr will assert its interrupt + * until the OVF flag in its CCCR is cleared. + * - LVTPC is masked on interrupt and must be + * unmasked by the LVTPC handler. + */ + rdmsrl(wd->cccr_msr, dummy); + dummy &= ~P4_CCCR_OVF; + wrmsrl(wd->cccr_msr, dummy); + apic_write(APIC_LVTPC, APIC_DM_NMI); + /* start the cycle over again */ + write_watchdog_counter(wd->perfctr_msr, NULL, nmi_hz); +} + +static struct wd_ops p4_wd_ops = { + .reserve = p4_reserve, + .unreserve = p4_unreserve, + .setup = setup_p4_watchdog, + .rearm = p4_rearm, + .stop = stop_p4_watchdog, + /* RED-PEN this is wrong for the other sibling */ + .perfctr = MSR_P4_BPU_PERFCTR0, + .evntsel = MSR_P4_BSU_ESCR0, + .checkbit = 1ULL<<39, +}; + +/* Watchdog using the Intel architected PerfMon. Used for Core2 and hopefully + all future Intel CPUs. */ + +#define ARCH_PERFMON_NMI_EVENT_SEL ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL +#define ARCH_PERFMON_NMI_EVENT_UMASK ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK + +static int setup_intel_arch_watchdog(unsigned nmi_hz) +{ + unsigned int ebx; + union cpuid10_eax eax; + unsigned int unused; + unsigned int perfctr_msr, evntsel_msr; + unsigned int evntsel; + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + + /* + * Check whether the Architectural PerfMon supports + * Unhalted Core Cycles Event or not. + * NOTE: Corresponding bit = 0 in ebx indicates event present. + */ + cpuid(10, &(eax.full), &ebx, &unused, &unused); + if ((eax.split.mask_length < (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX+1)) || + (ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT)) + return 0; + + perfctr_msr = MSR_ARCH_PERFMON_PERFCTR1; + evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL1; + + wrmsrl(perfctr_msr, 0UL); + + evntsel = ARCH_PERFMON_EVENTSEL_INT + | ARCH_PERFMON_EVENTSEL_OS + | ARCH_PERFMON_EVENTSEL_USR + | ARCH_PERFMON_NMI_EVENT_SEL + | ARCH_PERFMON_NMI_EVENT_UMASK; + + /* setup the timer */ + wrmsr(evntsel_msr, evntsel, 0); + nmi_hz = adjust_for_32bit_ctr(nmi_hz); + write_watchdog_counter32(perfctr_msr, "INTEL_ARCH_PERFCTR0", nmi_hz); + apic_write(APIC_LVTPC, APIC_DM_NMI); + evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsr(evntsel_msr, evntsel, 0); + + wd->perfctr_msr = perfctr_msr; + wd->evntsel_msr = evntsel_msr; + wd->cccr_msr = 0; //unused + wd_ops->checkbit = 1ULL << (eax.split.bit_width - 1); + return 1; +} + +static struct wd_ops intel_arch_wd_ops = { + .reserve = single_msr_reserve, + .unreserve = single_msr_unreserve, + .setup = setup_intel_arch_watchdog, + .rearm = p6_rearm, + .stop = single_msr_stop_watchdog, + .perfctr = MSR_ARCH_PERFMON_PERFCTR0, + .evntsel = MSR_ARCH_PERFMON_EVENTSEL0, +}; + +static void probe_nmi_watchdog(void) +{ + switch (boot_cpu_data.x86_vendor) { + case X86_VENDOR_AMD: + if (boot_cpu_data.x86 != 6 && boot_cpu_data.x86 != 15 && + boot_cpu_data.x86 != 16) + return; + wd_ops = &k7_wd_ops; + break; + case X86_VENDOR_INTEL: + if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { + wd_ops = &intel_arch_wd_ops; + break; + } + switch (boot_cpu_data.x86) { + case 6: + if (boot_cpu_data.x86_model > 0xd) + return; + + wd_ops = &p6_wd_ops; + break; + case 15: + if (boot_cpu_data.x86_model > 0x4) + return; + + wd_ops = &p4_wd_ops; + break; + default: + return; + } + break; + } +} + +/* Interface to nmi.c */ + +int lapic_watchdog_init(unsigned nmi_hz) +{ + if (!wd_ops) { + probe_nmi_watchdog(); + if (!wd_ops) + return -1; + } + + if (!(wd_ops->setup(nmi_hz))) { + printk(KERN_ERR "Cannot setup NMI watchdog on CPU %d\n", + raw_smp_processor_id()); + return -1; + } + + return 0; +} + +void lapic_watchdog_stop(void) +{ + if (wd_ops) + wd_ops->stop(NULL); +} + +unsigned lapic_adjust_nmi_hz(unsigned hz) +{ + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + if (wd->perfctr_msr == MSR_P6_PERFCTR0 || + wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR1) + hz = adjust_for_32bit_ctr(hz); + return hz; +} + +int lapic_wd_event(unsigned nmi_hz) +{ + struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); + u64 ctr; + rdmsrl(wd->perfctr_msr, ctr); + if (ctr & wd_ops->checkbit) { /* perfctr still running? */ + return 0; + } + wd_ops->rearm(wd, nmi_hz); + return 1; +} + +int lapic_watchdog_ok(void) +{ + return wd_ops != NULL; +} |