* Mikael Pettersson : PM converted to driver model. Disable/enable API.
*/
+#include <linux/nmi.h>
#include <linux/mm.h>
#include <linux/delay.h>
#include <linux/interrupt.h>
#include <linux/module.h>
#include <linux/sysdev.h>
-#include <linux/nmi.h>
#include <linux/sysctl.h>
#include <linux/kprobes.h>
+#include <linux/cpumask.h>
#include <asm/smp.h>
#include <asm/nmi.h>
#include <asm/proto.h>
#include <asm/kdebug.h>
#include <asm/mce.h>
+#include <asm/intel_arch_perfmon.h>
+
+int unknown_nmi_panic;
+int nmi_watchdog_enabled;
+int panic_on_unrecovered_nmi;
/* perfctr_nmi_owner tracks the ownership of the perfctr registers:
* evtsel_nmi_owner tracks the ownership of the event selection
static DEFINE_PER_CPU(unsigned, perfctr_nmi_owner);
static DEFINE_PER_CPU(unsigned, evntsel_nmi_owner[2]);
+static cpumask_t backtrace_mask = CPU_MASK_NONE;
+
/* 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)
*/
static DEFINE_PER_CPU(struct nmi_watchdog_ctlblk, nmi_watchdog_ctlblk);
/* local prototypes */
-static void stop_apic_nmi_watchdog(void *unused);
static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu);
/* converts an msr to an appropriate reservation bit */
case X86_VENDOR_AMD:
return (msr - MSR_K7_PERFCTR0);
case X86_VENDOR_INTEL:
- return (msr - MSR_P4_BPU_PERFCTR0);
+ if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+ return (msr - MSR_ARCH_PERFMON_PERFCTR0);
+ else
+ return (msr - MSR_P4_BPU_PERFCTR0);
}
return 0;
}
case X86_VENDOR_AMD:
return (msr - MSR_K7_EVNTSEL0);
case X86_VENDOR_INTEL:
- return (msr - MSR_P4_BSU_ESCR0);
+ if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+ return (msr - MSR_ARCH_PERFMON_EVENTSEL0);
+ else
+ return (msr - MSR_P4_BSU_ESCR0);
}
return 0;
}
case X86_VENDOR_AMD:
return boot_cpu_data.x86 == 15;
case X86_VENDOR_INTEL:
- return boot_cpu_data.x86 == 15;
+ if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+ return 1;
+ else
+ return (boot_cpu_data.x86 == 15);
}
return 0;
}
/* Run after command line and cpu_init init, but before all other checks */
-void __cpuinit nmi_watchdog_default(void)
+void nmi_watchdog_default(void)
{
if (nmi_watchdog != NMI_DEFAULT)
return;
nmi_watchdog = NMI_IO_APIC;
}
+static int endflag __initdata = 0;
+
#ifdef CONFIG_SMP
/* The performance counters used by NMI_LOCAL_APIC don't trigger when
* the CPU is idle. To make sure the NMI watchdog really ticks on all
*/
static __init void nmi_cpu_busy(void *data)
{
- volatile int *endflag = data;
local_irq_enable_in_hardirq();
/* Intentionally don't use cpu_relax here. This is
to make sure that the performance counter really ticks,
pause instruction. On a real HT machine this is fine because
all other CPUs are busy with "useless" delay loops and don't
care if they get somewhat less cycles. */
- while (*endflag == 0)
- barrier();
+ while (endflag == 0)
+ mb();
}
#endif
+static unsigned int adjust_for_32bit_ctr(unsigned int hz)
+{
+ unsigned int retval = hz;
+
+ /*
+ * On Intel CPUs with 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
+ */
+ if ((((u64)cpu_khz * 1000) / retval) > 0x7fffffffULL) {
+ retval = ((u64)cpu_khz * 1000) / 0x7fffffffUL + 1;
+ }
+ return retval;
+}
+
int __init check_nmi_watchdog (void)
{
- volatile int endflag = 0;
int *counts;
int cpu;
if (!atomic_read(&nmi_active)) {
kfree(counts);
atomic_set(&nmi_active, -1);
+ endflag = 1;
return -1;
}
endflag = 1;
/* now that we know it works we can reduce NMI frequency to
something more reasonable; makes a difference in some configs */
- if (nmi_watchdog == NMI_LOCAL_APIC)
+ if (nmi_watchdog == NMI_LOCAL_APIC) {
+ struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
nmi_hz = 1;
+ if (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0)
+ nmi_hz = adjust_for_32bit_ctr(nmi_hz);
+ }
kfree(counts);
return 0;
if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE))
return 0;
- if ((nmi == NMI_LOCAL_APIC) && (nmi_known_cpu() == 0))
- return 0; /* no lapic support */
nmi_watchdog = nmi;
return 1;
}
static int lapic_nmi_suspend(struct sys_device *dev, pm_message_t state)
{
+ /* only CPU0 goes here, other CPUs should be offline */
nmi_pm_active = atomic_read(&nmi_active);
- disable_lapic_nmi_watchdog();
+ stop_apic_nmi_watchdog(NULL);
+ BUG_ON(atomic_read(&nmi_active) != 0);
return 0;
}
static int lapic_nmi_resume(struct sys_device *dev)
{
- if (nmi_pm_active > 0)
- enable_lapic_nmi_watchdog();
+ /* only CPU0 goes here, other CPUs should be offline */
+ if (nmi_pm_active > 0) {
+ setup_apic_nmi_watchdog(NULL);
+ touch_nmi_watchdog();
+ }
return 0;
}
release_perfctr_nmi(wd->perfctr_msr);
}
+#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(void)
+{
+ 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))
+ goto fail;
+
+ perfctr_msr = MSR_ARCH_PERFMON_PERFCTR0;
+ evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL0;
+
+ if (!reserve_perfctr_nmi(perfctr_msr))
+ goto fail;
+
+ if (!reserve_evntsel_nmi(evntsel_msr))
+ goto fail1;
+
+ 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);
+ wrmsr(perfctr_msr, (u32)(-((u64)cpu_khz * 1000 / nmi_hz)), 0);
+
+ 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->check_bit = 1ULL << (eax.split.bit_width - 1);
+ return 1;
+fail1:
+ release_perfctr_nmi(perfctr_msr);
+fail:
+ return 0;
+}
+
+static void stop_intel_arch_watchdog(void)
+{
+ unsigned int ebx;
+ union cpuid10_eax eax;
+ unsigned int unused;
+ 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;
+
+ wrmsr(wd->evntsel_msr, 0, 0);
+
+ release_evntsel_nmi(wd->evntsel_msr);
+ release_perfctr_nmi(wd->perfctr_msr);
+}
+
void setup_apic_nmi_watchdog(void *unused)
{
+ struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
/* only support LOCAL and IO APICs for now */
if ((nmi_watchdog != NMI_LOCAL_APIC) &&
(nmi_watchdog != NMI_IO_APIC))
return;
+ if (wd->enabled == 1)
+ return;
+
+ /* cheap hack to support suspend/resume */
+ /* if cpu0 is not active neither should the other cpus */
+ if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0))
+ return;
+
if (nmi_watchdog == NMI_LOCAL_APIC) {
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_AMD:
return;
break;
case X86_VENDOR_INTEL:
+ if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+ if (!setup_intel_arch_watchdog())
+ return;
+ break;
+ }
if (!setup_p4_watchdog())
return;
break;
return;
}
}
- __get_cpu_var(nmi_watchdog_ctlblk.enabled) = 1;
+ wd->enabled = 1;
atomic_inc(&nmi_active);
}
-static void stop_apic_nmi_watchdog(void *unused)
+void stop_apic_nmi_watchdog(void *unused)
{
+ struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
/* only support LOCAL and IO APICs for now */
if ((nmi_watchdog != NMI_LOCAL_APIC) &&
(nmi_watchdog != NMI_IO_APIC))
return;
+ if (wd->enabled == 0)
+ return;
+
if (nmi_watchdog == NMI_LOCAL_APIC) {
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_AMD:
stop_k7_watchdog();
break;
case X86_VENDOR_INTEL:
+ if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+ stop_intel_arch_watchdog();
+ break;
+ }
stop_p4_watchdog();
break;
default:
return;
}
}
- __get_cpu_var(nmi_watchdog_ctlblk.enabled) = 0;
+ wd->enabled = 0;
atomic_dec(&nmi_active);
}
{
int sum;
int touched = 0;
+ int cpu = smp_processor_id();
struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
u64 dummy;
int rc=0;
touched = 1;
}
+ if (cpu_isset(cpu, backtrace_mask)) {
+ static DEFINE_SPINLOCK(lock); /* Serialise the printks */
+
+ spin_lock(&lock);
+ printk("NMI backtrace for cpu %d\n", cpu);
+ dump_stack();
+ spin_unlock(&lock);
+ cpu_clear(cpu, backtrace_mask);
+ }
+
#ifdef CONFIG_X86_MCE
/* Could check oops_in_progress here too, but it's safer
not too */
*/
local_inc(&__get_cpu_var(alert_counter));
if (local_read(&__get_cpu_var(alert_counter)) == 5*nmi_hz)
- die_nmi("NMI Watchdog detected LOCKUP on CPU %d\n", regs);
+ die_nmi("NMI Watchdog detected LOCKUP on CPU %d\n", regs,
+ panic_on_timeout);
} else {
__get_cpu_var(last_irq_sum) = sum;
local_set(&__get_cpu_var(alert_counter), 0);
dummy &= ~P4_CCCR_OVF;
wrmsrl(wd->cccr_msr, dummy);
apic_write(APIC_LVTPC, APIC_DM_NMI);
- }
- /* start the cycle over again */
- wrmsrl(wd->perfctr_msr, -((u64)cpu_khz * 1000 / nmi_hz));
+ /* start the cycle over again */
+ wrmsrl(wd->perfctr_msr,
+ -((u64)cpu_khz * 1000 / nmi_hz));
+ } else if (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+ /*
+ * ArchPerfom/Core Duo needs to re-unmask
+ * the apic vector
+ */
+ apic_write(APIC_LVTPC, APIC_DM_NMI);
+ /* ARCH_PERFMON has 32 bit counter writes */
+ wrmsr(wd->perfctr_msr,
+ (u32)(-((u64)cpu_khz * 1000 / nmi_hz)), 0);
+ } else {
+ /* start the cycle over again */
+ wrmsrl(wd->perfctr_msr,
+ -((u64)cpu_khz * 1000 / nmi_hz));
+ }
rc = 1;
} else if (nmi_watchdog == NMI_IO_APIC) {
/* don't know how to accurately check for this.
char buf[64];
sprintf(buf, "NMI received for unknown reason %02x\n", reason);
- die_nmi(buf,regs);
+ die_nmi(buf, regs, 1); /* Always panic here */
+ return 0;
+}
+
+/*
+ * proc handler for /proc/sys/kernel/nmi
+ */
+int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file,
+ void __user *buffer, size_t *length, loff_t *ppos)
+{
+ int old_state;
+
+ nmi_watchdog_enabled = (atomic_read(&nmi_active) > 0) ? 1 : 0;
+ old_state = nmi_watchdog_enabled;
+ proc_dointvec(table, write, file, buffer, length, ppos);
+ if (!!old_state == !!nmi_watchdog_enabled)
+ return 0;
+
+ if (atomic_read(&nmi_active) < 0) {
+ printk( KERN_WARNING "NMI watchdog is permanently disabled\n");
+ return -EIO;
+ }
+
+ /* if nmi_watchdog is not set yet, then set it */
+ nmi_watchdog_default();
+
+ if (nmi_watchdog == NMI_LOCAL_APIC) {
+ if (nmi_watchdog_enabled)
+ enable_lapic_nmi_watchdog();
+ else
+ disable_lapic_nmi_watchdog();
+ } else {
+ printk( KERN_WARNING
+ "NMI watchdog doesn't know what hardware to touch\n");
+ return -EIO;
+ }
return 0;
}
#endif
+void __trigger_all_cpu_backtrace(void)
+{
+ int i;
+
+ backtrace_mask = cpu_online_map;
+ /* Wait for up to 10 seconds for all CPUs to do the backtrace */
+ for (i = 0; i < 10 * 1000; i++) {
+ if (cpus_empty(backtrace_mask))
+ break;
+ mdelay(1);
+ }
+}
+
EXPORT_SYMBOL(nmi_active);
EXPORT_SYMBOL(nmi_watchdog);
EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi);