#include <linux/percpu.h>
#include <linux/dmi.h>
#include <linux/kprobes.h>
+#include <linux/cpumask.h>
+#include <linux/kernel_stat.h>
#include <asm/smp.h>
#include <asm/nmi.h>
static DEFINE_PER_CPU(unsigned long, perfctr_nmi_owner);
static DEFINE_PER_CPU(unsigned long, evntsel_nmi_owner[3]);
+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)
*/
{
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_AMD:
- return ((boot_cpu_data.x86 == 15) || (boot_cpu_data.x86 == 6));
+ return ((boot_cpu_data.x86 == 15) || (boot_cpu_data.x86 == 6)
+ || (boot_cpu_data.x86 == 16));
case X86_VENDOR_INTEL:
if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
return 1;
return 0;
}
+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)
+{
+ 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 int __init check_nmi_watchdog(void)
{
- volatile int endflag = 0;
unsigned int *prev_nmi_count;
int cpu;
/* Enable NMI watchdog for newer systems.
- Actually it should be safe for most systems before 2004 too except
- for some IBM systems that corrupt registers when NMI happens
- during SMM. Unfortunately we don't have more exact information
- on these and use this coarse check. */
- if (nmi_watchdog == NMI_DEFAULT && dmi_get_year(DMI_BIOS_DATE) >= 2004)
+ Probably safe on most older systems too, but let's be careful.
+ IBM ThinkPads use INT10 inside SMM and that allows early NMI inside SMM
+ which hangs the system. Disable watchdog for all thinkpads */
+ if (nmi_watchdog == NMI_DEFAULT && dmi_get_year(DMI_BIOS_DATE) >= 2004 &&
+ !dmi_name_in_vendors("ThinkPad"))
nmi_watchdog = NMI_LOCAL_APIC;
if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DEFAULT))
struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
nmi_hz = 1;
- /*
- * 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 (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0 &&
- ((u64)cpu_khz * 1000) > 0x7fffffffULL) {
- u64 count = (u64)cpu_khz * 1000;
- do_div(count, 0x7fffffffUL);
- nmi_hz = count + 1;
+
+ if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
+ wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+ nmi_hz = adjust_for_32bit_ctr(nmi_hz);
}
}
if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE))
return 0;
- /*
- * If any other x86 CPU has a local APIC, then
- * please test the NMI stuff there and send me the
- * missing bits. Right now Intel P6/P4 and AMD K7 only.
- */
- if ((nmi == NMI_LOCAL_APIC) && (nmi_known_cpu() == 0))
- return 0; /* no lapic support */
+
nmi_watchdog = nmi;
return 1;
}
}
}
+static void __acpi_nmi_disable(void *__unused)
+{
+ apic_write_around(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED);
+}
+
+/*
+ * Disable timer based NMIs on all CPUs:
+ */
+void acpi_nmi_disable(void)
+{
+ if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
+ on_each_cpu(__acpi_nmi_disable, NULL, 0, 1);
+}
+
+static void __acpi_nmi_enable(void *__unused)
+{
+ apic_write_around(APIC_LVT0, APIC_DM_NMI);
+}
+
+/*
+ * Enable timer based NMIs on all CPUs:
+ */
+void acpi_nmi_enable(void)
+{
+ if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
+ on_each_cpu(__acpi_nmi_enable, NULL, 0, 1);
+}
+
#ifdef CONFIG_PM
static int nmi_pm_active; /* nmi_active before suspend */
wrmsrl(perfctr_msr, 0 - count);
}
+static void write_watchdog_counter32(unsigned int perfctr_msr,
+ const char *descr)
+{
+ 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);
+}
+
/* Note that these events don't tick when the CPU idles. This means
the frequency varies with CPU load. */
/* setup the timer */
wrmsr(evntsel_msr, evntsel, 0);
- write_watchdog_counter(perfctr_msr, "P6_PERFCTR0");
+ nmi_hz = adjust_for_32bit_ctr(nmi_hz);
+ write_watchdog_counter32(perfctr_msr, "P6_PERFCTR0");
apic_write(APIC_LVTPC, APIC_DM_NMI);
evntsel |= P6_EVNTSEL0_ENABLE;
wrmsr(evntsel_msr, evntsel, 0);
/* setup the timer */
wrmsr(evntsel_msr, evntsel, 0);
- write_watchdog_counter(perfctr_msr, "INTEL_ARCH_PERFCTR0");
+ nmi_hz = adjust_for_32bit_ctr(nmi_hz);
+ write_watchdog_counter32(perfctr_msr, "INTEL_ARCH_PERFCTR0");
apic_write(APIC_LVTPC, APIC_DM_NMI);
evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
wrmsr(evntsel_msr, evntsel, 0);
if (nmi_watchdog == NMI_LOCAL_APIC) {
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_AMD:
- if (boot_cpu_data.x86 != 6 && boot_cpu_data.x86 != 15)
+ if (boot_cpu_data.x86 != 6 && boot_cpu_data.x86 != 15 &&
+ boot_cpu_data.x86 != 16)
return;
if (!setup_k7_watchdog())
return;
void touch_nmi_watchdog (void)
{
- int i;
+ if (nmi_watchdog > 0) {
+ unsigned cpu;
- /*
- * Just reset the alert counters, (other CPUs might be
- * spinning on locks we hold):
- */
- for_each_possible_cpu(i)
- alert_counter[i] = 0;
+ /*
+ * Just reset the alert counters, (other CPUs might be
+ * spinning on locks we hold):
+ */
+ for_each_present_cpu (cpu)
+ alert_counter[cpu] = 0;
+ }
/*
* Tickle the softlockup detector too:
touched = 1;
}
- sum = per_cpu(irq_stat, cpu).apic_timer_irqs;
+ 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);
+ }
+
+ /*
+ * Take the local apic timer and PIT/HPET into account. We don't
+ * know which one is active, when we have highres/dyntick on
+ */
+ sum = per_cpu(irq_stat, cpu).apic_timer_irqs + kstat_irqs(0);
- /* if the apic timer isn't firing, this cpu isn't doing much */
+ /* if the none of the timers isn't firing, this cpu isn't doing much */
if (!touched && last_irq_sums[cpu] == sum) {
/*
* Ayiee, looks like this CPU is stuck ...
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);
}
else if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
* 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);
+ } else {
+ /* start the cycle over again */
+ write_watchdog_counter(wd->perfctr_msr, NULL);
}
- /* start the cycle over again */
- write_watchdog_counter(wd->perfctr_msr, NULL);
rc = 1;
} else if (nmi_watchdog == NMI_IO_APIC) {
/* don't know how to accurately check for this.
#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);
git.openpandora.org / pandora-kernel.git / blobdiff