On Fri, 29 Sep 2006 21:54:30 +0200
Andi Kleen <[EMAIL PROTECTED]> wrote:

> On Friday 29 September 2006 21:18, Ingo Molnar wrote:
> > 
> > * Andrew Morton <[EMAIL PROTECTED]> wrote:
> > 
> > > gad, there have been so many all-CPU-backtrace patches over the years.
> > > 
> > > <optimistically cc's Ingo>
> > > 
> > > Ingo, do you think that's something which we shuld have in the 
> > > spinlock debugging code?  A trace to let us see which CPU is holding 
> > > that lock, and where from?  I guess if the other cpu is stuck in 
> > > spin_lock_irqsave() then we'll get stuck delivering the IPI, so it'd 
> > > need to be async.
> > 
> > used to have this in -rt for i686 and x86_64 for the NMI watchdog tick 
> > to print on all CPUs, in the next tick (i.e. no need to actually 
> > initiate an IPI) - but it was all a bit hacky [but worked]. It fell 
> > victim to some recent flux in that area.
> 
> You mean spinlock debugging setting a global variable and the NMI
> watchdog testing that?  Makes sense. I can put it on my todo list.

It does make sense.

<type, type>

Something like this?  (compiled only)

From: Andrew Morton <[EMAIL PROTECTED]>

When a spinlock lockup occurs, arrange for the NMI code to emit an all-cpu
backtrace, so we get to see which CPU is holding the lock, and where.

Cc: Andi Kleen <[EMAIL PROTECTED]>
Cc: Ingo Molnar <[EMAIL PROTECTED]>
Cc: Badari Pulavarty <[EMAIL PROTECTED]>
Signed-off-by: Andrew Morton <[EMAIL PROTECTED]>
---

 arch/i386/kernel/nmi.c   |   14 ++++++++++++++
 arch/x86_64/kernel/nmi.c |   17 ++++++++++++++++-
 include/asm-i386/nmi.h   |    3 +++
 include/asm-x86_64/nmi.h |    3 +++
 include/linux/nmi.h      |    5 +++++
 lib/spinlock_debug.c     |    4 ++++
 6 files changed, 45 insertions(+), 1 deletion(-)

diff -puN lib/spinlock_debug.c~spinlock-debug-all-cpu-backtrace 
lib/spinlock_debug.c
--- a/lib/spinlock_debug.c~spinlock-debug-all-cpu-backtrace
+++ a/lib/spinlock_debug.c
@@ -7,6 +7,7 @@
  */
 
 #include <linux/spinlock.h>
+#include <linux/nmi.h>
 #include <linux/interrupt.h>
 #include <linux/debug_locks.h>
 #include <linux/delay.h>
@@ -116,6 +117,9 @@ static void __spin_lock_debug(spinlock_t
                                raw_smp_processor_id(), current->comm,
                                current->pid, lock);
                        dump_stack();
+#ifdef CONFIG_SMP
+                       trigger_all_cpu_backtrace();
+#endif
                }
        }
 }
diff -puN arch/i386/kernel/nmi.c~spinlock-debug-all-cpu-backtrace 
arch/i386/kernel/nmi.c
--- a/arch/i386/kernel/nmi.c~spinlock-debug-all-cpu-backtrace
+++ a/arch/i386/kernel/nmi.c
@@ -23,6 +23,7 @@
 #include <linux/percpu.h>
 #include <linux/dmi.h>
 #include <linux/kprobes.h>
+#include <linux/cpumask.h>
 
 #include <asm/smp.h>
 #include <asm/nmi.h>
@@ -40,6 +41,8 @@
 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)
  */
@@ -905,6 +908,12 @@ __kprobes int nmi_watchdog_tick(struct p
                touched = 1;
        }
 
+       if (cpu_isset(cpu, backtrace_mask)) {
+               cpu_clear(cpu, backtrace_mask);
+               printk("NMI backtrace for cpu %d\n", cpu);
+               dump_stack();
+       }
+
        sum = per_cpu(irq_stat, cpu).apic_timer_irqs;
 
        /* if the apic timer isn't firing, this cpu isn't doing much */
@@ -1031,6 +1040,11 @@ int proc_nmi_enabled(struct ctl_table *t
 
 #endif
 
+void __trigger_all_cpu_backtrace(void)
+{
+       backtrace_mask = CPU_MASK_ALL;
+}
+
 EXPORT_SYMBOL(nmi_active);
 EXPORT_SYMBOL(nmi_watchdog);
 EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi);
diff -puN arch/x86_64/kernel/nmi.c~spinlock-debug-all-cpu-backtrace 
arch/x86_64/kernel/nmi.c
--- a/arch/x86_64/kernel/nmi.c~spinlock-debug-all-cpu-backtrace
+++ a/arch/x86_64/kernel/nmi.c
@@ -12,14 +12,15 @@
  *  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>
@@ -37,6 +38,8 @@
 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)
  */
@@ -778,6 +781,7 @@ int __kprobes nmi_watchdog_tick(struct p
 {
        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;
@@ -795,6 +799,12 @@ int __kprobes nmi_watchdog_tick(struct p
                touched = 1;
        }
 
+       if (cpu_isset(cpu, backtrace_mask)) {
+               cpu_clear(cpu, backtrace_mask);
+               printk("NMI backtrace for cpu %d\n", cpu);
+               dump_stack();
+       }
+
 #ifdef CONFIG_X86_MCE
        /* Could check oops_in_progress here too, but it's safer
           not too */
@@ -927,6 +937,11 @@ int proc_nmi_enabled(struct ctl_table *t
 
 #endif
 
+void __trigger_all_cpu_backtrace(void)
+{
+       backtrace_mask = CPU_MASK_ALL;
+}
+
 EXPORT_SYMBOL(nmi_active);
 EXPORT_SYMBOL(nmi_watchdog);
 EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi);
diff -puN include/linux/nmi.h~spinlock-debug-all-cpu-backtrace 
include/linux/nmi.h
--- a/include/linux/nmi.h~spinlock-debug-all-cpu-backtrace
+++ a/include/linux/nmi.h
@@ -14,9 +14,14 @@
  * disables interrupts for a long time. This call is stateless.
  */
 #ifdef ARCH_HAS_NMI_WATCHDOG
+#include <asm/nmi.h>
 extern void touch_nmi_watchdog(void);
 #else
 # define touch_nmi_watchdog() do { } while(0)
 #endif
 
+#ifndef trigger_all_cpu_backtrace
+#define trigger_all_cpu_backtrace() do { } while (0)
+#endif
+
 #endif
diff -puN include/asm-i386/nmi.h~spinlock-debug-all-cpu-backtrace 
include/asm-i386/nmi.h
--- a/include/asm-i386/nmi.h~spinlock-debug-all-cpu-backtrace
+++ a/include/asm-i386/nmi.h
@@ -36,4 +36,7 @@ extern unsigned int nmi_watchdog;
 #define NMI_LOCAL_APIC 2
 #define NMI_INVALID    3
 
+void __trigger_all_cpu_backtrace(void);
+#define trigger_all_cpu_backtrace() __trigger_all_cpu_backtrace()
+
 #endif /* ASM_NMI_H */
diff -puN include/asm-x86_64/nmi.h~spinlock-debug-all-cpu-backtrace 
include/asm-x86_64/nmi.h
--- a/include/asm-x86_64/nmi.h~spinlock-debug-all-cpu-backtrace
+++ a/include/asm-x86_64/nmi.h
@@ -70,4 +70,7 @@ extern unsigned int nmi_watchdog;
 #define NMI_LOCAL_APIC 2
 #define NMI_INVALID    3
 
+void __trigger_all_cpu_backtrace(void);
+#define trigger_all_cpu_backtrace() __trigger_all_cpu_backtrace()
+
 #endif /* ASM_NMI_H */
_


-
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to