[patch] ratelimit performance counter interrupts

[Mike Galbraith <efault@gmx.de>]

While fiddling around with kerneltop, I found that an unwary user can
lock his box up quite thoroughly if he does something really clever like
starting kerneltop with -c 100. 

Ratelimit performance counter interrupts to 100KHz per CPU.

Signed-off-by: Mike Galbraith <efault@gmx.de>

---
 arch/x86/kernel/apic.c             |    2 +
 arch/x86/kernel/cpu/perf_counter.c |   43 ++++++++++++++++++++++++++++++-------
 include/linux/perf_counter.h       |    2 +
 3 files changed, 40 insertions(+), 7 deletions(-)

Index: linux-2.6/arch/x86/kernel/apic.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/apic.c
+++ linux-2.6/arch/x86/kernel/apic.c
@@ -781,6 +781,8 @@ static void local_apic_timer_interrupt(v
 	inc_irq_stat(apic_timer_irqs);
 
 	evt->event_handler(evt);
+
+	perf_counter_unthrottle();
 }
 
 /*
Index: linux-2.6/arch/x86/kernel/cpu/perf_counter.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_counter.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_counter.c
@@ -33,6 +33,8 @@ static int nr_counters_fixed __read_most
 struct cpu_hw_counters {
 	struct perf_counter	*counters[X86_PMC_IDX_MAX];
 	unsigned long		used[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
+	unsigned long		interrupts;
+	u64			global_enable;
 };
 
 /*
@@ -468,23 +470,26 @@ perf_handle_group(struct perf_counter *s
 }
 
 /*
+ * Mamimum 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:
  */
 static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
 {
 	int bit, cpu = smp_processor_id();
-	u64 ack, status, saved_global;
-	struct cpu_hw_counters *cpuc;
+	u64 ack, status;
+	struct cpu_hw_counters *cpuc = &per_cpu(cpu_hw_counters, cpu);
 
-	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, saved_global);
+	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable);
 
 	/* Disable counters globally */
 	wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0);
 	ack_APIC_irq();
 
-	cpuc = &per_cpu(cpu_hw_counters, cpu);
-
 	rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
 	if (!status)
 		goto out;
@@ -533,9 +538,33 @@ again:
 		goto again;
 out:
 	/*
-	 * Restore - do not reenable when global enable is off:
+	 * Restore - do not reenable when global enable is off or throttled:
 	 */
-	wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, saved_global);
+	if (++cpuc->interrupts < PERFMON_MAX_INTERRUPTS)
+		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable);
+}
+
+void perf_counter_unthrottle(void)
+{
+	struct cpu_hw_counters *cpuc;
+	u64 global_enable;
+
+	if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+		return;
+
+	if (unlikely(!perf_counters_initialized))
+		return;
+
+	cpuc = &per_cpu(cpu_hw_counters, smp_processor_id());
+	if (cpuc->interrupts >= PERFMON_MAX_INTERRUPTS) {
+		if (printk_ratelimit())
+			printk(KERN_WARNING "PERFMON: max interrupts exceeded!\n");
+		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable);
+	}
+	rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, global_enable);
+	if (unlikely(cpuc->global_enable && !global_enable))
+		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable);
+	cpuc->interrupts = 0;
 }
 
 void smp_perf_counter_interrupt(struct pt_regs *regs)
Index: linux-2.6/include/linux/perf_counter.h
===================================================================
--- linux-2.6.orig/include/linux/perf_counter.h
+++ linux-2.6/include/linux/perf_counter.h
@@ -254,6 +254,7 @@ extern void perf_counter_init_task(struc
 extern void perf_counter_exit_task(struct task_struct *child);
 extern void perf_counter_notify(struct pt_regs *regs);
 extern void perf_counter_print_debug(void);
+extern void perf_counter_unthrottle(void);
 extern u64 hw_perf_save_disable(void);
 extern void hw_perf_restore(u64 ctrl);
 extern int perf_counter_task_disable(void);
@@ -281,6 +282,7 @@ static inline void perf_counter_init_tas
 static inline void perf_counter_exit_task(struct task_struct *child)	{ }
 static inline void perf_counter_notify(struct pt_regs *regs)		{ }
 static inline void perf_counter_print_debug(void)			{ }
+static inline void perf_counter_unthrottle(void)			{ }
 static inline void hw_perf_restore(u64 ctrl)			{ }
 static inline u64 hw_perf_save_disable(void)		      { return 0; }
 static inline int perf_counter_task_disable(void)	{ return -EINVAL; }