nmi_backtrace: add more trigger_*_cpu_backtrace() methods
authorChris Metcalf <cmetcalf@mellanox.com>
Sat, 8 Oct 2016 00:02:45 +0000 (17:02 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 8 Oct 2016 01:46:30 +0000 (18:46 -0700)
Patch series "improvements to the nmi_backtrace code" v9.

This patch series modifies the trigger_xxx_backtrace() NMI-based remote
backtracing code to make it more flexible, and makes a few small
improvements along the way.

The motivation comes from the task isolation code, where there are
scenarios where we want to be able to diagnose a case where some cpu is
about to interrupt a task-isolated cpu.  It can be helpful to see both
where the interrupting cpu is, and also an approximation of where the
cpu that is being interrupted is.  The nmi_backtrace framework allows us
to discover the stack of the interrupted cpu.

I've tested that the change works as desired on tile, and build-tested
x86, arm, mips, and sparc64.  For x86 I confirmed that the generic
cpuidle stuff as well as the architecture-specific routines are in the
new cpuidle section.  For arm, mips, and sparc I just build-tested it
and made sure the generic cpuidle routines were in the new cpuidle
section, but I didn't attempt to figure out which the platform-specific
idle routines might be.  That might be more usefully done by someone
with platform experience in follow-up patches.

This patch (of 4):

Currently you can only request a backtrace of either all cpus, or all
cpus but yourself.  It can also be helpful to request a remote backtrace
of a single cpu, and since we want that, the logical extension is to
support a cpumask as the underlying primitive.

This change modifies the existing lib/nmi_backtrace.c code to take a
cpumask as its basic primitive, and modifies the linux/nmi.h code to use
the new "cpumask" method instead.

The existing clients of nmi_backtrace (arm and x86) are converted to
using the new cpumask approach in this change.

The other users of the backtracing API (sparc64 and mips) are converted
to use the cpumask approach rather than the all/allbutself approach.
The mips code ignored the "include_self" boolean but with this change it
will now also dump a local backtrace if requested.

Link: http://lkml.kernel.org/r/1472487169-14923-2-git-send-email-cmetcalf@mellanox.com
Signed-off-by: Chris Metcalf <cmetcalf@mellanox.com>
Tested-by: Daniel Thompson <daniel.thompson@linaro.org> [arm]
Reviewed-by: Aaron Tomlin <atomlin@redhat.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/arm/include/asm/irq.h
arch/arm/kernel/smp.c
arch/mips/include/asm/irq.h
arch/mips/kernel/process.c
arch/sparc/include/asm/irq_64.h
arch/sparc/kernel/process_64.c
arch/x86/include/asm/irq.h
arch/x86/kernel/apic/hw_nmi.c
include/linux/nmi.h
lib/nmi_backtrace.c

index 1bd9510..e53638c 100644 (file)
@@ -36,8 +36,9 @@ extern void set_handle_irq(void (*handle_irq)(struct pt_regs *));
 #endif
 
 #ifdef CONFIG_SMP
-extern void arch_trigger_all_cpu_backtrace(bool);
-#define arch_trigger_all_cpu_backtrace(x) arch_trigger_all_cpu_backtrace(x)
+extern void arch_trigger_cpumask_backtrace(const cpumask_t *mask,
+                                          bool exclude_self);
+#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
 #endif
 
 static inline int nr_legacy_irqs(void)
index 937c892..5abc569 100644 (file)
@@ -760,7 +760,7 @@ static void raise_nmi(cpumask_t *mask)
        smp_cross_call(mask, IPI_CPU_BACKTRACE);
 }
 
-void arch_trigger_all_cpu_backtrace(bool include_self)
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
 {
-       nmi_trigger_all_cpu_backtrace(include_self, raise_nmi);
+       nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_nmi);
 }
index 15e0fec..6bf10e7 100644 (file)
@@ -51,7 +51,8 @@ extern int cp0_fdc_irq;
 
 extern int get_c0_fdc_int(void);
 
-void arch_trigger_all_cpu_backtrace(bool);
-#define arch_trigger_all_cpu_backtrace arch_trigger_all_cpu_backtrace
+void arch_trigger_cpumask_backtrace(const struct cpumask *mask,
+                                   bool exclude_self);
+#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
 
 #endif /* _ASM_IRQ_H */
index d2d0615..9514e5f 100644 (file)
@@ -569,9 +569,16 @@ static void arch_dump_stack(void *info)
        dump_stack();
 }
 
-void arch_trigger_all_cpu_backtrace(bool include_self)
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
 {
-       smp_call_function(arch_dump_stack, NULL, 1);
+       long this_cpu = get_cpu();
+
+       if (cpumask_test_cpu(this_cpu, mask) && !exclude_self)
+               dump_stack();
+
+       smp_call_function_many(mask, arch_dump_stack, NULL, 1);
+
+       put_cpu();
 }
 
 int mips_get_process_fp_mode(struct task_struct *task)
index 3f70f90..1d51a11 100644 (file)
@@ -86,8 +86,9 @@ static inline unsigned long get_softint(void)
        return retval;
 }
 
-void arch_trigger_all_cpu_backtrace(bool);
-#define arch_trigger_all_cpu_backtrace arch_trigger_all_cpu_backtrace
+void arch_trigger_cpumask_backtrace(const struct cpumask *mask,
+                                   bool exclude_self);
+#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
 
 extern void *hardirq_stack[NR_CPUS];
 extern void *softirq_stack[NR_CPUS];
index fa14402..47ff558 100644 (file)
@@ -239,7 +239,7 @@ static void __global_reg_poll(struct global_reg_snapshot *gp)
        }
 }
 
-void arch_trigger_all_cpu_backtrace(bool include_self)
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
 {
        struct thread_info *tp = current_thread_info();
        struct pt_regs *regs = get_irq_regs();
@@ -255,15 +255,15 @@ void arch_trigger_all_cpu_backtrace(bool include_self)
 
        memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
 
-       if (include_self)
+       if (cpumask_test_cpu(this_cpu, mask) && !exclude_self)
                __global_reg_self(tp, regs, this_cpu);
 
        smp_fetch_global_regs();
 
-       for_each_online_cpu(cpu) {
+       for_each_cpu(cpu, mask) {
                struct global_reg_snapshot *gp;
 
-               if (!include_self && cpu == this_cpu)
+               if (exclude_self && cpu == this_cpu)
                        continue;
 
                gp = &global_cpu_snapshot[cpu].reg;
@@ -300,7 +300,7 @@ void arch_trigger_all_cpu_backtrace(bool include_self)
 
 static void sysrq_handle_globreg(int key)
 {
-       arch_trigger_all_cpu_backtrace(true);
+       trigger_all_cpu_backtrace();
 }
 
 static struct sysrq_key_op sparc_globalreg_op = {
index e7de5c9..16d3fa2 100644 (file)
@@ -50,8 +50,9 @@ extern int vector_used_by_percpu_irq(unsigned int vector);
 extern void init_ISA_irqs(void);
 
 #ifdef CONFIG_X86_LOCAL_APIC
-void arch_trigger_all_cpu_backtrace(bool);
-#define arch_trigger_all_cpu_backtrace arch_trigger_all_cpu_backtrace
+void arch_trigger_cpumask_backtrace(const struct cpumask *mask,
+                                   bool exclude_self);
+#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
 #endif
 
 #endif /* _ASM_X86_IRQ_H */
index f29501e..c73c9fb 100644 (file)
@@ -26,32 +26,32 @@ u64 hw_nmi_get_sample_period(int watchdog_thresh)
 }
 #endif
 
-#ifdef arch_trigger_all_cpu_backtrace
+#ifdef arch_trigger_cpumask_backtrace
 static void nmi_raise_cpu_backtrace(cpumask_t *mask)
 {
        apic->send_IPI_mask(mask, NMI_VECTOR);
 }
 
-void arch_trigger_all_cpu_backtrace(bool include_self)
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
 {
-       nmi_trigger_all_cpu_backtrace(include_self, nmi_raise_cpu_backtrace);
+       nmi_trigger_cpumask_backtrace(mask, exclude_self,
+                                     nmi_raise_cpu_backtrace);
 }
 
-static int
-arch_trigger_all_cpu_backtrace_handler(unsigned int cmd, struct pt_regs *regs)
+static int nmi_cpu_backtrace_handler(unsigned int cmd, struct pt_regs *regs)
 {
        if (nmi_cpu_backtrace(regs))
                return NMI_HANDLED;
 
        return NMI_DONE;
 }
-NOKPROBE_SYMBOL(arch_trigger_all_cpu_backtrace_handler);
+NOKPROBE_SYMBOL(nmi_cpu_backtrace_handler);
 
-static int __init register_trigger_all_cpu_backtrace(void)
+static int __init register_nmi_cpu_backtrace_handler(void)
 {
-       register_nmi_handler(NMI_LOCAL, arch_trigger_all_cpu_backtrace_handler,
+       register_nmi_handler(NMI_LOCAL, nmi_cpu_backtrace_handler,
                                0, "arch_bt");
        return 0;
 }
-early_initcall(register_trigger_all_cpu_backtrace);
+early_initcall(register_nmi_cpu_backtrace_handler);
 #endif
index 4630eea..a78c35c 100644 (file)
@@ -35,21 +35,34 @@ static inline void hardlockup_detector_disable(void) {}
  * base function. Return whether such support was available,
  * to allow calling code to fall back to some other mechanism:
  */
-#ifdef arch_trigger_all_cpu_backtrace
+#ifdef arch_trigger_cpumask_backtrace
 static inline bool trigger_all_cpu_backtrace(void)
 {
-       arch_trigger_all_cpu_backtrace(true);
-
+       arch_trigger_cpumask_backtrace(cpu_online_mask, false);
        return true;
 }
+
 static inline bool trigger_allbutself_cpu_backtrace(void)
 {
-       arch_trigger_all_cpu_backtrace(false);
+       arch_trigger_cpumask_backtrace(cpu_online_mask, true);
+       return true;
+}
+
+static inline bool trigger_cpumask_backtrace(struct cpumask *mask)
+{
+       arch_trigger_cpumask_backtrace(mask, false);
+       return true;
+}
+
+static inline bool trigger_single_cpu_backtrace(int cpu)
+{
+       arch_trigger_cpumask_backtrace(cpumask_of(cpu), false);
        return true;
 }
 
 /* generic implementation */
-void nmi_trigger_all_cpu_backtrace(bool include_self,
+void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
+                                  bool exclude_self,
                                   void (*raise)(cpumask_t *mask));
 bool nmi_cpu_backtrace(struct pt_regs *regs);
 
@@ -62,6 +75,14 @@ static inline bool trigger_allbutself_cpu_backtrace(void)
 {
        return false;
 }
+static inline bool trigger_cpumask_backtrace(struct cpumask *mask)
+{
+       return false;
+}
+static inline bool trigger_single_cpu_backtrace(int cpu)
+{
+       return false;
+}
 #endif
 
 #ifdef CONFIG_LOCKUP_DETECTOR
index 26caf51..df347e3 100644 (file)
 #include <linux/kprobes.h>
 #include <linux/nmi.h>
 
-#ifdef arch_trigger_all_cpu_backtrace
+#ifdef arch_trigger_cpumask_backtrace
 /* For reliability, we're prepared to waste bits here. */
 static DECLARE_BITMAP(backtrace_mask, NR_CPUS) __read_mostly;
 
-/* "in progress" flag of arch_trigger_all_cpu_backtrace */
+/* "in progress" flag of arch_trigger_cpumask_backtrace */
 static unsigned long backtrace_flag;
 
 /*
- * When raise() is called it will be is passed a pointer to the
+ * When raise() is called it will be passed a pointer to the
  * backtrace_mask. Architectures that call nmi_cpu_backtrace()
  * directly from their raise() functions may rely on the mask
  * they are passed being updated as a side effect of this call.
  */
-void nmi_trigger_all_cpu_backtrace(bool include_self,
+void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
+                                  bool exclude_self,
                                   void (*raise)(cpumask_t *mask))
 {
        int i, this_cpu = get_cpu();
@@ -44,13 +45,13 @@ void nmi_trigger_all_cpu_backtrace(bool include_self,
                return;
        }
 
-       cpumask_copy(to_cpumask(backtrace_mask), cpu_online_mask);
-       if (!include_self)
+       cpumask_copy(to_cpumask(backtrace_mask), mask);
+       if (exclude_self)
                cpumask_clear_cpu(this_cpu, to_cpumask(backtrace_mask));
 
        if (!cpumask_empty(to_cpumask(backtrace_mask))) {
-               pr_info("Sending NMI to %s CPUs:\n",
-                       (include_self ? "all" : "other"));
+               pr_info("Sending NMI from CPU %d to CPUs %*pbl:\n",
+                       this_cpu, nr_cpumask_bits, to_cpumask(backtrace_mask));
                raise(to_cpumask(backtrace_mask));
        }