nmi_backtrace: do a local dump_stack() instead of a self-NMI
authorChris Metcalf <cmetcalf@mellanox.com>
Sat, 8 Oct 2016 00:02:49 +0000 (17:02 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 8 Oct 2016 01:46:30 +0000 (18:46 -0700)
Currently on arm there is code that checks whether it should call
dump_stack() explicitly, to avoid trying to raise an NMI when the
current context is not preemptible by the backtrace IPI.  Similarly, the
forthcoming arch/tile support uses an IPI mechanism that does not
support generating an NMI to self.

Accordingly, move the code that guards this case into the generic
mechanism, and invoke it unconditionally whenever we want a backtrace of
the current cpu.  It seems plausible that in all cases, dump_stack()
will generate better information than generating a stack from the NMI
handler.  The register state will be missing, but that state is likely
not particularly helpful in any case.

Or, if we think it is helpful, we should be capturing and emitting the
current register state in all cases when regs == NULL is passed to
nmi_cpu_backtrace().

Link: http://lkml.kernel.org/r/1472487169-14923-3-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: Petr Mladek <pmladek@suse.com>
Acked-by: Aaron Tomlin <atomlin@redhat.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>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/arm/kernel/smp.c
lib/nmi_backtrace.c

index 5abc5697e4e595f7126840696a9352aaf2da4657..7dd14e8395e62976b3083f677b0e4d2ea0d71d95 100644 (file)
@@ -748,15 +748,6 @@ core_initcall(register_cpufreq_notifier);
 
 static void raise_nmi(cpumask_t *mask)
 {
-       /*
-        * Generate the backtrace directly if we are running in a calling
-        * context that is not preemptible by the backtrace IPI. Note
-        * that nmi_cpu_backtrace() automatically removes the current cpu
-        * from mask.
-        */
-       if (cpumask_test_cpu(smp_processor_id(), mask) && irqs_disabled())
-               nmi_cpu_backtrace(NULL);
-
        smp_cross_call(mask, IPI_CPU_BACKTRACE);
 }
 
index df347e355267b176330aabc746b88787032e88e3..393a3cca1f47660e436131e6dbcf718d125ff6bb 100644 (file)
@@ -49,6 +49,15 @@ void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
        if (exclude_self)
                cpumask_clear_cpu(this_cpu, to_cpumask(backtrace_mask));
 
+       /*
+        * Don't try to send an NMI to this cpu; it may work on some
+        * architectures, but on others it may not, and we'll get
+        * information at least as useful just by doing a dump_stack() here.
+        * Note that nmi_cpu_backtrace(NULL) will clear the cpu bit.
+        */
+       if (cpumask_test_cpu(this_cpu, to_cpumask(backtrace_mask)))
+               nmi_cpu_backtrace(NULL);
+
        if (!cpumask_empty(to_cpumask(backtrace_mask))) {
                pr_info("Sending NMI from CPU %d to CPUs %*pbl:\n",
                        this_cpu, nr_cpumask_bits, to_cpumask(backtrace_mask));