[PATCH v6 1/4] nmi_backtrace: add more trigger_*_cpu_backtrace() methods

On 8/8/2016 9:57 AM, Petr Mladek wrote:

On Thu 2016-07-14 16:50:29, Chris Metcalf wrote:

quoted

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 either the old "all/all_but_self" arch methods, or the new
"cpumask" method, depending on which is available.

I triggered this function using
   echo l >/proc/sysrq-trigger


and got

[  270.791328] -----------  All but itself: ---------------------

[  270.791331] ===============================
[  270.791331] [ INFO: suspicious RCU usage. ]
[  270.791333] 4.8.0-rc1-4-default+ #3086 Not tainted
[  270.791333] -------------------------------
[  270.791335] ./include/linux/rcupdate.h:556 Illegal context switch in RCU read-side critical section!

Ah hah, you tested this with CPUMASK_OFFSTACK, which I didn't.
That explains why you got RCU kmalloc warnings.

quoted

+	cpumask_copy(mask, cpu_online_mask);
+	cpumask_clear_cpu(cpu, mask);
+	arch_trigger_cpumask_backtrace(mask);
+	put_cpu();
+	free_cpumask_var(mask);
+	return true;

Also this looks too much code for an inlined function.
It is rather slow and there is not a big gain. I would move
the definition to lib/nmi_backtrace.c.

After some thought, I ended up just removing both cpumask allocation
sites.  For the allbutself() case, I just re-introduced the "include_self"
boolean that the code used to have.  If it is false when we get into the inner
nmi_trigger_cpumask_backtrace(), I just clear the cpu bit of the current
cpu.  It requires passing a funny boolean around with the mask, but the
alternative (if we don't want to allocate a mask on this path) is to
break apart the nmi_trigger_cpumask_backtrace() function so we can
piggy-back on its locking and its cpumask and set up the cpumask the
way we want, which I think is too much added ugliness.

For the trigger_single_cpu_backtrace() case, I remembered that there was
a cpumask_of() function that we can use that is fast and doesn't allocate,
even with CPUMASK_OFFSTACK, so I just used that instead.

PS: I am sorry for sending this so late in the game. I was
curious why the patch had not been upstream yet and. I made
a closer look to give a Reviewed-by tag...

No worries - even a late review is much better than none!  I'll
send v7 shortly and please do let me know if it works for you.

-- 
Chris Metcalf, Mellanox Technologies
http://www.mellanox.com