In the current code, we use the atomic_cmpxchg() to serialize the
output of the dump_stack(), but this implementation suffers the
thundering herd problem. We have observed such kind of livelock on a
Marvell cn96xx board(24 cpus) when heavily using the dump_stack() in
a kprobe handler. Actually we can let the competitors to wait for
the releasing of the lock before jumping to atomic_cmpxchg(). This
will definitely mitigate the thundering herd problem. Thanks Linus
for the suggestion.
Fixes: b58d977432c8 ("dump_stack: serialize the output from dump_stack()")
Cc: [email protected]
Signed-off-by: Kevin Hao <[email protected]>
---
v2:
1. Use the method suggested by Linus.
2. Since this fix seems pretty simple, maybe it deserve to be
backported to stable kernel. So CC the stable.
lib/dump_stack.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/lib/dump_stack.c b/lib/dump_stack.c
index 5cff72f18c4a..0062dfcafd63 100644
--- a/lib/dump_stack.c
+++ b/lib/dump_stack.c
@@ -106,7 +106,11 @@ asmlinkage __visible void dump_stack(void)
was_locked = 1;
} else {
local_irq_restore(flags);
- cpu_relax();
+ /*
+ * Wait the lock to release before jumping to atomic_cmpxchg()
+ * in order to mitigate the thundering herd problem.
+ */
+ do { cpu_relax(); } while (atomic_read(&dump_lock) != -1);
goto retry;
}
--
2.14.4