tracing: Add BUG_ON when stack end location is over written

It is difficult to detect a stack overrun when it actually occurs.

We have observed that this type of corruption is often silent and can go unnoticed. Once the corrupted region is examined, the outcome is undefined and often results in sporadic system crashes.

When the stack tracing feature is enabled, let's check for this condition and take appropriate action.

Note: init_task doesn't get its stack end location set to STACK_END_MAGIC.

---
 kernel/trace/trace_stack.c | 3 +++
 1 file changed, 3 insertions(+)

index e6be585..21b320e 100644
@@ -13,6 +13,7 @@
 #include <linux/sysctl.h>
 #include <linux/init.h>
 #include <linux/fs.h>
+#include <linux/magic.h>
 
 #include <asm/setup.h>
 

@@ -144,6 +145,8 @@ check_stack(unsigned long ip, unsigned long *stack)
 			i++;
 	}
 
+	BUG_ON(current != &init_task &&
+		*(end_of_stack(current)) != STACK_END_MAGIC);
  out:
 	arch_spin_unlock(&max_stack_lock);
 	local_irq_restore(flags);

1.8.5.3

I added this to my 3.15 queue, and if it survives testing, I'll push it to my for-next branch.

Thanks,

-- Steve