On Fri, Dec 02, 2016 at 10:09:03PM +0100, Pavel Machek wrote:

On Fri 2016-12-02 11:42:21, Josh Poimboeuf wrote:

quoted

Resuming from a suspend operation is showing a KASAN false positive
warning:


Reported-by: Scott Bauer <redacted>
Signed-off-by: Josh Poimboeuf <redacted>

Acked-by: Pavel Machek <redacted>

quoted

diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
index 0e9505f..b2a0cff 100644
--- a/mm/kasan/kasan.c
+++ b/mm/kasan/kasan.c

@@ -80,7 +80,14 @@ void kasan_unpoison_task_stack(struct task_struct *task)
 /* Unpoison the stack for the current task beyond a watermark sp value. */
 asmlinkage void kasan_unpoison_task_stack_below(const void *watermark)
 {
-	__kasan_unpoison_stack(current, watermark);
+	/*
+	 * Calculate the task stack base address.  Avoid using 'current'
+	 * because this function is called by early resume code which hasn't
+	 * yet set up the percpu register (%gs).
+	 */
+	void *base = (void *)((unsigned long)watermark & ~(THREAD_SIZE - 1));
+
+	kasan_unpoison_shadow(base, watermark - base);
 }
 

I know you modified this code to be arch-independend... but is it
really? I guess it is portable enough across architectures that run
kasan today..

Yes, it's arch-independent as far as I know.

All the implementations of alloc_thread_stack_node() in kernel/fork.c
create THREAD_SIZE sized/aligned stacks.

ia64 has its own implementation of alloc_thread_stack_node(), which also
has a THREAD_SIZE sized/aligned stack, with task_struct stored at the
beginning.

For those architectures for which stack grows up, they would need to
call a different helper which unpoisons the stack above the watermark,
but that was also the case before my patch.

-- 
Josh