Re: [PATCH v13 10/14] unwind: Clear unwind_mask on exit back to user space

On Tue, Jul 15, 2025 at 01:20:16PM -0400, Steven Rostedt wrote:

On Tue, 15 Jul 2025 12:29:12 +0200
Peter Zijlstra [off-list ref] wrote:

quoted

@@ -170,41 +193,62 @@ static void unwind_deferred_task_work(st
 int unwind_deferred_request(struct unwind_work *work, u64 *cookie)
 {
 	struct unwind_task_info *info = &current->unwind_info;
-	int ret;
+	unsigned long bits, mask;
+	int bit, ret;
 
 	*cookie = 0;
 
-	if (WARN_ON_ONCE(in_nmi()))
-		return -EINVAL;
-
 	if ((current->flags & (PF_KTHREAD | PF_EXITING)) ||
 	    !user_mode(task_pt_regs(current)))
 		return -EINVAL;
 
+	/* NMI requires having safe cmpxchg operations */
+	if (WARN_ON_ONCE(!UNWIND_NMI_SAFE && in_nmi()))
+		return -EINVAL;

I don't think we want to have a WARN_ON() here as the perf series tries
to first do the deferred unwinding and if that fails, it will go back
to it's old method.

The thing is, I don't think we have an architecture that supports NMIs
and does not have NMI safe cmpxchg. And if we do have one such -- I
don't think it has perf; perf very much assumes cmpxchg is NMI safe.

Calling this from NMI context and not having an NMI safe cmpxchg is very
much a dodgy use case. Please leave the WARN, if it ever triggers, we'll
look at who manages and deal with it then.