On 12/09/14 18:03, Russell King - ARM Linux wrote:

On Thu, Sep 11, 2014 at 12:31:14PM +0100, Daniel Thompson wrote:

quoted

-	.macro	svc_entry, stack_hole=0
+	.macro	svc_entry, stack_hole=0, call_trace=1
  UNWIND(.fnstart		)
  UNWIND(.save {r0 - pc}		)
 	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)

@@ -183,7 +183,9 @@ ENDPROC(__und_invalid)
 	stmia	r7, {r2 - r6}
 
 #ifdef CONFIG_TRACE_IRQFLAGS
+	.if \call_trace
 	bl	trace_hardirqs_off
+	.endif
 #endif

Good, you picked this up from my patch.  But what about the call into
lockdep from usr_entry?

That was writen from your review comment rather than taken from your patch.

Yes, it should be safe if we're entering from user mode, because by
definition, the kernel can't be holding any locks at that point.
However, I'd much prefer to keep to a set of simple rules here: avoid
lockdep in FIQ code altogether.

Ok. You're right that I followed the "can't be holding any locks" logic
when I didn't update usr_entry in reaction to the original review comment.

I'm also happy with the "avoid lockdep in FIQ code altogether" approach.
I'll do this.

That's much easier to understand than "we can call into lockdep provided
we've been entered from user mode".

The other thing you miss is that /potentially/ call into the scheduler
as well from a FIQ.  Do we /really/ want to do that kind of work here?

Not happy.

Sorry. I will fix these.


Daniel.