Re: [PATCH] tracing: Fix tracing_marker may trigger page fault during preempt_disable

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 1 Sep 2025 10:43:36 +0100
Mark Rutland <mark.rutland@arm.com> wrote:

> > So, __copy_from_user_inatomic() is supposed to be called if
> > pagefault_disable() has already been called? If this is the case, can we
> > add more comments to this code?  
> 
> Just to check, you're asking for better comments in <linux/uaccess.h>,
> right?

Yes.

> 
> > I've been using the inatomic() version this
> > way in preempt disabled locations since 2016.
> > 
> > Looks like it needs to be converted to copy_from_user_nofault().
> > 
> > Luo, this version of the patch looks legit, no need for a v2.
> > 
> > I just wanted to figure out why __copy_from_user_inatomic() wasn't atomic.
> > If anything, it needs to be better documented.  
> 
> If that had roughly the same kerneldoc comment as for
> __copy_to_user_inatomic(), would that be sufficient, or do you think
> both need to be made more explicit?

Yeah, it would have been very helpful to have that. Note, my use of
__copy_from_user_inatomic() was added before copy_from_user_nofault()
existed. So it would likely not have been helpful back then ;-)

I think even adding a comment to the kerneldoc of these functions with
something like:

  You probably want to use copy_from/to_user_nofault(). These are more
  for internal handling, and should be avoided unless you really know
  what you are doing.

-- Steve