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

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 15 Jul 2025 08:49:32 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> > >   *
> > > - * Return: 1 if the the callback was already queued.
> > > - *         0 if the callback successfully was queued.
> > > + * Return: 0 if the callback successfully was queued.
> > > + *         UNWIND_ALREADY_PENDING if the the callback was already queued.
> > > + *         UNWIND_ALREADY_EXECUTED if the callback was already called
> > > + *                (and will not be called again)
> > >   *         Negative if there's an error.
> > >   *         @cookie holds the cookie of the first request by any user
> > >   */    
> > 
> > Lots of babbling in the Changelog, but no real elucidation as to why you
> > need this second return value.
> > 
> > AFAICT it serves no real purpose; the users of this function should not
> > care. The only difference is that the unwind reference (your cookie)
> > becomes a backward reference instead of a forward reference. But why
> > would anybody care?  
> 
> Older versions of the code required it. I think I can remove it now.

Ah it is still used in the perf code:

perf_callchain() has:

        if (defer_user) {
                int ret = deferred_request(event);
                if (!ret)
                        local_inc(&event->ctx->nr_no_switch_fast);
                else if (ret < 0)
                        defer_user = false;
        }

Where deferred_requests() is as static function that returns the result
of the unwind request. If it is zero, it means the callback will be
called, if it is greater than zero it means it has already been called,
and negative is an error (and use the old method).

It looks like when the callback is called it expects nr_no_switch_fast
to be incremented and it will decrement it. This is directly from
Josh's patch and I don't know perf well enough to know if that update
to nr_no_switch_fast is needed.

If it's not needed, we can just return 0 on success and negative on
failure. What do you think?

Here's the original patch:

  https://lore.kernel.org/all/20250708020050.928524258@kernel.org/

-- Steve