RE: [RFC PATCH 6/6] Drivers: hv: vmbus: Refactor the ring-buffer iterator functions

["Michael Kelley (LINUX)" <mikelley@microsoft.com>]

From: Andrea Parri <parri.andrea@gmail.com> Sent: Friday, April 15, 2022 9:28 AM
> 
> On Fri, Apr 15, 2022 at 09:00:31AM +0200, Andrea Parri wrote:
> > > > @@ -470,7 +471,6 @@ struct vmpacket_descriptor *hv_pkt_iter_first_raw(struct
> > > > vmbus_channel *channel)
> > > >
> > > >  	return (struct vmpacket_descriptor *)(hv_get_ring_buffer(rbi) + rbi-
> > > > >priv_read_index);
> > > >  }
> > > > -EXPORT_SYMBOL_GPL(hv_pkt_iter_first_raw);
> > >
> > > Does hv_pkt_iter_first_raw() need to be retained at all as a
> > > separate function?  I think after these changes, the only caller
> > > is hv_pkt_iter_first(), in which case the code could just go
> > > inline in hv_pkt_iter_first().  Doing that combining would
> > > also allow the elimination of the duplicate call to
> > > hv_pkt_iter_avail().
> 
> Back to this, can you clarify what you mean by "the elimination of..."?
> After moving the function "inline", hv_pkt_iter_avail() would be called
> in to check for a non-NULL descriptor (in the inline function) and later
> in the computation of bytes_avail.

I was thinking something like this:

bytes_avail = hv_pkt_iter_avail(rbi);
if (bytes_avail < sizeof(struct vmpacket_descriptor))
	return NULL;
bytes_avail = min(rbi->pkt_buffer_size, bytes_avail);

desc = (struct vmpacket_descriptor *)(hv_get_ring_buffer(rbi) + rbi->priv_read_index);

And for that matter, hv_pkt_iter_avail() is now only called in one place.
It's a judgment call whether to keep it as a separate helper function vs.
inlining it in hv_pkt_iter_first() as well.  I'm OK either way.


Michael