Re: [PATCH V5 14/22] arm/ioreq: Introduce arch specific bits for IOREQ/DM features

[Oleksandr <olekstysh@xxxxxxxxx>]

On 26.01.21 11:15, Jan Beulich wrote:

Hi Jan

> On 25.01.2021 20:08, Oleksandr Tyshchenko wrote:
> > --- a/xen/include/xen/dm.h
> > +++ b/xen/include/xen/dm.h
> > @@ -19,6 +19,8 @@
> >   
> >   #include <xen/sched.h>
> >   
> > +#include <public/hvm/dm_op.h>
> > +
> >   struct dmop_args {
> >       domid_t domid;
> >       unsigned int nr_bufs;
> How come this becomes necessary at this point in the series, when
> nothing else in this header changes, and nothing changes in the
> public headers at all? Is it perhaps a .c file that needs the
> #include instead? Headers shouldn't pull in other headers without
> clear need - as indicated in reply to a prior version, we have
> way too many bad examples (causing headaches in certain cases),
> and we'd like to avoid gaining more.
Yes, I understand this and completely agree. I remember last discussion 
on that, this is not forgotten. The only reason I made this (non 
entirely correct) change is to make
series compilable on Arm with IOREQ support enabled. If I considered 
this change as a correct one I would make it from the very beginning (in 
patch #9 which adds this common header)...
I added remark to draw reviewer's attention on the fact that I got stuck 
with resolving that, what I did wrong and how it should be done 
properly. The problem is that I didn't manage to make it properly.
Of course, I tried to include it directly by dm.c, but this didn't help 
much without violating "alphabetical order" rule. Details here:
https://lore.kernel.org/xen-devel/e0bc7f80-974e-945d-4605-173bd05302af@xxxxxxxxx/

I would appreciate any input on that.

> (Oh, I notice you actually
> have a post-commit-message remark about this, but then this
> patch should be marked RFC until the issue was resolved.)
Agree, I should have marked this patch as RFC to avoid misunderstanding.


--
Regards,

Oleksandr Tyshchenko