Re: [PATCH 0/7] floppy: build as modules.

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

Hi John,

On 8/16/21 11:55 PM, John Snow wrote:
> On Thu, Aug 5, 2021 at 3:12 AM Gerd Hoffmann <kraxel@redhat.com
> <mailto:kraxel@redhat.com>> wrote:
> 
>     On Wed, Aug 04, 2021 at 05:19:02PM +0200, Philippe Mathieu-Daudé wrote:
>     > +Mark
>     >
>     > On 8/4/21 4:27 PM, Gerd Hoffmann wrote:
>     > > Some code shuffling needed beforehand due to floppy being part of
>     > > several platforms.  While being at it also make floppy optional
>     > > in pc machine type.
>     >
>     > >   floppy: move fdctrl_init_sysbus
>     > >   floppy: move sun4m_fdctrl_init
>     >
>     > https://www.mail-archive.com/qemu-block@nongnu.org/msg84008.html
>     <https://www.mail-archive.com/qemu-block@nongnu.org/msg84008.html>
>     >
>     > Mark suggested:
>     >
>     >   You may be able to simplify this further by removing the
>     >   global legacy init functions fdctrl_init_sysbus() and
>     >   sun4m_fdctrl_init(): from what I can see fdctrl_init_sysbus()
>     >   is only used in hw/mips/jazz.c and sun4m_fdctrl_init() is only
>     >   used in hw/sparc/sun4m.c so you might as well inline them or
>     >   move the functions to the relevant files.
>     >
>     > I did it and plan to send during 6.2. Sounds simpler than module.
>     > You could easily rebase your series on top (or I can include your
>     > patches while sending).
> 
>     Feel free to include them.  But I can also rebase when your patches
>     landed upstream.  Your choice ;)
> 
> What's the plan here, what are we trying to solve with this series
> *exactly*?
> If Phil sends his cleanups, do we still want/need the modularization here?

Both series are orthogonal, but if my cleanups get merged first, there
is less floppy code to modularize.

> For now I'm gonna shuffle these off of my review queue and I assume I'll
> see a respin/rebase from either you or phil during the 6.2 window, let
> me know if this is wrong.

This is OK. Probably easier for everybody if I can rebase/include Gerd's
patches along. I'm still not convinced FDC modularization is the right
way to go; but the PC machine is one of machines I know the less, and
has inherited a lot of odd things, so I need to carefully audit few more
things.

I'd rather have faithful chipsets modelled. Long term I don't think
FDC are going away from QEMU, as they are used by happy hobbyist running
old DOS programs from the 80's. But being able to build QEMU without
FDC would be nice indeed.