Hi All,

This is a follow up in my long running effort of making inb()/outb() and
similar I/O port accessors compile-time optional. After initially
sending this as a treewide series with the latest revision at[0]
we switched to per subsystem series. Now though as we're left with only
5 patches left I'm going back to a single series with Arnd planning
to take this via the the asm-generic tree.

This series may also be viewed for your convenience on my git.kernel.org
tree[1] under the b4/has_ioport branch. As for compile-time vs runtime
see Linus' reply to my first attempt[2].

Thanks,
Niklas

[0] https://lore.kernel.org/all/20230522105049.1467313-1-schnelle@linux.ibm.com/
[1] https://git.kernel.org/pub/scm/linux/kernel/git/niks/linux.git
[2] https://lore.kernel.org/lkml/CAHk-=wg80je=K7madF4e7WrRNp37e3qh6y10Svhdc7O8SZ_-8g@mail.gmail.com/

Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
---
Changes in v9:
- In drm patch sort includes and reformat if (Thomas). Use IS_ENABELD()
  and add a helper instead of #ifdef (Arnd, Thomas)
- Rebased on v6.12-rc4
- Compile tested applied to next. There are a few conflicts with drm
  next but they're all just context changes
- Link to v8: https://lore.kernel.org/r/20241008-b4-has_ioport-v8-0-793e68aeadda@linux.ibm.com

Changes in v8:
- Don't remove "depends on !S390" for SERIAL_8250
- Link to v7: https://lore.kernel.org/r/20241008-b4-has_ioport-v7-0-8624c09a4d77@linux.ibm.com

Changes in v7:
- Renamed serial_8250_need_ioport() helper to
  serial_8250_warn_need_ioport() and move it to 8250_pcilib.c so it can
  be used in serial8250_pci_setup_port()
- Flattened if in serial8250_pci_setup_port() (Maciej)
- Removed gratuituous changes (Maciej)
- Removed is_upf_fourport() helper in favor of zeroing UPF_FOURPORT
  if CONFIG_HAS_IOPORT is not set (Maciej)
- Link to v6: https://lore.kernel.org/r/20241007-b4-has_ioport-v6-0-03f7240da6e5@linux.ibm.com

Changes since v5 / per subsystem patches:

drm:
- Add HAS_IOPORT dependency for GMA500
tty: serial:
- Make 8250 PCI driver emit an error message when trying to use devices
  which require I/O ports without CONFIG_HAS_IOPORT (Maciej)
- Use early returns + dead code elimination to skip inb()/outb() uses
  in quirks (Arnd)
- In 8250 PCI driver also handle fintek and moxi quirks
- In 8250 ports code handle um's defined(__i385__) &&
  defined(CONFIG_HAS_IOPORT) case
- Use IS_ENABLED() early return also in is_upf_fourport()
  __always_inline to force constant folding

---
Niklas Schnelle (5):
      hexagon: Don't select GENERIC_IOMAP without HAS_IOPORT support
      Bluetooth: add HAS_IOPORT dependencies
      drm: handle HAS_IOPORT dependencies
      tty: serial: handle HAS_IOPORT dependencies
      asm-generic/io.h: Remove I/O port accessors for HAS_IOPORT=n

 arch/hexagon/Kconfig                  |  1 -
 drivers/bluetooth/Kconfig             |  6 ++--
 drivers/gpu/drm/gma500/Kconfig        |  2 +-
 drivers/gpu/drm/qxl/Kconfig           |  2 +-
 drivers/gpu/drm/tiny/bochs.c          | 19 ++++++++---
 drivers/gpu/drm/tiny/cirrus.c         |  2 ++
 drivers/gpu/drm/xe/Kconfig            |  2 +-
 drivers/tty/Kconfig                   |  4 +--
 drivers/tty/serial/8250/8250_early.c  |  4 +++
 drivers/tty/serial/8250/8250_pci.c    | 40 +++++++++++++++++++++++
 drivers/tty/serial/8250/8250_pcilib.c | 12 ++++++-
 drivers/tty/serial/8250/8250_pcilib.h |  2 ++
 drivers/tty/serial/8250/8250_port.c   | 27 +++++++++++++---
 drivers/tty/serial/8250/Kconfig       |  4 +--
 drivers/tty/serial/Kconfig            |  2 +-
 include/asm-generic/io.h              | 60 +++++++++++++++++++++++++++++++++++
 include/linux/serial_core.h           |  4 +++
 17 files changed, 171 insertions(+), 22 deletions(-)
---
base-commit: 42f7652d3eb527d03665b09edac47f85fb600924
change-id: 20241004-b4-has_ioport-60ac6ce1deb6

Best regards,
-- 
Niklas Schnelle

On Thu, Oct 24, 2024, at 17:54, Niklas Schnelle wrote:
> Hi All,
>
> This is a follow up in my long running effort of making inb()/outb() and
> similar I/O port accessors compile-time optional. After initially
> sending this as a treewide series with the latest revision at[0]
> we switched to per subsystem series. Now though as we're left with only
> 5 patches left I'm going back to a single series with Arnd planning
> to take this via the the asm-generic tree.
>
> This series may also be viewed for your convenience on my git.kernel.org
> tree[1] under the b4/has_ioport branch. As for compile-time vs runtime
> see Linus' reply to my first attempt[2].

Hi Niklas,

Thanks for your endless work on this. I have now pulled it into
the asm-generic tree as I want to ensure we get enough time to
test this as part of linux-next before the merge window.

If minor issues still come up, I would try to fix those as
add-on patches to avoid rebasing my tree.

I also expect that we will continue with add-on patches in
the future, in particular I hope to make HAS_IOPORT optional
on arm, arm64 and powerpc, and only enabled for
configurations that actually want it.

     Arnd

On Fri, 2024-10-25 at 13:41 +0000, Arnd Bergmann wrote:
> On Thu, Oct 24, 2024, at 17:54, Niklas Schnelle wrote:
> > Hi All,
> > 
> > This is a follow up in my long running effort of making inb()/outb() and
> > similar I/O port accessors compile-time optional. After initially
> > sending this as a treewide series with the latest revision at[0]
> > we switched to per subsystem series. Now though as we're left with only
> > 5 patches left I'm going back to a single series with Arnd planning
> > to take this via the the asm-generic tree.
> > 
> > This series may also be viewed for your convenience on my git.kernel.org
> > tree[1] under the b4/has_ioport branch. As for compile-time vs runtime
> > see Linus' reply to my first attempt[2].
> 
> Hi Niklas,
> 
> Thanks for your endless work on this. I have now pulled it into
> the asm-generic tree as I want to ensure we get enough time to
> test this as part of linux-next before the merge window.
> 
> If minor issues still come up, I would try to fix those as
> add-on patches to avoid rebasing my tree.
> 
> I also expect that we will continue with add-on patches in
> the future, in particular I hope to make HAS_IOPORT optional
> on arm, arm64 and powerpc, and only enabled for
> configurations that actually want it.
> 
>      Arnd
> 

Thanks for taking it and sticking by my side through this! Now let's
just hope there won't be too much fallout but I will be here to help if
needed. As for arm, arm64, and powerpc I like it, having more
!HAS_IOPORT targets will help to share the load of new inb()/outb()
which "worked for me on x86". I definitely learned a lot in the
process. Of course I wished and originally expected it to go a lot
faster but hey looks like we might persevere in the end. And yes, I
will pour myself a drink when this finally made it into Linus' tree :-)
And then when we meet at some conference in the future we can laugh
about how this turned from a 5 line patch into at least 53 commits over
3 years.

Thanks,
Niklas