Re: [GIT PULL 1/2] asm-generic: rework PCI I/O space access

From: Niklas Schnelle <schnelle@linux.ibm.com>
To: John Garry <john.garry@huawei.com>, Arnd Bergmann <arnd@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	linux-arch <linux-arch@vger.kernel.org>,
	linux-pci <linux-pci@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [GIT PULL 1/2] asm-generic: rework PCI I/O space access
Date: Tue, 21 Dec 2021 17:57:06 +0100	[thread overview]
Message-ID: <3b1b7881bbecc25a2d70f54743f1f6a9decbaa45.camel@linux.ibm.com> (raw)
In-Reply-To: <3d543c90-383f-647a-5cd4-f7fd4e7246ad@huawei.com>

On Tue, 2021-12-21 at 16:48 +0000, John Garry wrote:
> On 20/12/2021 09:27, Niklas Schnelle wrote:
> > >   > My feeling is that in this case we want some other dependency, e.g. a
> > >   > new CONFIG_LPC. It should actually be possible to use this driver on
> > >   > any machine with an LPC bus, which would by definition be the primary
> > >   > I/O space, so it should be possible to load it on Arm64.
> > > 
> > > You did suggest HARDCODED_IOPORT earlier in this thread, and the
> > > definition/premise there seemed sensible to me.
> > > 
> > > Anyway it seems practical to make all these changes in a single series,
> > > so need a way forward as Niklas has no such changes for this additional
> > > kconfig option.
> > > 
> > > As a start, may I suggest we at least have Niklas' patch committed to a
> > > dev branch based on -next or latest mainline release for further analysis?
> > > 
> > > Thanks,
> > > John
> > > 
> > > 
> > My plan would be to split the patch up into more manageable pieces as
> > suggested by Arnd plus of course fixes like the missing ARM select. As
> > Arnd suggested I'll split the HAS_IOPORT additions into the initial
> > introduction plus arch selects and then the HAS_IOPORT dependencies per
> > subsytem. I think these per subsystem dependency patches then would be
> > a great place to find drivers which should have a different dependency
> > be it on LPC or a newly introduced HARDCODED_IOPORT. The thing is we
> > can find and check HAS_IOPORT dependencies easily but it's hard to find
> > HARDCODED_IOPORT so I think the lattter should be a refinement of the
> > former. It can of course still go in as a single series. I'll
> > definitely make the next iteration available as a git branch.
> 
> I'll do an audit for what would require HARDCODED_IOPORT to understand 
> the scope while you can continue the work on your current path.
> 
> Thanks,
> john
> 

Sounds good, I'm open to adding such a config option given a clear
enough picture of what drivers it would affect. Meanwhile I've made
some progress splitting things up. I still need to do a bit more
testing and refining of comments before sending an RFC but if you're
curious you can check out the 'has_ioport' branch on my GitHub here:

https://github.com/niklas88/linux.git  (still figuring out if/how I can
get a proper git.kernel.org branch/repository).

Thanks,
Niklas