On Mon, Jan 26, 2026 at 10:14:46AM +0000, Wilken Gottwalt wrote:
> On Mon, 26 Jan 2026 11:59:43 +0200
> Andy Shevchenko <[email protected]> wrote:
> 
> > On Sun, Jan 25, 2026 at 07:46:51PM +0100, Wolfram Sang wrote:
> > > TLDR: I want to create a hwspinlock provider outside of the hwspinlock
> > > directory. So, I refactored the headers into a provider/consumer pair.
> > > Which seems to me like a reasonable seperation anyhow. No functional
> > > changes. My build tests went fine and buildbots are happy, too.
> > > 
> > > Longer explanation:
> > > 
> > > There is a device (MFIS) in newer Renesas SoCs which combines various
> > > things like hwspinlocks, mailboxes and other stuff. Sadly, these are not
> > > strictly separated. Registers are kind of mixed and its register
> > > unprotection scheme will need one of its own locks. I tried various
> > > paths to handle this device (MFD, auxiliary bus) but I concluded that
> > > the sub-device dependencies give enough reasons for a single driver in
> > > drivers/soc/. So, this series will allow me to instantiate a hwspinlock
> > > provider from the other directory.
> > > 
> > > Patches 1+2 do the actual refactoring with a fallback being in place. I
> > > used '-B' with git-format-patch in this RFC, so the actual changes are
> > > more visible when the headers are moved.
> > > 
> > > Patch 3 converts all the users. There are not many. We could try to get
> > > all the acks for this single patch. Or I can break it into single
> > > patches and send them to subsystems. I don't mind.
> > > 
> > > Patch 4 simply removes the fallback.
> > > 
> > > Looking forward to comments on this approach. If the hwspinlock
> > > maintainers like it as is, I would kindly propose to apply patches 1+2
> > > after 7.0-rc1 comes out. This might sound a bit hasty, but a) I want to
> > > avoid chasing a moving target and b) this would remove one dependency of
> > > the hwspinlock driver I originally intend to upstream, of course.
> > > 
> > > I would take care of patches 3+4 as needed.
> > > 
> > > A branch can be found here:
> > > 
> > > git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux.git 
> > > renesas/hwspinlock/refactor-includes
> > > 
> > > Patches are based on linux-next as of 2026-01-21.
> > > 
> > > Opinions?
> > 
> > I don't like the idea of sharing internal stuff. Why would we need to have
> > a struct hwspinlock to be visible?
> > 
> 
> I see what Wolfram wants to achieve. It is the same issue I encounterd while I
> wrote the sun6i hwspinlock driver. Currently it is impossible to write 
> external
> (out-of-kernel-tree) drivers because of internal structures. And it was a pain
> in the ass for testing purposes. I prefer to be able to write external 
> hwspinlock
> drivers.

I am not against _that_. I'm against the implementation. At least I can't see
the impediment in making that struct opaque. Maybe I missed something. That's
why I'm asking why we need it to be visible to the entire world.

-- 
With Best Regards,
Andy Shevchenko



Reply via email to