Re: [PATCH v1 05/30] soc: sifive: l2 cache: Convert to platform driver

[Hal Feng]

On Wed, 5 Oct 2022 15:05:03 +0100, Conor Dooley wrote:
> On Wed, Oct 05, 2022 at 03:55:17PM +0200, Emil Renner Berthing wrote:
> > On Wed, 5 Oct 2022 at 15:48, Ben Dooks <ben.dooks@xxxxxxxxxxxxxxx> wrote:
> > >
> > > On 05/10/2022 14:44, Emil Renner Berthing wrote:
> > > > On Thu, 29 Sept 2022 at 19:59, Ben Dooks <ben.dooks@xxxxxxxxxx> wrote:
> > > >>
> > > >> On 29/09/2022 15:32, Hal Feng wrote:
> > > >>> From: Emil Renner Berthing <kernel@xxxxxxxx>
> > > >>>
> > > >>> This converts the driver to use the builtin_platform_driver_probe macro
> > > >>> to initialize the driver. This macro ends up calling device_initcall as
> > > >>> was used previously, but also allocates a platform device which gives us
> > > >>> access to much nicer APIs such as platform_ioremap_resource,
> > > >>> platform_get_irq and dev_err_probe.
> > > >>
> > > >> This is useful, but also there are other changes currently being sorted
> > > >> out by Zong Li (cc'd into this message) which have already been reviewed
> > > >> and are hopefully queued for the next kernel release.
> > > >>
> > > >>> Signed-off-by: Emil Renner Berthing <kernel@xxxxxxxx>
> > > >>> Signed-off-by: Hal Feng <hal.feng@xxxxxxxxxxxxxxxxxxxxxx>
> > > >
> > > > I'm ok with something like this being merged, but please note that if
> > > > we ever want to support the JH7100 which uses registers in this
> > > > peripheral to flush the cache for its non-coherent DMAs then this
> > > > driver needs to be loaded before other peripherals or we will trigger
> > > > the 2nd warning in arch/riscv/mm/dma-noncoherent.c. I'm not sure we
> > > > can do that when it's a platform driver. See this patch for an
> > > > alternative to support the JH71x0s:
> > > > https://github.com/esmil/linux/commit/9c5b29da56ae29159c9572c5bb195fe3a1b535c5
> > > >
> > > > /Emil
> > >
> > > Are you replying to your own patch that does the conversion to
> > > platform driver and then saying that it could actually cause
> > > issues?
> > 
> > Yes, I can see it seems odd, but this patch lived for a while in the
> > kernel repo for the JH7100 until I rebased on 6.0-rc1 and realized the
> > above.
> > Hal Feng must have based his patches on a version of the code before
> > that when preparing this series.
> > 
> > > I'm all for dropping this for the moment and keeping the old
> > > early init for the ccache.
> > 
> > Cool.
> 
> FWIW, if converting to a platform driver will inhibit using the driver
> for doing non-coherent stuff I would like to NAK the patch :)
> 

Yeah, I agree, and this patch will be dropped on the next version.