Re: [PATCH 04/12] soc: sifive: ccache: Add non-coherent DMA handling

[Conor Dooley]

On Sun, Feb 19, 2023 at 10:32:52PM +0100, Emil Renner Berthing wrote:
> On Thu, 16 Feb 2023 at 19:51, Conor Dooley <conor@xxxxxxxxxx> wrote:
> >
> > Emil,
> >
> > +CC Daire
> >
> > On Sat, Feb 11, 2023 at 05:18:13AM +0200, Cristian Ciocaltea wrote:
> > > From: Emil Renner Berthing <kernel@xxxxxxxx>
> > >
> > > Add functions to flush the caches and handle non-coherent DMA.
> > >
> > > Signed-off-by: Emil Renner Berthing <kernel@xxxxxxxx>
> > > [replace <asm/cacheflush.h> with <linux/cacheflush.h>]
> > > Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea@xxxxxxxxxxxxx>
> > > ---
> >
> > > +void *sifive_ccache_set_uncached(void *addr, size_t size)
> > > +{
> > > +     phys_addr_t phys_addr = __pa(addr) + uncached_offset;
> > > +     void *mem_base;
> > > +
> > > +     mem_base = memremap(phys_addr, size, MEMREMAP_WT);
> > > +     if (!mem_base) {
> > > +             pr_err("%s memremap failed for addr %p\n", __func__, addr);
> > > +             return ERR_PTR(-EINVAL);
> > > +     }
> > > +
> > > +     return mem_base;
> > > +}
> >
> > The rest of this I either get b/c we did it, or will become moot so I
> > amn't worried about it, but can you please explain this, in particular
> > the memremap that you're doing here?
> 
> No, I can't really. As we talked about it's also based on a prototype
> by Atish. I'm sure you know that the general idea is that we want to
> return a pointer that accesses the same physical memory, but through
> the uncached alias.

Yah, I follow all the rest of what's going on - it's just this bit of it
that I don't.

> I can't tell you exactly why it's done this way
> though, sorry.

I had a bit of a look on lore, but don't really see anything there that
contained any discussion of what was going on here.

Adding Atish in the off-chance that he remembers!

Attachment: signature.asc
Description: PGP signature