Re: [PATCH v4 2/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt

From: Anup Patel
Date: Thu Oct 20 2022 - 03:39:31 EST


On Thu, Oct 20, 2022 at 2:20 AM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Wed, Oct 19, 2022, at 18:10, Anup Patel wrote:
> > On Wed, Oct 19, 2022 at 7:49 PM Heiko Stuebner <heiko@xxxxxxxxx> wrote:
> >>
> >> Am Mittwoch, 19. Oktober 2022, 15:11:26 CEST schrieb Anup Patel:
> >> > Currently, all flavors of ioremap_xyz() function maps to the generic
> >> > ioremap() which means any ioremap_xyz() call will always map the
> >> > target memory as IO using _PAGE_IOREMAP page attributes. This breaks
> >> > ioremap_cache() and ioremap_wc() on systems with Svpbmt because memory
> >> > remapped using ioremap_cache() and ioremap_wc() will use _PAGE_IOREMAP
> >> > page attributes.
> >> >
> >> > To address above (just like other architectures), we implement RISC-V
> >> > specific ioremap_cache() and ioremap_wc() which maps memory using page
> >> > attributes as defined by the Svpbmt specification.
> >> >
> >> > Fixes: ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")
> >> > Co-developed-by: Mayuresh Chitale <mchitale@xxxxxxxxxxxxxxxx>
> >> > Signed-off-by: Mayuresh Chitale <mchitale@xxxxxxxxxxxxxxxx>
> >> > Signed-off-by: Anup Patel <apatel@xxxxxxxxxxxxxxxx>
> >>
> >> Wasn't there discussion around those functions in general in v2?
> >
> > Yes, there was discussion about a few drivers using ioremap_xyz()
> > which is discouraged and drivers should use memremap().
> >
> > We still need the arch specific ioremap_xyz() functions/macros
> > added by this patch because these are required by the generic
> > kernel memremap() implementation (refer, kernel/iomem.c).
>
> There is a difference between the strongly discouraged
> ioremap_cache() that pretty much has no valid users, and
> the ioremap_wt/ioremap_wc functions that are sometimes used
> for mapping video framebuffer or similar.
>
> It should be sufficient to provide a arch_memremap_wb()
> and no ioremap_cache() to make memremap() work correctly.
>

Okay, I will simplify this patch to only implement arch_memremap_wb().

This will make the MEMREMAP_WB flag to work correctly but other
MEMREMAP_xyz flags will always use non-cacheable IO mappings.

Regards,
Anup