Re: [PATCH 05/12] riscv: Implement non-coherent DMA support via SiFive cache flushing

[Conor Dooley <conor@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2211 bytes --]

On Tue, Feb 14, 2023 at 08:06:49PM +0200, Cristian Ciocaltea wrote:
> On 2/13/23 10:30, Ben Dooks wrote:
> > On 11/02/2023 03:18, Cristian Ciocaltea wrote:
> > > From: Emil Renner Berthing <kernel@esmil.dk>

> > > diff --git a/arch/riscv/mm/dma-noncoherent.c
> > > b/arch/riscv/mm/dma-noncoherent.c
> > > index d919efab6eba..e07e53aea537 100644
> > > --- a/arch/riscv/mm/dma-noncoherent.c
> > > +++ b/arch/riscv/mm/dma-noncoherent.c
> > > @@ -9,14 +9,21 @@
> > >   #include <linux/dma-map-ops.h>
> > >   #include <linux/mm.h>
> > >   #include <asm/cacheflush.h>
> > > +#include <soc/sifive/sifive_ccache.h>
> > >   static bool noncoherent_supported;
> > >   void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,
> > >                     enum dma_data_direction dir)
> > >   {
> > > -    void *vaddr = phys_to_virt(paddr);
> > > +    void *vaddr;
> > > +    if (sifive_ccache_handle_noncoherent()) {
> > > +        sifive_ccache_flush_range(paddr, size);
> > > +        return;
> > > +    }
> > > +
> > > +    vaddr = phys_to_virt(paddr);
> > >       switch (dir) {
> > >       case DMA_TO_DEVICE:
> > >           ALT_CMO_OP(clean, vaddr, size, riscv_cbom_block_size);
> > > @@ -35,8 +42,14 @@ void arch_sync_dma_for_device(phys_addr_t paddr,
> > > size_t size,
> > >   void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size,
> > >                  enum dma_data_direction dir)
> > >   {
> > > -    void *vaddr = phys_to_virt(paddr);
> > > +    void *vaddr;
> > > +
> > > +    if (sifive_ccache_handle_noncoherent()) {
> > > +        sifive_ccache_flush_range(paddr, size);
> > > +        return;
> > > +    }
> > 
> > ok, what happens if we have an system where the ccache and another level
> > of cache also requires maintenance operations?

TBH, I'd hope that a system with that complexity is also not trying to
manage the cache in this manner!

> According to [1], the handling of non-coherent DMA on RISC-V is currently
> being worked on, so I will respin the series as soon as the proper support
> arrives.

But yeah, once that stuff lands we can carry out these operations only
for the platforms that need/"need" it.

Cheers,
Conor.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]