Re: [PATCH V10 04/22] LoongArch: Add writecombine support for drm

[Huacai Chen]

Hi, Xuerui,

On Sun, May 15, 2022 at 12:19 PM WANG Xuerui <kernel@xxxxxxxxxx> wrote:
>
>
> On 5/14/22 16:03, Huacai Chen wrote:
> > LoongArch maintains cache coherency in hardware, but its WUC attribute
> > (Weak-ordered UnCached, which is similar to WC) is out of the scope of
> > cache coherency machanism. This means WUC can only used for write-only
> > memory regions.
> >
> > Signed-off-by: Huacai Chen <chenhuacai@xxxxxxxxxxx>
> > ---
> >   drivers/gpu/drm/drm_vm.c         | 2 +-
> >   drivers/gpu/drm/ttm/ttm_module.c | 2 +-
> >   include/drm/drm_cache.h          | 8 ++++++++
> >   3 files changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/drm_vm.c b/drivers/gpu/drm/drm_vm.c
> > index e957d4851dc0..f024dc93939e 100644
> > --- a/drivers/gpu/drm/drm_vm.c
> > +++ b/drivers/gpu/drm/drm_vm.c
> > @@ -69,7 +69,7 @@ static pgprot_t drm_io_prot(struct drm_local_map *map,
> >       pgprot_t tmp = vm_get_page_prot(vma->vm_flags);
> >
> >   #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || \
> > -    defined(__mips__)
> > +    defined(__mips__) || defined(__loongarch__)
> >       if (map->type == _DRM_REGISTERS && !(map->flags & _DRM_WRITE_COMBINING))
> >               tmp = pgprot_noncached(tmp);
> >       else
> > diff --git a/drivers/gpu/drm/ttm/ttm_module.c b/drivers/gpu/drm/ttm/ttm_module.c
> > index a3ad7c9736ec..b3fffe7b5062 100644
> > --- a/drivers/gpu/drm/ttm/ttm_module.c
> > +++ b/drivers/gpu/drm/ttm/ttm_module.c
> > @@ -74,7 +74,7 @@ pgprot_t ttm_prot_from_caching(enum ttm_caching caching, pgprot_t tmp)
> >   #endif /* CONFIG_UML */
> >   #endif /* __i386__ || __x86_64__ */
> >   #if defined(__ia64__) || defined(__arm__) || defined(__aarch64__) || \
> > -     defined(__powerpc__) || defined(__mips__)
> > +     defined(__powerpc__) || defined(__mips__) || defined(__loongarch__)
> >       if (caching == ttm_write_combined)
> >               tmp = pgprot_writecombine(tmp);
> >       else
> > diff --git a/include/drm/drm_cache.h b/include/drm/drm_cache.h
> > index 22deb216b59c..08e0e3ffad13 100644
> > --- a/include/drm/drm_cache.h
> > +++ b/include/drm/drm_cache.h
> > @@ -67,6 +67,14 @@ static inline bool drm_arch_can_wc_memory(void)
> >        * optimization entirely for ARM and arm64.
> >        */
> >       return false;
> > +#elif defined(CONFIG_LOONGARCH)
> > +     /*
> > +      * LoongArch maintains cache coherency in hardware, but its WUC attribute
> > +      * (Weak-ordered UnCached, which is similar to WC) is out of the scope of
> > +      * cache coherency machanism. This means WUC can only used for write-only
> > +      * memory regions.
> > +      */
> > +     return false;
> >   #else
> >       return true;
> >   #endif
>
> The code changes look reasonable, given the adequate comments, but have
> the drm people given acks? This seems to exclusively touch drm bits and
> not directly related to arch bring-up. (You may get scrambled screen
> output but everything else is working, I'm running my LoongArch devbox
> headlessly ever since I first set it up last year.)
>
> If anything, IMO you could even take this patch out and still get the
> arch properly brought up. What do others think?
Thanks, I think I should add Cc:  dri-devel@xxxxxxxxxxxxxxxxxxxxx

Huacai
>
> Nevertheless,
>
> Reviewed-by: WANG Xuerui <git@xxxxxxxxxx>
>