On Sun, Apr 26, 2020 at 12:26:42AM -0700, Christoph Hellwig wrote:
> > diff --git a/arch/arc/mm/highmem.c b/arch/arc/mm/highmem.c
> > index 4db13a6b9f3b..1cae4b911a33 100644
> > --- a/arch/arc/mm/highmem.c
> > +++ b/arch/arc/mm/highmem.c
> > @@ -53,11 +53,10 @@ void *kmap_atomic(struct page *page)
> > {
> > int idx, cpu_idx;
> > unsigned long vaddr;
> > + void *addr = kmap_atomic_fast(page);
> >
> > - preempt_disable();
> > - pagefault_disable();
> > - if (!PageHighMem(page))
> > - return page_address(page);
> > + if (addr)
> > + return addr;
>
> Wouldn't it make sense to just move kmap_atomic itelf to common code,
> and call out to a kmap_atomic_high for the highmem case, following the
> scheme in kmap?
>
Sure I do like that symmetry between the calls.
>
> Same for the unmap side.
FWIW that would simply be renaming __kunmap_atomic() to kunmap_atomic_high()
>
> That might require to support
> kmap_atomic_prot everywhere first, which sounds like a really good
> idea anyway, and would avoid the need for strange workaround in drm.
Having a kmap_atomic_prot() seems like a good idea. But I'm not exactly sure
why CONFIG_x86 is being called out specifically in the DRM code?
Ira
