Re: [PATCH] mm/highmem: Align-down to page the address for kunmap_flush_on_unmap()

[Ira Weiny <ira.weiny@intel.com>]

Matthew Wilcox wrote:
> On Thu, Jan 26, 2023 at 09:37:08PM +0100, Helge Deller wrote:
> > > This 'fixes' looks correct to me.  I don't know how many folks are running
> > > highmem with parisc but if they are I am sure they would appreciate the
> > > extra knowledge.
> > 
> > It seems nobody is running highmem on parisc, because it can't be enabled.
> > AFAICS, it's not in any parisc related Kconfig file.
> 
> But this isn't being used for highmem on parisc; it's being used for
> cache coherency.

<sigh> right!  I missed that.  Thanks to you and Al for setting me
straight.

> 
> > > I do wonder if this should be cc'ed to stable to ensure it gets
> > > backported?  Helge do you think there is a need for that?
> > 
> > For correctness I think it's nevertheless good to backport it.
> 
> I cc'd stable on my version of this patch, and included a Fixes tag
> to indicate how far back to backport it.

Thanks!

> 
> > That would be another possibility:
> > 
> > diff --git a/arch/parisc/include/asm/cacheflush.h b/arch/parisc/include/asm/cacheflush.h
> > index 0bdee6724132..ce5d1f8a23bd 100644
> > --- a/arch/parisc/include/asm/cacheflush.h
> > +++ b/arch/parisc/include/asm/cacheflush.h
> > @@ -77,6 +77,7 @@ void flush_anon_page(struct vm_area_struct *vma, struct page *page, unsigned lon
> >  #define ARCH_HAS_FLUSH_ON_KUNMAP
> >  static inline void kunmap_flush_on_unmap(const void *addr)
> >  {
> > +       addr = PTR_ALIGN_DOWN(addr, PAGE_SIZE);
> 
> I considered that, but thought it a shame to do it here when all the
> other users are passing in a page-aligned address.
> 

I think this also keeps the kunmap_local() API intact better.

Patch reviewed.

Thanks!
Ira