Huge page(contiguous bit) slow down

[will.deacon@arm.com (Will Deacon)]

On Wed, Sep 19, 2018 at 04:37:08PM +0100, Steve Capper wrote:
> On Wed, Sep 19, 2018 at 11:29:37AM +0100, Will Deacon wrote:
> > diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
> > index 192b3ba07075..edae6774132d 100644
> > --- a/arch/arm64/mm/hugetlbpage.c
> > +++ b/arch/arm64/mm/hugetlbpage.c
> > @@ -324,7 +324,7 @@ int huge_ptep_set_access_flags(struct vm_area_struct *vma,
> >  			       unsigned long addr, pte_t *ptep,
> >  			       pte_t pte, int dirty)
> >  {
> > -	int ncontig, i, changed = 0;
> > +	int ncontig, i;
> >  	size_t pgsize = 0;
> >  	unsigned long pfn = pte_pfn(pte), dpfn;
> >  	pgprot_t hugeprot;
> > @@ -336,9 +336,14 @@ int huge_ptep_set_access_flags(struct vm_area_struct *vma,
> >  	ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
> >  	dpfn = pgsize >> PAGE_SHIFT;
> >  
> > +	for (i = 0; i < ncontig; i++)
> > +		if (!pte_same(huge_ptep_get(ptep + i), pte))
> > +			break;
> > +
> 
> This comparison will always fail as our pfns will be different. I have a work
> in progress fix (a separate function to go through the contiguous range
> to preserve the dirty information) that I'll test and post shortly.

Ha, yes of course. I forgot about the most useful part of the pte!
I'll take a look at your patch when you post it.

Will