Re: [PATCH 04/10] hugetlb: hugepage migration core

[Mel Gorman <mel@csn.ul.ie>]

On Wed, Sep 22, 2010 at 01:59:07PM +0900, Naoya Horiguchi wrote:
> On Mon, Sep 20, 2010 at 12:10:55PM +0100, Mel Gorman wrote:
> > On Wed, Sep 08, 2010 at 10:19:35AM +0900, Naoya Horiguchi wrote:
> ...
> > > @@ -95,26 +96,34 @@ static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
> > >  	pte_t *ptep, pte;
> > >   	spinlock_t *ptl;
> > >  
> > > - 	pgd = pgd_offset(mm, addr);
> > > -	if (!pgd_present(*pgd))
> > > -		goto out;
> > > +	if (unlikely(PageHuge(new))) {
> > > +		ptep = huge_pte_offset(mm, addr);
> > > +		if (!ptep)
> > > +			goto out;
> > > +		ptl = &mm->page_table_lock;
> > > +	} else {
> > > +		pgd = pgd_offset(mm, addr);
> > > +		if (!pgd_present(*pgd))
> > > +			goto out;
> > >  
> > > -	pud = pud_offset(pgd, addr);
> > > -	if (!pud_present(*pud))
> > > -		goto out;
> > > +		pud = pud_offset(pgd, addr);
> > > +		if (!pud_present(*pud))
> > > +			goto out;
> > >  
> > 
> > Why are the changes to teh rest of the walkers necessary? Instead, why
> > did you not identify which PTL lock you needed and then goto the point
> > where spin_lock(ptl) is called? Similar to what page_check_address()
> > does for example.
> 
> This is because Andi-san commented to avoid using goto sentense.
> But honestly I'm not sure which is a clear way.
> 

Ok, personally I would prefer to have matched page_check_address() so we use
similar code patterns when the intention is the same but as functionally it
seems fine, I'm not going to fight over it either.

Thanks.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab