About get_page for compound page

About get_page for compound page

[Fengwei Yin]

Hi,
In get_page(), the parameter page is changed by
         page = compound_head(page);
if the page is not the first page of compound pages.

My question: is this behavior correct?

Regards
Yin, Fengwei

Re: About get_page for compound page

[Peter Zijlstra]

On Thu, 2007-07-19 at 23:54 +0800, Fengwei Yin wrote:
> Hi,
> In get_page(), the parameter page is changed by
>          page = compound_head(page);
> if the page is not the first page of compound pages.
> 
> My question: is this behavior correct?

Yes it is, in general the page state of a compound page is kept in the
first page of the set.

Re: About get_page for compound page

[Fengwei Yin]

On 7/20/07, Peter Zijlstra <peterz@infradead.org> wrote:
> On Thu, 2007-07-19 at 23:54 +0800, Fengwei Yin wrote:
> > Hi,
> > In get_page(), the parameter page is changed by
> >          page = compound_head(page);
> > if the page is not the first page of compound pages.
> >
> > My question: is this behavior correct?
>
> Yes it is, in general the page state of a compound page is kept in the
> first page of the set.
>
Yes. I know this. But
static inline void get_page(struct page *page)
                                                     ~~~~~~ changed by
this function.
{
         page = compound_head(page);
                    ~~~~~~~~~~~~~~~~~~~~~~~
         VM_BUG_ON(atomic_read(&page->_count) == 0);
         atomic_inc(&page->_count);
}

As I understand, just need update the first page reference count.
Or the caller aware of this bahavior?

Or change it to
static inline void get_page(struct page *page)
{
         struct page *head_page = compound_head(page);
         VM_BUG_ON(atomic_read(&head_page->_count) == 0);
         atomic_inc(&head_page->_count);
}

Regards
Yin, Fengwei

Re: About get_page for compound page

[Peter Zijlstra]

On Fri, 2007-07-20 at 07:59 +0800, Fengwei Yin wrote:
> On 7/20/07, Peter Zijlstra <peterz@infradead.org> wrote:
> > On Thu, 2007-07-19 at 23:54 +0800, Fengwei Yin wrote:
> > > Hi,
> > > In get_page(), the parameter page is changed by
> > >          page = compound_head(page);
> > > if the page is not the first page of compound pages.
> > >
> > > My question: is this behavior correct?
> >
> > Yes it is, in general the page state of a compound page is kept in the
> > first page of the set.
> >
> Yes. I know this. But
> static inline void get_page(struct page *page)
>                                                      ~~~~~~ changed by
> this function.
> {
>          page = compound_head(page);
>                     ~~~~~~~~~~~~~~~~~~~~~~~
>          VM_BUG_ON(atomic_read(&page->_count) == 0);
>          atomic_inc(&page->_count);
> }
> 
> As I understand, just need update the first page reference count.
> Or the caller aware of this bahavior?
> 
> Or change it to
> static inline void get_page(struct page *page)
> {
>          struct page *head_page = compound_head(page);
>          VM_BUG_ON(atomic_read(&head_page->_count) == 0);
>          atomic_inc(&head_page->_count);
> }

Arguments are local to the function.