Re: [RFC PATCH 4/4] perf: Use folios for the aux ringbuffer & pagefault path

[Matthew Wilcox <willy@xxxxxxxxxxxxx>]

On Wed, Aug 23, 2023 at 03:38:13PM +0800, Yin Fengwei wrote:
> On 8/22/23 04:20, Matthew Wilcox (Oracle) wrote:
> > -	get_page(vmf->page);
> > -	vmf->page->mapping = vmf->vma->vm_file->f_mapping;
> > -	vmf->page->index   = vmf->pgoff;
> > +	folio_get(folio);
> > +	rcu_read_unlock();
> > +	folio_lock(folio);
> > +	if (!folio->mapping)
> > +		folio->mapping = vmf->vma->vm_file->f_mapping;
> >  
> > -	ret = 0;
> > +	return VM_FAULT_LOCKED;
> In __do_fault():
> 
>         if (unlikely(!(ret & VM_FAULT_LOCKED)))                                 
>                 lock_page(vmf->page);                                           
>         else                                                                    
>                 VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); 
> 
> As we lock folio, not sure whether !PageLocked(vmf->page) can be true
> here. My understanding is yes if vmf->pgoff belongs to tail pages. Did
> I can miss something here?

There's only one lock bit per folio; there's no lock bit for individual
pages.  When we check PageLocked() on a tail page, it redirects to the
head page.

__PAGEFLAG(Locked, locked, PF_NO_TAIL)

#define PF_NO_TAIL(page, enforce) ({                                    \
                VM_BUG_ON_PGFLAGS(enforce && PageTail(page), page);     \
                PF_POISONED_CHECK(compound_head(page)); })

#define TESTPAGEFLAG(uname, lname, policy)                              \
static __always_inline int Page##uname(struct page *page)               \
{ return test_bit(PG_##lname, &policy(page, 0)->flags); }

and that expands out to:

static __always_inline int PageLocked(struct page *page)
{ return test_bit(PG_##locked, &compound_head(page)->flags); }