On Wed, Sep 13, 2023 at 03:31:05PM +0200, Peter Zijlstra wrote:
> On Sun, Aug 27, 2023 at 08:33:39AM +0100, Lorenzo Stoakes wrote:
>
> > > @@ -785,25 +785,23 @@ __perf_mmap_to_page(struct perf_buffer *rb, unsigned long pgoff)
> > > return virt_to_page(rb->data_pages[pgoff - 1]);
> > > }
> > >
> > > -static void *perf_mmap_alloc_page(int cpu)
> > > +static void *perf_mmap_alloc_page(int node)
> >
> > Nitty point but since we're dealing with folios here maybe rename to
> > perf_mmap_alloc_folio()?
>
> Since it's an explicit order-0 allocation, does that really make sense?
>
True, it does ultimately yield a single page and doesn't reference its
metadata so it's not the end of the world to keep it as-is (it was very
nitty after all!)
> > > {
> > > - struct page *page;
> > > - int node;
> > > + struct folio *folio;
> > >
> > > - node = (cpu == -1) ? cpu : cpu_to_node(cpu);
> > > - page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
> > > - if (!page)
> > > + folio = __folio_alloc_node(GFP_KERNEL | __GFP_ZERO, 0, node);
> > > + if (!folio)
> > > return NULL;
> > >
> > > - return page_address(page);
> > > + return folio_address(folio);
> > > }
>
