On 2021/5/12 23:57, Matthew Wilcox wrote:
> On Tue, May 11, 2021 at 05:25:36PM +0300, Ilias Apalodimas wrote:
>> Nope not at all, either would work. we'll switch to that
>
> You'll need something like this because of the current use of
> page->index to mean "pfmemalloc".
>
>>From ecd6d912056a21bbe55d997c01f96b0b8b9fbc31 Mon Sep 17 00:00:00 2001
> From: "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx>
> Date: Fri, 16 Apr 2021 18:12:33 -0400
> Subject: [PATCH] mm: Indicate pfmemalloc pages in compound_head
>
> The net page_pool wants to use a magic value to identify page pool pages.
> The best place to put it is in the first word where it can be clearly a
> non-pointer value. That means shifting dma_addr up to alias with ->index,
> which means we need to find another way to indicate page_is_pfmemalloc().
> Since page_pool doesn't want to set its magic value on pages which are
> pfmemalloc, we can use bit 1 of compound_head to indicate that the page
> came from the memory reserves.
>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>
> ---
> include/linux/mm.h | 12 +++++++-----
> include/linux/mm_types.h | 7 +++----
> 2 files changed, 10 insertions(+), 9 deletions(-)
>
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index bd21864449bf..4f9b2007efad 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -1670,10 +1670,12 @@ struct address_space *page_mapping(struct page *page);
> static inline bool page_is_pfmemalloc(const struct page *page)
> {
> /*
> - * Page index cannot be this large so this must be
> - * a pfmemalloc page.
> + * This is not a tail page; compound_head of a head page is unused
> + * at return from the page allocator, and will be overwritten
> + * by callers who do not care whether the page came from the
> + * reserves.
> */
> - return page->index == -1UL;
> + return page->compound_head & 2;
> }
>
> /*
> @@ -1682,12 +1684,12 @@ static inline bool page_is_pfmemalloc(const struct page *page)
> */
> static inline void set_page_pfmemalloc(struct page *page)
> {
> - page->index = -1UL;
> + page->compound_head = 2;
Is there any reason why not use "page->compound_head |= 2"? as
corresponding to the "page->compound_head & 2" in the above
page_is_pfmemalloc()?
Also, this may mean we need to make sure to pass head page or
base page to set_page_pfmemalloc() if using
"page->compound_head = 2", because it clears the bit 0 and head
page ptr for tail page too, right?
> }
>
> static inline void clear_page_pfmemalloc(struct page *page)
> {
> - page->index = 0;
> + page->compound_head = 0;
> }
>
> /*
> diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
> index 5aacc1c10a45..1352e278939b 100644
> --- a/include/linux/mm_types.h
> +++ b/include/linux/mm_types.h
> @@ -96,10 +96,9 @@ struct page {
> unsigned long private;
> };
> struct { /* page_pool used by netstack */
> - /**
> - * @dma_addr: might require a 64-bit value on
> - * 32-bit architectures.
> - */
> + unsigned long pp_magic;
> + struct page_pool *pp;
> + unsigned long _pp_mapping_pad;
> unsigned long dma_addr[2];
It seems the dma_addr[1] aliases with page->private, and
page_private() is used in skb_copy_ubufs()?
It seems we can avoid using page_private() in skb_copy_ubufs()
by using a dynamic allocated array to store the page ptr?
> };
> struct { /* slab, slob and slub */
>
