Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sun, 11 Apr 2021 11:43:07 +0200
From:   Jesper Dangaard Brouer <brouer@redhat.com>
To:     "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc:     linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        netdev@vger.kernel.org, linuxppc-dev@lists.ozlabs.org,
        linux-arm-kernel@lists.infradead.org, linux-mips@vger.kernel.org,
        Ilias Apalodimas <ilias.apalodimas@linaro.org>,
        Matteo Croce <mcroce@linux.microsoft.com>,
        Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>,
        Grygorii Strashko <grygorii.strashko@ti.com>,
        Arnd Bergmann <arnd@kernel.org>, brouer@redhat.com,
        Christoph Hellwig <hch@lst.de>
Subject: Re: [PATCH 1/1] mm: Fix struct page layout on 32-bit systems
Message-ID: <20210411114307.5087f958@carbon>
In-Reply-To: <20210410205246.507048-2-willy@infradead.org>
References: <20210410205246.507048-1-willy@infradead.org>
        <20210410205246.507048-2-willy@infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Sat, 10 Apr 2021 21:52:45 +0100
"Matthew Wilcox (Oracle)" <willy@infradead.org> wrote:

> 32-bit architectures which expect 8-byte alignment for 8-byte integers
> and need 64-bit DMA addresses (arc, arm, mips, ppc) had their struct
> page inadvertently expanded in 2019.  When the dma_addr_t was added,
> it forced the alignment of the union to 8 bytes, which inserted a 4 byte
> gap between 'flags' and the union.
> 
> We could fix this by telling the compiler to use a smaller alignment
> for the dma_addr, but that seems a little fragile.  Instead, move the
> 'flags' into the union.  That causes dma_addr to shift into the same
> bits as 'mapping', so it would have to be cleared on free.  To avoid
> this, insert three words of padding and use the same bits as ->index
> and ->private, neither of which have to be cleared on free.
> 
> Fixes: c25fff7171be ("mm: add dma_addr_t to struct page")
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>  include/linux/mm_types.h | 38 ++++++++++++++++++++++++++------------
>  1 file changed, 26 insertions(+), 12 deletions(-)
> 
> diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
> index 6613b26a8894..45c563e9b50e 100644
> --- a/include/linux/mm_types.h
> +++ b/include/linux/mm_types.h
> @@ -68,16 +68,22 @@ struct mem_cgroup;
>  #endif
>  
>  struct page {
> -	unsigned long flags;		/* Atomic flags, some possibly
> -					 * updated asynchronously */
>  	/*
> -	 * Five words (20/40 bytes) are available in this union.
> -	 * WARNING: bit 0 of the first word is used for PageTail(). That
> -	 * means the other users of this union MUST NOT use the bit to
> +	 * This union is six words (24 / 48 bytes) in size.
> +	 * The first word is reserved for atomic flags, often updated
> +	 * asynchronously.  Use the PageFoo() macros to access it.  Some
> +	 * of the flags can be reused for your own purposes, but the
> +	 * word as a whole often contains other information and overwriting
> +	 * it will cause functions like page_zone() and page_node() to stop
> +	 * working correctly.
> +	 *
> +	 * Bit 0 of the second word is used for PageTail(). That
> +	 * means the other users of this union MUST leave the bit zero to
>  	 * avoid collision and false-positive PageTail().
>  	 */
>  	union {
>  		struct {	/* Page cache and anonymous pages */
> +			unsigned long flags;
>  			/**
>  			 * @lru: Pageout list, eg. active_list protected by
>  			 * lruvec->lru_lock.  Sometimes used as a generic list
> @@ -96,13 +102,14 @@ struct page {
>  			unsigned long private;
>  		};
>  		struct {	/* page_pool used by netstack */
> -			/**
> -			 * @dma_addr: might require a 64-bit value even on
> -			 * 32-bit architectures.
> -			 */
> -			dma_addr_t dma_addr;

The original intend of placing member @dma_addr here is that it overlap
with @LRU (type struct list_head) which contains two pointers.  Thus, in
case of CONFIG_ARCH_DMA_ADDR_T_64BIT=y on 32-bit architectures it would
use both pointers.

Thinking more about this, this design is flawed as bit 0 of the first
word is used for compound pages (see PageTail and @compound_head), is
reserved.  We knew DMA addresses were aligned, thus we though this
satisfied that need.  BUT for DMA_ADDR_T_64BIT=y on 32-bit arch the
first word will contain the "upper" part of the DMA addr, which I don't
think gives this guarantee.

I guess, nobody are using this combination?!?  I though we added this
to satisfy TI (Texas Instrument) driver cpsw (code in
drivers/net/ethernet/ti/cpsw*).  Thus, I assumed it was in use?


> +			unsigned long _pp_flags;
> +			unsigned long pp_magic;
> +			unsigned long xmi;

Matteo notice, I think intent is we can store xdp_mem_info in @xmi.

> +			unsigned long _pp_mapping_pad;
> +			dma_addr_t dma_addr;	/* might be one or two words */
>  		};

Could you explain your intent here?
I worry about @index.

As I mentioned in other thread[1] netstack use page_is_pfmemalloc()
(code copy-pasted below signature) which imply that the member @index
have to be kept intact. In above, I'm unsure @index is untouched.

[1] https://lore.kernel.org/lkml/20210410082158.79ad09a6@carbon/
-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer

/*
 * Return true only if the page has been allocated with
 * ALLOC_NO_WATERMARKS and the low watermark was not
 * met implying that the system is under some pressure.
 */
static inline bool page_is_pfmemalloc(const struct page *page)
{
	/*
	 * Page index cannot be this large so this must be
	 * a pfmemalloc page.
	 */
	return page->index == -1UL;
}

/*
 * Only to be called by the page allocator on a freshly allocated
 * page.
 */
static inline void set_page_pfmemalloc(struct page *page)
{
	page->index = -1UL;
}

static inline void clear_page_pfmemalloc(struct page *page)
{
	page->index = 0;
}