Re: [PATCH] mm: fix struct page layout on 32-bit systems

From: Matthew Wilcox <willy@infradead.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: stable <stable@vger.kernel.org>,
	Ilias Apalodimas <ilias.apalodimas@linaro.org>,
	Jesper Dangaard Brouer <brouer@redhat.com>,
	Vlastimil Babka <vbabka@suse.cz>,
	Matteo Croce <mcroce@linux.microsoft.com>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [PATCH] mm: fix struct page layout on 32-bit systems
Date: Sun, 16 May 2021 20:09:20 +0100	[thread overview]
Message-ID: <YKFt4Njj5au/JEhT@casper.infradead.org> (raw)
In-Reply-To: <CAHk-=wihKAt+Wz6=nccQAXxi_VWFJpx4JwWTJSwT0UvUs1RtZw@mail.gmail.com>

On Sun, May 16, 2021 at 11:27:10AM -0700, Linus Torvalds wrote:
> On Sun, May 16, 2021 at 11:22 AM Matthew Wilcox <willy@infradead.org> wrote:
> >
> > Nobody's been willing to guarantee that all 32-bit architectures keep the
> > top 20 bits clear for their DMA addresses.  I've certainly seen hardware
> > (maybe PA-RISC?  MIPS?) which uses the top few bits of the DMA address to
> > indicate things like "coherent" or "bypasses IOMMU".  Rather than trying
> > to find out, I thought this was the safer option.
> 
> Fair enough. I just find it somewhat odd.
> 
> But I still find this a bit ugly. Maybe we could just have made that
> one sub-structure "__aligned(4)", and avoided this all, and let the
> compiler generate the split load (or, more likely, just let the
> compiler generate a regular load from an unaligned location).
> 
> IOW, just
> 
>                 struct {        /* page_pool used by netstack */
>                         /**
>                          * @dma_addr: might require a 64-bit value even on
>                          * 32-bit architectures.
>                          */
>                         dma_addr_t dma_addr;
>                 } __aligned((4));
> 
> without the magic shifting games?

That was the other problem fixed by this patch -- on big-endian 32-bit
platforms with 64-bit dma_addr_t (mips, ppc), a DMA address with bit 32 set
inadvertently sets the PageTail bit.  So we need to store the low bits
in the first word, even on big-endian platforms.

There's an upcoming patch to move dma_addr out of the union with
compound_head, but that requires changing page_is_pfmemalloc() to
use an indicator other than page->index == -1.  Once we do that,
we can have fun with __aligned().  Since we knew it would have to
be backported, this felt like the least risky patch to start with.