On Sat, Apr 10, 2021 at 09:10:47PM +0200, Arnd Bergmann wrote:
> On Sat, Apr 10, 2021 at 4:44 AM Matthew Wilcox <willy@infradead.org> wrote:
> > +                       dma_addr_t dma_addr __packed;
> >                 };
> >                 struct {        /* slab, slob and slub */
> >                         union {
> >
> > but I don't know if GCC is smart enough to realise that dma_addr is now
> > on an 8 byte boundary and it can use a normal instruction to access it,
> > or whether it'll do something daft like use byte loads to access it.
> >
> > We could also do:
> >
> > +                       dma_addr_t dma_addr __packed __aligned(sizeof(void *));
> >
> > and I see pahole, at least sees this correctly:
> >
> >                 struct {
> >                         long unsigned int _page_pool_pad; /*     4     4 */
> >                         dma_addr_t dma_addr __attribute__((__aligned__(4))); /*     8     8 */
> >                 } __attribute__((__packed__)) __attribute__((__aligned__(4)));
> >
> > This presumably affects any 32-bit architecture with a 64-bit phys_addr_t
> > / dma_addr_t.  Advice, please?
> 
> I've tried out what gcc would make of this:  https://godbolt.org/z/aTEbxxbG3
> 
> struct page {
>     short a;
>     struct {
>         short b;
>         long long c __attribute__((packed, aligned(2)));
>     } __attribute__((packed));
> } __attribute__((aligned(8)));
> 
> In this structure, 'c' is clearly aligned to eight bytes, and gcc does
> realize that
> it is safe to use the 'ldrd' instruction for 32-bit arm, which is forbidden on
> struct members with less than 4 byte alignment. However, it also complains
> that passing a pointer to 'c' into a function that expects a 'long long' is not
> allowed because alignof(c) is only '2' here.
> 
> (I used 'short' here because I having a 64-bit member misaligned by four
> bytes wouldn't make a difference to the instructions on Arm, or any other
> 32-bit architecture I can think of, regardless of the ABI requirements).

So ... we could do this:

+++ b/include/linux/types.h
@@ -140,7 +140,7 @@ typedef u64 blkcnt_t;
  * so they don't care about the size of the actual bus addresses.
  */
 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
-typedef u64 dma_addr_t;
+typedef u64 __attribute__((aligned(sizeof(void *)))) dma_addr_t;
 #else
 typedef u32 dma_addr_t;
 #endif

but I'm a little scared that this might have unintended consequences.
And Jesper points out that a big-endian 64-bit dma_addr_t can impersonate
a PageTail page, and we should solve that problem while we're at it.
So I don't think we should do this, but thought I should mention it as
a possibility.