From: Matthew Wilcox

Sent: 14 April 2021 22:36

On Wed, Apr 14, 2021 at 09:13:22PM +0200, Jesper Dangaard Brouer wrote:

quoted

(If others want to reproduce).  First I could not reproduce on ARM32.
Then I found out that enabling CONFIG_XEN on ARCH=arm was needed to
cause the issue by enabling CONFIG_ARCH_DMA_ADDR_T_64BIT.

hmmm ... you should be able to provoke it by enabling ARM_LPAE,
which selects PHYS_ADDR_T_64BIT, and

config ARCH_DMA_ADDR_T_64BIT
        def_bool 64BIT || PHYS_ADDR_T_64BIT

quoted

 struct page {
        long unsigned int          flags;                /*     0     4 */

        /* XXX 4 bytes hole, try to pack */

        union {
                struct {
                        struct list_head lru;            /*     8     8 */
                        struct address_space * mapping;  /*    16     4 */
                        long unsigned int index;         /*    20     4 */
                        long unsigned int private;       /*    24     4 */
                };                                       /*     8    20 */
                struct {
                        dma_addr_t dma_addr

Adding __packed here will remove the 4 byte hole before the union
and the compiler seems clever enough to know that anything following
a 'long' must also be 'long' aligned.
So you don't get anything horrid like byte accesses.
On 64bit dma_addr will remain 64bit aligned.
On arm32 dma_addr will be 32bit aligned - but forcing two 32bit access
won't make any difference.

So definitely the only simple fix.

	David

quoted

                                           ;             /*     8     8 */
                };                                       /*     8     8 */

[...]

quoted

        } __attribute__((__aligned__(8)));               /*     8    24 */
        union {
                atomic_t           _mapcount;            /*    32     4 */
                unsigned int       page_type;            /*    32     4 */
                unsigned int       active;               /*    32     4 */
                int                units;                /*    32     4 */
        };                                               /*    32     4 */
        atomic_t                   _refcount;            /*    36     4 */

        /* size: 40, cachelines: 1, members: 4 */
        /* sum members: 36, holes: 1, sum holes: 4 */
        /* forced alignments: 1, forced holes: 1, sum forced holes: 4 */
        /* last cacheline: 40 bytes */
} __attribute__((__aligned__(8)));

If you also enable CONFIG_MEMCG or enough options to make
LAST_CPUPID_NOT_IN_PAGE_FLAGS true, you'll end up with another 4-byte
hole at the end.

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel