On Thu, Dec 03, 2015 at 05:47:08PM +0000, Catalin Marinas wrote:

On Wed, Dec 02, 2015 at 12:29:04AM +0300, Yury Norov wrote:

quoted

My question. Why aarch64 defines COMPAT_SHMLBA as 0x4000?

This was done to match the arch/arm value of 4 * 4K. The historical
32-bit reason for 4 pages is to cope with aliasing VIPT caches (see
https://git.kernel.org/cgit/linux/kernel/git/tglx/history.git/commit?id=4197692eef113eeb8e3e413cc70993a5e667e5b8)

quoted

If there's
no specific reason for it, it looks like a bug, and we should
define it like in arch/arm:
        #define SHMLBA  (4 * PAGE_SIZE)          /* attach addr a multiple of this */

I guess you meant COMPAT_SHMLBA.

I citated arm code here. In aarch64 it's COMPAT_SHMLBA, of course.

I'm not sure there is much value in
keeping 4*PAGE_SIZE for larger page sizes but I agree that the current
16K value doesn't work well with 64K pages.

Arnd told there will be a workaround for arm v6 caches. Than this
header will not be needed at all. Until that, this is simpliest
fix as it doesn't affect userspace.

-- 
Catalin