Re: x86-64: Maintain 16-byte stack alignment

On 12 January 2017 at 06:12, Herbert Xu [off-list ref] wrote:

On Tue, Jan 10, 2017 at 05:30:48PM +0000, Ard Biesheuvel wrote:

quoted

Apologies for introducing this breakage. It seemed like an obvious and
simple cleanup, so I didn't even bother to mention it in the commit
log, but if the kernel does not guarantee 16 byte alignment, I guess
we should revert to the old method. If SSE instructions are the only
ones that require this alignment, then I suppose not having a ABI
conforming stack pointer should not be an issue in general.

BTW Ard, what is the stack alignment on ARM64?

[From memory] the arm64 ELF psABI mandates a 16 byte stack alignment
at function entry, and 8 byte alignment at all other times. This means
compiled code will typically preserve 16 byte alignment, and
__aligned(16) on a stack variable will likely not result in an
explicit alignment of the stack pointer *. But the arm64 ISA does not
have any load/store instructions that would trigger a misalignment
fault on an address that is 8 byte aligned but not 16 byte aligned, so
the situation is very different from x86 (assuming I am correct in
asserting that there are no failure modes resulting from a misaligned
stack other than this one and a potential performance hit)

* I didn't check whether the exception handling realigns the stack
pointer on nested exceptions (arm64 has separate IRQ stacks)