Re: [RFC PATCH glibc 4/8] glibc: Perform rseq(2) registration at C startup and thread creation (v15)

[Florian Weimer <fw@deneb.enyo.de>]

* Mathieu Desnoyers:

>> You also need to add an assert that the compiler supports
>> __attribute__ ((aligned)) because ignoring it produces an
>> ABI-incompatible header.
>
> Are you aware of some helper macro I should use to do this, or
> is it done elsewhere in glibc ?

I don't think we have any such GCC-only types yet.  max_align_t is
provided by GCC itself.

>> The struct rseq/struct rseq_cs definitions
>> are broken, they should not try to change the alignment.
>
> AFAIU, this means we should ideally not have used __attribute__((aligned))
> in the uapi headers in the first place. Why is it broken ?

Compilers which are not sufficiently GCC-compatible define
__attribute__(X) as the empty expansion, so you silently get a
different ABI.

There is really no need to specify 32-byte alignment here.  Is not
even the size of a standard cache line.  It can result in crashes if
these structs are heap-allocated using malloc, when optimizing for
AVX2.

For example, clang turns

void
clear (struct rseq *p)
{
  memset (p, 0, sizeof (*p));
}

into:

	vxorps	%xmm0, %xmm0, %xmm0
	vmovaps	%ymm0, (%rdi)
	vzeroupper
	retq

My understanding is that vmovaps will trap if the pointer is
misaligned (“When the source or destination operand is a memory
operand, the operand must be aligned on a 32-byte boundary or a
general-protection exception (#GP) will be generated.”).

> However, now that it is in the wild, it's a bit late to change that.

I had forgotten about the alignment crashes.  I think we should
seriously consider changing the types. 8-(