[CRIU] [PATCH 3/4] arm64: provide own memcpy for pie

[Dmitry Safonov]

On 11/08/2016 12:23 AM, Kir Kolyshkin wrote:
> Chris, Dmitry,
>
> Can you please take a look at this (and, ideally, test on amd64 with
> both clang and gcc)?

Well, looks ok to me, the small nip is that we can omit if(n) check
here, if I read asm rightly:

>> +    if (n)
>> +        memcpy_arm64(to, from, n);
>> +    return to;

I don't have arm64 board, but have a qemu VM.
Anyway, it's quite big pile of code and if we can't see gain - and
don't have arm64 boards to prove it's perf impact (I don't think it
will be visible on qemu), what the fuss about?
Don't get me wrong - I don't object the patch, it's ok to me, but
why not simplify at this moment and just use for(...) a[i] = b[i]?

I mean, if anything got broken in this asm (by any reason, you could
see the history on the source link had some fixes already) - the people
start comming with "help me, criu segfaults/corrupts memory/silently
dies on arm64", which in my POV is harder to fix than for() loop,
which I'm sure is rightly handled by any compiler.

So, again I don't object, I'm ok if the patch applied - just don't get
a reason for it.

-- 
              Dmitry