[tarantool-patches] Re: [PATCH] lib/core/fiber: Initialize stack_watermark where appropriate

[Cyrill Gorcunov]

On Mon, Mar 18, 2019 at 09:55:29PM +0300, Konstantin Osipov wrote:
> * Vladimir Davydov <vdavydov.dev at gmail.com> [19/03/18 20:56]:
> 
> > On Mon, Mar 18, 2019 at 08:23:52PM +0300, Cyrill Gorcunov wrote:
> > > The stack_watermark member declared with HAVE_MADV_DONTNEED wrap,
> > > so need to guard it here the same way.
> 
> What is the reason to keep the poison pool 8 elements now that it
> is only used to save on madvise() invocations and is put in a
> random position? Shouldn't 1 element be enough?

Even 8 elements are not guarantee us from false positives and
with a sole element the situation would be even worse.

> What is the actual stack size for 99.9% of cases? Why not put the
> poison at 16kb of stack?

Well, I think we can't answer this question without gathering
statistics.