From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gorcunov@gmail.com>
Return-Path: <gorcunov@gmail.com>
Date: Tue, 5 Mar 2019 22:17:55 +0300
From: Cyrill Gorcunov <gorcunov@gmail.com>
Subject: Re: [tarantool-patches] [rfc 3/4] core/fiber: Put static watermark
 into stack to track its usage
Message-ID: <20190305191755.GN13301@uranus.lan>
References: <20190302125524.279852704@gmail.com>
 <20190302130034.050488016@gmail.com>
 <20190305081013.GK21955@chai>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190305081013.GK21955@chai>
To: Konstantin Osipov <kostja@tarantool.org>
Cc: tarantool-patches@freelists.org, vdavydov.dev@gmail.com
List-ID: <tarantool-patches.dev.tarantool.org>

On Tue, Mar 05, 2019 at 11:10:13AM +0300, Konstantin Osipov wrote:
> * Cyrill Gorcunov <gorcunov@gmail.com> [19/03/03 23:25]:
> > +/*
> > + * Random values generated with uuid.
> > + * Try to fit a cache line.
> > + */
> > +static const uint64_t poison_pool[] = {
> > +	0x74f31d37285c4c37, 0xb10269a05bf10c29,
> > +	0x0994d845bd284e0f, 0x9ffd4f7129c184df,
> > +	0x357151e6711c4415, 0x8c5e5f41aafe6f28,
> > +	0x6917dd79e78049d5, 0xba61957c65ca2465,
> > +};
> 
> With randomness incorporated into each fiber I think we can reduce
> the poison pool to 4 8-byte integers or even less.

Kostja, you know, randoms values is a bad idea. For example I just got

[cyrill@uranus tarantool.git] ./src/tarantool
poison_gaps[0] = 736
poison_gaps[1] = 952
poison_gaps[2] = 976
poison_gaps[3] = 2016
Tarantool 2.1.1-360-g6666db8d5
type 'help' for interactive help

So the first gap almost close to 1K value :( Which means we won't detect
the page dirtifying while someone is putting values on start of the
page and even more, we're not controlling what exactly random values
come to us. I think putting poisons at 128 byte bound a way better

The pretty draf code I used is
---
static void
stack_poison_gaps_init(void)
{
	size_t i, j, m = page_size / 2;
	uint16_t v;

	assert(POISON_SIZE > 1);

	/*
	 * Fill the gaps with random (8 byte aligned)
	 * values in a half page range. This should decrease
	 * probability that someone manage to sneak out
	 * the poison values.
	 */

	random_bytes((void *)poison_gaps, sizeof(poison_gaps));
	for (i = 0; i < POISON_SIZE; i++) {
		v = poison_gaps[i] % m;
		poison_gaps[i] = (v + 8) & ~7;
	}

	for (i = 0; i < POISON_SIZE; i++) {
		v = poison_gaps[i], j = i;
		while (j > 0 && poison_gaps[j-1] > v)
			poison_gaps[j] = poison_gaps[j-1], j--;
		poison_gaps[j] = v;
	}

	for (i = 0; i < POISON_SIZE; i++) {
		say_info("poison_gaps[%d] = %d", i, (unsigned)poison_gaps[i]);
		if (i == 0)
			continue;
		assert(poison_gaps[i-1] != poison_gaps[i]);
	}
}
---