netdev - Re: Improving accuracy of PHC readings

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20181023164832.GB12019@localhost>
Date:   Tue, 23 Oct 2018 18:48:32 +0200
From:   Miroslav Lichvar <mlichvar@...hat.com>
To:     "Keller, Jacob E" <jacob.e.keller@...el.com>
Cc:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Richard Cochran <richardcochran@...il.com>
Subject: Re: Improving accuracy of PHC readings

On Fri, Oct 19, 2018 at 04:52:13PM +0000, Keller, Jacob E wrote:
> > This should significantly improve the accuracy of the synchronization,
> > reduce the uncertainty in the readings to less than a half or third,
> > and also reduce the jitter as there are fewer register reads sensitive
> > to the PCIe delay.
> > 
> > What do you think?
> > 
> 
> Nice! I think this is good. I'd love to see some data to back it up, but it makes sense to me.

I tried a quick hack with an X550 and I219. The delay dropped from
about 2940 ns to 1040 ns on the first port of the X550, from 1920 ns
to 660 ns on the second port of the X550, and from 2500 ns to 1300 ns
on the I219.

The I219 supports the SYS_OFFSET_PRECISE ioctl (cross timestamping),
which we can use for comparison. The difference between the offsets
calculated using the two ioctls was about 500-600 ns before and now it
is about -50--150 ns.

I was not able to find any information on how accurate cross
timestamping on this HW is actually supposed to be, so I'm wondering
which of the two is closer to the truth.

Here is an output from phc2sys with the I219:

Before:
phc offset       -59 s2 freq     +40 delay   2527
phc offset        19 s2 freq    +101 delay   2526
phc offset       -23 s2 freq     +64 delay   2522
phc offset        46 s2 freq    +126 delay   2535
phc offset       -32 s2 freq     +62 delay   2530
phc offset       -10 s2 freq     +75 delay   2526
phc offset       102 s2 freq    +184 delay   2523

After:
phc offset        17 s2 freq    +105 delay   1298
phc offset        47 s2 freq    +140 delay   1299
phc offset       -42 s2 freq     +65 delay   1293
phc offset        -6 s2 freq     +88 delay   1299
phc offset        34 s2 freq    +127 delay   1300
phc offset       -14 s2 freq     +89 delay   1301
phc offset       -86 s2 freq     +13 delay   1296
phc offset       -21 s2 freq     +52 delay   1298

-- 
Miroslav Lichvar