Unveiling The Zero Ping bug

[BruteForce]

Last week Unibrain participated in an interoperability and testing
event organized by the 1394 Trade Association in Taipei, Taiwan.

Unibrain was one of the four "tester" companies at the event. The
other three were Quantum Parametrics, Fraunhofer and Microsoft.

During our testing we uncovered a new hardware bug in 1394 chips/
adapters.
We call it the Zero Ping bug.

The 1394 specification supports a special type of PHY packet that can
be used to do pings. The response time you get back has sub-
microsecond accuracy (49.152MHz) and the time measurement is done by
the hardware, not the software.

So during our testing we found 3 adapters that almost always reported
all ping times as zero.

In the case of two of these adapters, if you made the adapter the root
AND put around 9.5KB of isochronous traffic per isochronous cycle
(roughly 77MB/sec) then they would report valid measurements.
The interesting part is that up to 8KB of iso traffic per cycle all
measurements were zero (there were 28 devices on the 1394 bus), at
around 8.5KB you start getting some non-zero results, at 9KB even more
and finally at 9.5KB you get valid timings for all nodes.

That's a first for Unibrain and a big success I believe!

Btw, pings are actually being used by the Serial Bus Manager in order
to optimize bus performance, so it is not a useless feature that noone
uses.

Warm Regards,
Dimitris Staikos
Unibrain