Level 2 I/O Protocol

The topic here is chiefly how data gets into L2 SLIC's and MBT's. The basic answer is Cypress Hotlinks on Unshielded Twisted Pair. Both Maryland and Nevis have observed error rates less than 1 in 10^14 on the selected cable.

Here is a brief summary of the protocol. As the draft is developed, it will include the following. For current information see the FIC and MBT TDR's.

• frequency of oscillator is any 16.0 MHz .01% crystal (Spec in MBT TDR).
• list of special characters defined (in Cypress notation and Hex)
  • pad character K28.5 05
  • start event K28.0 00 (BOE)
  • end event K23.7 08 (EOE) [choice not quite final yet]
• how the links get started up
  • reframe on power-up or SCL_INIT or 2 (or more?) consecutive errors
• how self-test is initiated and terminated
  • VME commands for both; all active channels on card
• target bit error rates 1 in 10^15 is a goal, 6 in 10^14 achieved
• STP cables and connectors RJ45 and UTP (see MBT TDR)
• source responsibilities:
  • message length
  • 16 B padding required after trailer
  • header is standard L2 header, all fields filled including correct Data Type
  • trailer is standard L2 trailer (may contain 2B of longitudinal parity)
  • physical trailer of up to 2 Bytes (typically error Byte and 1B long. parity)
  • occurs AFTER padding to 16B ???
  • sequence for starting an event
    • BOE then data may follow on next byte
  • sequence for stopping an event
    • 5 pads, EOE, 5 pads
• receiver responsibilities (Below is the minimum, for "stateless" input. More complex 2-state using both BOE and EOE is allowed; see FIC and MBT TDR's)
  • receipt of pad character: recognize as non-data
  • receipt of end event
    • prepare physical trailer with parity and prepare for new event
  • receipt of start event
    • realign fifo if not done by end event
  • receipt of illegal character
    • count error
    • ignore or accept as data, depending if stateless or 2-state
  • receipt of undefined special character
    • as for illegal character
  • resynchronization
    • on multiple successive errors, begin attempt to reframe
    • if fielding an unbuffered data source, L1 busy may be raised

J. Linnemann's slides from December 98, revised January 99, close to convergence.

J-F. Renardy's table of Cypress characters 1 byte distant from control characters, which will help us think about which control characters we would like to use in L2.

---

Older Information

Muon circuits and documentation (see especially MRC2 specification and MRC1 CSR description)

SLIC proposed circuit and top and bottom layout with (Coax so far)

A proposed state machine by Mike Fortner for receiving Cypress inputs assuming special characters for begin, end

• There are only 2 long-lived states, IDLE and EVENT (whether you are between events or in one)
• the 2 states allow you to ignore bad characters between events, which is most of them
• getting unexpected begin/end are handled assuming it is more likely to miss a special character than for noise to give you a specific character
• the extra detail in the diagram shows details of handling data, noise, and special characters in these contexts
• a separate state machine would handle reframing

notes from the FIC and L2 protocol meetings of September 1998.

J-F. Renardy's presentation on the FIC and protocol questions

J. Linnemann's (unpresented) slides on protocol questions from September 98.

---

From the G-link, which is the input to the FIC, all the issues appear again. The source is the L1CFT system. Currently we refer to the L1CFT TDR Draft. Latest information may be found from Manuel's latest new presentation (which may still not be fully little-endian in data details). The FIC TDR also has relevant detail.

Also of interest are G-link resynch measurements by the Barsotti group and their resynch protocol (see G-link Usage) for a unidirectional link.

---