Tags:
view all tags
THIS MANUAL IS INCOMPLETE but what is written here should be more accurate and up to date than the old "functional specification". Please e-mail me if you find mistakes/omissions here (Eric H).

See Also

Contents

DAQ Outputs

The AMC13 implements the S-Link Express protocol using a core provided by CMS CDAQ. This is currently a 5.0Gb/s link, received by the FedKit (see CMSFedkitManual) or by an AMC13v1 (Virtex-6 version only) running special firmware version 0xff.

The AMC13 has 3 SFP transceivers which may be used for DAQ. There are 3 possible configurations, determined by the value written to CONF.SFP.ENABLE_MASK. In addition, CONF.EVB.ENABLE_DAQLSC must be set to '1'. The AMC13Tool2 command daq may be used to accomplish this configuration.

Configuration Mask value Notes
SFP0 only 1 (top) DAQ fiber for AMC1-AMC12 readout
SFP0, 1 3 SFP0 for AMC1-AMC6, SFP1 for AMC7-AMC12
SFP0, 1, 2 7 SFP0 for AMC1-AMC4, SFP1 for AMC5-AMC8, SFP2 for AMC9-AMC12

After changing the DAQ output configuration the logic and high-speed transceivers must be reset by writing to ACTION.RESETS.DAQ (AMC13Tool2 command rd).

The DAQ sender firmware itself was provided by the CMS cDAQ group. Links to the firmware in the CERN SVN and it's documentation may be found here:

TTS

The TTS (Trigger Throttling System) output from the AMC13 is a four-bit code transmitted over the transmit half of the bottom SFP fiber transciever, and is normally sent to the TCDS system to control the trigger rate in response to pending buffer overflows. This system is intended to be logically compatible with the legacy system documented HERE.

The AMC13 outputs the TTS state as the four-bit code described in the link above:

0000 Disconnected Hardware Failure or broken cable
0001 Overflow Warning Imminent buffer overflow
0010 Sync Lost AMC13 is not synchronized with DAQ due to buffer overflow
0100 Busy Cannot accept triggers
1000 REady Ready to accept triggers
1100 Error Any other state that prevents functioning
1111 Disconnected Hardware failure or broken cable

Any time the AMC13 is not in run mode (such as after power up) the AMC13 sends state "0100" (busy).

Internally the AMC13 manages the TTS state using a 5 bit internal format; this is exposed in a few monitoring registers. This format should be decoded as text in the latest AMC13 software, but is documented here for completeness:

Bit 4 Disconnected
Bit 3 Error
Bit 2 Sync Lost
Bit 1 Busy
Bit 0 Overflow Warning

(A value of 0 means "ready")

The TTS state is sent over the same 5.0Gb/s link as the DAQ data. There's an input port in our link logic on the AMC side: 

-- TTS port
    TTSclk            : in  std_logic;  -- clock source which clocks TTS signals
    TTS               : in  std_logic_vector (3 downto 0);

When the TTS state changes, a control word is sent across the link to transfer the information with minimum latency to the AMC13.

Here are some more details on how the TTS state is generated in the AMC13:

The AMC13 contains an L1A FIFO into which L1A are stored when received by the TTC or generated by the internal L1A generator. The event builders (up to 3) read L1A from the FIFO and wait for available event fragments from each enabled AMC input. Then the event is built and sent out the corresponding DAQ link and/or stored in the SDRAM monitor buffer. The AMC13 internal TTS state is generated exclusively based on the number of L1A stored in the L1A FIFO.

Additionally, up to 12 TTS states are received as described above from AMC cards. The final TTS output state is simply a priority encoding of the 12 AMC states plus the local AM13 one.

Transition FIFO level
RDY->OFW 96
OFW->RDY 63
OFW->BSY 224
BSY->OFW 223
BSY->SYN 225

TTC Simulator

The AMC13 has the ability to generate simulated TTC signals and distribute them to AMC cards in the crate. This allows operation of a stand-alone test setup with only an AMC13 and AMC cards in a single crate without requiring any external TTC hardware. The simulated TTC signal will always include a BC0 sent once per LHC orbit. Four "BGO" channels are provided which can send programmed short or long format TTC commands either once under program command or periodically. In addition, ECR (event count reset) and OCR (orbit count reset) TTC commands may be sent by writing to the ACTION.LOCAL_TRIG.SEND_ECR and ACTION.LOCAL_TRIG.SEND_OCR registers, respectively.

N.B. the OCR and ECR mentioned above will not be reflected in the AMC13 registers until after the next L1A, as the current EvN and OrN in the AMC13 are not visible; they are used only to stamp an event in in response to L1A.

The "BGO" channels are programmed using registers 0x24-0x27 (CONF.TTC.BGOn). Each of the four channels requires the following settings, where BGOn is BGO0, BGO1, BGO2 or BGO3.

Register Function
CONF.TTC.BGOn.COMMAND Short (bits 0-7) or long (bits 0-31) format TTC command
CONF.TTC.BGOn.LONG_CMD Bit '1' for long-format command, '0' for short format
CONF.TTC.BGOn.ENABLE_SINGLE Bit '1' to enable single command (trigger with ACTION.TTC.SINGLE_CMD) [1]
CONF.TTC.BGOn.ORBIT_PRESCALE Orbit prescale (prescale is value + 1)
CONF.TTC.BGOn.BX Bunch crossing number on which to send command

[1] Only one of four ENABLE_SINGLE may be set at one time. If bit is set to 0 the commands are sent periodically

Locally-generated triggers may be sent in a rather flexible way. See the next section for details.

The simulated TTC function requires a clock which is transmitted on the output side of the TTC SFP optical transceiver and must be received on the input side, so a short loop-back cable must be plugged between the input and output sides of the bottom SFP.

This feature is enabled by setting CONF.DIAG.FAKE_TTC_ENABLE to 1. The local L1A generator must also be enabled (CONF.TTC.ENABLE_INTERNAL_L1A set to 1).

TTC Command Details

The AMC13 can transmit and take action on several specific TTC commands.

Command Default Value Programmable? Notes
BC0 1 No Bunch Count Reset (sent every orbit)... CMS standard?
EC0 3 No[1] Event Count Reset (send at the start of each lumisection to set the EvN to 1)
OC0 9 Yes Orbit Count Reset - reset orbit count to 0
Resync 0x28 Yes Resync after error

Note that for the programmable commands, there is both a command value register (8 bits) and a mask value register (8 bits) which allows only specific bits to be matched when decoding a received command.

[1] This should be programmable

These are taken originally from this table maintained by HCAL:

Local L1A Generator

This feature allows the AMC13 to generate L1A and transmit them over the TTC backplane signals to AMC cards. It may be used in conjunction with the TTC Simulator described above, or with an external TTC input.

There are 4 modes of operation available:

  • Individual triggers under software control
  • Burst with count and spacing in BX or orbits specified
  • Continuous triggers equally spaced by BX or orbits
  • Random triggers from 2Hz to about 130kHz rate with CMS trigger rules respected

Various registers are used to control the local L1A generator. The easiest way to control this feature is using the method AMC13::configureLocalL1A() in the AMC13 class. Then, call AMC13::sendL1ABurst() to send a single software-triggered burst, or AMC13::startContinuousL1A() and AMC13::stopContinuousL1A() to start or stop continuous triggers.

AMC13::configureLocalL1A( bool ena, int mode, uint32_t burst, uint32_t rate, int rules) documentation:

Parameter Description
ena true to enable the L1A generator
mode 0 - periodic triggers spaced by rate orbits at BX=500
1 - periodic triggers spaced by rate bx
2 - random trigger at 2* rate Hz
burst number of triggers in a burst (1-4096)
rate sets the rate based on mode (1-65536)
rules set to 0 normally to enforce the "standard" CMS trigger rules

Brief register-level documentation follows. Register at offset 0x1c controls the local L1A generation through the following bit fields:

  • CONF.LOCAL_TRIG.RATE sets the rate or spacing (0 means spacing=1)
  • CONF.LOCAL_TRIG.NUM_TRIG sets the burst count (0 means count=1)
  • CONF.LOCAL_TRIG.TYPE is the mode (0 for orbit, 2 for BX, 3 for random)
  • CONF.LOCAL_TRIG.RULES specifies which CMS trigger rules are followed

Rule 1 is always enforced. The rules parameter to AMC13::configureLocalL1A or the CONF.LOCAL_TRIG.RULES item may be set as follows to suppress other rules:

  • 0 means enforce all rules (1-4)
  • 1 means all except rule 4
  • 2 means enforce rules 1 and 2
  • 3 means enforce only rule 1

TTC History Capture

This feature added to the T2 (Spartan) firmware starting in version 0x26 allows the capture of up to 512 TTC short-format broadcast commands in a buffer. The commands may originate in the AMC13 itself (if the TTC simulator is being used) or received externally on the TTC fiber input.

A filter feature is provided which checks incoming commands against a list of up to 16 entries, and if a match is found the command is discarded rather than being stored in the history.

Each filter item has the following 3 fields: 

  bits 0-7   TTC command value to match
  bits 8-15  Mask applied before match.  '1' to ignore specified bit
  bit 16     This filter item is enabled if '1'

Several C++ functions are provided which are briefly listed below. See the nightly API documentation for details. 

    void setTTCHistoryEna( bool ena);                      // enable/disable history capture
    void setTTCFilterEna( bool ena);                       // enable/disable history filter
    void setTTCHistoryFilter( int n, uint32_t filterVal);  // set individual filter item
    uint32_t getTTCHistoryFilter( int n);                  // get individual filter item
    void clearTTCHistoryFilter();                          // clear entire filter list
    void clearTTCHistory();                                // clear capture history (reset count)
    void getTTCHistory( uint32_t* buffer, int nhist);      // get TTC history list (READ DOCS)
    int getTTCHistoryCount();                              // get TTC hsitory count

Local Trigger Logic (DT)

The following section describes a local trigger implemented for the DT group according to the following specification: DT AMC13 requirement (rev 2015-06-05). This trigger logic is contained entirely in the T2 (Spartan) FPGA and is introduced in version 0x29.

This trigger uses Fabric B inputs from 12 AMC modules and TRIG0 and TRIG1 from special T3 board as inputs to a 14-bit Look Up Table to generate a trigger at every TTC clock cycle.

To align the trigger inputs, there is an 8 bit delay line at each trigger input. The unit of the delay is one eighth of a TTC clock cycle. To help adjusting the delays, there is a sampling buffer of 14-bit and 1024 deep which samples the delayed input trigger at eight times of the TTC clock frequency.

Before using the trigger, delay adjustment is necessary.

  • First, write 1 to register 0x101 to enable the trigger.
  • Second, fill the LUT with 0xffffffff except 0x200 which should be loaded with 0xfffffffe. This results a trigger of simple OR of all fourteen trigger inputs.
  • Then write 1 to register 0x100 to enable the sampling.
  • After that send a signal to all fourteen trigger source so that LUT will receive trigger from all of them.
  • Read out the sampling buffer and first adjust the three LSB of the input delay so that the trigger will be recorded with the same seven MSB of the sample buffer read address. (assuming the input trigger signal is 25 ns wide, otherwise the trigger should be centered in the bins at least), this ensures the LUT clock edge is always optimally centered.
  • Next adjust the seven MSB of the input delay so that all input trigger have the same seven MSB address of the sample buffer.
This calibration should be repeated whenever possible to correct for possible timing drift due to temperature/voltage changes.

LUT trigger uses registers in the range of 0x100-0x10f and 0x200-0x7ff. The bits are numbered LSB-MSB within each 32-bit word, and thus may be treated as a single vector of 16384 bits. The address within this vector is formed using a 14-bit address as follows:

LUT Address Bit 13 12 11 ... 1 0
Input TRIG1 TRIG0 AMC12 ... AMC2 AMC1

This feature is controlled by the following registers on the T2 board (so use the writeT2 or ws commands in AMC13Tool2.exe).

Register Name Bits Function
CONF.DTTRIG.ENABLE 1 Enable the trigger (1) or disable (0)
CONF.DTTRIG.AMC_DELAY_00 8 Set delay for AMC1 (sorry, 0-based numbering!)
CONF.DTTRIG.AMC_DELAY_11 8 Set delay for AMC12 (sorry, 0-based numbering!)
CONF.DTTRIG.TRIG0_DELAY 8 set delay for TRIG0 input
CONF.DTTRIG.TRIG1_DELAY 8 set delay for TRIG1 input
CONF.DTTRIG.SAMPLE_BUFFER_ENABLE 1 Start capture of trigger inputs for time alignment
CONF.DTTRIG.LUT 32x512 Look-up table with 2^14 bits
STATUS.DTTRIG.SAMPLE_BUFFER 14x1024 14-bit capture buffer with 1k words

Local Trigger Logic (HCAL)

The following section applies only to the HCAL firmware series (Kintex v0x4000 and up). A local trigger may be formed from 8 bits supplied each BX from each AMC card. There are a total of 8 independent logic triggers which are evaluated every BX and output on an optical fiber at 1.6 Gb/s (actually the TTC clock times 40) with 8b10b encoding. (Fabric B is not used because the HCAL uHTR did not connect it!)

Each of the 8 individual logic triggers works as follows:

  • Apply a mask to each of 8 bits from each of 12 AMCs (96 bits programmable). A '1' bit disables the corresponding input
  • Count the number of non-zero AMC bytes after masking (result is 0-12)
  • Apply a programmable threshold to this value, producing a '0' or '1' resut

So there are a total of 96 * 8 programmable mask bits and 8 programmable 0-12 thresholds.

Output format (to be confirmed):

Byte Use
0 Comma character (0xBC k-char)
1 BX0 [7] VER[6:4]=1 BX ID [11:8]
2 Bx ID [7:0]
3 local trigger word

The registers which control this feature are as follows:

Name Use
CONF.LTRIG.AMCxx.BITy.TRIGGER_MASK Set 8-bit mask for AMC number xx (01-11) trigger bit y (0-7)
CONF.LTRIG.BITy.TRIGGER_THRESHOLD Set 4-bit threshold (0-12) for trigger bit y (0-7)

External Clock / Trigger Inputs (g-2)

The following section applies only to the g-2 firmware series (Kintex 0x8000 and up). The AMC13 clock and external trigger can come from one of 3 sources (optical fiber, internal or external copper signal). The choices are enumerated in the following table.

Clock Source Trigger Source Hardware T1 0x1 bit 15 T1 0x1 bit 8 T1 0x1 bit 2
Internal Internal Loop-back fiber (Note 1) 0 1 1
Fiber Internal Fiber with TTC input 0 0 1
Lemo Internal Special T3 with clock input 1 0 1
- - Not used 1 0 0

Notes:

  1. SFP transceiver with loop-back fiber from Tx to Rx must be installed in bottom site

HCAL Orbit Gap Calibration

The AMC13 implements several features to facilitate triggers for calibration purposes during the LHC "Orbit Gap" during which no normal L1A should occur. The details of this have to a certain extent been lost in time.

See 2009 CMS Note by Jeremy et al about this. Here is a table of TTC command used by HCAL extracted from the document.

Code Name Source MeaningSorted ascending AMC13 Action
10000000 (0x80) Gap-Sequence-Step BGo-13 Advance laser Increment current laser position
00001001 (0x09) BCZero BGo-1 Bunch Counter Zero (any with LSB set) BcR
11100010 (0xe2) ECR BGo-7 Event Counter reset ECR (programmable code)
01000000 (0x40) Gap-Trigger BGo-13 Laser/LED in next gap Enable gap trigger this orbit only
01100000 (0x60) Gap-Pedestal BGo-13 Pedestal in next gap Enable gap trigger this orbit only
00000100 (0x04) SOG BGo-11 Start-of-Gap (QIE Reset) -none-
10001000 (0x88) Start BGo-9 TPG generator start -none-
10101000 (0xa8) Stop BGo-10 TPG generator stop -none-

Wu's Debugging Guide 

AMC13 quick trobleshooting with register dump
	Last updated on 3/19/2015

Following description is accurate only for T1 versions
0x4020, 0x225, T2 version 0x27 and later.
Also make sure bit 11-0 of T1 reg 0x5 are all 0, any bit
set to 1 indicates that AMC module has a different backplane
link version as that of the AMC13 T1 firmware.

T1 version is bit 31-16 of T1 reg 0x1
T2 version is bit 15-0 of T2 reg 0x0

a)Keep firmware up to date
  Always check for the latest firmware and upgrade
your system. New versions are released to fix bugs or
adding debugging information, so it is important to
keep your firmware up to date. If you have problems,
upgrade to the latest version and see if that solves
your problem.

b)TTC problems
  Once set up right and TTC works correctly, check regularly
  the following T2 registers:
  0x7 counts bcnt errors, it should have no more than couple
      of counts.
  0x8 counts TTC single errors, it should have no more than couple
      of counts.
  0x9 counts TTC multiple errors, it should have no more than couple
      of counts.
  If TTC does not work, 
    first check T1 reg 0x4:
    if bit0 is 1, TTC optical receiver is absent.
    if bit7 is 1, there's no TTC input signal, check the
                  cabling to TTC source. 
  If AMC13 registers look OK, but AMC modules have TTC
  problem, make sure your TTC decoder has the right timing.
  AMC13 output TTC clock's edge is in the middle of the TTC
  data on the backplane.

c)run stopped because of AMC13
  If run stopped and bit 15-12 of T1 reg 0x19 is not 0x8, check T1 registers
  0xe1a, 0xe1b and 0xe1c. If any of them is non-zero, at least one AMC is
  causing the problem. Each AMC uses one byte, AMC1 using bit7-0 of reg 0xe1a
  and AMC2 using bit15-8 of reg 0xe1a and AMC3 .... The definition of the byte is
  bit7 if set, AMC has been in disconnected state
  bit6 if set, AMC has been in error state
  bit5 if set, AMC has been in out of sync state
  bit4 if set, AMC is in disconnected state
  bit3 if set, AMC is in error state
  bit2 if set, AMC is in out of sync state
  bit1 if set, AMC is in busy state
  bit0 if set, AMC is in overflow warning state

  If AMC13 is in overflow warning or busy states, first check T1 reg 0x0. If bit0
  is set, cDAQ is down. Otherwise, check T1 reg 0xd4, if bit 18-16 are not all 0.
  cDAQ full stopped sending data out. If cDAQ is neither down nor full, and bit 10-8
  are all 0, then it is event builder not building events. Next check T1 register
  0xe0c, if any bit of bit 11-0 is not 0, the corresponding AMC(bit 0 AMC1) has no
  data.

d) data integrity problems
  T1 register 0xb3-0xb5 counts event cmsCRC error for SFP0,SFP1 and SFP2
  T1 register 0xb6-0xb8 counts event length error for SFP0,SFP1 and SFP2
  T1 register space 0x800-0xdff are monitoring counters for AMC modules,
  each AMC module occupies 0x80 32 bit space, AMC1 uses 0x800-0x87F.
  following registers' address is offset address in their own space. Each counter
  occupies two 32 bit space. Even address is the lower 32 bits and odd address is
  the upper 16 bits of the counter.
  offset 0x6-7 is event number of the event mismatch counter
  offset 0x8-9 is Orbit count of the event mismatch counter
  offset 0xa-b is BC count of the event mismatch counter
  offset 0x12-13 is bad EventLength counter
  offset 0x14-15 is trailer Evn mismatch counter
  offset 0x1e-1f is link input Evn skip counter
  offset 0x3a-3b short event at input counter(less than three 64bit words)
  offset 0x3c-3d number padded words for short event
  (offset less than 0x40 are from the backplane link module built inside the AMC module.)
  offset 0x6a-6b is the same as offset 0x6-7, but counted inside AMC13
  offset 0x6c-6d is the same as offset 0xa-b, but counted inside AMC13
  offset 0x6e-6f is the same as offset 0x8-9, but counted inside AMC13
  offset 0x70-71 is the same as offset 0x12-13, but counted inside AMC13
  offset 0x78-79 is bad AMC event CRC counter
  offset 0x7a-7b is TTS state is error counter
  offset 0x7c-7d is TTS state is out of sync counter
  offset 0x7e-7f is TTS state is disconnect counter

  If all these counters are 0, there is no data integrity problem detected.

e) Other T1 registers containing important run information
  0x46 number of L1A received
  0xba	low word of SFP0 sum of event length from CDF trailer
  0xbb	bit 55-32 of SFP0 sum of event length from CDF trailer
  0xbc	low word of SFP1 sum of event length from CDF trailer
  0xbd	bit 55-32 of SFP1 sum of event length from CDF trailer
  0xbe	low word of SFP2 sum of event length from CDF trailer
  0xbf	bit 55-32 of SFP2 sum of event length from CDF trailer
  0xc0	SFP0 built event count
  0xc1	SFP1 built event count
  0xc2	SFP2 built event count
  0xc4	SFP0 built event word count(lower 32 bit)
  0xc5	SFP1 built event word count(lower 32 bit)
  0xc6	SFP2 built event word count(lower 32 bit)
  0xc8	SFP0 built event block count
  0xc9	SFP1 built event block count
  0xca	SFP2 built event block count

  in the range of 0x800-0xdff, for each AMC module:
  offset 0xc-d  number of events received at link input
  offset 0x18-19  number of words received at link input
  offset 0x40-41  number of words received by AMC13 from AMC module
  offset 0x52-53  number of events received by AMC13 from AMC module
  offset 0x72-73  number of event blocks received by AMC13 from AMC module

f) Monitor buffer can buffer up to 0x400 events/blocks, each buffer occupies 
   0x20000 32 bit words.
  you can read any buffer using the following command:
  rv [starting address] [length]
  where starting address = 0x8000000 + (offset x 0x20000)
  e.g. the starting address of the first buffer is 0x8000000
  and the starting address of the second buffer is 0x8020000, etc.
  maximum offset is 0x3ff and the length is in 32 bit words.

Please email comments and suggestions to
wusx@bu.edu with subject as amc13debug

Monitoring Registers

This section provides documentation on a few of the monitoring registers which have proven to be confusing. Eventually there should be detailed description for all registers, but who knows when this will get done! 


AMC_EvN_Mismatch (offset 0x6/0x7)
 This checks that the EvN in bits 32-55 of the first word sent by the AMC
 Matches the current EvN in the AMC13 (reset to 1 on EcR, increment each L1A)
 This check is performed in our link firmware in the AMC card.

AMC_Trailer_EvN_Bad
 This checks if the 8 bits of EvN in the last word bits 24-31 match bit 32-39
 of the first word.

AMC_EvN_Errors
 This check if the EvN supplied by the AMC increments by one each L1A

AMC13_EvN_Mismatch
 This is essentially the same as AMC_EvN_Mismatch except that the check
 is performed in the AMC13 firmware.
-- EricHazen - 23 Mar 2015
Edit | Attach | Watch | Print version | History: r46 | r26 < r25 < r24 < r23 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r24 - 08 Sep 2015 - EricHazen
 
  • Edit
  • Attach
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback