GitLab

Reworked the daqd stall signal test to be a python application.

See merge request cds/advligorts!165

This was done to allow the use of pyepics instead of caget to check for epics status.

Also removes the bash based test.

Work on #186 parameter sanitation for daqd and transports.

See merge request cds/advligorts!163

Adjusting the signedness of variables for overflow checks in response to review comments in producer_shmem.cc

Adding some more checks for bounds/cycle issues.

This does some more checks in the daqd producer to verify that dcu id/counts are sane as well as there being no overflows.

Updates to dix_xmit and to the daq_core files to make total models and fulldatablocksize unsigned.  This allows a check for size to skip if the send length is < 0 before writing to dolphin.

The daqd stall test often fails when it shouldn't

See merge request cds/advligorts!164

The issue is the timing of epics updates a caget.  Update the test
to put in bounded caget loops to wait for the daqd to get to a
non-stalled state and then a stalled state.

This is a test improvement, not a daqd code change.

the IOC main.c code.

Stop models from restarting over and over when the kernel module fails to load.

See merge request cds/advligorts!162

This makes awgtpman fail if the module service is not loaded, but does systemd will not attempt to start the module.

Previously if the module failed, awgtpman would eventually restart, triggering a restart of the whole model.  This would lead to cluttered logs full of errors from each restart.
This chang fixes that issue.

- Modified counting of cards found during IO mapping to account for
  Contec6464 cards counting at 2 (upper and lower word addresses).
- Code will exit on incorrect card count with message to syslog on
  type of cards not found.
- RCG modified to add REQUIRE_IO_CNT flag for IOP models with H1 or
  L1 IFO designator to automatically invoke this card checking and
  exit on card count error.

Daqd channel hash with broadcast

See merge request cds/advligorts!161

The broadcast daqds modify their channel list to mark only the broadcast channels as active/acquired.  This is what they should do, but it means a straight hash of the channel list creates a different channel hash than the other daqds.  This fix calculates the channel hash two ways, one w/o modification, and one w/ the broadcast list explicitly ignored.  This will all the broadcast daqd to declare that it has the same base channel hash (before applying the broadcast list to the channel list) as other daqds and will allow gds broadcasters to have their broadcast list checked as well.

Move the hashing from the start main handler to the channels begin/end handler.  Add a {PREFIX}CHANNEL_LIST_CHECK_SUM_BCAST pv to show the channel hash after processing for the broadcast list.  On non-broadcast daqds these will be identical to the {PREFIX}CHANNEL_LIST_CHECK_SUM pv.

Pub sub restart

See merge request cds/advligorts!160

Added DIO_TEST compile option. This was added to provide Contec6464
testing by reading back the Contec output register to verify
correct mapping and writing of data.

This is an interface to allow users to selectively restart subscriptions.  Still in testing.

With previous change for ADC/DAC mapping to allow DAC models in any order, BIO
mapping needed to change to same structure.

- Updated all code that uses the above structure.
- Fixed user space code to use dac_info routine instead of their own.

Daqd run number hash fix

Closes #171

See merge request cds/advligorts!159

It truncates the name/units.  This code adds an updated hash function that correctly uses the name/units.  It keeps the old function availble for use when a parameter is set for compatibility.

This adds {prefix}CHANNEL_LIST_CHECK_SUM as a 32bit crc of the channel list.

Fixed build_edc_ini to add newline at end of ini file if not present.

See merge request cds/advligorts!157

Daqd diagnostics

Closes #177

See merge request cds/advligorts!158

This defaults the channel to stalled and forces the system to take action to mark it as not stalled.  In a bad state we want to default to bad, not a bad state hidden because the producer wasn't run.

This adds code to the shmem receiver to have it signal when it has taken 2s to look for new data.

This is meant to be a simple flag to show when the producer is 'stuck' waiting for data.

- IOP card counting for model/bus comparison fixed. IOP code was
  double counting Contec6464 and counting IRIG-B card.
- Common module load changed to properly shutdown Dolphin if exit
  on incorrect card count by IOP.

Fixed build_edc_ini to add newline at end of ini file if not present.  Also added some simple tests for build_edc_ini.

running on Dolphin timing.

all ADC cards advertised by IOP were actually there. Example: IOP
model has 6 ADCs defined, but only found 3 on bus. Not normal in
production systems , but often the case  in test systems.

core locking, which is not really supported (lots of messages
fill dmesg). Changed this to use udelay call instead, as in
the rest of the FE code.