Bug: `saftd` fails to start

[anragsi]

When trying to start the demon with a Pexarria5 with a DM image, the follwoing happens:

dm_user@acopc999:~/projects/saftlib$ sudo saftd tr0:dev/ttyUSB0
configurable chunck allocator
creating allocator with  16384 chuncks of size    128
creating allocator with   1024 chuncks of size   1024
creating allocator with     64 chuncks of size  16384
libsaft-service.so is plugin
tr0:dev/ttyUSB0 is argument
OpenDevice::OpenDevice("dev/ttyUSB0")
OpenDevice first,last,mask = 0,ffff,ffff
msi_target_adr for poll check: 00024870
LM32Cluster::LM32Cluster
found 4 lm32 cpus
Error: no SDB device with VENDOR_ID=0x00000651 and DEVICE_ID=0x5f3eaf43

[anragsi]

Doesn't work neither via PCIE nor USB.

[mkreider]

Issue seems a mix of several problems

• saftd expects a serdes IO module and fails if it finds none on pexarria5 DM GW. This holds for udp, pcie and usb.
• ECA present, IO channels are deactivated

creating allocator with  16384 chuncks of size    128
creating allocator with   1024 chuncks of size   1024
creating allocator with     64 chuncks of size  16384
libsaft-service.so is plugin
tr0:dev/ttyUSB0 is argument
OpenDevice::OpenDevice("dev/ttyUSB0")
OpenDevice first,last,mask = 0,ffff,ffff
msi_target_adr for poll check: 00024870
LM32Cluster::LM32Cluster
found 1 lm32 cpus
Error: no SDB device with VENDOR_ID=0x00000651 and DEVICE_ID=0x5f3eaf43`

• If LVDS channels are completely removed from GW, saftd fails with an etherbone error. GW fails to reply to queries to ECA control unit. build verified by @alyxazon

16.7           0000000000000651:b2afc251           4000200  ECA_UNIT:CONTROL   

mkreider@acopc050:~/projects/saftlib$ eb-read dev/ttyUSB0 0x4000200/4
eb-read: etherbone cycle error: one or more operations failed

• If saftlib is built using -O0 instead of -O2 for debugging, etherbone error leads to segfault. Possible semi dirty implementation relying on compiler optimization to handle unsafe free

• current pexarria5 DM does work with saftlib, if ECA with all IO bells and whistles is synthesized (luckily achieves timing closure)

[anragsi]

Referenced version of FTM firmware without LVDS as mentioned above van be found in the following branch:
https://github.com/GSI-CS-CO/bel_projects/tree/fallout_ftm_no_lvds_no_eca_channel

[anragsi]

Different behavior between different optimization levels can be confirmed.