Ille, Ondrej, Ing. (3b28c211) at 21 Dec 16:41
src: Make width of RX Buffer pointers generic to avoid uncoverable ...
Implementation-wise, this can be done like so:
FRAME_FORMAT_W
word will get another bit at position 8 (LLBP) - Local Loopback mode.
In TXT Buffer when this bit is set, it will cause the frame to be transmitted in
Loopback mode (the same as MODE[ILBP]
). Thus, on TX, there would be
"Global loopback" (MODE[ILBP]
) and "local loopback" or "per-frame loopback" that would
be given by FRAME_FORMAT[LLBP]
.
On RX, this would indicate that the frame was received via loopback.
Ille, Ondrej, Ing. (09175a2d) at 19 Dec 15:48
src: Reduce init_vect of crc_calc to single bit. Rest is anyway dri...
Ille, Ondrej, Ing. (d422ea79) at 18 Dec 17:19
src: Change ssp* registers in ssp_generator to unsigne. Use resize ...
Ille, Ondrej, Ing. (d032e579) at 17 Dec 21:44
src: Disable Nominal bit time counters when PC FSM is off, and mr_s...
Coverage analysis towards v2.5 has shown several places where additional tests would be good. Several were written, others are pending.
Following test would be good to be written / extended:
retr_limit
, retr_limit_2
and retr_limit_3
to iterate over all available TXT Buffers.tst_mem_acc_rx
to write the memory also via March pattern, not only random data pattern. This will give us full toggle coverage on all bits on RX Buffer even if 4096 width is tested.Ille, Ondrej, Ing. (0f7862da) at 17 Dec 20:48
src: Remove txtb_unlock as a part of txtb_hw_cmd record. Re-create ...
The NVC now supports code coverage reasonably.
PSL support is still not sufficient, but code coverage is more important.
There are following options:
Simpler approach would be manipulation directly via SETTINGS[ENA].
Enabling:
When SETTINGS[ENA]=0
, and user writes SETTINGS[ENA]=1
, then SETTINGS[ENA]
becomes 1
immediately. User polls on FAULT_STATE
to get the moment in time when the unit joined
the bus.
I would not change enabling.
Disabling:
When SETTINGS[ENA]=1
and user writes SETTINGS[ENA]=0
, the SETINGS[ENA]
would stay
1 until nearest bus idle/intermission, and it would turn 0 only then. This would mean,
that if user/driver writes SETTINGS[ENA]
in the middle of frame transmission, no
error frame.
This would allow "turning off" the core also without clean-up of TXT Buffers first.
Currently, there is only single bit which enables operation of CTU CAN FD (SETTINGS[ENA]). All modes (ROM, LOM, STM, etc...) are handled by dedicated bits in MODE register, and the protection of these bits is added to disallow their change during run-time.
It would be better, to have operating mode FSM, which would be communicating with protocol control FSM, and have this FSM-state (device operating mode) exposed to user.
Right now it is unclear what is the behavior if multiple special modes are combined together.
It still needs to be re-though what will be the granularity of operating modes of the device.
This approach might be overly complex and backwards incompatible. For now, this will not be pursued.
Ille, Ondrej, Ing. (859a1495) at 17 Dec 00:25
The coverage analysis executed with VCS during verification towards v2.5 revealed potential design optimizations to reduce number of unreachable coverage bins in various blocks.
This is a tracking issue to work on these.
The main point is to optimize the code in a way where too generic design patterns cause lot of unreachable code to be generated. This leads to requirement for tedious exclude files if high coverage numbers are needed.
Following design patterns causing unreachable design were identified:
txtb_hw_cmd
has unlock
that is active together with either of failed
, error
or arbl
.
Due to the design of next state decoder in protocol control FSM, and encoding of txtb_hw_cmd
,
there are unreachable expressions and branches in the TXT Buffer FSM.crc_calc
.Ille, Ondrej, Ing. (859a1495) at 15 Dec 16:26
Merge branch '462-flip-to-vcs-flow' into 'master'
Ille, Ondrej, Ing. (859a1495) at 15 Dec 16:19
Merge branch '462-flip-to-vcs-flow' into 'master'
... and 2 more commits
ci: Do not copy the full regression results to avoid pages size overflow. Fix paths to functional coverage.
Closes #462
ci: Do not copy the full regression results to avoid pages size overflow. Fix paths to functional coverage.
Closes #462
Ille, Ondrej, Ing. (4a0a7ed5) at 15 Dec 16:18
test: Fix-up GLS compliance parameters
... and 1 more commit
The aim of this issue is to debug last model mismatches compared to ISO compliance tests.
This will need modifications of ISO compliance model and test-sequence.
All compliance tests are now debugged in master branch.