We've reversed core launch order to workaround issue of
ARConnect initialization interfere with secondary cores
startup (we don't want to workaround it in runtime as it's
only possible in case of debug session).
However it bring us new issues with the simulation run:
- mismatch arcnum (core ID) with ARConnect ID
- mismatch arcnum (core ID) with CPU name in nSIM instruction traces
To avoid these issues let's use direct core order for simulation
runs.
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Bump number of cores limit in mdb west runner.
As we are here - adjust core number in runner test closer to
the limit.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
This significantly improve user experience as 32 bit mdb binary
require to install multiple libraries before it can be used on
modern linux distros.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Add hostlink-awareness to mdb-hw runner. The mdb-nsim and arc-nsim
runners (as well as cmake scripting for nSIM boards) doesn't require
any changes.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Currently we just pass in case of the unsupported jtag adapter
is selected. Fix that by rising ValueError as we do in case of
other incorrect arguments.
While I'm at it add few missing test-cases for mdb west runners.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Currently, in case of multi-core target or in case jtag adapter
auto-selectio, we pass unexpected empty argument (empty quotes `''`)
to MDB binary due to argument generation error.
Fix that.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Remove code which looking for `cld` process pid - this functionality
isn't use anymore by twister. And as manual MDB termination is also
fixed now this functionality isn't required for anything now.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Currently we try to start MDB (MetaWare Debugger) in background
with ignored sigint no matter if we are planing to do debug
session (via `west debug`) or just run something on HW or on
nSIM (via `west flash`)
That cause really bad user experience as after we do
`west debug` or `west flash` we can't terminate the debugger
from the console. Moreover even if we terminate the debugger
process from separate terminal the console stil would be
broken so we need to do `stty sane` in it to make it usable
again.
Fix all that by proper starting the debugger process.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Make the build folder the place where MDB is run. Otherwise mdb
will store the .sc.project folder in the place where twister is
launched (so .sc.project folder will be shared across the runs)
For that we add **kwargs argument to popen_ignore_int method
so we can pass a cwd argument to the Popen constructor.
Signed-off-by: Evgeniy Paltsev <PaltsevEvgeniy@gmail.com>
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
The ARConnect Inter-core Debug Unit (ICD) provides
additional debug assist features in multi-core scenarios.
In master core(core 0) initial stage, we will program ICD to halt
all other cores based on a halt occurring in one ore more core.
And all cores are in halt mode on reset, so we need to make
sure other slave cores have launched and in running mode
before we enable ICD in master core.
Currently we launch master first, Let's reverse the launch
order, launch master last, to make sure slave cores have
launched before we program and enable ICD.
Signed-off-by: Watson Zeng <zhiwei@synopsys.com>
Runner implementations are only allowed to unconditionally import
modules in the python standard library. They are not allowed to import
anything that comes from pip or other third party sources unless they
catch the ImportError and gracefully do nothing.
Fix the imports in the mdb runner accordingly, sorting the imports
into sections to make it clearer what's stdlib, what's runners, and
what's third party while we're here.
Signed-off-by: Martí Bolívar <marti.bolivar@nordicsemi.no>
mdb binary starts several subproceses and one of them is cld process.
In runners/mdb.py we record process id of cld on each mdb launch
to terminate simulator correctly later. However we can finish test
and terminate mdb before the cld process was found (so cld won't
be terminated correctly by sanitycheck infrastructure). It may happen
if we launch mdb on fast host machine.
That leads to several issues. First of all we get ugly error in
sanitycheck output:
------------------------>8--------------------------------
FileNotFoundError: [Errno 2] No such file or directory: '/xxxx/mdb.pid'
------------------------>8--------------------------------
Secondly (and it's more important) we terminate simulator incorrectly.
We terminate mdb leaving cld process alive, running and consuming one
cpu core permanently (until we kill it manually)
So, let's increase granularity of lookups and don't wait extra 0.5
seconds before the first lookup.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Record MDB's CLD process pid to file so this process can be
terminated by sanitycheck infrastructure.
Update mdb runners test to be able to handle changes.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
mdb runner is quite special as it can be used to run Zephyr on
both simulator (nSIM) and real hardware.
However it is really misleading as same command (west flash)
will run Zephyr in simulation for one board and try to run it
on HW for another board. Things are getting worse for boards
supporting both runs in simulation and on real hardware.
Let's split mdb runner for mdb-hw (for runs on HW) and mdb-nsim
(for runs in simulation) runners.
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Require all implementations to provide a do_create(), a new
ZephyrBinaryRunner abstract class method, and make create() itself
concrete.
This allows us to enforce common conventions related to individual
runner capabilities as each runner provides to the core via
RunnerCaps.
For now, just enforce that:
- common options related to capabilities are always added, so runners
can't reuse them for different ends
- common options provided for runners which don't support them emit
sensible error messages that should be easy to diagnose and support
Signed-off-by: Martí Bolívar <marti.bolivar@nordicsemi.no>
We can just call super() instead of super(MyClassName, self). The
original extra verbosity is likely due to old habits of mine from
Python 2 which are no longer necessary, but got copy/pasted around.
Signed-off-by: Martí Bolívar <marti.bolivar@nordicsemi.no>
* add the runner script for metaware debugger(mdb).
* mdb is required for SMP case
* mdb also can provides a GUI interface
Signed-off-by: Wayne Ren <wei.ren@synopsys.com>
