Multicore support for echo server #255
Conversation
34f6aaf to
5ecd421
Compare
5ecd421 to
e912784
Compare
e912784 to
47710c8
Compare
47710c8 to
203309d
Compare
203309d to
e2afb65
Compare
c0aa4ce to
161eaac
Compare
4e1afc5 to
69045af
Compare
161eaac to
cc0cd2c
Compare
cc0cd2c to
ad35a2b
Compare
8ea0ff8 to
9caea02
Compare
f8ab8e9 to
a982ab9
Compare
5ab5ab9 to
576e061
Compare
include/sddf/benchmark/config.h
Outdated
|
|
||
| #define BENCHMARK_MAX_CHILDREN 64 // TODO: Can we have a higher upper bound on this? | ||
| #define BENCHMARK_MAX_CORES 64 |
There was a problem hiding this comment.
We should know that CONFIG_MAX_NUM_NODES from seL4 generated configuration, why do we need a max of 64?
There was a problem hiding this comment.
Yes you are right, we do know this, the problem is that this number is used to define the size of arrays in the benchmark configuration structs, which are eventually serialised by the python meta program. To serialise correctly (leave the correct amount of empty array entries), the meta program must know the size of the C array, or CONFIG_MAX_NUM_NODES. I'm sure there is a way to give the python meta program access to this macro, but I'm not familiar with how to do it gracefully which is why it it's easier to hardcode a large number. If you can think of a graceful way to expose this macro in the meta program, feel free to change it.
There was a problem hiding this comment.
There should be a gen_config.json file in the board directory somewhere which can be imported in python.
It's possible this makes sense to expose from the microkit end of sdfgen.
It's probably a better idea to make this a future issue for sdfgen rather than forcing you to add this.
But a comment would be nice just explaining why.
This would also solve the other review comment too.
There was a problem hiding this comment.
Sure, I can add a comment.
Signed-off-by: Courtney Darville <courtneydarville94@outlook.com>
Signed-off-by: Courtney Darville <courtneydarville94@outlook.com>
|
Not sure if we need to include the license prefix for the |
You just need to add the files to here to get the CI to pass. |
|
The JSON files are really annoying for the license check since you can't have comments in JSON so you can't put the license headers there. I'll have more of a look into the license checker to see if there's anything smarter we can do to about this, it's really annoying having to add an entry into that file every time we add a new driver. The reason I don't pattern match on all |
This PR adds multicore support to the echo server example.
To build for SMP, first a core configuration file must be created associating a core value to each PD in the system. There are three example core configuration files added by this PR in the
core_configdirectory:single_core.json,two_core.jsonandfour_core.json.single_core.jsonsets each PD's core to 0, but will use the SMP microkit board.Once the core configuration
jsonhas been created, it must be passed tomakeby setting theSMP_CONFIGflag to its path. If this flag is set,makewill automatically add the_4_coressuffix toMICROKIT_BOARDwhere required. The flag is optional, and if it is not set the non-SMP microkit board will be used.Changes introduced by this PR:
benchmark.chas been restructured so that it is aware of which core it is running on for better benchmark result output.benchmarkPD signals the benchmark on the next core to propagate thestartandstopsignals.utilization_socket.cis now aware of how many cores are being utilised, and tracks the cycle counts for each active core. Thus it now has a variable number of cycle counter memory regions mapped in.meta.pyto enable additionalbenchmarkandidlePDs to be created depending on the number of active cores. This also requires that a variable number of.elfand.datafiles be created, thus these files are now created and formatted bymeta.pyrather thanecho.mk.These changes enable the build system to automatically restructure the system file and outputted elfs based on the core configuration file passed, so that no additional code changes are required.
Other changes required for this to be merged:
Note that the struct serialisation performed by
meta.pyassumes the target architecture is 8 byte aligned.