[GSoC 2020] : BSP Buildset for EPICS (next steps)

Sat Aug 8 17:40:33 UTC 2020

On Sat, Aug 8, 2020 at 11:08 AM Mritunjay Sharma
<mritunjaysharma394 at gmail.com> wrote:
>
>
>
> On Sat, Aug 8, 2020 at 9:46 PM Heinz Junkes <junkes at fhi-berlin.mpg.de> wrote:
>>
>> Hallo Mritunjay,
>> everything looks pretty good. I'm commenting on the text. I also send this mail
>> to two EPICS experts (Andrew Johnson and Michael Davidsaver). Maybe they also have some ideas.
>
>
> Thank you so much Heinz! It will be really a great help!
>>
>>
>>
>> >
>> > Current Status:
>> >
>> > 1) Successfully built EPICS7 with RTEMS5 by hand for pc-386
>> > 2) Worked for RSB recipe.
>> >    In its due process, I Wrote:
>> >     i) rsb/rtems/config/epics/epics-7-1.cfg
>> >     ii)rsb/rtems/config/epics/epics-base.bset
>> >     iii)rsb/source-builder/config/epics-7-1.cfg
>> > 3) Added Patch for RTEMS-pc-386 support which made the above recipe work successfully.
>> > 4) Therefore, Successully built EPICS7 with RTEMS5 by using RSB recipe as well for pc-386 as of now.
>> > 5) Sent 4 Patches for review of the same.
>> >
>> > What problems are in the next steps?
>> >
>> > 1) How to make it work across different architectures?
>> > 2) Exisiting EPICS works on the old legacy network stack.
>> > 3) I am not using EPICS upstream branch. It is being built
>> > by Heinz's epics playground.
>> > 4) Doubts in how to start with testing.
>> >
>> > My Resarch work for the Problem no: 1
>> >
>> > I have gone through the EPICS developer guide from here
>> > exhaustively in the past couple of day and here are few interesting things
>> > that I found which can help:
>> >
>> > 1) "The main ingredients of the build system are:
>> > • A set of configuration files and tools provided in the EPICS base/configure directory
>> > • A corresponding set of configuration files in the <top>/configure directory of a non-base <top> directory
>> > structure to be built. The makeBaseApp.pl and makeBaseExt.pl scripts create these configuration files. Many of
>> > these files just include a file of the same name from the base/configure directory.
>> > • Makefiles in each directory of the <top> directory structure to be built
>> > • User created configuration files in build created $(INSTALL_LOCATION)/cfg directories.
>> > "
>> >
>> > Remarks: Now since it is also mentioned in the guide that "makeBaseApp.pl
>> > creates directories and then copies template files into the newly created directories
>> > while expanding macros in the template files. EPICS base provides two sets of template files: simple and example."
>> > Can we think of using makeBaseApp.pl to that end? Making the user allow
>> > to change the configurations from the terminal?
>>
>> I don't think that makeBaseApp.pl will help you. This is intended to build an example IOC. It takes the settings from the above mentioned configuration files.
>>
>> You "only" need to specify the location of the RTEMS installation in configure/os/CONFIG_SITE.Common.RTEMS.
>> RTEMS_VERSION =
>> RTEMS_BASE =
>>
>> Then you have to define the target in configure/CONFIG_SITE:
>> ...
>> # Which target architectures to cross-compile for.
>> #  Definitions in configure/os/CONFIG_SITE.<host>.Common
>> #  may override this setting.
>> CROSS_COMPILER_TARGET_ARCHS=
>> …
>> e.g. “CROSS_COMPILER_TARGET_ARCHS=RTEMS-xilinx_zynq_a9_qemu"
>>
>> And for each target there must be a file in configure/os:
>> e.g. CONFIG_Common.RTEMS-xilinx_zynq_a9_qemu
>> If it is not provided by EPICS, the RSB should install it there (adapted to the target to be used by epics make).
>

Probably they should be provided by EPICS for known-to-work
configurations. If they are not, we should push upstream.

>
> Yes, Heinz. I followed the above steps and created a patch which I applied in the configuration files of RSB recipe.
> The problem with it is that it's made only for pc-386 and I have to hardcode there about location of the RTEMS installation in
> configure/os/CONFIG_SITE.Common.RTEMS. My doubt is how to modify the patch that can it offer user-specific location of the RTEMS
> installation and bsp?
>>

I still think this should be done through some kind of pre-processing
(scripting) over these configuration files for a given target, using
some fixed pattern, rather than by patching. A patch is fine for
proof-of-concept, but I don't think we want to have x patches for x
BSP targets of EPICS.  Maybe it is fine, there aren't that many BSP
targets right now, but I can see this kind of patch-based
configuration getting a little unwieldy.

If you proceed with the patch-based approach, you need to figure out
how to pick the right patch to apply based on the target/bsp build. So
that would be your next step. Create a patch for the zynq, and see if
you can dynamically determine which one to apply (zynq or pc386) based
on RSB internal state/variables.

>>
>> >
>> > 2) "The startup directory in EPICS base contains a perl script, EpicsHostArch.pl, which can be used to define
>> > EPICS_HOST_ARCH. This script can be invoked with a command line parameter defining the alternate compiler (e.g.
>> > if invoking EpicsHostArch.pl yields solaris-sparc, then invoking EpicsHostArch.pl gnu will yield
>> > solaris-sparc-gnu).
>> > The startup directory also contains scripts to help users set the path and other environment variables”
>> This has nothing to do with 2)
>
>
> I am sorry for the misunderstanding. All the 4 points mentioned here are my observations only for the Problem No.1
> `1) How to make it work across different architectures?`
>>
>>
>> There's no need to adjust anything here. The EPICS make recognizes the architecture on which it is started.
>>
>> > Remarks: As EPICS_HOST_ARCH, can we do something similar for CROSS_COMPILER_TARGET_ARCHS?
>> >
>> > 3) ") The following is a summary of targets that can be specified for gnumake:
>> > • <action>
>> > • <arch>
>> > • <action>.<arch>
>> > • <dir>
>> > • <dir>.<action>
>> > • <dir>.<arch>
>> > • <dir>.<action>.<arch>
>> > where:
>> > <arch> is an architecture such as solaris-sparc, vxWorks-68040, win32-x86, etc.
>> > <action> is help, clean, realclean, distclean, inc, install, build, rebuild, buildInstall, realuninstall, or uninstall"
>> >
>> > Remarks: Now similar to the above stated, can we work for Cross Compiler target Architecture?
>> >
>>
>> But this does not refer to 3) ?
>
>
> No no, this remark is also for the problem 1 only as told above. Slight misunderstanding here :)
>>
>>
>>
>>
>> 3) You have to take "my" repo at the moment, because the adaptations to RTEMS5 are not yet included in the official epics-base. This is a hen-and-egg problem because RTEMS is only now in the release phase, so my changes have not been implemented yet.
>
>
> Ok, I hope Dr. Gedare and Chris can help you with that.

We just need to be ready for when RTEMS 5.1 is officially released.
Hopefully soon, but I don't have a timeline. Releases are mostly
volunteer time, so they happen when they happen. We're trying to get
better about that, but it is hard (due to lack of incentives).

>>
>>
>> 2) I'm just about to figure out in the Epics Makefiles whether the target was built with the legacy-stack or libbsd-stack. It's working already.
>> Now I also have to adjust the rtems_init.c accordingly. Here I have to clean up a little bit. But I hope to have this finished by the middle of the week.
>
>
> Thank you so much for the update!
>
> So I would like to ask my other mentors - what can I do for the time being? What should be the next steps for this week?

Prepare the zynq patch and try to work out the logic of how to select
the right patch to apply.

Then similar logic might be usable to script the configuration changes
of EPICS so we don't need patches.

> And yes how to begin the testing part?

This was suggested by Heinz earlier to look at the CI test scripts
that EPICS maintainers use.

> I have tried to find some resources but I think it will be
> better if you can help somewhere to look at.
>
> Thanks
> Mritunjay Sharma
>>
>>
>> > These were the little doubts that originated from the research work I did.
>> > I will like the opinion of mentors that what can be the optimal way now to approach the
>> > project after this? What can be some resources for better research work of the
>> > above problems?
>> >
>> > Also, for the reference:
>> > Link to the changes in commits of rsb can be found here: https://github.com/RTEMS/rtems-source-builder/compare/master...mritunjaysharma394:epics-support
>> >
>> > The patch for epics can be found here: https://github.com/mritunjaysharma394/epics-mritunjay/tree/master/patches
>> >
>> >
>> > Thanks
>> > Mritunjay Sharma
>> >
>> >
>> >
>> >
>> Heinz