[PATCH 00/10] Raspberry Pi 2 (BCM2836) SMP support

[Pavel Pisa]

Hello Alan,

On Sunday 04 of September 2016 16:38:36 Alan Cudmore wrote:
> Hi Pavel,
> I applied your patches, and verified that my apps still work on the
> Raspberry Pi 1 ( Pi Zero ). Now I am moving to the Pi 2 for some tests.
>
> When building for the Pi 2, I need to use the —enable-smp configure switch,
> correct?

Yes. There is even added configure option RASPBERRYPI_CPUS which controls
number of CPUs reported trough linker script to the application.
The default is all four (4) when --enable-smp is set. If it is not set
then it is set to 1. I have not solved misconfiguration which can
be caused by selection --enable-smp for RPi1. It fails horribly
because some peripherals referenced by SMP code are not available.

> When I build the raspberrypi2 BSP with —enable-smp and —enable-tests, the
> smpfatal08 test does not compile. It has multiple definition errors like
> the following:
> ../../../../../raspberrypi2/lib/librtemsbsp.a(libbsp_a-bspsmp.o): In
> function `_CPU_SMP_Get_current_processor':
> /Users/alan/rtems/build/build-pavel-tmp/arm-rtems4.12/c/raspberrypi2/lib/li
>bbsp/arm/raspberrypi/../../../../../.././raspberrypi2/lib/include/rtems/scor
>e/cpu.h:522: multiple definition of `_CPU_SMP_Start_processor'
> init.o:/Users/alan/rtems/build/build-pavel-tmp/arm-rtems4.12/c/raspberrypi2
>/testsuites/smptests/smpfatal08/../../../../../../../../rtems-src/rtems-pave
>l/c/src/../../testsuites/smptests/smpfatal08/init.c:60: first defined here

I have tried complete tests build and this one fails for me as well.
I have removed references to smpfatal08 from lists in the file

  arm-rtems4.12/c/raspberrypi2/testsuites/smptests/Makefile

The test seems to check cpukit core by replacing BSP provided
CPU management functions by some dummy/test ones and for given
setup the original ones are pull in together with some another
symbol and result is clash.

Then all other tests build OK. I have tried many of them. I have
noticed only one failure (testsuites/smptests/smpmutex01)

*** BEGIN OF TEST SMPMUTEX 1 ***
../../../../../../../../../git/rtems/c/src/../../testsuites/smptests/smpmutex01/init.c: 184 expected == actual

Has that problem been spotted on Zynq or Altera too?

> But I tried several of the SMP tests that did compile before this one and
> they are working on my Raspberry Pi 2!
>
> When I build with just the samples, rather than the full test suite, the
> build completes, and I can link and run my application just like the Pi.

If you build own application then you need to to include something
like

  #define CONFIGURE_SMP_MAXIMUM_PROCESSORS  2
  #define CONFIGURE_SCHEDULER_PRIORITY_AFFINITY_SMP

before

  #include <rtems/confdefs.h>

to configure application to actually use more CPUs. Example is to use
only two. If you select limit above 4 then it is corrected by BSP.

I have prepared simple OMK example which runs two tasks on two CPUs

  https://rtime.felk.cvut.cz/gitweb/rtems-devel.git/tree/refs/heads/master:/rtems-omk-template/appsmptest

I used it during SMP mailboxes and timer debugging.

By the way, have you Raspberry Pi 3?
I have not tested that but there is some chance that RPi2
build could run on RPi3 unmodified for 32-bit mode.
According to scattered information it seems that
there are no differences in base addresses of critical
peripherals. But may it be that there would be some problem
in core code support or elsewhere.

Best wishes,

              Pavel