PowerPC PPC32 CONFIG_HIGHMEM dependency question

PowerPC PPC32 CONFIG_HIGHMEM dependency question

[Robert Johnson]

I am using an embedded single board computer (SBC) that has a Freescale P2020 processor.  The vendor board support package (BSP) for this product currently uses a 3.0.4 Linux kernel.  In looking through the RT patch for various Linux kernel versions even up through 3.12.13-rt21, I see that CONFIG_HIGHMEM is not allowed if CONFIG_PREEMPT_RT_FULL is selected (see arch/powerpc/Kconfig).

The result for my SBC is that if I enable CONFIG_PREEMPT_RT_FULL in the kernel, I am then only able to access 1GB of RAM instead of the 4GB that is resident.

Is there any way to get CONFIG_PREEMPT_RT_FULL to work with CONFIG_HIGHMEM?

If I want to use CONFIG_HIGHMEM, should I consider the use of CONFIG_PREEMPT_RTB which does not appear to have a dependency?

Re: PowerPC PPC32 CONFIG_HIGHMEM dependency question

[Sebastian Andrzej Siewior]

* Robert Johnson | 2014-03-17 17:41:25 [+0000]:

>I am using an embedded single board computer (SBC) that has a Freescale P2020 processor.  The vendor board support package (BSP) for this product currently uses a 3.0.4 Linux kernel.  In looking through the RT patch for various Linux kernel versions even up through 3.12.13-rt21, I see that CONFIG_HIGHMEM is not allowed if CONFIG_PREEMPT_RT_FULL is selected (see arch/powerpc/Kconfig).
>
>The result for my SBC is that if I enable CONFIG_PREEMPT_RT_FULL in the kernel, I am then only able to access 1GB of RAM instead of the 4GB that is resident.

For 2 GiB you could move the memory mapping to something like 1.5GiB
userland, 2.GiB kernel and the remaining 500MiB for virtual.

>Is there any way to get CONFIG_PREEMPT_RT_FULL to work with CONFIG_HIGHMEM?

Yes. The trick is to look either at the ARM code or at the x86 code and
do something similar for powerpc.

>If I want to use CONFIG_HIGHMEM, should I consider the use of CONFIG_PREEMPT_RTB which does not appear to have a dependency?

You could CONFIG_PREEMPT_RT_BASE with CONFIG_HIGHMEM, yes. But then
don't complain about your latency which will be higher compared to
_FULL. Again: CONFIG_PREEMPT_RT_BASE is only debugging option.

Sebastian

RE: PowerPC PPC32 CONFIG_HIGHMEM dependency question

[Robert Johnson]

>> Robert Johnson | 2014-03-17 17:41:25 [+0000]:
>> Is there any way to get CONFIG_PREEMPT_RT_FULL to work with CONFIG_HIGHMEM?

> Sebastian Andrzej Siewior | 2014-03-18 03:55:
> Yes. The trick is to look either at the ARM code or at the x86 code and do something similar for powerpc.

Which ARM and x86 source files should be referenced?

Which powerpc source files would need to be modified?

Re: PowerPC PPC32 CONFIG_HIGHMEM dependency question

[Sebastian Andrzej Siewior]

* Robert Johnson | 2014-03-21 15:12:20 [+0000]:

>>> Robert Johnson | 2014-03-17 17:41:25 [+0000]:
>>> Is there any way to get CONFIG_PREEMPT_RT_FULL to work with CONFIG_HIGHMEM?
>
>> Sebastian Andrzej Siewior | 2014-03-18 03:55:
>> Yes. The trick is to look either at the ARM code or at the x86 code and do something similar for powerpc.
>
>Which ARM and x86 source files should be referenced?
>
>Which powerpc source files would need to be modified?

In the queue there is

| # KMAP/HIGHMEM
| mm-rt-kmap-atomic-scheduling.patch
| 0002-x86-highmem-add-a-already-used-pte-check.patch
| 0003-arm-highmem-flush-tlb-on-unmap.patch
| arm-enable-highmem-for-rt.patch

which should cover x86 and arm. And now you need to do something like
that for powerpc.

Sebastian