[Xenomai] Questions about __ipipe_pin_range_globally

[Xenomai] Questions about __ipipe_pin_range_globally

[Hongfei Cheng]

While porting Xenomai 3 to an ARMv8 SoC, we're seeing peculiar errors,
such as "level 1 translation fault", in the Linux domain when running
a specific application. The same app runs fine when I-pipe is
disabled.

One function, __ipipe_pin_range_globally, is of particular interest to
us (mostly because I don't think I fully understand its purpose and
impact).

Searching the mailing list archives, I see the following comment
(http://www.xenomai.org/pipermail/xenomai/2014-September/031731.html)
made by Gilles on Sep 22, 2014:
"... ipipe_pin_range_globally probably has never been used with 3
level page tables, which are enabled by LPAE ..."

The code change logs (in branch ipipe-3.14) show that, about 2 months
later, on Nov 30, 2014, Philippe updated the code of
__ipipe_pin_range_globally and pgd_holder, with a nice comment section
(ccbd22da).

Our armv8 SoC uses 4KB page configuration and 3-level translation
tables. I have the following questions:
1). Has anyone tested the latest __ipipe_pin_range_globally code
against ARMv7 with LPAE enabled (and 3-level page tables)?
2). Is __ipipe_pin_range_globally mainly for the purpose of preventing
"(minor) page faults" for applications running in primary/real-time
domain?

Thanks,
Hongfei

Re: [Xenomai] Questions about __ipipe_pin_range_globally

[Philippe Gerum]

On 06/25/2015 02:12 AM, Hongfei Cheng wrote:
> While porting Xenomai 3 to an ARMv8 SoC, we're seeing peculiar errors,
> such as "level 1 translation fault", in the Linux domain when running
> a specific application. The same app runs fine when I-pipe is
> disabled.
> 
> One function, __ipipe_pin_range_globally, is of particular interest to
> us (mostly because I don't think I fully understand its purpose and
> impact).
> 
> Searching the mailing list archives, I see the following comment
> (http://www.xenomai.org/pipermail/xenomai/2014-September/031731.html)
> made by Gilles on Sep 22, 2014:
> "... ipipe_pin_range_globally probably has never been used with 3
> level page tables, which are enabled by LPAE ..."
> 
> The code change logs (in branch ipipe-3.14) show that, about 2 months
> later, on Nov 30, 2014, Philippe updated the code of
> __ipipe_pin_range_globally and pgd_holder, with a nice comment section
> (ccbd22da).
> 
> Our armv8 SoC uses 4KB page configuration and 3-level translation
> tables. I have the following questions:
> 1). Has anyone tested the latest __ipipe_pin_range_globally code
> against ARMv7 with LPAE enabled (and 3-level page tables)?

Yes, A15.

> 2). Is __ipipe_pin_range_globally mainly for the purpose of preventing
> "(minor) page faults" for applications running in primary/real-time
> domain?
> 

Correct.

-- 
Philippe.

Re: [Xenomai] Questions about __ipipe_pin_range_globally

[Hongfei Cheng]

On Thu, Jun 25, 2015 at 12:10 AM, Philippe Gerum <rpm@xenomai.org> wrote:
> On 06/25/2015 02:12 AM, Hongfei Cheng wrote:
>> While porting Xenomai 3 to an ARMv8 SoC, we're seeing peculiar errors,
>> such as "level 1 translation fault", in the Linux domain when running
>> a specific application. The same app runs fine when I-pipe is
>> disabled.
>>
>> One function, __ipipe_pin_range_globally, is of particular interest to
>> us (mostly because I don't think I fully understand its purpose and
>> impact).
>>
>> Searching the mailing list archives, I see the following comment
>> (http://www.xenomai.org/pipermail/xenomai/2014-September/031731.html)
>> made by Gilles on Sep 22, 2014:
>> "... ipipe_pin_range_globally probably has never been used with 3
>> level page tables, which are enabled by LPAE ..."
>>
>> The code change logs (in branch ipipe-3.14) show that, about 2 months
>> later, on Nov 30, 2014, Philippe updated the code of
>> __ipipe_pin_range_globally and pgd_holder, with a nice comment section
>> (ccbd22da).
>>
>> Our armv8 SoC uses 4KB page configuration and 3-level translation
>> tables. I have the following questions:
>> 1). Has anyone tested the latest __ipipe_pin_range_globally code
>> against ARMv7 with LPAE enabled (and 3-level page tables)?
>
> Yes, A15.
>
>> 2). Is __ipipe_pin_range_globally mainly for the purpose of preventing
>> "(minor) page faults" for applications running in primary/real-time
>> domain?
>>
>
> Correct.
>
> --
> Philippe.

Thank you, Philippe, for your quick and helpful responses.