RT_PREEMPT autodetection from user program

RT_PREEMPT autodetection from user program

[Michael Haberler]

Hi,

I need to autodetect the RT kernel for an application which runs under RTAI, Xenomai, RT_PREEMPT and vanilla kernels. 

To support a unified binary, I need to runtime detect the RT OS - what are my best options for RT_PREEMPT? from https://rt.wiki.kernel.org/index.php/RT_PREEMPT_HOWTO#Checking_the_Kernel I gather the single method:

parsing 'ps' output for IRQ handlers being kthreads, i.e. for '[IRQ' or '[irq'

any chance of doing better than that, or parsing of utsname.release for "-rt"?  A false positive on a vanilla kernel could be a safety hazard.

thanks in advance,

Michael

Re: RT_PREEMPT autodetection from user program

[Sven-Thorsten Dietrich]

On Jan 19, 2013, at 4:49 PM, Michael Haberler <mail17@mah.priv.at> wrote:

> Hi,
> 
> I need to autodetect the RT kernel for an application which runs under RTAI, Xenomai, RT_PREEMPT and vanilla kernels. 
> 
> To support a unified binary, I need to runtime detect the RT OS - what are my best options for RT_PREEMPT? from https://rt.wiki.kernel.org/index.php/RT_PREEMPT_HOWTO#Checking_the_Kernel I gather the single method:
> 
> parsing 'ps' output for IRQ handlers being kthreads, i.e. for '[IRQ' or '[irq'
> 
> any chance of doing better than that, or parsing of utsname.release for "-rt"?  A false positive on a vanilla kernel could be a safety hazard.
> 

Not sure what distro you are supporting and so on, but zcat /proc/config.gz | grep PREEMPT_RT might work.


Sven

Re: RT_PREEMPT autodetection from user program

[Stanislav Meduna]

On 20.01.2013 01:13, John Morris wrote:

> LCNC drives machines that weigh many tons and spins
> spindles at 24k RPM, so this is important!
...
> We haven't found /proc/config.gz in any vendor-provided
> PREEMPT_RT kernels, so that is not an option either.

On 20.01.2013 01:49, Michael Haberler wrote:

> A false positive on a vanilla kernel could be a safety hazard.

If we are speaking safety here I'd say that you have to support
specific version of kernel on specific hardware that _you_ have
quality assured. Anything else is a recipe for problems.

I've seen OSADL Linux deadlock, RT PREEMPT leaking memory,
RT PREEMPT on but without high-res timers, IDE driver
monopolizing processor so that the RT throttler was activated,
priority inversion issues between kernel and application
threads, interrupt latencies due to interrupt routing
on the specific hardware etc.

There is no way you can guarantee anything if you are using
vendor-provided kernel on a hardware you have never seen.
With your kernel, a list of supported hardware, with initscripts
that are checking if the environment matches and with the help
from this list (thank you guys, the level of support here
is excellent!) you will sleep a lot better.

Just my two cents

Regards
-- 
                                        Stano

Re: RT_PREEMPT autodetection from user program

[Michael Mueller]

On Sun, Jan 20, 2013 at 4:49 AM, Stanislav Meduna <stano@meduna.org> wrote:
>
> On 20.01.2013 01:13, John Morris wrote:
>
> > LCNC drives machines that weigh many tons and spins
> > spindles at 24k RPM, so this is important!

> On 20.01.2013 01:49, Michael Haberler wrote:
>
> > A false positive on a vanilla kernel could be a safety hazard.
>
> If we are speaking safety here I'd say that you have to support
> specific version of kernel on specific hardware that _you_ have
> quality assured. Anything else is a recipe for

catastrophic failure.

Re: RT_PREEMPT autodetection from user program

[Clark Williams]

[-- Attachment #1: Type: text/plain, Size: 809 bytes --]

On Sun, 20 Jan 2013 01:49:59 +0100
Michael Haberler <mail17@mah.priv.at> wrote:

> Hi,
> 
> I need to autodetect the RT kernel for an application which runs under RTAI, Xenomai, RT_PREEMPT and vanilla kernels. 
> 
> To support a unified binary, I need to runtime detect the RT OS - what are my best options for RT_PREEMPT? from https://rt.wiki.kernel.org/index.php/RT_PREEMPT_HOWTO#Checking_the_Kernel I gather the single method:
> 
> parsing 'ps' output for IRQ handlers being kthreads, i.e. for '[IRQ' or '[irq'
> 
> any chance of doing better than that, or parsing of utsname.release for "-rt"?  A false positive on a vanilla kernel could be a safety hazard.
> 
>

A PREEMPT_RT kernel creates /sys/kernel/realtime when it boots. You
can just check for the existence of that file.

Clark

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]