[Xenomai] the Linux kernel and operating system

[Xenomai] the Linux kernel and operating system

[ali hagigat]

What kernel subsystems are effecting the real time functionality of an
x86 PC system? Is there a complete list of the CONFIG_ variables
regarding that for 2.6.38.8?

Re: [Xenomai] the Linux kernel and operating system

[abhrajit]

I am new to this, is x86 PC real time?

AB

On Sat, Jun 9, 2012 at 2:43 PM, ali hagigat <hagigatali@gmail.com> wrote:

> What kernel subsystems are effecting the real time functionality of an
> x86 PC system? Is there a complete list of the CONFIG_ variables
> regarding that for 2.6.38.8?
>
> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> http://www.xenomai.org/mailman/listinfo/xenomai
>

Re: [Xenomai] the Linux kernel and operating system

[Philippe Gerum]

On 06/09/2012 11:13 AM, ali hagigat wrote:
> What kernel subsystems are effecting the real time functionality of an
> x86 PC system? Is there a complete list of the CONFIG_ variables
> regarding that for 2.6.38.8?
>

http://www.xenomai.org/index.php/FAQs#Tips_and_tricks_setting_up_your_x86_kernel

> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> http://www.xenomai.org/mailman/listinfo/xenomai
>


-- 
Philippe.

Re: [Xenomai] the Linux kernel and operating system

[Lennart Sorensen]

On Sat, Jun 09, 2012 at 01:43:37PM +0430, ali hagigat wrote:
> What kernel subsystems are effecting the real time functionality of an
> x86 PC system? Is there a complete list of the CONFIG_ variables
> regarding that for 2.6.38.8?

Well in general I would say that using an x86 PC to do real time can't
be done.  Most x86 PCs use SMI, which destroys any hope of actually
getting real time to provide any real response time promises.

Now if you happen to have a PC that doesn't use SMI, or can have it
disabled, then you can get real time.

On things that are not PCs this is pretty much never a problem.

There are some server vendors that provide information on turning off
SMI because they have customers that want to run real time software on
the machine.

-- 
Len Sorensen

Re: [Xenomai] the Linux kernel and operating system

[ali hagigat]

I think SMI needs some initialization from BIOS and then Linux kernel.
The Linux kernel has some CONFIG_ variables to disable/enable SMI.
In theory a motherboard can be built to generate SMI without the
software help. But I think OEMs do not build such a board. So I think
an x86 PC hard real time system is possible for today's PCs.

Please correct me if I am wrong.

On Mon, Jun 11, 2012 at 6:44 PM, Lennart Sorensen
<lsorense@csclub.uwaterloo.ca> wrote:
> On Sat, Jun 09, 2012 at 01:43:37PM +0430, ali hagigat wrote:
>> What kernel subsystems are effecting the real time functionality of an
>> x86 PC system? Is there a complete list of the CONFIG_ variables
>> regarding that for 2.6.38.8?
>
> Well in general I would say that using an x86 PC to do real time can't
> be done.  Most x86 PCs use SMI, which destroys any hope of actually
> getting real time to provide any real response time promises.
>
> Now if you happen to have a PC that doesn't use SMI, or can have it
> disabled, then you can get real time.
>
> On things that are not PCs this is pretty much never a problem.
>
> There are some server vendors that provide information on turning off
> SMI because they have customers that want to run real time software on
> the machine.
>
> --
> Len Sorensen

Re: [Xenomai] the Linux kernel and operating system

[Gilles Chanteperdrix]

On 06/17/2012 02:55 PM, ali hagigat wrote:
> I think SMI needs some initialization from BIOS and then Linux kernel.
> The Linux kernel has some CONFIG_ variables to disable/enable SMI.

Last time I checked, the linux kernel did not have such CONFIG_
variables, which is why xenomai adds them.

> In theory a motherboard can be built to generate SMI without the
> software help. But I think OEMs do not build such a board. So I think
> an x86 PC hard real time system is possible for today's PCs.
> 
> Please correct me if I am wrong.

An SMI is an interrupt, so, there must be software to handle the SMI,
so, no, a motherboard can not be built to generate SMI without software
help.

Intel chipsets provide a register to disable SMIs, but also provide a
register to lock usage of the first register. So yes, there are some PC
today that you can not use for real-time. Unless, of course, you are
ready to use your own BIOS instead of the one provided by the hardware
vendor, or your hardware vendor is willing to change the BIOS.

-- 
                                                                Gilles.

Re: [Xenomai] the Linux kernel and operating system

[Lennart Sorensen]

On Sun, Jun 17, 2012 at 05:25:08PM +0430, ali hagigat wrote:
> I think SMI needs some initialization from BIOS and then Linux kernel.
> The Linux kernel has some CONFIG_ variables to disable/enable SMI.
> In theory a motherboard can be built to generate SMI without the
> software help. But I think OEMs do not build such a board. So I think
> an x86 PC hard real time system is possible for today's PCs.
> 
> Please correct me if I am wrong.

The majority of systems I have seen do have SMI enabled at all times
and no way to turn it off.

IBM even released a kernel driver to allow linux users to disable SMI on
some xSeries servers, but they must also handle certain fan controllers
from linux in that case since the SMI BIOS code is no longer doing it.
For those doing real time, this is acceptable to have to do, certainly
much better than having the SMI BIOS code interrupting real time just
to check on the fans.

I do not believe linux has any way to turn it off.  It is rather system
specific.  All I see is CONFIG_GOOGLE_SMI which is specific to google's
machines.  Also CONFIG_IBM_RTL (IBM Premium Real Time Mode).

-- 
Len Sorensen

Re: [Xenomai] the Linux kernel and operating system

[ali hagigat]

So if BIOS does not lock the SMI disabling so Xenomai kernel can disable SMI.
Then if the x86 PC system can work this way, there will be no problem
I think because the SMBus code is used for a better performance only I
think.

The fans start rotating after pressing the power-on switch.
I wonder if the fan speed control is compulsory for the correct
function of a computer.  ACPI spec does not talk about any necessity
of the running of the SMBus code.

I wonder if laptop computers work with Xenomai without SMI.

Regards

On Mon, Jun 18, 2012 at 6:38 PM, Lennart Sorensen
<lsorense@csclub.uwaterloo.ca> wrote:
> On Sun, Jun 17, 2012 at 05:25:08PM +0430, ali hagigat wrote:
>> I think SMI needs some initialization from BIOS and then Linux kernel.
>> The Linux kernel has some CONFIG_ variables to disable/enable SMI.
>> In theory a motherboard can be built to generate SMI without the
>> software help. But I think OEMs do not build such a board. So I think
>> an x86 PC hard real time system is possible for today's PCs.
>>
>> Please correct me if I am wrong.
>
> The majority of systems I have seen do have SMI enabled at all times
> and no way to turn it off.
>
> IBM even released a kernel driver to allow linux users to disable SMI on
> some xSeries servers, but they must also handle certain fan controllers
> from linux in that case since the SMI BIOS code is no longer doing it.
> For those doing real time, this is acceptable to have to do, certainly
> much better than having the SMI BIOS code interrupting real time just
> to check on the fans.
>
> I do not believe linux has any way to turn it off.  It is rather system
> specific.  All I see is CONFIG_GOOGLE_SMI which is specific to google's
> machines.  Also CONFIG_IBM_RTL (IBM Premium Real Time Mode).
>
> --
> Len Sorensen