[Xenomai] Hardware I/O from Userspace

[Xenomai] Hardware I/O from Userspace

[Drasko DRASKOVIC]

Hi all,
I would like to access to various peripheral and CPU regs of my FPGA
from userspace.

Is mmap of /dev/mem best way to it, or I should do it from kernelspace
via kernel module?

I am looking at  examples/common/hw_direct_io.c of Xenomai examples.
Can somebody explain me how Xenomai is used in this case, as I see
only standard pthread interface. Was this patched by Xenomai patches?
I do not see calls to RT tasks or some other Xenomai-related API.

I can see:
struct sched_param param;

printf("%s: %s %s\n", __FUNCTION__, __DATE__, __TIME__ );
param.__sched_priority = 99;
res = pthread_setschedparam(pthread_self(),  SCHED_FIFO, &param);

How these lines affect Xenomai?

Thanks and best regards,
Drasko

Re: [Xenomai] Hardware I/O from Userspace

[Gilles Chanteperdrix]

On 10/16/2013 06:06 PM, Drasko DRASKOVIC wrote:
> Hi all,
> I would like to access to various peripheral and CPU regs of my FPGA
> from userspace.
> 
> Is mmap of /dev/mem best way to it, or I should do it from kernelspace
> via kernel module?

Hi,

The RTDM kernel module is the recommended way, among other things, it may
be used to prevent several applications to access the registers at the
same time.

> 
> I am looking at  examples/common/hw_direct_io.c of Xenomai examples.
> Can somebody explain me how Xenomai is used in this case, as I see
> only standard pthread interface. Was this patched by Xenomai patches?
> I do not see calls to RT tasks or some other Xenomai-related API.
> 
> I can see:
> struct sched_param param;
> 
> printf("%s: %s %s\n", __FUNCTION__, __DATE__, __TIME__ );
> param.__sched_priority = 99;
> res = pthread_setschedparam(pthread_self(),  SCHED_FIFO, &param);
> 
> How these lines affect Xenomai?

Xenomai has more skins than just the native skin, it has notably a 
POSIX skin which attempts to implement the POSIX API. How 
pthread_setschedparam is hijacked by Xenomai is explained here:

https://www.xenomai.org/index.php/Porting_POSIX_applications_to_Xenomai#Compilation_flags

Regards.


-- 
					    Gilles.

Re: [Xenomai] Hardware I/O from Userspace

[Drasko DRASKOVIC]

Hi Gilles,

On Wed, Oct 16, 2013 at 7:14 PM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> On 10/16/2013 06:06 PM, Drasko DRASKOVIC wrote:
>> Hi all,
>> I would like to access to various peripheral and CPU regs of my FPGA
>> from userspace.
>>
>> Is mmap of /dev/mem best way to it, or I should do it from kernelspace
>> via kernel module?
>
> Hi,
>
> The RTDM kernel module is the recommended way, among other things, it may
> be used to prevent several applications to access the registers at the
> same time.

I will most probably have only one application who access these regs.

Is not putting this application into the user-space safer - if it
breaks it will not block system?
Did i misunderstand something, or RTDM kernel module is like standard
kernel module - if it has a bug it can break the whole kernel.

>
>>
>> I am looking at  examples/common/hw_direct_io.c of Xenomai examples.
>> Can somebody explain me how Xenomai is used in this case, as I see
>> only standard pthread interface. Was this patched by Xenomai patches?
>> I do not see calls to RT tasks or some other Xenomai-related API.
>>
>> I can see:
>> struct sched_param param;
>>
>> printf("%s: %s %s\n", __FUNCTION__, __DATE__, __TIME__ );
>> param.__sched_priority = 99;
>> res = pthread_setschedparam(pthread_self(),  SCHED_FIFO, &param);
>>
>> How these lines affect Xenomai?
>
> Xenomai has more skins than just the native skin, it has notably a
> POSIX skin which attempts to implement the POSIX API. How
> pthread_setschedparam is hijacked by Xenomai is explained here:
>
> https://www.xenomai.org/index.php/Porting_POSIX_applications_to_Xenomai#Compilation_flags

Usage of the compilation flags is pretty-much clear. What I was asing
is can anybody elaborate of how Posix calls became Xenomai ones in
this example (if they did).

BR,
Drasko

Re: [Xenomai] Hardware I/O from Userspace

[Lennart Sorensen]

On Thu, Oct 17, 2013 at 04:57:43PM +0200, Drasko DRASKOVIC wrote:
> I will most probably have only one application who access these regs.
> 
> Is not putting this application into the user-space safer - if it
> breaks it will not block system?
> Did i misunderstand something, or RTDM kernel module is like standard
> kernel module - if it has a bug it can break the whole kernel.

What do you think using /dev/mem can do?

-- 
Len Sorensen

Re: [Xenomai] Hardware I/O from Userspace

[Drasko DRASKOVIC]

On Thu, Oct 17, 2013 at 5:00 PM, Lennart Sorensen
<lsorense@csclub.uwaterloo.ca> wrote:
> On Thu, Oct 17, 2013 at 04:57:43PM +0200, Drasko DRASKOVIC wrote:
>> I will most probably have only one application who access these regs.
>>
>> Is not putting this application into the user-space safer - if it
>> breaks it will not block system?
>> Did i misunderstand something, or RTDM kernel module is like standard
>> kernel module - if it has a bug it can break the whole kernel.
>
> What do you think using /dev/mem can do?

I guess it opens the file for reading... right? Is there more than
meets the eye? Can it do much harm when file is opened from userspace
and how?

BR,
Drasko

Re: [Xenomai] Hardware I/O from Userspace

[Drasko DRASKOVIC]

On Thu, Oct 17, 2013 at 4:57 PM, Drasko DRASKOVIC
<drasko.draskovic@gmail.com> wrote:
https://www.xenomai.org/index.php/Porting_POSIX_applications_to_Xenomai#Compilation_flags
>
> Usage of the compilation flags is pretty-much clear. What I was asing
> is can anybody elaborate of how Posix calls became Xenomai ones in
> this example (if they did).

Whoops, just saw that "Under the hood: the --wrap flag" section
actually explains this quite well. Thanks.

BR,
Drasko

Re: [Xenomai] Hardware I/O from Userspace

[Gilles Chanteperdrix]

On 10/17/2013 04:57 PM, Drasko DRASKOVIC wrote:
> Hi Gilles,
> 
> On Wed, Oct 16, 2013 at 7:14 PM, Gilles Chanteperdrix
> <gilles.chanteperdrix@xenomai.org> wrote:
>> On 10/16/2013 06:06 PM, Drasko DRASKOVIC wrote:
>>> Hi all,
>>> I would like to access to various peripheral and CPU regs of my FPGA
>>> from userspace.
>>>
>>> Is mmap of /dev/mem best way to it, or I should do it from kernelspace
>>> via kernel module?
>>
>> Hi,
>>
>> The RTDM kernel module is the recommended way, among other things, it may
>> be used to prevent several applications to access the registers at the
>> same time.
> 
> I will most probably have only one application who access these regs.
> 
> Is not putting this application into the user-space safer - if it
> breaks it will not block system?
> Did i misunderstand something, or RTDM kernel module is like standard
> kernel module - if it has a bug it can break the whole kernel.

Yes, but having to separate application from driver is a future proof
pattern: the same driver can be used with different applications, and
changing hardware does not break the application, you only have to write
a driver for the new hardware.

-- 
                                                                Gilles.

Re: [Xenomai] Hardware I/O from Userspace

[Drasko DRASKOVIC]

On Thu, Oct 17, 2013 at 8:39 PM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> On 10/17/2013 04:57 PM, Drasko DRASKOVIC wrote:
>> Hi Gilles,
>>
>> On Wed, Oct 16, 2013 at 7:14 PM, Gilles Chanteperdrix
>> <gilles.chanteperdrix@xenomai.org> wrote:
>>> On 10/16/2013 06:06 PM, Drasko DRASKOVIC wrote:
>>>> Hi all,
>>>> I would like to access to various peripheral and CPU regs of my FPGA
>>>> from userspace.
>>>>
>>>> Is mmap of /dev/mem best way to it, or I should do it from kernelspace
>>>> via kernel module?
>>>
>>> Hi,
>>>
>>> The RTDM kernel module is the recommended way, among other things, it may
>>> be used to prevent several applications to access the registers at the
>>> same time.
>>
>> I will most probably have only one application who access these regs.
>>
>> Is not putting this application into the user-space safer - if it
>> breaks it will not block system?
>> Did i misunderstand something, or RTDM kernel module is like standard
>> kernel module - if it has a bug it can break the whole kernel.
>
> Yes, but having to separate application from driver is a future proof
> pattern: the same driver can be used with different applications, and
> changing hardware does not break the application, you only have to write
> a driver for the new hardware.
>

Here is my scenario: I have to access HW regs to set up and initialize
peripherals and FPGA. There is a lot of initialization, and currently
I am doing this from userspace via /dev/mem.

You are suggesting to move this as RTDM into the kernel space.

But I also have some code that is executed in this application that
might use floating point calculation, etc, i.e. pure data processing,
no HW access - but it has to be executed with high priority and in RT.

What would be the best  scenario in that case:
a) keep all as userspace
b) move all to kernelspace
c) divide this somehow into kernel-spce part of HW access and small
userspace app (of data processing) and make them communicate somehow
(how?). Would this affect responsiveness?


Best regards,
Drasko