[Xenomai-help] Passing data from device init to open function

[Xenomai-help] Passing data from device init to open function

[Johan Borkhuis]

Hello,

I am facing the following problem. During device initialization I am 
filling a structure with device specific data. This structure is 
provided to the Linux driver using pci_set_drvdata. Now I want my 
Xenomai driver to have access to this data as well.
I am planning to use the dev_private area for this, and have it contain 
a pointer to the same structure. The rtdm_dev_context structure is only 
available after the open call and not in the device init function. So I 
am searching for a safe way to transfer data from the device init 
function to the open function.

Is there a way to transfer data from the device init function to the 
open (or other) functions in such a way that it is possible to have 
multiple devices that can be opened multiple times?

Kind regards,
    Johan Borkhuis

Re: [Xenomai-help] Passing data from device init to open function

[Johan Borkhuis]

[-- Attachment #1: Type: text/plain, Size: 1528 bytes --]

This is something I try to avoid (replying to my own messages), but I'll 
do it this time.

Johan Borkhuis wrote:
> Hello,
>
> I am facing the following problem. During device initialization I am 
> filling a structure with device specific data. This structure is 
> provided to the Linux driver using pci_set_drvdata. Now I want my 
> Xenomai driver to have access to this data as well.
>   

I made a fix for the problem I described, but I would also like to 
propose a change to the rtdm_device structure.

What I did to solve this problem is add the following statement to my 
device initialization function (the PCI probe function) after the 
pci_set_drvdata:

    MyDevice *unit;
    ....
    pci_set_drvdata(dev, unit);
    my_rtdm_device.device_id = (int)dev;

In my ops-functions I can now use the following statement to access the 
device data:

int myrtdm_open_nrt(struct rtdm_dev_context    *context,
                 rtdm_user_info_t              *user_info,
                 int                            oflags)
{
    MyDevice *unit = pci_get_drvdata((struct pci_dev 
*)context->device->device_id);

This does work, but only as long as sizeof(int) == sizeof(<pointer>), 
which is true on 32 bit systems, but not on 64 bit systems. So I would 
like to add an extra pointer field to the rtdm_device structure, 
pointing to some device-data to allow this to work on both 32 and 64 bit 
systems.

I attached a diff-file containing a patch to rtdm_driver.h that makes 
this change.

Kind regards,
    Johan Borkhuis


[-- Attachment #2: xeno.diff --]
[-- Type: text/plain, Size: 597 bytes --]

diff -u -r org/xenomai-2.3.2/include/rtdm/rtdm_driver.h xenomai-2.3.2/include/rtdm/rtdm_driver.h
--- org/xenomai-2.3.2/include/rtdm/rtdm_driver.h	2007-02-19 15:09:49.000000000 +0100
+++ xenomai-2.3.2/include/rtdm/rtdm_driver.h	2007-10-11 09:15:36.000000000 +0200
@@ -428,6 +428,9 @@
     /** Driver definable device ID */
     int                             device_id;
 
+    /** Driver definable device data */
+    void                            *device_data;
+
     /** Data stored by RTDM inside a registered device (internal use only) */
     struct rtdm_dev_reserved        reserved;
 };

Re: [Xenomai-help] Passing data from device init to open function

[Jan Kiszka]

Johan Borkhuis wrote:
> This is something I try to avoid (replying to my own messages), but I'll
> do it this time.
> 
> Johan Borkhuis wrote:
>> Hello,
>>
>> I am facing the following problem. During device initialization I am
>> filling a structure with device specific data. This structure is
>> provided to the Linux driver using pci_set_drvdata. Now I want my
>> Xenomai driver to have access to this data as well.
>>   
> 
> I made a fix for the problem I described, but I would also like to
> propose a change to the rtdm_device structure.
> 
> What I did to solve this problem is add the following statement to my
> device initialization function (the PCI probe function) after the
> pci_set_drvdata:
> 
>    MyDevice *unit;
>    ....
>    pci_set_drvdata(dev, unit);
>    my_rtdm_device.device_id = (int)dev;
> 
> In my ops-functions I can now use the following statement to access the
> device data:
> 
> int myrtdm_open_nrt(struct rtdm_dev_context    *context,
>                 rtdm_user_info_t              *user_info,
>                 int                            oflags)
> {
>    MyDevice *unit = pci_get_drvdata((struct pci_dev
> *)context->device->device_id);
> 
> This does work, but only as long as sizeof(int) == sizeof(<pointer>),
> which is true on 32 bit systems, but not on 64 bit systems. So I would
> like to add an extra pointer field to the rtdm_device structure,
> pointing to some device-data to allow this to work on both 32 and 64 bit
> systems.
> 
> I attached a diff-file containing a patch to rtdm_driver.h that makes
> this change.

Hmm, that's what we had in an early RTDM version, but the more common
use-case turned out to be

per_device_data = my_per_device_data[device->device_id];

where my_per_device_data is a global variable, typical a data structure
containing all the required per-device information.

Anyway, as you are adding this field, not replacing the ID with it, I
think I'm going to merge this.

Thanks,
Jan

-- 
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux

Re: [Xenomai-help] Passing data from device init to open function

[Johan Borkhuis]

Jan Kiszka wrote:
>> This does work, but only as long as sizeof(int) == sizeof(<pointer>),
>> which is true on 32 bit systems, but not on 64 bit systems. So I would
>> like to add an extra pointer field to the rtdm_device structure,
>> pointing to some device-data to allow this to work on both 32 and 64 bit
>> systems.
>>     
> Hmm, that's what we had in an early RTDM version, but the more common
> use-case turned out to be
>
> per_device_data = my_per_device_data[device->device_id];
>
> where my_per_device_data is a global variable, typical a data structure
> containing all the required per-device information.
>   

I considered that as well. The main problem is that you have to 
statically allocate the memory based on the maximum number of devices 
you want to support, and I prefer to allocate the memory for each 
available device.

> Anyway, as you are adding this field, not replacing the ID with it, I
> think I'm going to merge this.
>   
One should have a very good reason to replace or remove a field in a 
system-structure, and this is just an addition to the structure, so no 
real need to replace anything.

Thanks.

Kind regards,
    Johan Borkhuis