/dev/fw major not constant (embedded mdev devtmpfs ieee1394)

/dev/fw major not constant (embedded mdev devtmpfs ieee1394)

[Paul Chavent]

Hi.

I'am working on an embedded gnu/linux system that allow to capture an ieee1394 stream.

The system is compound of a kernel, a basic initrd (libc,libgcc,busybox,libdc1394,...) and my app.

I use the new firewire stack.

As my system is very minimalist, i tought to create the nodes in /dev at the moment of building my initrd.

But when i needed to modify my kernel configuration, it has changed the major number of /dev/fw*.

It's not really important for the definitive system (it will be freeze), but i wonder how to solve this issue for my future works ?

I don't need hotplug, i just need to fill /dev at startup (coldplug). The target is an embedded device, so it should be easy to build/configure/install, and lightweight.

For those reasons i don't want to use udev.

I tried mdev (i run mdev -s in my startup script) from busybox. The problem is that mdev only look in /sys/class for filling /dev (i have notified their mailing list http://lists.busybox.net/pipermail/busybox/2010-January/071384.html).

The linux ieee1394 wiki people give me some idea (http://marc.info/?t\x126390679500003&r=1&w=2) :
  - maintain a local patch for making the major constant
  - use devtmpfs

I prefer to try devtmpfs, but i would like to have your opinion on how would you do for me ?

Thank for your reading.

Paul.

Re: /dev/fw major not constant (embedded mdev devtmpfs ieee1394)

[Greg KH]

On Thu, Jan 21, 2010 at 09:56:10PM +0100, Paul Chavent wrote:
> Hi.
> 
> I'am working on an embedded gnu/linux system that allow to capture an ieee1394 stream.
> 
> The system is compound of a kernel, a basic initrd (libc,libgcc,busybox,libdc1394,...) and my app.
> 
> I use the new firewire stack.
> 
> As my system is very minimalist, i tought to create the nodes in /dev at the moment of building my initrd.
> 
> But when i needed to modify my kernel configuration, it has changed the major number of /dev/fw*.
> 
> It's not really important for the definitive system (it will be
> freeze), but i wonder how to solve this issue for my future works ?
> 
> I don't need hotplug, i just need to fill /dev at startup (coldplug).
> The target is an embedded device, so it should be easy to
> build/configure/install, and lightweight.
> 
> For those reasons i don't want to use udev.

Why, is udev somehow not "lightweight"?

> I tried mdev (i run mdev -s in my startup script) from busybox. The
> problem is that mdev only look in /sys/class for filling /dev (i have
> notified their mailing list
> http://lists.busybox.net/pipermail/busybox/2010-January/071384.html).
> 
> The linux ieee1394 wiki people give me some idea (http://marc.info/?t\x126390679500003&r=1&w=2) :
>   - maintain a local patch for making the major constant
>   - use devtmpfs
> 
> I prefer to try devtmpfs, but i would like to have your opinion on how
> would you do for me ?

I would recommend using devtmpfs, this is exactly why it was created,
and is how lots of others are using it.  Why not try it out?

thanks,

greg k-h

Re: /dev/fw major not constant (embedded mdev devtmpfs ieee1394)

[Grant Likely]

On Thu, Jan 21, 2010 at 1:56 PM, Paul Chavent <paul.chavent@fnac.net> wrote:
> Hi.
>
> I'am working on an embedded gnu/linux system that allow to capture an
> ieee1394 stream.
[...]
> But when i needed to modify my kernel configuration, it has changed the
> major number of /dev/fw*.
[...]
> The linux ieee1394 wiki people give me some idea
> (http://marc.info/?t\x126390679500003&r=1&w=2) :
>  - maintain a local patch for making the major constant
>  - use devtmpfs
>
> I prefer to try devtmpfs, but i would like to have your opinion on how would
> you do for me ?

devtmpfs is probably your best bet as that is what it is designed for.
 You can also run a trivial script at boot to read /proc/devices and
create the device nodes as described in chapter 3, page 47 of LDD3:

http://lwn.net/images/pdf/LDD3/ch03.pdf

g.

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.

Re: /dev/fw major not constant (embedded mdev devtmpfs ieee1394)

[Mike Frysinger]

On Thu, Jan 21, 2010 at 15:57, Greg KH wrote:
> On Thu, Jan 21, 2010 at 09:56:10PM +0100, Paul Chavent wrote:
>> I'am working on an embedded gnu/linux system that allow to capture an ieee1394 stream.
>>
>> The system is compound of a kernel, a basic initrd (libc,libgcc,busybox,libdc1394,...) and my app.
>>
>> I use the new firewire stack.
>>
>> As my system is very minimalist, i tought to create the nodes in /dev at the moment of building my initrd.
>>
>> But when i needed to modify my kernel configuration, it has changed the major number of /dev/fw*.
>>
>> It's not really important for the definitive system (it will be
>> freeze), but i wonder how to solve this issue for my future works ?
>>
>> I don't need hotplug, i just need to fill /dev at startup (coldplug).
>> The target is an embedded device, so it should be easy to
>> build/configure/install, and lightweight.
>>
>> For those reasons i don't want to use udev.
>
> Why, is udev somehow not "lightweight"?

in comparison, it is absolutely not
-mike

Re: /dev/fw major not constant (embedded mdev devtmpfs ieee1394)

[Greg KH]

On Thu, Jan 21, 2010 at 04:21:26PM -0500, Mike Frysinger wrote:
> On Thu, Jan 21, 2010 at 15:57, Greg KH wrote:
> > On Thu, Jan 21, 2010 at 09:56:10PM +0100, Paul Chavent wrote:
> >> I'am working on an embedded gnu/linux system that allow to capture an ieee1394 stream.
> >>
> >> The system is compound of a kernel, a basic initrd (libc,libgcc,busybox,libdc1394,...) and my app.
> >>
> >> I use the new firewire stack.
> >>
> >> As my system is very minimalist, i tought to create the nodes in /dev at the moment of building my initrd.
> >>
> >> But when i needed to modify my kernel configuration, it has changed the major number of /dev/fw*.
> >>
> >> It's not really important for the definitive system (it will be
> >> freeze), but i wonder how to solve this issue for my future works ?
> >>
> >> I don't need hotplug, i just need to fill /dev at startup (coldplug).
> >> The target is an embedded device, so it should be easy to
> >> build/configure/install, and lightweight.
> >>
> >> For those reasons i don't want to use udev.
> >
> > Why, is udev somehow not "lightweight"?
> 
> in comparison, it is absolutely not

Heh, in comparison to devtmpfs, yes, good point :)

thanks,

greg k-h