Best practice for loading platform modules

Best practice for loading platform modules

[Chris Hallinan]

I have an OMAP3 board, and want usb (host mode) configured as a
loadable module.  Using a generic rule such as this:
RUN+="/sbin/modprobe -b $env{MODALIAS}"

ends up passing "platform:musb_hdrc" to modprobe, which obviously
fails.  Here is what the kernel is passing to udevd, which in turn is
passed to modprobe.  Notice the "platform:" prefix prepended to the
module name:

ACTION="add"
DEVPATH="/devices/platform/musb_hdrc"
DRIVER="musb_hdrc"
HOME="/"
MODALIAS="platform:musb_hdrc"
<snip...>
SEQNUM="1302"
SHLVL="1"
SUBSYSTEM="platform"

All of the platform drivers are passed with the "platform:" prefix, so
the problem applies of these drivers that I wish to load as modules.

The way I see it with my limited understanding of udev (I am not a
kernel developer) I have a few choices.

1) Write an explicit udev rule passing the literal "musb_hdrc" to modprobe
2) Use a startup script to load the module instead of letting udev do
it (not my first choice)
3) Provide a wrapper for modprobe to strip "platform:" from the string
it is passed.

There are probably other options.

It's more likely I'm missing some understanding and there is a
"proper" way to write this rule which I haven't figured out.

Any advice/comment will be much appreciated.

Regards,

Chris Hallinan


--
Life is like Linux - it never stands still.

Re: Best practice for loading platform modules

[Greg KH]

On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
> I have an OMAP3 board, and want usb (host mode) configured as a
> loadable module.  Using a generic rule such as this:
> RUN+="/sbin/modprobe -b $env{MODALIAS}"
> 
> ends up passing "platform:musb_hdrc" to modprobe, which obviously
> fails.

Why would that fail?  Do you have the proper modalias in the kernel
module to handle this?  If not, can you add it?

thanks,

greg k-h

Re: Best practice for loading platform modules

[Greg KH]

On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
> > I have an OMAP3 board, and want usb (host mode) configured as a
> > loadable module.  Using a generic rule such as this:
> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
> > 
> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
> > fails.
> 
> Why would that fail?  Do you have the proper modalias in the kernel
> module to handle this?  If not, can you add it?

I see:
	MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
"musb_hdrc" somehow?

thanks,

greg k-h

Re: Best practice for loading platform modules

[Chris Hallinan]

On Wed, Apr 29, 2009 at 2:12 PM, Greg KH <greg@kroah.com> wrote:
> On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
>> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
>> > I have an OMAP3 board, and want usb (host mode) configured as a
>> > loadable module.  Using a generic rule such as this:
>> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
>> >
>> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
>> > fails.
>>
>> Why would that fail?  Do you have the proper modalias in the kernel
>> module to handle this?  If not, can you add it?
>
> I see:
>        MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
> in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
> "musb_hdrc" somehow?

From musb_core.c:

#define MUSB_DRIVER_NAME "musb_hdrc"
const char musb_driver_name[] = MUSB_DRIVER_NAME;

So somewhere else in kernel code (common Kobject/platform code, I
don't know) must be prepending the string "platform:" to the driver
name.  Is this what you think the problem is? All of the platform
drivers display this way.  From a simple script I wrote that dumps
env:

export MODALIAS="platform:twl4030_usb"
export MODALIAS="platform:twl4030_gpio"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_rtc"
export MODALIAS="platform:twl4030_gpio"
export MODALIAS="platform:twl4030_usb"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:gpio-keys"
export MODALIAS="platform:i2c_omap"
export MODALIAS="platform:leds-gpio"
export MODALIAS="platform:i2c_omap"
export MODALIAS="platform:mmci-omap-hs"
export MODALIAS="platform:musb_hdrc"
export MODALIAS="platform:omap3beagle_lcd"
export MODALIAS="platform:omap2-nand"
export MODALIAS="platform:omapfb"
export MODALIAS="platform:serial8250"
export MODALIAS="platform:serial8250"
export MODALIAS="platform:i2c_omap"
export MODALIAS="platform:gpio-keys"
export MODALIAS="platform:i2c_omap"
export MODALIAS="platform:leds-gpio"
export MODALIAS="platform:omap2-nand"
export MODALIAS="platform:mmci-omap-hs"
export MODALIAS="platform:musb_hdrc"
export MODALIAS="platform:omap3beagle_lcd"
export MODALIAS="platform:serial8250"
export MODALIAS="platform:omapfb"
export MODALIAS="platform:serial8250"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_reg"
export MODALIAS="platform:twl4030_rtc"


-Chris


-- 
Life is like Linux - it never stands still.

Re: Best practice for loading platform modules

[Greg KH]

On Wed, Apr 29, 2009 at 02:31:49PM -0400, Chris Hallinan wrote:
> On Wed, Apr 29, 2009 at 2:12 PM, Greg KH <greg@kroah.com> wrote:
> > On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
> >> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
> >> > I have an OMAP3 board, and want usb (host mode) configured as a
> >> > loadable module.  Using a generic rule such as this:
> >> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
> >> >
> >> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
> >> > fails.
> >>
> >> Why would that fail?  Do you have the proper modalias in the kernel
> >> module to handle this?  If not, can you add it?
> >
> > I see:
> >        MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
> > in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
> > "musb_hdrc" somehow?
> 
> >From musb_core.c:
> 
> #define MUSB_DRIVER_NAME "musb_hdrc"
> const char musb_driver_name[] = MUSB_DRIVER_NAME;

Good, so you should be fine.

Can you run:
	modprobe -n -v platform:musb_hdrc
on your system?

It should try to load that module, as it looks like you have it set up
properly, based on the module alias.

> So somewhere else in kernel code (common Kobject/platform code, I
> don't know) must be prepending the string "platform:" to the driver
> name.  Is this what you think the problem is? All of the platform
> drivers display this way.  From a simple script I wrote that dumps
> env:

Yes, that is correct, this is what is supposed to be happening.

thanks,

greg k-h

Re: Best practice for loading platform modules

[Chris Hallinan]

On Wed, Apr 29, 2009 at 2:44 PM, Greg KH <greg@kroah.com> wrote:
> On Wed, Apr 29, 2009 at 02:31:49PM -0400, Chris Hallinan wrote:
>> On Wed, Apr 29, 2009 at 2:12 PM, Greg KH <greg@kroah.com> wrote:
>> > On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
>> >> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
>> >> > I have an OMAP3 board, and want usb (host mode) configured as a
>> >> > loadable module.  Using a generic rule such as this:
>> >> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
>> >> >
>> >> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
>> >> > fails.
>> >>
>> >> Why would that fail?  Do you have the proper modalias in the kernel
>> >> module to handle this?  If not, can you add it?
>> >
>> > I see:
>> >        MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
>> > in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
>> > "musb_hdrc" somehow?
>>
>> >From musb_core.c:
>>
>> #define MUSB_DRIVER_NAME "musb_hdrc"
>> const char musb_driver_name[] = MUSB_DRIVER_NAME;
>
> Good, so you should be fine.
>
> Can you run:
>        modprobe -n -v platform:musb_hdrc
> on your system?
>

# modprobe -n -v platform:musb_hdrc
FATAL: Module platform:musb_hdrc not found.
# modprobe --version
module-init-tools version 3.7

But...

# modprobe -n -v musb_hdrc
insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/core/usbcore.ko
insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/musb/musb_hdrc.ko

Seems to be complaining about the "platform:" prefix, I think.  This
is the very latest version of module-init-tools from git tree a couple
days ago.

-Chris

Re: Best practice for loading platform modules

[Greg KH]

On Wed, Apr 29, 2009 at 03:15:13PM -0400, Chris Hallinan wrote:
> On Wed, Apr 29, 2009 at 2:44 PM, Greg KH <greg@kroah.com> wrote:
> > On Wed, Apr 29, 2009 at 02:31:49PM -0400, Chris Hallinan wrote:
> >> On Wed, Apr 29, 2009 at 2:12 PM, Greg KH <greg@kroah.com> wrote:
> >> > On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
> >> >> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
> >> >> > I have an OMAP3 board, and want usb (host mode) configured as a
> >> >> > loadable module.  Using a generic rule such as this:
> >> >> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
> >> >> >
> >> >> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
> >> >> > fails.
> >> >>
> >> >> Why would that fail?  Do you have the proper modalias in the kernel
> >> >> module to handle this?  If not, can you add it?
> >> >
> >> > I see:
> >> >        MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
> >> > in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
> >> > "musb_hdrc" somehow?
> >>
> >> >From musb_core.c:
> >>
> >> #define MUSB_DRIVER_NAME "musb_hdrc"
> >> const char musb_driver_name[] = MUSB_DRIVER_NAME;
> >
> > Good, so you should be fine.
> >
> > Can you run:
> >        modprobe -n -v platform:musb_hdrc
> > on your system?
> >
> 
> # modprobe -n -v platform:musb_hdrc
> FATAL: Module platform:musb_hdrc not found.
> # modprobe --version
> module-init-tools version 3.7
> 
> But...
> 
> # modprobe -n -v musb_hdrc
> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/core/usbcore.ko
> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/musb/musb_hdrc.ko
> 
> Seems to be complaining about the "platform:" prefix, I think.  This
> is the very latest version of module-init-tools from git tree a couple
> days ago.

But your kernel is over a year old, not much we can do about that :)

I suggest poking your distro, it seems they need to provide a fix
for your kernel, nothing we can do here, everything works just fine with
the latest kernel releases from what I can tell.

good luck,

greg k-h

Re: Best practice for loading platform modules

[Kay Sievers]

On Wed, Apr 29, 2009 at 21:15, Chris Hallinan <challinan@gmail.com> wrote:

> # modprobe -n -v musb_hdrc
> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/core/usbcore.ko
> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/musb/musb_hdrc.ko

The kernel module itself needs an alias like this:
  modinfo pcspkr
  filename:
/lib/modules/2.6.30-rc3-00425-g27b1833-dirty/kernel/drivers/input/misc/pcspkr.ko
  alias:          platform:pcspkr
  license:        GPL
  description:    PC Speaker beeper driver
  ...

If it does not show that, you need to add it to the kernel module
itself, as Greg pointed out earlier with the MODULE_ALIAS() thing.

Thanks,
Kay

Re: Best practice for loading platform modules

[Chris Hallinan]

On Wed, Apr 29, 2009 at 3:25 PM, Greg KH <greg@kroah.com> wrote:
> On Wed, Apr 29, 2009 at 03:15:13PM -0400, Chris Hallinan wrote:
>> On Wed, Apr 29, 2009 at 2:44 PM, Greg KH <greg@kroah.com> wrote:
>> > On Wed, Apr 29, 2009 at 02:31:49PM -0400, Chris Hallinan wrote:
>> >> On Wed, Apr 29, 2009 at 2:12 PM, Greg KH <greg@kroah.com> wrote:
>> >> > On Wed, Apr 29, 2009 at 11:11:35AM -0700, Greg KH wrote:
>> >> >> On Wed, Apr 29, 2009 at 02:05:37PM -0400, Chris Hallinan wrote:
>> >> >> > I have an OMAP3 board, and want usb (host mode) configured as a
>> >> >> > loadable module.  Using a generic rule such as this:
>> >> >> > RUN+="/sbin/modprobe -b $env{MODALIAS}"
>> >> >> >
>> >> >> > ends up passing "platform:musb_hdrc" to modprobe, which obviously
>> >> >> > fails.
>> >> >>
>> >> >> Why would that fail?  Do you have the proper modalias in the kernel
>> >> >> module to handle this?  If not, can you add it?
>> >> >
>> >> > I see:
>> >> >        MODULE_ALIAS("platform:" MUSB_DRIVER_NAME);
>> >> > in the musb code, perhaps MUSB_DRIVER_NAME is not set correctly to be
>> >> > "musb_hdrc" somehow?
>> >>
>> >> >From musb_core.c:
>> >>
>> >> #define MUSB_DRIVER_NAME "musb_hdrc"
>> >> const char musb_driver_name[] = MUSB_DRIVER_NAME;
>> >
>> > Good, so you should be fine.
>> >
>> > Can you run:
>> >        modprobe -n -v platform:musb_hdrc
>> > on your system?
>> >
>>
>> # modprobe -n -v platform:musb_hdrc
>> FATAL: Module platform:musb_hdrc not found.
>> # modprobe --version
>> module-init-tools version 3.7
>>
>> But...
>>
>> # modprobe -n -v musb_hdrc
>> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/core/usbcore.ko
>> insmod /lib/modules/2.6.24_mvl5024/kernel/drivers/usb/musb/musb_hdrc.ko
>>
>> Seems to be complaining about the "platform:" prefix, I think.  This
>> is the very latest version of module-init-tools from git tree a couple
>> days ago.
>
> But your kernel is over a year old, not much we can do about that :)
>
> I suggest poking your distro, it seems they need to provide a fix
> for your kernel, nothing we can do here, everything works just fine with
> the latest kernel releases from what I can tell.
>

Erm, I am the distro maintainer!  I tried just today w/ 2.6.29.  The
problem seems like a mismatch.  Kernel passes "platform:musb_hdrc".
Udev receives and passes same to modprobe.  The answer I'm seeking is
"what is the correct/proper approach?"  Should I seek to modify the
kernel and remove the "platform:" prefix, or should I seek a
modification to modprobe to be able to understand and strip off the
"platform:" prefix?  Or should udev have a construct/rules syntax to
strip the "platform:" prefix?

Thanks,

Chris


-- 
Life is like Linux - it never stands still.

Re: Best practice for loading platform modules

[Kay Sievers]

On Wed, Apr 29, 2009 at 21:35, Chris Hallinan <challinan@gmail.com> wrote:

> Erm, I am the distro maintainer!  I tried just today w/ 2.6.29.  The
> problem seems like a mismatch.  Kernel passes "platform:musb_hdrc".
> Udev receives and passes same to modprobe.  The answer I'm seeking is
> "what is the correct/proper approach?"  Should I seek to modify the
> kernel and remove the "platform:" prefix, or should I seek a
> modification to modprobe to be able to understand and strip off the
> "platform:" prefix?  Or should udev have a construct/rules syntax to
> strip the "platform:" prefix?

The module itself, if that's still needed.

Kay

Re: Best practice for loading platform modules

[Chris Hallinan]

On Wed, Apr 29, 2009 at 3:46 PM, Kay Sievers <kay.sievers@vrfy.org> wrote:
> On Wed, Apr 29, 2009 at 21:35, Chris Hallinan <challinan@gmail.com> wrote:
>
>> Erm, I am the distro maintainer!  I tried just today w/ 2.6.29.  The
>> problem seems like a mismatch.  Kernel passes "platform:musb_hdrc".
>> Udev receives and passes same to modprobe.  The answer I'm seeking is
>> "what is the correct/proper approach?"  Should I seek to modify the
>> kernel and remove the "platform:" prefix, or should I seek a
>> modification to modprobe to be able to understand and strip off the
>> "platform:" prefix?  Or should udev have a construct/rules syntax to
>> strip the "platform:" prefix?
>
> The module itself, if that's still needed.
>

OK, the light bulb just came on!

Thanks, Kay, and Greg for hanging in there with me.  I added
MODULE_ALIAS("platform:musb_hdrc) to musb_core.c, and all works as advertised.

Thanks again,

Chris



-- 
Life is like Linux - it never stands still.

Re: Best practice for loading platform modules

[Greg KH]

On Wed, Apr 29, 2009 at 03:53:53PM -0400, Chris Hallinan wrote:
> On Wed, Apr 29, 2009 at 3:46 PM, Kay Sievers <kay.sievers@vrfy.org> wrote:
> > On Wed, Apr 29, 2009 at 21:35, Chris Hallinan <challinan@gmail.com> wrote:
> >
> >> Erm, I am the distro maintainer!

Heh, my condolences...

You should possibly post from your company email address if you expect
us to know this, gmail.com doesn't exactly specify much :)

> >> I tried just today w/ 2.6.29.  The
> >> problem seems like a mismatch.  Kernel passes "platform:musb_hdrc".
> >> Udev receives and passes same to modprobe.  The answer I'm seeking is
> >> "what is the correct/proper approach?"  Should I seek to modify the
> >> kernel and remove the "platform:" prefix, or should I seek a
> >> modification to modprobe to be able to understand and strip off the
> >> "platform:" prefix?  Or should udev have a construct/rules syntax to
> >> strip the "platform:" prefix?
> >
> > The module itself, if that's still needed.
> >
> 
> OK, the light bulb just came on!
> 
> Thanks, Kay, and Greg for hanging in there with me.  I added
> MODULE_ALIAS("platform:musb_hdrc) to musb_core.c, and all works as advertised.

But that alias is already there, it was added in the 2.6.27-rc kernel,
in commit 550a7375.

So I'm confused as to why 2.6.29 would not work properly for you.

thanks,

greg k-h

Re: Best practice for loading platform modules

[Chris Hallinan]

On Wed, Apr 29, 2009 at 4:18 PM, Greg KH <greg@kroah.com> wrote:
> On Wed, Apr 29, 2009 at 03:53:53PM -0400, Chris Hallinan wrote:
>> On Wed, Apr 29, 2009 at 3:46 PM, Kay Sievers <kay.sievers@vrfy.org> wrote:
>> > On Wed, Apr 29, 2009 at 21:35, Chris Hallinan <challinan@gmail.com> wrote:
>> >
>> >> Erm, I am the distro maintainer!
>
> Heh, my condolences...
>
> You should possibly post from your company email address if you expect
> us to know this, gmail.com doesn't exactly specify much :)

Well, my inquiries are related to my research while preparing the
second edition of my book "Embedded Linux Primer", as opposed to my
duties as an FAE for MontaVista!  Thus the posting from my personal
address.

>
>> >> I tried just today w/ 2.6.29.  The
>> >> problem seems like a mismatch.  Kernel passes "platform:musb_hdrc".
>> >> Udev receives and passes same to modprobe.  The answer I'm seeking is
>> >> "what is the correct/proper approach?"  Should I seek to modify the
>> >> kernel and remove the "platform:" prefix, or should I seek a
>> >> modification to modprobe to be able to understand and strip off the
>> >> "platform:" prefix?  Or should udev have a construct/rules syntax to
>> >> strip the "platform:" prefix?
>> >
>> > The module itself, if that's still needed.
>> >
>>
>> OK, the light bulb just came on!
>>
>> Thanks, Kay, and Greg for hanging in there with me.  I added
>> MODULE_ALIAS("platform:musb_hdrc) to musb_core.c, and all works as advertised.
>
> But that alias is already there, it was added in the 2.6.27-rc kernel,
> in commit 550a7375.
>
> So I'm confused as to why 2.6.29 would not work properly for you.

Erm, my own misunderstanding of the problem contributed to the
confusion.  I booted 2.6.29, saw that the kernel was passing
"platform:<driver>", using my test script that dumps env, (instead of
passing it to modprobe) and assumed it was still broken.  Yes, my
2.6.29 does indeed contain those changes, and I'm sure it will work
correctly.  Learn something new every day!!!

Thanks again for the help.  I think I know enough about udev for
Chapter 18: Udev!  ;)

Best,

Chris

PS: I enjoy your own writing, Greg!

-- 
Life is like Linux - it never stands still.