Writing modules for 2.5

Writing modules for 2.5

[Måns Rullgård]

What magic needs to be done when writing modules for linux 2.5.x?
Insmod tells me "Invalid module format" and the kernel log says "No
module found in object".  I've tried to mimic the foo.mod.c stuff in
the kernel tree, but I can't figure out the right way to do it.

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Jonathan Corbet]

> What magic needs to be done when writing modules for linux 2.5.x?

May I humbly suggest LWN's "porting drivers to 2.5" series?

	http://lwn.net/Articles/driver-porting

There's 20-some articles there now on various API changes in the 2.5
kernel.  The series is getting reasonably complete at this point; hopefully
it will be helpful.

jon

Jonathan Corbet
Executive editor, LWN.net
corbet@lwn.net

Re: Writing modules for 2.5

[Sam Ravnborg]

On Tue, Apr 15, 2003 at 01:15:17PM +0200, Måns Rullgård wrote:
> 
> What magic needs to be done when writing modules for linux 2.5.x?

First of all you need to follow Documentation/modules.txt when
compiling modules.
If you try to make your own build system you are sure to hit problems.

	Sam

Re: Writing modules for 2.5

[Russell King]

On Tue, Apr 15, 2003 at 05:42:07PM +0200, Måns Rullgård wrote:
> Does this work in 2.4 also

Yes.

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: Writing modules for 2.5

[Måns Rullgård]

Russell King <rmk@arm.linux.org.uk> writes:

> > My situation is like this: I am converting a char device driver to
> > work with linux 2.5.  In the open and close functions there are
> > MOD_INC/DEC_USECOUNT calls.  The question is what they should be
> > replaced with.  Will it be handled correctly without them?
> 
> If it's a character device driver using the struct file_operations,
> set the owner field as Alan mentioned, and remove the
> MOD_{INC,DEC}_USE_COUNT macros from the open/close methods.  This
> allows chrdev_open() (in fs/char_dev.c) to increment your module use
> count automatically.

Does this work in 2.4 also, or is the owner field used for something
else there?

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Alan Cox]

On Maw, 2003-04-15 at 13:39, Måns Rullgård wrote:
> My situation is like this: I am converting a char device driver to
> work with linux 2.5.  In the open and close functions there are
> MOD_INC/DEC_USECOUNT calls.  The question is what they should be
> replaced with.  Will it be handled correctly without them?

Remove the MOD_INC/DEC counts in the open/close path and add
owner: THIS_MODULE in the file ops, and all is happy.

Re: Writing modules for 2.5

[Dave Jones]

On Tue, Apr 15, 2003 at 01:57:58PM +0100, Russell King wrote:
 > On Tue, Apr 15, 2003 at 02:39:14PM +0200, M?ns Rullg?rd wrote:
 > > My situation is like this: I am converting a char device driver to
 > > work with linux 2.5.  In the open and close functions there are
 > > MOD_INC/DEC_USECOUNT calls.  The question is what they should be
 > > replaced with.  Will it be handled correctly without them?
 > 
 > If it's a character device driver using the struct file_operations,
 > set the owner field as Alan mentioned, and remove the
 > MOD_{INC,DEC}_USE_COUNT macros from the open/close methods.  This
 > allows chrdev_open() (in fs/char_dev.c) to increment your module use
 > count automatically.

Unless the open/close functions are doing funky things (like the
watchdog drivers do).

		Dave

Re: Writing modules for 2.5

[Russell King]

On Tue, Apr 15, 2003 at 02:15:18PM +0200, Andreas Henriksson wrote:
> On Tue, Apr 15, 2003 at 02:05:48PM +0200, Måns Rullgård wrote:
> > Next question:  what is the correct replacement for MOD_INC_USE_COUNT?
> 
> "SET_MODULE_OWNER(dev);" or simply "dev->owner=THIS_MODULE;" and the
> counting should then be done automatically AFAIK.

You don't need that for char drivers.

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: Writing modules for 2.5

[Russell King]

On Tue, Apr 15, 2003 at 02:39:14PM +0200, Måns Rullgård wrote:
> My situation is like this: I am converting a char device driver to
> work with linux 2.5.  In the open and close functions there are
> MOD_INC/DEC_USECOUNT calls.  The question is what they should be
> replaced with.  Will it be handled correctly without them?

If it's a character device driver using the struct file_operations,
set the owner field as Alan mentioned, and remove the
MOD_{INC,DEC}_USE_COUNT macros from the open/close methods.  This
allows chrdev_open() (in fs/char_dev.c) to increment your module use
count automatically.

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: Writing modules for 2.5

[Måns Rullgård]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> > Next question:  what is the correct replacement for MOD_INC_USE_COUNT?
> 
> If you know the use count is already one there isnt one, jut use
> MOD_INC_USE_COUNT and flame Rusty
> 
> If you can't prove the use count is already one for your own module
> you have a problem (and did in 2.4 in truth). The layers try and 
> lock the layer below. So char drivers are locked by core code,
> block drivers likewise, tty drivers and so on...
> 
> Most driver layers now have an owner: field in the file ops or other
> structure they register
> 
> 	owner: THIS_MODULE,
> 
> (THIS_MODULE comes out currently in a non modular buils too)

My situation is like this: I am converting a char device driver to
work with linux 2.5.  In the open and close functions there are
MOD_INC/DEC_USECOUNT calls.  The question is what they should be
replaced with.  Will it be handled correctly without them?

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Måns Rullgård]

Andreas Henriksson <andreas@fjortis.info> writes:

> > Next question:  what is the correct replacement for MOD_INC_USE_COUNT?
> 
> "SET_MODULE_OWNER(dev);" or simply "dev->owner=THIS_MODULE;" and the
> counting should then be done automatically AFAIK.

OK, but what might dev be and when should this be done?  What about
MOD_DEC_USE_COUNT?

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Andreas Henriksson]

On Tue, Apr 15, 2003 at 02:05:48PM +0200, Måns Rullgård wrote:
> Next question:  what is the correct replacement for MOD_INC_USE_COUNT?

"SET_MODULE_OWNER(dev);" or simply "dev->owner=THIS_MODULE;" and the
counting should then be done automatically AFAIK.

> -- 
> Måns Rullgård
> mru@users.sf.net

Regards,
Andreas Henriksson

Re: Writing modules for 2.5

[Måns Rullgård]

Andi Kleen <ak@suse.de> writes:

> > What magic needs to be done when writing modules for linux 2.5.x?
> > Insmod tells me "Invalid module format" and the kernel log says "No
> > module found in object".  I've tried to mimic the foo.mod.c stuff in
> > the kernel tree, but I can't figure out the right way to do it.
> 
> Welcome to the wonderful new world of in kernel module loading, finally
> with understandable error messages. Now bad error reporting is not limited
> to netlink anymore.
> 
> You need -DKBUILD_BASENAME=name
> 
> Also you need module_init/module_exit declarations.

Next question:  what is the correct replacement for MOD_INC_USE_COUNT?

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Andi Kleen]

On Tue, Apr 15, 2003 at 02:00:45PM +0200, M?ns Rullg?rd wrote:
> Andi Kleen <ak@suse.de> writes:
> 
> > > What magic needs to be done when writing modules for linux 2.5.x?
> > > Insmod tells me "Invalid module format" and the kernel log says "No
> > > module found in object".  I've tried to mimic the foo.mod.c stuff in
> > > the kernel tree, but I can't figure out the right way to do it.
> > 
> > Welcome to the wonderful new world of in kernel module loading, finally
> > with understandable error messages. Now bad error reporting is not limited
> > to netlink anymore.
> > 
> > You need -DKBUILD_BASENAME=name
> 
> Should I also add -DKBUILD_MODNAME=name?

Yes.

-Andi

Re: Writing modules for 2.5

[Måns Rullgård]

Andi Kleen <ak@suse.de> writes:

> > What magic needs to be done when writing modules for linux 2.5.x?
> > Insmod tells me "Invalid module format" and the kernel log says "No
> > module found in object".  I've tried to mimic the foo.mod.c stuff in
> > the kernel tree, but I can't figure out the right way to do it.
> 
> Welcome to the wonderful new world of in kernel module loading, finally
> with understandable error messages. Now bad error reporting is not limited
> to netlink anymore.
> 
> You need -DKBUILD_BASENAME=name

Should I also add -DKBUILD_MODNAME=name?

-- 
Måns Rullgård
mru@users.sf.net

Re: Writing modules for 2.5

[Alan Cox]

On Maw, 2003-04-15 at 13:05, Måns Rullgård wrote:
> Next question:  what is the correct replacement for MOD_INC_USE_COUNT?

If you know the use count is already one there isnt one, jut use
MOD_INC_USE_COUNT and flame Rusty

If you can't prove the use count is already one for your own module
you have a problem (and did in 2.4 in truth). The layers try and 
lock the layer below. So char drivers are locked by core code,
block drivers likewise, tty drivers and so on...

Most driver layers now have an owner: field in the file ops or other
structure they register

	owner: THIS_MODULE,

(THIS_MODULE comes out currently in a non modular buils too)

Re: Writing modules for 2.5

[Andi Kleen]

mru@users.sourceforge.net (Måns Rullgård) writes:

> What magic needs to be done when writing modules for linux 2.5.x?
> Insmod tells me "Invalid module format" and the kernel log says "No
> module found in object".  I've tried to mimic the foo.mod.c stuff in
> the kernel tree, but I can't figure out the right way to do it.

Welcome to the wonderful new world of in kernel module loading, finally
with understandable error messages. Now bad error reporting is not limited
to netlink anymore.

You need -DKBUILD_BASENAME=name

Also you need module_init/module_exit declarations.

-Andi