No stable driver/module api?

No stable driver/module api?

[Per Strandh]

Hi,

I'm working for a company with arm-based embedded-linux products.
Both the hardware and software is developed by ourselves.




In a slightly simplified description our products is moduled
(1) Assembler bootloader in the first partition.
(2) Compressed kernel in the second partion
(3) Root file system in the third partition

The kernel is ported, configured and compiled for our own hardware.
We have selected kernel version 3.0 since it is a LTS track.

The Root file system holds one driver for our hardware.





I have now encountered one problem with this layout.
When upgrading the kernel, our driver in the root file system isn't
always compatible/loading.

When upgrading from 3.0.18 to 3.0.22 it worked out fine.
When upgrading from 3.0.22 to 3.0.23 the driver isn't loading.
When trying to load it if fails with the message:
- "disagrees about version of symbol module_layout"

We have configured the kernel with the option
 CONFIG_MODULE_FORCE_LOAD=y



Isn't the driver/module api stable within the 3.0 track?

Regards
/Per/

No stable driver/module api?

[Arnd Bergmann]

On Sunday 23 June 2013 19:29:52 Per Strandh wrote:
> 
> I have now encountered one problem with this layout.
> When upgrading the kernel, our driver in the root file system isn't
> always compatible/loading.
> 
> When upgrading from 3.0.18 to 3.0.22 it worked out fine.
> When upgrading from 3.0.22 to 3.0.23 the driver isn't loading.
> When trying to load it if fails with the message:
> - "disagrees about version of symbol module_layout"
> 
> We have configured the kernel with the option
>  CONFIG_MODULE_FORCE_LOAD=y
> 
> 
> 
> Isn't the driver/module api stable within the 3.0 track?

The API is generally stable, unless there is a serious bug,
the ABI is not. If the API ever changes, you should get a build
warning when recompiling your module.

The easiest way to deal with those changes is to keep your
own code as a patch to the mainline kernel and build it all
together. That also makes it easier for your to show that
you are shipping the matching source code for all the
binary kernel code.

	Arnd

No stable driver/module api?

[Russell King - ARM Linux]

On Sun, Jun 23, 2013 at 08:13:16PM +0200, Arnd Bergmann wrote:
> On Sunday 23 June 2013 19:29:52 Per Strandh wrote:
> > 
> > I have now encountered one problem with this layout.
> > When upgrading the kernel, our driver in the root file system isn't
> > always compatible/loading.
> > 
> > When upgrading from 3.0.18 to 3.0.22 it worked out fine.
> > When upgrading from 3.0.22 to 3.0.23 the driver isn't loading.
> > When trying to load it if fails with the message:
> > - "disagrees about version of symbol module_layout"
> > 
> > We have configured the kernel with the option
> >  CONFIG_MODULE_FORCE_LOAD=y
> > 
> > 
> > 
> > Isn't the driver/module api stable within the 3.0 track?
> 
> The API is generally stable, unless there is a serious bug,
> the ABI is not. If the API ever changes, you should get a build
> warning when recompiling your module.
> 
> The easiest way to deal with those changes is to keep your
> own code as a patch to the mainline kernel and build it all
> together. That also makes it easier for your to show that
> you are shipping the matching source code for all the
> binary kernel code.

Also remember that there's: Documentation/stable_api_nonsense.txt
which covers this topic.

No stable driver/module api?

[Ezequiel Garcia]

Hello Per,

On Sun, Jun 23, 2013 at 2:29 PM, Per Strandh <per.strandh@gmail.com> wrote:
>
> I'm working for a company with arm-based embedded-linux products.
> Both the hardware and software is developed by ourselves.
>

Have you considered mainlining your drivers and the other changes made
to the kernel?

This way you won't need to bother with API changes, and only provide some help
maintaining the code from time to time. In my own experience, this is
less painful
than having to port your code on every release.
-- 
    Ezequiel