Re: [PATCH 10/11] iwlwifi: rely on API version read from firmware

[reinette chatre <reinette.chatre@intel.com>]

On Tue, 2008-12-02 at 13:51 -0800, Marcel Holtmann wrote:
> Hi Reinette,
> 
> > > > This adds the infrastructure to support older firmware APIs.
> > > > The API version number is stored as part of the filename, we first try to
> > > > load the most recent firmware and progressively try lower versions.
> > > > The API version is also read from the firmware self and stored as part
> > > > of the iwl_priv structure. Only firmware that is supported by driver will
> > > > be loaded. The version number read from firmware is compared
> > > > to supported versions in the driver not the API version used as part of
> > > > filename.
> > > 
> > > thanks for doing this. This can really help smooth upgrade path in case
> > > the firmware API changes.
> > >  
> > > > -MODULE_FIRMWARE(IWL5000_MODULE_FIRMWARE);
> > > > -MODULE_FIRMWARE(IWL5150_MODULE_FIRMWARE);
> > > > +MODULE_FIRMWARE(IWL5000_MODULE_FIRMWARE(IWL5000_UCODE_API_MAX));
> > > > +MODULE_FIRMWARE(IWL5150_MODULE_FIRMWARE(IWL5150_UCODE_API_MAX));
> > > 
> > > So we don't have clear semantics on how MODULE_FIRMWARE should be used.
> > > The current way is that it list all potential firmware versions. So in
> > > cases it support API -1 and -2, then it has to list both. And not only
> > > the latest.
> > 
> > Even though the driver supports older firmware we really want the user
> > to use the latest firmware. The driver will also print a clear error if
> > the user is using API -1 and the latest available is -2. For this reason
> > we do want to make clear through MODULE_FIRMWARE which firmware the
> > driver would like to work with: the latest.
> 
> as I said, we never clearly defined the semantics of MODULE_FIRMWARE for
> these cases. The only problem that I see is that some initrd tools are
> actually using MODULE_FIRMWARE to figure which files need to be included
> for the builtin kernel drivers. So on a system with old firmware only,
> it will not include these.

This seems to fit in with the other suggestion you had to add a big
warning to modules_install when MODULE_FIRMWARE is not present on the
filesystem. We want users to have a working wireless system at the time
they find out they need new firmware. So, if initrd figures out it needs
new firmware the user can just go and download it at that time.

> 
> Maybe we need to extend MODULE_FIRMWARE. Maybe something like
> MODULE_ALTERNATE_FIRMWARE or some extra flags.

I don't think this is necessary.  

Reinette