Re: [PATCH] MMC host class

[Pierre Ossman <drzeus-list@drzeus.cx>]

Russell King wrote:

>On Fri, Jul 15, 2005 at 10:21:43PM +0200, Pierre Ossman wrote:
>  
>
>>Russell King wrote:
>>    
>>
>>>Also note that since we have a class_dev, the mmc_host 'dev' field can
>>>be removed.  However, we'll probably have to update the host drivers
>>>to do this, so it should be a separate patch.
>>>      
>>>
>>I believe there's a bit of abstraction to be gained from not poking
>>around inside the class_dev struct in too many places. It's not like
>>we're wasting any large amounts of memory.
>>    
>>
>
>I still don't like the needless duplication.  How about doing it this
>way (see the attached patch.)
>  
>

The mmc_hostname macro seems like a good solution. It'll keep the
abstraction even if stuff needs to be moved around.

I see a problem with waiting until mmc_add_host() until initialising the
kobject though. If a driver calls mmc_alloc_host() and then
mmc_free_host(), perhaps because of some error, then the structure won't
be freed since we rely on release getting called. That's why I tried to
get the kobject stuff set up with the allocation.

Perhaps it is possible to test if a kobject is initialised and if not
free the structure directly?

Rgds
Pierre