Re: device struct bloat

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

[-- Attachment #1: Type: text/plain, Size: 1334 bytes --]

On Sun, 2007-11-04 at 19:58 -0800, Greg KH wrote:
> On Sun, Nov 04, 2007 at 09:29:18PM +0100, Peter Zijlstra wrote:
> > 
> > On Sat, 2007-11-03 at 12:48 -0700, Stephen Hemminger wrote:
> > > The sizeof(struct device) is way too big, especially in the network device case.
> > > We want to support 1000's of device's and the change from class_device to
> > > net_device has caused needless bloat.
> > > 
> > > sizeof(struct device) = 272
> > > sizeof(struct class_device) = 92
> > >   * not the class_id in class_device could also be removed or changed to
> > >      a ptr, since it is redundant for net_devices.
> > 
> > The thing that surprised me most was that it contains a struct
> > semaphore, Greg, is that really needed?
> 
> Yes, it serializes bind and unbind stuff for the device.  There are
> comments about it in drivers/base/dd.c if you want to look into it.

Ah, that is not what I meant. My question was, why a semaphore, not a
mutex? Semaphores should not be used if at all possible - for one you
don't get lockdep checking your code.

/me converts the stuff over, and boots

Ah, that probably is the reason. This device tree is hard to annotate.
I'll try and think of something..

Anyway, Andy, could you make checkpatch whine on something like:
   ^\+[[::space:]]*struct semaphore




[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]