Re: [PATCH 5/8] staging: vme: add functions for bridge module refcounting

[Martyn Welch <martyn.welch@ge.com>]

On 04/08/11 17:34, Emilio G. Cota wrote:
> On Thu, Aug 04, 2011 at 08:23:48 +0100, Martyn Welch wrote:
>> On 03/08/11 16:23, Emilio G. Cota wrote:
>>> On Wed, Aug 03, 2011 at 16:06:30 +0200, Manohar Vanga wrote:
>>>>> Can we not do this inside vme_master_request, vme_slave_request, etc?
>>>>> I.e. Add reference count when resources are given out.
>>>>> This would then be done in the vme_*_free routines.
>>>>
>>>> I agree this would be much better. I will change this and resend :)
>>>
>>> To me it seems better to keep this as is, to be
>>> used by .probe and .release methods, in the same way
>>> usb_get_dev() and usb_put_dev() are used by usb drivers.
>>
>> The description of the functions in usb.c suggests, to me at least, that these
>> functions are the equivalent to the vme_*_request functions in the vme code.
>> Being called when a function binds to an interface and when it is finished
>> with it.
> 
> Which functions? usb_get_dev and usb_put_dev? These _only_
> in/de-crement the refcounts. They're called by .probe and
> .release methods of usb drivers. The "resources" or "services"
> the callers acquire from the usb bridge is orthogonal.
> 
> The point is to separate incrementing the refcounts from other
> functionality. We have currently five "request" functions:
> vme_{master,slave,dma,irq,lm}_request. The average driver
> would call two or three of these for each device. This would
> result in a fairly large refcount for the bridge, that would
> tell us nothing about how many users (devices) are hanging in
> there--"yeah, we have lots" is not good enough.
> 

Actually, it would give you a good indication of how many of the resources
provided by each VME bridge chip were used. I don't see the refcount
accurately reflecting the number of users as being important, more as a means
of tracking which bridges have resources that are being used (and therefore
can't be removed).

> So no, usb_get_dev and usb_put_dev are not equivalent to our
> vme_*_request functions. The get/put functions only operate
> on the refcounts, to mark the dependency of a struct device
> on another device (the bridge) so that the parent cannot
> be removed. This is the way things are done in other much
> more mature subsystems, I don't see why we should do it
> differently.
> 

The USB bus and VME bus are very different entities.

Martyn

-- 
Martyn Welch (Principal Software Engineer) | Registered in England and
GE Intelligent Platforms                   | Wales (3828642) at 100
T +44(0)127322748                          | Barbirolli Square, Manchester,
E martyn.welch@ge.com                      | M2 3AB  VAT:GB 927559189