[PATCH 0/9] staging: comedi: update auto-attach mechanism

[Ian Abbott]

On 2012-10-24 05:16, Greg Kroah-Hartman wrote:
> On Tue, Oct 23, 2012 at 10:53:22AM +0100, Ian Abbott wrote:
>> On 2012-10-22 20:44, Greg Kroah-Hartman wrote:
>>> On Mon, Oct 15, 2012 at 01:07:30PM +0100, Ian Abbott wrote:
>>>> The auto-configuration/auto-attach mechanism in Comedi used by most of
>>>> the low-level PCI and USB Comedi hardware drivers currently uses
>>>> separate hooks in `struct comedi_driver` for each bus type -
>>>> `->attach_usb()` and `->attach_pci()`.  I'd like to move away from that
>>>> and use a single auto-attach hook, `->auto_attach()` and make it easy
>>>> for an implementation of this hook in a low-level driver to get a
>>>> pointer to the `struct pci_dev`, `struct usb_interface` or whatever
>>>> device wrapper is relevant to the device being auto-attached to comedi.
>>>>
>>>> Also, the Comedi core exports nice wrapper functions for auto-attaching
>>>> PCI or USB devices (`comedi_pci_auto_config()` and
>>>> `comedi_usb_auto_config()`) along with functions for auto-detaching
>>>> them, but there is currently no way to auto-attach devices on some
>>>> arbitrary type of bus.  Export functions `comedi_auto_config()`
>>>> (different to the current static function of the same name) and
>>>> `comedi_auto_unconfig()`) to allow this.
>>>>
>>>> Eventually, there will be nice wrapper functions for some more commonly
>>>> used device types such as PCMCIA, but it should be easy enough to
>>>> implement those as static inline wrappers around `comedi_auto_config()`
>>>> and `comedi_auto_unconfig()`.
>>>>
>>>> This patch series only changes one PCI driver and one USB driver to use
>>>> the new `auto_attach()` hook, but it will be easy enough to change the
>>>> others.  Once that is done, the `attach_pci()` and `attach_usb()` hooks
>>>> can be removed from `struct comedi_driver`.
>>>
>>> I took the first two patches here only, care to rework the rest and
>>> resend them?
>>
>> I'm prepared to discuss the patches.  One thing I want to allow is
>> for drivers to call comedi_auto_config() directly using whatever
>> context value they want (as long as it can be represented by an
>> unsigned long), and to have the same value passed back in their
>> auto_attach routine.
>
> Ok, as long as nothing else outside that driver is trying to intrepret
> that value as meaning anything.

Only the driver would interpret the context value, although most would 
just ignore it.  If the driver called comedi_auto_config() directly, it 
would get the context back in its ->auto_config() hook.  If the driver 
used one of the wrappers around comedi_auto_config() such as 
comedi_pci_auto_config(), it would get the predefined magic value as the 
context "for free"; well, the context parameter would have to be set to 
_something_ and some magic value that depends on which wrapper was 
called is possibly more useful than a plain old 0 or NULL value.

Having comedi_pci_auto_config() pass a pointer to the `struct pci_dev` 
in the context (and similar for other wrappers around 
comedi_auto_config() handling different device types) wouldn't be 
terribly useful to an ->auto_config() hook that could handle multiple 
types of device wrappers as it wouldn't be able to tell what sort of 
device type it should be dealing with as one random pointer value looks 
pretty much like any other.

-- 
-=( Ian Abbott @ MEV Ltd.    E-mail: <abbotti at mev.co.uk>        )=-
-=( Tel: +44 (0)161 477 1898   FAX: +44 (0)161 718 3587         )=-