On 2012/04/10 07:24 PM, Greg KH wrote:
> On Mon, Apr 02, 2012 at 11:48:54AM +0100, Ian Abbott wrote:
>> On 2012-03-19 23:26, Greg KH wrote:
>>> On Mon, Mar 19, 2012 at 03:46:49PM -0700, Gerard Snitselaar wrote:
>>>> I looked at this a bit more. It looks like they lose visibility to the
>>>> pci_dev structure.
>>>>
>>>> *_probe()
>>>>    comedi_pci_auto_config()     pci_dev
>>>>       comedi_auto_config()      pci_dev->dev
>>>>          comedi_device_attach() ??
>>>>             driv->attach()      ??<= iterate through pci devices.
>>>>
>>>> Most of the examples I have looked at so far use for_each_pci_dev() to
>>>> find the device, and s626 shortcuts it a bit by directly making calls
>>>> to pci_get_subsys() with specific ids. They all verify they have the
>>>> right device by checking the bus and slot that are grabbed from the
>>>> pci_dev in comedi_pci_auto_config() and passed down.
>>>
>>> Ugh, surely there's a way to keep the pci dev through the
>>> comedi_device_attach() call, right?
>>
>> comedi_device_attach() is also called for the COMEDI_DEVCONFIG ioctl
>> for "manually" configuring a comedi device, and that has no idea
>> about struct pci_dev, etc.
>>
>> I recently posted a series of patches that allows lower-level comedi
>> drivers to supply separate hooks for auto-configuring PCI devices or
>> USB devices without abusing the old "manual configuration" code
>> paths, see <http://driverdev.linuxdriverproject.org/pipermail/devel/2012-March/025331.html>.
>>
>> The old loop that searches the PCI bus is still needed for the
>> "manual configuration" code path.
>
> So, now that I've applied your patches, this patch isn't needed anymore,
> right?  Or should it be reworked to use the new interfaces?

I think it's still needed as that the comedi_driver->attach() hook is 
still needed to support manual configuration by the COMEDI_DEVCONFIG 
ioctl, and the existing code really shouldn't be looking in 
s626_pci_table[] as it's tagged __devinitconst.

I never saw any mismatched section warnings when I compiled a linux-next 
kernel on my system with a full set of comedi drivers, even with 
CONFIG_DEBUG_SECTION_MISMATCH configured.  I'm not sure why not, unless 
it has something to do with me building the kernel with a separate build 
directory.

The new comedi_driver->attach_pci() hook should be implemented for this
driver as well, and it's on my "todo" list (in my head!).

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