Re: avoiding pci_disable_device()...

[Greg KH]

On Sun, Feb 13, 2005 at 08:42:55PM -0500, Jeff Garzik wrote:
> 
> Currently, in almost every PCI driver, if pci_request_regions() fails -- 
> indicating another driver is using the hardware -- then 
> pci_disable_device() is called on the error path, disabling a device 
> that another driver is using
> 
> To call this "rather rude" is an understatement :)
> 
> Fortunately, the ugliness is mitigated in large part by the PCI layer 
> helping to make sure that no two drivers bind to the same PCI device. 
> Thus, in the vast majority of cases, pci_request_regions() -should- be 
> guaranteed to succeed.
> 
> However, there are oddball cases like mixed PCI/ISA devices (hello IDE) 
> or cases where a driver refers a pci_dev other than the primary, where 
> pci_request_regions() and request_regions() still matter.

But this is a very small subset of pci devices, correct?

> As a result, I have committed the attached patch to libata-2.6.  In many 
> cases, it is a "semantic fix", addressing the case
> 
> 	* pci_request_regions() indicates hardware is in use
> 	* we rudely disable the in-use hardware
> 
> that would not occur in practice.
> 
> But better safe than sorry.  Code cuts cut-n-pasted all over the place.
> 
> I'm hoping one or two things will happen now:
> * janitors fix up the other PCI drivers along these lines
> * improve the PCI API so that pci_request_regions() is axiomatic

Do you have any suggestions for how to do this?

thanks,

greg k-h
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/