Re: 2.6.7-rc1 breaks forcedeth

[Jeff Garzik <jgarzik@pobox.com>]

Linus Torvalds wrote:
> But it's usually a good thing to try to reset a device as much as possible 
> when you probe for it. If for no other reason than the fact that then 
> you'll have it in a "known state", and hopefully there won't be as many 
> surprises..


Strongly agreed.  I stress this, in Linux driver writing talks and 
rants, to whomever will listen as "the proper way to do things in Linux".

Presuming things about a device's state upon entry to the driver has led 
to bugs in the past.  Popular bugs include assuming (a) MAC address 
registers hold a valid/useful value or (b) ethernet NIC's DMA engine is 
not active.  Both of these are quite often not true when you take into 
account driver re-loads, warm reboots, and firmware features such as USB 
kbd/mouse/storage or PXE booting from a network.

A good driver unconditionally makes sure the device is inactive in its 
probe function (struct pci_driver::probe), before registering itself 
with any kernel subsystems.  This must also be done before request_irq 
and before enabling the bus-mastering bit in PCI_COMMAND register.

	Jeff