Greg KH wrote:
> On Tue, Jul 15, 2003 at 04:10:38PM -0400, Trever L. Adams wrote:
> 
>>kernel: ehci_hcd 0000:00:02.2: PCI device 10de:0068 (nVidia Corporation)
>>kernel: irq 3: nobody cared!
>>kernel: Call Trace:
>>...
> 
> 
> Hm, but usb_hcd_irq() reports back the proper interrupt return value.  I
> don't see how this could happen, unless the ehci driver was somehow
> halted...
> 
> David, any ideas?

See if this patch resolves it.

The patch adds an explicit reset to HCD initialization, and then makes
EHCI use it.  (OHCI could do so even more easily ... but nobody's reported
firmware acting that type of strange with OHCI.)   It should prevent IRQs
being enabled while the HC is still in an indeterminate state.

This also fixes a missing local_irq_restore() that was generating some
annoying might_sleep() messages, and a missing readb() that affects some
ARM (and other) PCI systems.

- Dave