Bill Nottingham wrote:
> With a 2.6.6-rc1 based kernel. Happened when loading ehci_hcd some
> 10 hours after booting, couldn't reproduce in initial attempts. I
> suppose the question is also why it failed to init, but it certainly
> didn't like the failure...

Hmm, no it didn't.  The "illegal capability" is the hardware acting
broken (what kind of EHCI hardware?); I've had reports of similar
stuff happening after ACPI resume (bogus PCI config space values,
in this case zero).

The "-19" means -ENODEV, which seem likely to be another case of a PCI
request not responding correctly:  handshake() failing because reading
a register returned 0xffffffff.

Looks like a cleanup path needs to handle early failure a bit better;
likely just having ehci_stop test for ehci->async non-null (before
calling scan-async to clean up any pending work) would suffice.

- Dave


> Bill
> 
> PCI: Enabling device 0000:00:1d.7 (0000 -> 0002)
> ehci_hcd 0000:00:1d.7: EHCI Host Controller
> ehci_hcd 0000:00:1d.7: illegal capability!
> PCI: Setting latency timer of device 0000:00:1d.7 to 64
> ehci_hcd 0000:00:1d.7: irq 11, pci mem 22946000
> ehci_hcd 0000:00:1d.7: new USB bus registered, assigned bus number 1
> ehci_hcd 0000:00:1d.7: init error -19ehci_hcd 0000:00:1d.7: remove, state 0
> Unable to handle kernel NULL pointer dereference at virtual address 00000048
> ...