Bug#642911: TX watchdog timeout on RTL8168B

[Francois Romieu <romieu@fr.zoreil.com>]

Jonathan Nieder <jrnieder@gmail.com> :
[...]
>  - Is the PCI commit after writing to IntrStatus needed in the
>    context of 3.0.y which does not have 92fc43b4159b ("r8169: modify
>    the flow of the hw reset") but does call irq_mask_and_ack from time
>    to time without a PCI commit immediately afterward?  What symptoms
>    does it address?

MMIO writes are posted. One can only say that the write will be committed
somewhere between the time it is requested and the next MMIO read for
this register range.

>  - Does init_one need the same "don't ack uninteresting events" change
>    (0xffff -> tp->intr_event)?

No. Init is different. It happens once and the driver is only in control
afterwards.

>  - Am I understanding correctly that patch 3/4 (remove now-redundant
>    case statements) is a code cleanup / no-op?

Yes.

-- 
Ueimor