Re: 3.2.11: PCI Express card cannot be re-detected withing cca 60sec timeframe

[Yinghai Lu <yinghai@kernel.org>]

On Fri, Apr 20, 2012 at 4:23 AM, Martin Mokrejs
<mmokrejs@fold.natur.cuni.cz> wrote:
> Yinghai Lu wrote:
>> On Thu, Apr 19, 2012 at 2:48 PM, Yinghai Lu <yinghai@kernel.org> wrote:
>>> On Thu, Apr 19, 2012 at 4:16 PM, Martin Mokrejs
>>>> I just tried something similar under linux, with all the patches I accumulated for 3.4-rc3
>>>> and posted in my previous emails. It seems linux flips some value so on every second card
>>>> removal it really does realize it was just unplugged.
>>>
>>> that is really looks like silicon problem.
>>>
>> please check attached patch. Hope link change bit is flip around...
>> if there is change, we could add some logic:
>>    check if the device is there, or will reset link to normal state.
>
> I don't think it helped but maybe you will see some difference:

No, the interrupt still get delayed.

>
> Complete dmesg is attached. OK, the "60sec" delay is when xhci_hcd gives away its attempts.
> I am not knowlegeable of the kernel at all but it is my impression that because of the
> card presence detection there is no way to prevent that. But, if I get it right, now when I
> pluging the card+USB disk in it into the computer before those "60sec" it gets detected,
> so my problem is gone (xhci cancels the timeout because a "new card" was popped in with a "new"
> device.
>
> So, although I am now not sure whether I need your last patch, for the practical purpose that
> I want to plugin a new card with a device into my computer, it works now. (added Sarah Sharp
> into CC: just to make her aware of this scenario, no problem with xhci_hcd in my eyes)
> ;)

please try to "safely remove the card" before removing the card physically like:

echo 1 > /sys/bus/pci/devices/0000:11:00.0/remove

that will stop the drivers and remove pci device in the kernel.

Yinghai