[SLOF] xhci in slof issues
Nikunj A Dadhania
nikunj at linux.vnet.ibm.com
Tue Apr 26 20:19:35 AEST 2016
Nikunj A Dadhania <nikunj at linux.vnet.ibm.com> writes:
> Thomas Huth <thuth at redhat.com> writes:
>
>> On 21.04.2016 12:36, Dinar Valeev wrote:
>>> Hi,
>>>
>>> While trying to fix my usb over vnc automated issues, I switched to
>>> xhci, which helped a lot OS wise, but I see a regression in SLOF.
>>>
>>> As you can see [1] we want to type an install repo url. But at some
>>> point capitalization of some characters are lost. Nope
>>> openSUsE-tumbleweed instead of openSUSE-Tumbleweed.
>>
>> I recently also hit a problem with XHCI keyboard ... and maybe they are
>> related. My problem is that "key?" is incredibly slow when using XHCI.
>> For example, run the following Forth code at the SLOF prompt:
>>
>> get-msecs 10 0 do key? . cr loop get-msecs swap - .d cr
>>
>> With XHCI, this takes more than 7.5 seconds to finish here, while with
>> OHCI, this loop finishs within 400ms!
>>
>> Nikunj, do you have a clue what might be wrong here?
>
> Nothing that comes out immediately, i was reviewing the code, here is
> the call chain:
>
> usb_key_available()
> ->usb_poll_key()
> ->usb_poll_intr()
> ->xhci_poll_intr()
> ->xhci_poll_event() (there is a 5sec timeout, which obviously
> its not hitting).
>
> Either this is taking around 700msec per
> call, which looks very wierd.
There is a transfer report packet that is getting generated every 500ms,
thats where the poll_event is exiting, and because of which we are
seeing this delay. I am trying to rework the interrupt handling here to
exit earlier.
>
> Will spend some more time and debug.
>
> Regards
> Nikunj
More information about the SLOF
mailing list