Re: USB3 link PM policy design

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Feb 24, 2012 at 10:53:26AM +0100, Oliver Neukum wrote:
> Am Donnerstag, 23. Februar 2012, 21:41:25 schrieb Sarah Sharp:
> > 3. When a new configuration is selected, save the old U1 and U2
> > timeouts, and then disable the timeouts and device-initiated U1/U2 (if
> > the device is configured).  When the USB core calls into the xHCI driver
> > to add a new periodic endpoint for a bandwidth change, add the service
> > interval and endpoint type to a small linked list of endpoints, sorted
> > by configuration and interface numbers.  If the USB core removes an
> > endpoint, delete it from the list.  We might even be able to reuse some
> > of those awful bandwidth tables I had to create for the Panther Point
> > xHCI host controller.
> > 
> > 4. Once the bandwidth change has successfully (but before the interfaces
> > are bound), call into the xHCI driver twice to calculate the new U1 and
> > U2 timeouts.  It will attempt to submit an Evaluate Context command to
> > change the max exit latency for the device.  If the xHC host agrees to
> > that latency, the xHCI driver returns a timeout.  Otherwise the call
> > returns an error.  (We have to increase the max exit latency stored by
> > the xHC before we change any timeouts in parent hubs, so that the xHC
> > will have added extra latency in its schedule before we enable U1/U2.)
> 
> Why do you want to take driver binding into account? It seems to me that
> that is too early. Periodic transfers matter only when the driver decides
> to actually use them. I don't think you can calculate U1/2 latencies merely
> from descriptors without taking into account whether endpoints are used.

That's what we currently do with the bandwidth calculations.  Without
some API to let drivers indicate what periodic endpoints they're using
and what service interval they're using, that's what we have to do.

I know that there are some xHCI hosts that change their schedules to
accommodate isochronous ping requests if an isochronous ring is added,
even if it isn't used.  So as long as we add the endpoints in
usb_set_configuration and usb_set_interface, we need to set the link
timeout policies in the same place.

Sarah Sharp
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Linux Media]     [Linux Input]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Old Linux USB Devel Archive]

  Powered by Linux