Hi Greg,
On Mon, Dec 18, 2017 at 12:43:48PM +0100, Greg Kroah-Hartman wrote:
> On Fri, Dec 15, 2017 at 07:05:39PM -0800, Matthias Kaehlcke wrote:
> > We identified the above patch as the culprit, in combination with USB
> > autosuspend being enabled for the Bluetooth controller.
> >
> > We found that the following (recent) upstream patch fixes the issue on
> > most devices (we are still investigating one case on a specific device):
A key word to highlight here is "most". I have at least one device that
is consistently having problems even with both patches included; the
only way things work reliably so far is to simply revert the $subject
patch in 4.4.x.
I'll try to investigate further, since this result is a bit confusing
still. If there's a real problem with the patch (and I suspect there
might be), it probably shouldn't be in mainline either...
> > commit a0085f2510e8976614ad8f766b209448b385492f
> > Author: Sukumar Ghorai <sukumar.ghorai@xxxxxxxxx>
> > Date: Wed Aug 16 14:46:55 2017 -0700
> >
> > Bluetooth: btusb: driver to enable the usb-wakeup feature
[...]
> > stable branches are currently broken for BTUSB_QCA_ROME with USB
> > autosuspend enabled, since the above patch is not included (I only
> > checked v4.4 and v4.9), so we probably want to integrate it.
>
> Now queued up, thanks for letting me know.
I think you have almost as much information as I do at this point, and
I'll try to reply back here if I figure out anything more, so I'll leave
you to your decisions. But I would personally revert, not backport more
patches.
Among the reasons:
(a) I have at least one device that does not work better with both
patches [1]
(b) So far, I haven't seen any explanation why commit a0085f2510e8
should be a dependency for $subject patch; the closest I can imagine
would be that commit a0085f2510e8 could effectively neutralize
$subject patch -- if the device is marked as a wakeup source, then
it will never try to RESET on resume -- but that's still a fuzzy
signal; just because it's marked a wakeup source sometimes doesn't
mean it always will be. We can disable it in user space too.
(c) Isn't it safer to revert than to backport more? You're delving into
feature-land, not bugfix-land...
Brian
[1] Before you ask: this particular device is not quite fully supported
on upstream yet, though its sibling devices are. With a bit of effort, I
might be able to test a clean upstream. At the moment, I'm using our
Chrom{e,ium}OS 4.4 kernel, where we regularly merge in -stable updates.