[Bug] usb: dwc2: host function broken in OTG mode on bcm283x
stefan.wahren at i2se.com
Mon Aug 22 21:40:52 PDT 2016
> John Youn <John.Youn at synopsys.com> hat am 23. August 2016 um 00:08
> On 8/22/2016 1:47 PM, Stefan Wahren wrote:
> > Hi John,
> >> John Youn <John.Youn at synopsys.com> hat am 22. August 2016 um 22:06
> >> geschrieben:
> >> On 8/20/2016 6:03 AM, Stefan Wahren wrote:
> >>> Hi John,
> >> Hi Stefan,
> >> Why doesn't DT work? I think all the properties are there to set
> >> these.
> > it only works for the future releases, not for existing DT blobs. The DT is
> > part
> > of the ABI.
> > Yes, i already send a patch to fix the DT , but it would be better to fix
> > the
> > issue in deep.
> I don't know much about DT issues. So I'm not sure what the issue is
> with this.The properties for the gadget fifo sizes already exist so
> there's no change to the ABI, correct?
It's correct the ABI doesn't change, but it also means that newer kernel must
work with older DTB files. So the fix should be better in the driver and not in
the DT sources.
> Also, the patch you linked doesn't seem to have settings for the FIFO
It wouldn't make sense to provide gadget fifo sizes for a host-only setup. The
mentioned patch doesn't really fix this issue, that's why i mentioned that we
should fix this issue first.
> >> I hesitate to change the legacy/default settings in case it breaks any
> >> existing drivers that depend on them.
> > Platforms? I didn't expected that other drivers use these settings.
> Yes, I mean platforms.
> I believe these values were hard-coded since before the unified "dwc2"
> existed. So for sure there are platforms using these settings and to
> lower them will negatively impact those platforms. The DT properties
> were introduced to override these.
Unfortunately the gadget fifo sizes should have been marked as required for OTG
and gadget mode. But now it's too late. We only can improve DT binding here or
at least add a warning during probe.
> I think it is best to override them in the broken platforms. Either
> that or fix it so that the defaults are more intelligently determined
> in a way that maximizes the fifo sizes and cannot fail, as opposed to
> some other arbitrary values.
I don't have the knowledge for the second solution and i wouldn't prefer much
more detection intelligence.
> > Another option would be to add the gadget fifo sizes to platform data, but
> > it's
> > not a nice solution.
> I agree, that wouldn't be good.
More information about the linux-rpi-kernel