Hi,
> > > > > > > Until this is settled, swap the order to follow what Linux does.
> > > > > > > This is suboptimal, because it means controller drivers have different
> > > > > > > order of the operations and that you can't combine arbitrary PHYs and
> > > > > > > controllers, but it seems unlikely we will support combinations that
> > > > > > > aren't supported by Linux in the first place anyway.
> > > > > >
> > > > > > This is valuable information and I don't want it to be lost, so instead
> > > > > > of applying this as a fixup I rewrote the subject to:
> > > > > >
> > > > > > usb: dwc2: swap order of phy_init and phy_power_on to what Linux does
> > > > > >
> > > > > > and applied it as a separate patch.
> > > > >
> > > > > With this patch applied, some stm32mp1 do seem to timeout when
> > > > > addressing the core.
> > > > >
> > > > > WARNING: dwc2 49000000.usb-otg@xxxxxxxxxxx: dwc2_core_reset: Timeout! Waiting for Core Soft Reset
> > > > > ERROR: dwc2 49000000.usb-otg@xxxxxxxxxxx: probe failed: Connection timed out
> > > > >
> > > > > When I revert this patch. Everything works fine again.
> > > > >
> > > > > Is it possible that we can revert it in mainline now?
> > > >
> > > > Since 0e37f94fbe1b ("phy: stm32: sync with upstream"), the phy-stm32-usbphyc now
> > > > has no poweron/poweroff callbacks, so a revert seems to only slightly change
> > > > timing. I'd suggest you dig some more.
> > >
> > > That is not true. phy_power_on also enables the phy_regulator. Changing
> > > this order has the effect that the regulator is enabled after or before
> > > phy_init call.
> >
> > Oh, good point.
> >
> > > > > How can I reproduce the case this was fixing in the first place?
> > > >
> > > > The case was that barebox calls poweron/init in a different order than Linux.
> > > > This point is moot now though because poweron is no longer provided by the
> > > > driver.
> > >
> > > Still not true, though.
> >
> > Isn't this the same order Linux uses? Powering on regulator and then phy_init?
> > Why does it lead to issues in barebox, but not in Linux?
>
> I have no clue so far. Especially as this seems to have implications on
> the platform I am working on but not on the dk2. Although both use the
> same st,stpmic1 node vdd_usb for the phy_regulator.
In linux driver dwc2 (and dwc3) call phy_poweron then phy_init, which I
believe is not the right order, it should be init -> poweron.
The thing is that's a two part issue, first the usb drivers using phy
must use the correct call order, secondly the phy drivers must implement
the correct "flow"
_______________________________________________
barebox mailing list
barebox@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/barebox
