Re: [PATCH 1/2] usb: ohci: Add per-port overcurrent quirk

[Hamish Martin]

On Mon, 2020-09-07 at 10:59 -0400, stern@xxxxxxxxxxxxxxxxxxx wrote:
> On Mon, Sep 07, 2020 at 01:50:10AM +0000, Hamish Martin wrote:
> > Hi Alan,
> > 
> > Thanks for your quick feedback. My replies are inline below.
> > 
> > On Fri, 2020-09-04 at 11:45 -0400, Alan Stern wrote:
> > > On Fri, Sep 04, 2020 at 03:22:46PM +1200, Hamish Martin wrote:
> > > > Some integrated OHCI controller hubs do not expose all ports of
> > > > the
> > > > hub
> > > > to pins on the SoC. In some cases the unconnected ports
> > > > generate
> > > > spurious overcurrent events. For example the Broadcom
> > > > 56060/Ranger
> > > > 2 SoC
> > > > contains a nominally 3 port hub but only the first port is
> > > > wired.
> > > > 
> > > > Default behaviour for ohci-platform driver is to use "ganged"
> > > > overcurrent protection mode. This leads to the spurious
> > > > overcurrent
> > > > events affecting all ports in the hub.
> > > > 
> > > > Allow this to be rectified by specifying per-port overcurrent
> > > > protection
> > > > mode via the device tree.
> > > > 
> > > > Signed-off-by: Hamish Martin <hamish.martin@xxxxxxxxxxxxxxxxxxx
> > > > >
> > > > ---
> > > >  drivers/usb/host/ohci-hcd.c      | 4 ++++
> > > >  drivers/usb/host/ohci-platform.c | 3 +++
> > > >  drivers/usb/host/ohci.h          | 1 +
> > > >  3 files changed, 8 insertions(+)
> > > > 
> > > > diff --git a/drivers/usb/host/ohci-hcd.c
> > > > b/drivers/usb/host/ohci-
> > > > hcd.c
> > > > index dd37e77dae00..01e3d75e29d9 100644
> > > > --- a/drivers/usb/host/ohci-hcd.c
> > > > +++ b/drivers/usb/host/ohci-hcd.c
> > > > @@ -687,6 +687,10 @@ static int ohci_run (struct ohci_hcd
> > > > *ohci)
> > > >  		val |= RH_A_NPS;
> > > >  		ohci_writel (ohci, val, &ohci->regs-
> > > > >roothub.a);
> > > >  	}
> > > > +	if (ohci->flags & OHCI_QUIRK_PER_PORT_OC) {
> > > > +		val |= RH_A_OCPM;
> > > > +		ohci_writel(ohci, val, &ohci->regs->roothub.a);
> > > > +	}
> > > 
> > > I don't think this is right, for two reasons.  First, isn't per-
> > > port 
> > > overcurrent protection the default?
> > 
> > Not as far as I understand the current code. Just above where my
> > patch
> > applies, the RH_A_OCPM (and RH_A_PSM) bits are explicitly cleared
> > in
> > 'val' with:
> >     val &= ~(RH_A_PSM | RH_A_OCPM);
> > 
> > This, coupled with the OHCI_QUIRK_HUB_POWER being set by virtue of
> > the
> > 'distrust_firmware' module param defaulting true, reads to me like
> > the
> > default is for ganged over-current protection. And that is my
> > experience in this case. 
> 
> You're right about that.  I hadn't noticed before; it makes little
> sense 
> to have a quirk that defaults to true.
> 
> It's not easy to tell the full story from the kernel history; that 
> module parameter predates the Git era.  I did learn that it was
> modified 
> in 2.6.3-rc3 and goes back even farther: see
> 
> 	https://marc.info/?l=linux-usb-devel&m=110628457424684&w=2
> 
> > If none of the quirks are selected then all of the fiddling with
> > 'val'
> > never gets written to 'ohci->regs->roothub.a'
> > 
> > I'd appreciate your reading of that analysis because I'm by no
> > means
> > sure of it.
> > 
> > > 
> > > Second, RH_A_OCPM doesn't do anything unless RH_A_NOCP is clear.
> > 
> > Correct, and that is my mistake. If I progress to a v2 of this
> > patch I
> > will update accordingly.
> 
> Shall we try changing the parameter's default value?  The USB
> subsystem 
> is a lot more mature and reliable now than it was back in 2004.

That doesn't really help me in my particular case. I tried turning the
param off and that just leads to the roothub.a reg not being modified
at all (and ganged over-current protection being left in place).

So, I guess I'm still back to my original idea of adding a new quirk
(perhaps quirk is not the best name for it in this case) that allows
the per-port over-current to be selected.
If you would rather that this not be a quirk and I rework the code such
that if no other quirks are selected then we configure for per-port
over-current as the default then I can do that too. If you expect per-
port over-current to be the default then explicit code that enforces
that might be best.

What's the best approach?

Thanks,
Hamish M

> 
> Alan Stern