Hi Chris,
On Wed, Dec 04, 2013 at 03:16:21PM +0800, Chris Ruehl wrote:
> On Tuesday, December 03, 2013 04:15 PM, Heikki Krogerus wrote:
> >On Mon, Dec 02, 2013 at 03:05:19PM +0800, Chris Ruehl wrote:
> >>@@ -154,6 +164,27 @@ int usb_phy_gen_create_phy(struct device *dev, struct usb_phy_gen_xceiv *nop,
> >> {
> >> int err;
> >>
> >>+ if (nop->ulpi_vbus> 0) {
> >>+ unsigned int flags = 0;
> >>+
> >>+ if (nop->ulpi_vbus& 0x1)
> >>+ flags |= ULPI_OTG_DRVVBUS;
> >>+ if (nop->ulpi_vbus& 0x2)
> >>+ flags |= ULPI_OTG_DRVVBUS_EXT;
> >>+ if (nop->ulpi_vbus& 0x4)
> >>+ flags |= ULPI_OTG_EXTVBUSIND;
> >>+ if (nop->ulpi_vbus& 0x8)
> >>+ flags |= ULPI_OTG_CHRGVBUS;
> >>+
> >>+ nop->ulpi = otg_ulpi_create(&ulpi_viewport_access_ops, flags);
> >>+ if (!nop->ulpi) {
> >>+ dev_err(dev, "Failed create ULPI Phy\n");
> >>+ return -ENOMEM;
> >>+ }
> >>+ dev_dbg(dev, "Create ULPI Phy\n");
> >>+ nop->ulpi->io_priv = nop->viewport;
> >>+ }
> >
> >This is so wrong. You are registering one kind of usb phy driver from
> >an other. Change drivers/usb/phy/ulpi.c to be a platform device. The
> >whole flag system in it is pretty horrid. While you are at it, change
> >that so it sets the values based on boolean flags from OF properties
> >or platform data.
> >
> >NAK for the whole set.
> >
> >
>
> Heikki,
>
> Thanks for your comments, even not much positive to me.. any how.
> My intention on the "horrid" path was to reduce kernel code where
> one of_read32 vs. four of_boolean. And mentioned logic is simple.
> But that's history.
I should probable explain why I have problems with them. First of all,
things like driving the vbus should be a function that can be called
from upper layers. struct usb_otg has the set_vbus hook for that. You
can call it for example from your host controller's init routine. I'm
assuming you have a host controller since you are driving vbus.
You don't need to set the ULPI_OTG_CHRGVBUS. It's used for the VBUS
pulsing of SRP, which btw. is not anymore supported in OTG&EH2.0 spec,
so just don't use that bit even if you want to start SRP.
The only of_booleans you should need are for the DRV_VBUS_EXT and
USE_EXT_VBUS_IND. In my case I could not use even those. My controller
provides it's own control for them, so even if I set them to my ULPI
phy, the controller would simply override the values.
Secondly, why those silly flags in the first place. Those flags are
just bits in the registers. It would have been much easier and cleaner
to deliver a small struct with default values for the registers
instead.
> On my way to find a solution for my board I'd look around and found using of
> phy-ulpi.c functions in phy-tegra-usb.c and don't mind to use them too.
OK, IC. I have not followed what is happening with USB in linux for a
while.
The whole otg_ulpi_create() thing, and the flags with it, were
originally planned to be used from platform code. It's evil and it
should have never been accepted into upstream kernel. The time it was
introduced I was on vacation and nobody else seemed to care :(. All I
was able to do was to protest afterwards.
> I accept your NAK and will work on a patch to make phy-ulpi.c
> working as platform device.
>
> Last question to you. What you don't like on the patch to support
> chip-select gpio of my patch-set.. I ask because you NAK the whole
> set.
> I really need the ChipSelect function to make my hardware work!
OK, I did not explain my problem with that patch. I'm sorry about
that. It also looks like I made wrong assumption with it. I thought
that your phy (is was ISP1504 right) is just like isp1704 that I have
worked with. On isp1704 you only have the chip_sel pin (no reset pin),
so I thought you can not have any reason to add handler for an other
gpio to this driver. After a quick look at isp1504 data sheet, it
looks like you have both reset and chip_sel pins on it, which I guess
are both connected to gpios on your platform.
So I don't have a problem with that. Though I'm not sure is this
driver the right place to handle things like these gpios, which are
pretty phy specific, in the first place. The phy-generic was
originally meant to be pure NOP phy driver.
One comment about how to handle the gpios. You should move to the new
gpio descriptor API. The legacy gpio API is now just a wrapper on top
of it.
> Chris
>
Thanks,
--
heikki
--
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
