Re: [PATCH v1 1/1] usb: host: tegra: implement dts based xcvr setup

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 8564 bytes --]

On Sun, Aug 20, 2023 at 09:10:17PM +0200, Marek Vasut wrote:
> On 8/20/23 20:32, Svyatoslav Ryhel wrote:
> > 20 серпня 2023 р. 21:14:15 GMT+03:00, Marek Vasut <marex@denx.de> написав(-ла):
> > > On 8/20/23 09:13, Svyatoslav Ryhel wrote:
> > > > 20 серпня 2023 р. 05:23:14 GMT+03:00, Marek Vasut <marex@denx.de> написав(-ла):
> > > > > On 8/19/23 17:06, Svyatoslav Ryhel wrote:
> > > > > > Some devices (like ASUS TF201) may not have fuse based xcvr setup
> > > > > > which will result in not working USB line. To fix this situation
> > > > > > allow xcvr setup to be performed from device tree values if
> > > > > > nvidia,xcvr-setup-use-fuses is not defined.
> > > > > > 
> > > > > > Tested-by: Svyatoslav Ryhel <clamor95@gmail.com> # ASUS TF201
> > > > > 
> > > > > I would hope so. Usually T-B tags are not added by the patch author, but that's a detail, and here it has the added model value, so keep it.
> > > > > 
> > > > > > Signed-off-by: Svyatoslav Ryhel <clamor95@gmail.com>
> > > > > > ---
> > > > > >     drivers/usb/host/ehci-tegra.c | 53 +++++++++++++++++++++++++++++++----
> > > > > >     1 file changed, 48 insertions(+), 5 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/usb/host/ehci-tegra.c b/drivers/usb/host/ehci-tegra.c
> > > > > > index 2cf1625670..f24baf8f0c 100644
> > > > > > --- a/drivers/usb/host/ehci-tegra.c
> > > > > > +++ b/drivers/usb/host/ehci-tegra.c
> > > > > > @@ -81,6 +81,8 @@ struct fdt_usb {
> > > > > >        enum periph_id periph_id;/* peripheral id */
> > > > > >        struct gpio_desc vbus_gpio;     /* GPIO for vbus enable */
> > > > > >        struct gpio_desc phy_reset_gpio; /* GPIO to reset ULPI phy */
> > > > > > +    bool xcvr_setup_use_fuses; /* Indicates that the value is read from the on-chip fuses */
> > > > > > +    u32 xcvr_setup; /* Input of XCVR cell, HS driver output control */
> > > > > >     };
> > > > > >       /*
> > > > > > @@ -464,10 +466,21 @@ static int init_utmi_usb_controller(struct fdt_usb *config,
> > > > > >                /* Recommended PHY settings for EYE diagram */
> > > > > >                val = readl(&usbctlr->utmip_xcvr_cfg0);
> > > > > > -            clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MASK,
> > > > > > -                            0x4 << UTMIP_XCVR_SETUP_SHIFT);
> > > > > > -            clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MSB_MASK,
> > > > > > -                            0x3 << UTMIP_XCVR_SETUP_MSB_SHIFT);
> > > > > > +
> > > > > > +            if (!config->xcvr_setup_use_fuses) {
> > > > > > +                    clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MASK,
> > > > > > +                                    config->xcvr_setup <<
> > > > > > +                                    UTMIP_XCVR_SETUP_SHIFT);
> > > > > > +                    clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MSB_MASK,
> > > > > > +                                    config->xcvr_setup <<
> > > > > > +                                    UTMIP_XCVR_SETUP_MSB_SHIFT);
> > > > > > +            } else {
> > > > > > +                    clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MASK,
> > > > > > +                                    0x4 << UTMIP_XCVR_SETUP_SHIFT);
> > > > > 
> > > > > What is this hard-coded value ?
> > > > > 
> > > > 
> > > > 0x4 and 0x3 (not same) but those are not in the device tree. Mainline linux
> > > > driver seems not set this at all if use_fuses is enabled but I decided to keep
> > > > original setup just to not cause regressions.
> > > > 
> > > > https://github.com/clamor-s/linux/blob/transformer/drivers/usb/phy/phy-tegra-usb.c#L587-L590
> > > > 
> > > > > Why not place this value into config->xcvr_setup in e.g. probe() and drop this if/else statement ?
> > > > > 
> > > > 
> > > > Because config->xcvr_setup should matter only if use_fuses is disabled
> > > 
> > > Can it matter always instead ?
> > > 
> > 
> > No, it cannot. You are inversing hw design. Xcvr_setup matters only if use_fuses is disabled. If use_fuses is on (which is default state) xcvr values are taken from chip fuse.
> 
> The way I read this block of code is, some value is written into the
> register if config->xcvr_setup_use_fuses is false, another value if
> config->xcvr_setup_use_fuses is true . Why not do this determination once in
> probe and then just program the appropriate value instead ?
> 
> > > > > > +                    clrsetbits_le32(&val, UTMIP_XCVR_SETUP_MSB_MASK,
> > > > > > +                                    0x3 << UTMIP_XCVR_SETUP_MSB_SHIFT);
> > > > > > +            }
> > > > > > +
> > > > > >                clrsetbits_le32(&val, UTMIP_XCVR_HSSLEW_MSB_MASK,
> > > > > >                                0x8 << UTMIP_XCVR_HSSLEW_MSB_SHIFT);
> > > > > >                writel(val, &usbctlr->utmip_xcvr_cfg0);
> > > > > > @@ -522,7 +535,9 @@ static int init_utmi_usb_controller(struct fdt_usb *config,
> > > > > >        setbits_le32(&usbctlr->utmip_bat_chrg_cfg0, UTMIP_PD_CHRG);
> > > > > >        clrbits_le32(&usbctlr->utmip_xcvr_cfg0, UTMIP_XCVR_LSBIAS_SE);
> > > > > > -    setbits_le32(&usbctlr->utmip_spare_cfg0, FUSE_SETUP_SEL);
> > > > > > +
> > > > > > +    if (config->xcvr_setup_use_fuses)
> > > > > > +            setbits_le32(&usbctlr->utmip_spare_cfg0, FUSE_SETUP_SEL);
> > > > > >        /*
> > > > > >         * Configure the UTMIP_IDLE_WAIT and UTMIP_ELASTIC_LIMIT
> > > > > > @@ -843,6 +858,8 @@ static const struct ehci_ops tegra_ehci_ops = {
> > > > > >     static int ehci_usb_of_to_plat(struct udevice *dev)
> > > > > >     {
> > > > > >        struct fdt_usb *priv = dev_get_priv(dev);
> > > > > > +    u32 usb_phy_phandle;
> > > > > > +    ofnode usb_phy_node;
> > > > > >        int ret;
> > > > > >        ret = fdt_decode_usb(dev, priv);
> > > > > > @@ -851,6 +868,32 @@ static int ehci_usb_of_to_plat(struct udevice *dev)
> > > > > >        priv->type = dev_get_driver_data(dev);
> > > > > >     +  ret = ofnode_read_u32(dev_ofnode(dev), "nvidia,phy", &usb_phy_phandle);
> > > > > > +    if (ret) {
> > > > > > +            log_debug("%s: required usb phy node isn't provided\n", __func__);
> > > > > > +            priv->xcvr_setup_use_fuses = true;
> > > > > > +            return 0;
> > > > > > +    }
> > > > > > +
> > > > > > +    usb_phy_node = ofnode_get_by_phandle(usb_phy_phandle);
> > > > > > +    if (!ofnode_valid(usb_phy_node)) {
> > > > > > +            log_err("%s: failed to find usb phy node\n", __func__);
> > > > > > +            return -EINVAL;
> > > > > > +    }
> > > > > 
> > > > > dev_read_phandle() should replace the above
> > > > > 
> > > > 
> > > > Goal of this piece is to get phy of_node by the phandle provided in the
> > > > controller node. Controller node has not only single nvidia,phy phandle
> > > > reference but also clock and reset reference. How will dev_read_phandle
> > > > return me a phandle of "nvidia,phy"?
> > > > 
> > > > https://github.com/clamor-s/u-boot/blob/master/arch/arm/dts/tegra30.dtsi#L798-L834
> > > > 
> > > > > > +    priv->xcvr_setup_use_fuses = ofnode_read_bool(
> > > > > > +            usb_phy_node, "nvidia,xcvr-setup-use-fuses");
> > > > > 
> > > > > dev_read_bool()
> > > > > 
> > > > 
> > > > This value is set in phy node, not controllers unfortunately.
> > > 
> > > The question that comes to mind is, would it make sense to implement a PHY driver similar to what e.g. imx does -- drivers/phy/phy-imx8mq-usb.c -- for the tegra PHY ?
> > 
> > Yes, but not by me or at least not now. I have already invested too much of my time and effort in this and I will not invest even more into refactoring all this code into 2 separate drivers. Existing code satisfies me apart from this small tweak.
> 
> The PHY driver implementation is trivial, example is the imx driver above,
> then just call phy on/off in the right place. Linux also has a PHY driver
> for tegra, maybe you can reuse it ?

The Linux USB PHY driver is something that in retrospect I would've done
differently. The problem is that it shares resources (registers and
clock/reset lines) with the USB controller and it's caused a lot of
headaches to support it in Linux.

Maybe within U-Boot this isn't such a big deal, but for Linux I would've
preferred a single driver that is both a USB controller and a USB PHY. I
suppose it could expose some sort of PHY object for abstraction, but
that would probably be a bit of overkill since this is really only used
within the USB controller driver.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]