Re: [PATCH 15/21] usb: chipidea: msm: Mux over secondary phy at the right time

[Peter Chen]

On Wed, Jun 29, 2016 at 12:28:58PM -0700, Stephen Boyd wrote:
> Quoting Peter Chen (2016-06-29 01:08:52)
> > On Sun, Jun 26, 2016 at 12:28:32AM -0700, Stephen Boyd wrote:
> > > We need to pick the correct phy at runtime based on how the SoC
> > > has been wired onto the board. If the secondary phy is used, take
> > > it out of reset and mux over to it by writing into the TCSR
> > > register. Make sure to do this on reset too, because this
> > > register is reset to the default value (primary phy) after the
> > > RESET bit is set in USBCMD.
> > > 
> > 
> > I am curious when you need the secondary phy?
> > 
> > > Cc: Peter Chen <peter.chen@xxxxxxx>
> > > Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > > Signed-off-by: Stephen Boyd <stephen.boyd@xxxxxxxxxx>
> > > ---
> > >  drivers/usb/chipidea/ci_hdrc_msm.c | 78 +++++++++++++++++++++++++++++++++++---
> > >  1 file changed, 73 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/drivers/usb/chipidea/ci_hdrc_msm.c b/drivers/usb/chipidea/ci_hdrc_msm.c
> > > index 40249b0e3e93..df0f8b31db4f 100644
> > > --- a/drivers/usb/chipidea/ci_hdrc_msm.c
> > > +++ b/drivers/usb/chipidea/ci_hdrc_msm.c
> > > @@ -8,30 +8,40 @@
> > >  #include <linux/module.h>
> > >  #include <linux/platform_device.h>
> > >  #include <linux/pm_runtime.h>
> > > -#include <linux/usb/gadget.h>
> > >  #include <linux/usb/chipidea.h>
> > >  #include <linux/clk.h>
> > >  #include <linux/reset.h>
> > > +#include <linux/mfd/syscon.h>
> > > +#include <linux/regmap.h>
> > > +#include <linux/io.h>
> > >  
> > >  #include "ci.h"
> > >  
> > >  #define HS_PHY_AHB_MODE                      0x0098
> > > +#define HS_PHY_SEC_CTRL                      0x0278
> > > +# define HS_PHY_DIG_CLAMP_N          BIT(16)
> > >  
> > 
> > One space at the beginning, and keep alignment.
> > 
> > >  struct ci_hdrc_msm {
> > >       struct platform_device *ci;
> > >       struct clk *core_clk;
> > >       struct clk *iface_clk;
> > > +     bool secondary_phy;
> > > +     void __iomem *base;
> > >  };
> > >  
> > >  static void ci_hdrc_msm_notify_event(struct ci_hdrc *ci, unsigned event)
> > >  {
> > > -     struct device *dev = ci->gadget.dev.parent;
> > > +     struct device *dev = ci->dev->parent;
> > > +     struct ci_hdrc_msm *msm_ci = dev_get_drvdata(dev);
> > >  
> > >       switch (event) {
> > >       case CI_HDRC_CONTROLLER_RESET_EVENT:
> > >               dev_dbg(dev, "CI_HDRC_CONTROLLER_RESET_EVENT received\n");
> > >               /* use AHB transactor, allow posted data writes */
> > >               hw_write_id_reg(ci, HS_PHY_AHB_MODE, 0xffffffff, 0x8);
> > > +             if (msm_ci->secondary_phy)
> > > +                     hw_write_id_reg(ci, HS_PHY_SEC_CTRL, HS_PHY_DIG_CLAMP_N,
> > > +                                     HS_PHY_DIG_CLAMP_N);
> > >               break;
> > >       default:
> > >               dev_dbg(dev, "unknown ci_hdrc event\n");
> > > @@ -49,12 +59,58 @@ static struct ci_hdrc_platform_data ci_hdrc_msm_platdata = {
> > >       .notify_event           = ci_hdrc_msm_notify_event,
> > >  };
> > >  
> > > +static int ci_hdrc_msm_mux_phy(struct ci_hdrc_msm *ci,
> > > +                            struct platform_device *pdev)
> > > +{
> > > +     struct regmap *regmap;
> > > +     struct device_node *syscon;
> > > +     struct device *dev = &pdev->dev;
> > > +     u32 off, val;
> > > +     int ret;
> > > +
> > > +     syscon = of_parse_phandle(dev->of_node, "phy-select", 0);
> > > +     if (!syscon)
> > > +             return 0;
> > > +
> > > +     regmap = syscon_node_to_regmap(syscon);
> > > +     if (IS_ERR(regmap))
> > > +             return PTR_ERR(regmap);
> > > +
> > > +     ret = of_property_read_u32_index(dev->of_node, "phy-select", 1, &off);
> > > +     if (ret < 0) {
> > > +             dev_err(dev, "no offset in syscon\n");
> > > +             return -EINVAL;
> > > +     }
> > > +
> > > +     ret = of_property_read_u32_index(dev->of_node, "phy-select", 2, &val);
> > > +     if (ret < 0) {
> > > +             dev_err(dev, "no value in syscon\n");
> > > +             return -EINVAL;
> > > +     }
> > > +
> > > +     ret = regmap_write(regmap, off, val);
> > > +     if (ret)
> > > +             return ret;
> > > +
> > > +     ci->secondary_phy = !!val;
> > > +     if (ci->secondary_phy) {
> > > +             val = readl_relaxed(ci->base + HS_PHY_SEC_CTRL);
> > > +             val |= HS_PHY_DIG_CLAMP_N;
> > > +             writel_relaxed(val, ci->base + HS_PHY_SEC_CTRL);
> > > +     }
> > > +
> > > +     return 0;
> > > +}
> > > +
> > >  static int ci_hdrc_msm_probe(struct platform_device *pdev)
> > >  {
> > >       struct ci_hdrc_msm *ci;
> > >       struct platform_device *plat_ci;
> > >       struct clk *clk;
> > >       struct reset_control *reset;
> > > +     struct resource *res;
> > > +     void __iomem *base;
> > > +     resource_size_t size;
> > >       int ret;
> > >  
> > >       dev_dbg(&pdev->dev, "ci_hdrc_msm_probe\n");
> > > @@ -76,6 +132,15 @@ static int ci_hdrc_msm_probe(struct platform_device *pdev)
> > >       if (IS_ERR(clk))
> > >               return PTR_ERR(clk);
> > >  
> > > +     res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > > +     if (!res)
> > > +             return -ENODEV;
> > > +
> > > +     size = resource_size(res);
> > > +     ci->base = base = devm_ioremap(&pdev->dev, res->start, size);
> > > +     if (!base)
> > > +             return -ENOMEM;
> > > +
> > 
> > The core will do the ioremap too, you can't remap io address two times.
> 
> You can ioremap an address twice, but it's not a great solution. On ARM,
> at least, we detect the double mapping and return the same virtual
> address the second time.
> 
> > The offset larger than 0x200 is vendor specific, you can map it as
> > the second io region.
> 
> Ok. That would mean we need to adjust the binding then to have to reg
> properties?

Just using two regs at dts, the second one is dedicated for glue layer.

> Or we can limit the size of the resource in the core and the
> size of the resource here can be bumped up by 0x200. So DT still says
> one resource for the entire ci address space but we don't map anything
> more than what we use. Something like this?
> 
> ---8<----
> diff --git a/drivers/usb/chipidea/ci_hdrc_msm.c b/drivers/usb/chipidea/ci_hdrc_msm.c
> index 4c70fa6fdc34..1121cf3e3fdc 100644
> --- a/drivers/usb/chipidea/ci_hdrc_msm.c
> +++ b/drivers/usb/chipidea/ci_hdrc_msm.c
> @@ -190,8 +190,9 @@ static int ci_hdrc_msm_probe(struct platform_device *pdev)
>  	if (!res)
>  		return -ENODEV;
>  
> -	size = resource_size(res);
> -	ci->base = base = devm_ioremap(&pdev->dev, res->start, size);
> +	res->start += 0x200;
> +	res->end -= 0x200;
> +	ci->base = base = devm_ioremap_resource(&pdev->dev, res);
>  	if (!base)
>  		return -ENOMEM;
>  
> diff --git a/drivers/usb/chipidea/core.c b/drivers/usb/chipidea/core.c
> index 298029a9ffce..8f5782913b11 100644
> --- a/drivers/usb/chipidea/core.c
> +++ b/drivers/usb/chipidea/core.c
> @@ -887,6 +887,8 @@ static int ci_hdrc_probe(struct platform_device *pdev)
>  	}
>  
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (resource_size(res) > 0x200)
> +		res->end = res->start + 0x200;
>  	base = devm_ioremap_resource(dev, res);
>  	if (IS_ERR(base))
>  		return PTR_ERR(base);
> 

No, don't do it. Keep the core unchanging, and map the second regs at
glue layer.

-- 

Best Regards,
Peter Chen