Re: [PATCH 20/21] phy: Add support for Qualcomm's USB HSIC phy

[Neil Armstrong]

On 06/28/2016 11:58 PM, Stephen Boyd wrote:
> Quoting Neil Armstrong (2016-06-28 01:49:37)
>> On 06/26/2016 09:28 AM, Stephen Boyd wrote:
>>> +     uphy->cal_sleep_clk = clk = devm_clk_get(&ulpi->dev, "cal_sleep");
>>> +     if (IS_ERR(clk))
>>> +             return PTR_ERR(clk);
>>
>> Hi Stephen,
>>
>> In the bindings the cal_sleep is marked optional, and I think should be since AFAIK
>> it's not present on MDM9615 for example.
> 
> The cal_sleep clk is just the sleep clk then (should be a board clk in
> DT). Sometimes there's a gate in GCC to allow us to turn it off, other
> times there isn't. Either way, it's always wired there so I'll update
> the binding to say it isn't optional.

Sorry I don't understand !
What should I do if GCC does not provide a gate here ? And looking at the driver, it could be optional.

> 
>>
>> Also MDM9615 HSIC requires "core", "alt-core", "phy", "cal" and "iface" clocks.
>> I assume "core" can be attributed to the main chipidea node, but I think "alt-core" and "iface" should be also optionnal.
> 
> Looking at the downstream sources I see this:
> 
>         "core_clk"     -> usb_hsic_sys_clk
>         "iface_clk"    -> usb_hsic_p_clk
>         "alt_core_clk" -> usb_hsic_xcvr_clk
> 
>         "cal_clk"      -> usb_hsic_hsio_cal_clk
>         "phy_clk"      -> usb_hsic_clk
> 
> "core_clk" would be the core clk in ci_hdrc_msm. "iface_clk" would be
> the iface clk in ci_hdrc_msm. "cal_clk" would be the cal clk in the hsic
> phy and "phy_clk" would be the phy clk in the hsic phy.
> 
> That leaves alt_core_clk which seems to be a clock that needs to be on
> during the reset assert/deassert and possibly for LPM and USB1.1 FS
> modes. Sometimes it's referred to as the "housekeeping" clk. Due to the
> way resets are done on msm8974 and later SoCs it looks like this clk was
> removed. I can make this an optional clk in the ci_hdrc_msm driver, or
> we can have two versions of the ci_hdrc_msm compatible string, one for a
> device that has the housekeeping clk and one for the device that
> doesn't.

Having it optional would be the best solution I think.

> 
>>
>> Finally, it misses an optional reset line AFAIK mandatory on MDM9615.
>>
> 
> From what I can tell downstream, all those clks point to the same bit 0
> of HSIC_RESET register? So there isn't any phy reset, just the chipidea
> controller wrapper reset bit, which should go into the wrapper node?

Ok, makes sense.