Re: [PATCH v5 2/2] phy: rockchip-inno-usb2: add a new driver for Rockchip usb2phy

[Kishon Vijay Abraham I]

Hi,

On Friday 17 June 2016 10:16 PM, Heiko Stübner wrote:
> Hi Kishon,
> 
> Am Freitag, 17. Juni 2016, 17:24:46 schrieb Kishon Vijay Abraham I:
> 
>>> +	ret = of_clk_add_provider(node, of_clk_src_simple_get, rphy->clk480m);
>>> +	if (ret < 0)
>>> +		goto err_clk_provider;
>>> +
>>> +	ret = devm_add_action(rphy->dev, rockchip_usb2phy_clk480m_unregister,
>>> +			      rphy);
>>> +	if (ret < 0)
>>> +		goto err_unreg_action;
>>> +
>>> +	return 0;
>>> +
>>> +err_unreg_action:
>>> +	of_clk_del_provider(node);
>>> +err_clk_provider:
>>> +	clk_unregister(rphy->clk480m);
>>> +err_register:
>>> +	if (rphy->clk)
>>> +		clk_put(rphy->clk);
>>> +	return ret;
>>> +}
>>
>> I'm seeing lot of similarities specifically w.r.t to clock handling part in
>> drivers/phy/phy-rockchip-usb.c. Why not just re-use that driver?
> 
> It's a completely different phy block (Designware vs. Innosilicon) and a lot 
> of stuff also is handled differently.
> 
> For one on the old block, each phy was somewhat independent and had for examle 
> its own clock-supply, while on this one there is only one for both ports of 
> the phy. Similarly with the clock getting fed back to the clock-controller 
> (one clock per port on the old block, now one clock for the whole phy).
> 
> Then as you can see, the handling for power-up and down is a bit different and 
> I guess one big block might be the still missing special otg handling, Frank 
> wrote about.

All right then.
> 
> 
> [...]
> 
>>> +		/*
>>> +		 * we don't need to rearm the delayed work when the phy port
>>> +		 * is suspended.
>>> +		 */
>>> +		mutex_unlock(&rport->mutex);
>>> +		return;
>>> +	default:
>>> +		dev_dbg(&rport->phy->dev, "unknown phy state\n");
>>> +		break;
>>> +	}
>>> +
>>> +next_schedule:
>>> +	mutex_unlock(&rport->mutex);
>>> +	schedule_delayed_work(&rport->sm_work, SCHEDULE_DELAY);
>>
>> Why are you scheduling the work again? Interrupt handlers can invoke this
>> right?
> 
> Frank said, that the phy is only able to detect the plug-in event via 
> interrupts, not the removal, so once a plugged device is detected, this gets 
> rescheduled until the device gets removed.

okay.
> 
> [...]
> 
>>> +	/* find out a proper config which can be matched with dt. */
>>> +	index = 0;
>>> +	while (phy_cfgs[index].reg) {
>>> +		if (phy_cfgs[index].reg == reg) {
>>
>> Why not pass these config values from dt? Moreover finding the config using
>> register offset is bound to break.
> 
> As you have probably seen, this phy block is no stand-alone (mmio-)device, but 
> gets controlled through special register/bits in the so called "General 
> Register Files" syscon.
> 
> The values stored and accessed here, are the location and layout of those 
> control registers. Bits in those phy control registers at times move between 
> phy-versions in different socs (rk3036, rk3228, rk3366, rk3368, rk3399) and 
> some are even missing. So I don't really see a nice way to describe that in dt 
> without describing the register and offset of each of those 22 used bits 
> individually.
> 
> 
> I'm also not sure where you expect it to break? The reg-offset is the offset 
> of the phy inside the GRF and the Designware-phy also already does something 
> similar to select some appropriate values.

I'm concerned the phy can be placed in the different reg-offset within GRF
(like in the next silicon revision) and this driver can't be used.

Thanks
Kishon
>