Re: [PATCH v4 0/2] ARM: sun8i-r40: Enable usb otg support

From: Evgeny Boger <boger@wirenboard.com>
To: qianfanguijin@163.com
Cc: andre.przywara@arm.com, boger@wirenboard.com,
	devicetree@vger.kernel.org, jernej.skrabec@gmail.com,
	krzysztof.kozlowski+dt@linaro.org,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-sunxi@lists.linux.dev,
	robh+dt@kernel.org, samuel@sholland.org, wens@csie.org
Subject: Re: [PATCH v4 0/2] ARM: sun8i-r40: Enable usb otg support
Date: Thu, 19 May 2022 23:54:57 +0300	[thread overview]
Message-ID: <ec5f7fe8-d47d-ce7f-0e0a-2bdf41a88ba2@wirenboard.com> (raw)
In-Reply-To: <264cb004-677a-13df-cc68-676ef3c2e7d8@163.com>

Hi qianfan,

As Allwinner A40i user, let me first thank you for your effort for 
making better upstream support for R40!

However, I would strongly suggest *not* to add USB support to one more 
Allwinner SoC in this particular way.
The problem is, this approach consists of a number of carefully crafted 
hacks in device tree to make current drivers work on Allwinner hardware 
without modification to the drivers.

a few examples:

1) please notice how ohci0 and ehci0 nodes do not contain reference to 
usb phy. It is done intentionally, otherwise EHCI will reset musb mode.
Of course omitting phy reference here is also completely breaking power 
cycling in case of usb error and otherwise messes with a power management.

2) one must always enable ohci, ehci and usb_otg nodes at the same time. 
If one forgets to enable ohci/ehci nodes while enabling usb_otg node, 
the system will silently fail to work as USB host.

3) For host-only mode we still have to enable usb_otg node despite no 
role switching is needed. That's because phy reference is missing in 
ehci/ohci, so the ehci/ohci driver won't enable the PHY.
Also I might be wrong, but I think phy won't be routed to ehci/ohci 
controllers is this case.

4) musb host controller is initialized and present to hardware though 
never actually used

To summarize, not only the resulting device tree is not describing the 
hardware properly, it is creating device tree configuration which will 
be very hard to support in future, once proper driver support is in place.

At Wiren Board kernel tree we tried to untangle this issue [1-6]. 
Unfortunately I didn't have time to prepare it for kernel submission 
yet, but I think I better submit it as RFC to get a feedback from you 
and others.

[1] 
https://github.com/wirenboard/linux/commit/359abbbd86ddff4d3c61179c882c286de32bb089
[2] 
https://github.com/wirenboard/linux/commit/6327f9d7972c21b229fb83457fdde643b31553f9
[3] 
https://github.com/wirenboard/linux/commit/f01f4c66758bde460a4d8c5b54ecee3b585c0232
[4] 
https://github.com/wirenboard/linux/commit/c27598ad601e5a46f624b73412a531d6f1f63d37
[5] 
https://github.com/wirenboard/linux/commit/5796d6eebb86b32a3751b2038b63af46f94954b3
[6] 
https://github.com/wirenboard/linux/commit/0928a675d875f9c2849fd3a9888f718bbb673bda

-- 
Kind regards,
Evgeny Boger
CTO @ Wiren Board
+49 3046690053
https://wirenboard.com/