Re: [PATCH v4 5/5] usb: dwc3: rockchip: add devicetree bindings documentation

From: William Wu
Date: Mon Jun 27 2016 - 23:18:32 EST

Dear Heiko,

On 06/25/2016 03:50 AM, Heiko Stuebner wrote:
Hi William,

Am Dienstag, 21. Juni 2016, 17:11:44 schrieb William Wu:
On 06/20/2016 10:44 PM, Heiko Stübner wrote:
Am Freitag, 17. Juni 2016, 17:18:59 schrieb William Wu:
On 06/17/2016 07:15 AM, Heiko Stübner wrote:
Am Donnerstag, 2. Juni 2016, 20:34:56 schrieb William Wu:
This patch adds the devicetree documentation required for Rockchip
USB3.0 core wrapper consisting of USB3.0 IP from Synopsys.

It supports DRD mode, and could operate in device mode (SS, HS, FS)
and host mode (SS, HS, FS, LS).

Signed-off-by: William Wu <william.wu@xxxxxxxxxxxxxx>
[...]

+Optional clocks:
+ "aclk_usb3otg0" Aclk for specific usb controller clock.
what does this clock do? Also most likely same argument as below.
Here is partial rk3399 clk tree about usb3:

aclk_usb3 7 7 297000000 0 0
aclk_usb3_grf 1 1
297000000 0 0
aclk_usb3_rksoc_axi_perf 1 1
297000000 0 0
aclk_usb3otg1 1 1
297000000 0 0
aclk_usb3otg0 1 1
297000000 0 0
aclk_usb3_noc 1 1
297000000 0 0

from the clk tree, and check with our IC designers, we can see that:
1. aclk_usb3 is the parent clk of aclk_usb3****
2. aclk_usb3_grf is used for both otg0 and otg1 grf, and these usb3 grf
can be set
to control otg0 and otg1 controller, but not the phy.
3. aclk_usb3otg1 is otg1 controller clk, aclk_usb3otg0 is otg0
controller clk.
4. aclk_usb3_rksoc_axi_perf is the clk for usb3 performance monitor
module. 5. aclk_usb3_noc is the clk for soc bus interconnect.

+ "aclk_usb3_rksoc_axi_perf" USB AXI perf clock. Not present on
all
platforms.
The clock names looks pretty strange. What are they for? Especially as
nothing seems to use them right now.
"aclk_usb3_rksoc_axi_perf", it's the clk for usb3 performance monitor
module, you can refer to the GRF_USB3_PERF_xxx. And we don't use the
usb3
performance monitor control registers right now.
ok, then I'd suggest not defining the clock for now.

For one, there are more perf blocks in the GRF (usb3, pcie, hdcp22,
gmac, gpu, etc) which all seem to share a somewhat similar design, so
they will maybe result in a separate driver of some form for the
performance monitors.

And secondly, it is somewhat easy to add new optional properties, but
you
cannot remove anything defined previously. So if we later decide to
handle all the performance monitors differently, you can't remove that
clock from the binding again. (or at least only with quite a bit of
hassle).

So as this clock isn't used at all yet, I guess it should not get
included now.
Yes, I agree with you. We can remove the aclk_usb3_rksoc_axi_perf right
now.
+ "aclk_usb3_grf" USB grf clock. Not present on all platforms.
for my own education, which part of the GRF does this clock supply?
"aclk_usb3_grf", it's the clk for USB3 grf, e.g. GRF_USB3OTGX_CONX
Hmm, this looks more like it belongs to the otg phy?
Anyway, also seems unused right now, so same argument as above applies
here.
As I have described above, the "aclk_usb3_grf" is used for both otg0
and otg1 grf,
and these usb3 grf can be set to control otg0 and otg1 controller, but
not the phy.
And we have done a test that remove the grf clk, and the result showed
that usb3
controller can't work normally. So I think we need the usb3 grf clk.

So about the usb3 controller clk management, I think it should contain
the following clk:
1. aclk_usb3otg1
2. aclk_usb3otg0
3. aclk_usb3_grf
correct, aclk_usb3otgX would then be the busclk for each controller if I'm
not mistaken and the grf clock should also get enabled, like we also plan on
doing for the vio_grf clock in the display area.
OK, so aclk_usb3_grf should be marked as critical, right?

I found that most of the grf clocks haven't been marked as critical, apart from vio_grf. So may I keep the aclk_usb3_grf in usb3 dts, and remove it after clock manager adds it to critical clocks?

4. aclk_usb3_noc
For "aclk_usb3_noc", I discuss with our clk manager, the clk is always
on before,
but according to upstream maintainer's suggestion, we need to manage the
noc clk by each module.
can you point me to this discussion? The bus-interconnect is a very separate
component, which we don't model so far and thus we have all the noc clocks
simply marked as critical.

As this clock doesn't belong to the actual usb controller block, but said
separate component, handling it in the controller seems somehow wrong to me.

So my (current) opinion would again be to mark the noc clock as critical for
the time being.
Sorry, it seems that I get the new information about clk management too late.:-)

There's no dedicated discussion about noc clk, but similar to grf clock, please refer to "https://patchwork.kernel.org/patch/9171467/"; for add pclk_vio_grf to critical clock on the RK3399, and you have agreed on that mark vio grf clk as critical. So I agree with your opinion, thanks~

and the follow two clk can be removed:
1. aclk_usb3
2. aclk_usb3_rksoc_axi_perf

Is it ok?
yep, apart from the noc-clock, this looks great.


Heiko
Best regards,
William Wu