Re: [PATCH v2 1/5] usb: dwc2: bus suspend/resume for hosts with DWC2_POWER_DOWN_PARAM_NONE

[Artur Petrosyan]

On 5/1/2019 05:57, Doug Anderson wrote:
> Hi,
> 
> On Mon, Apr 29, 2019 at 11:06 PM Artur Petrosyan
> <Arthur.Petrosyan@xxxxxxxxxxxx> wrote:
>>
>> Hi,
>>
>> On 4/29/2019 21:34, Doug Anderson wrote:
>>> Hi,
>>>
>>> On Mon, Apr 29, 2019 at 1:43 AM Artur Petrosyan
>>> <Arthur.Petrosyan@xxxxxxxxxxxx> wrote:
>>>>
>>>> Hi,
>>>>
>>>> On 4/18/2019 04:15, Douglas Anderson wrote:
>>>>> This is an attempt to rehash commit 0cf884e819e0 ("usb: dwc2: add bus
>>>>> suspend/resume for dwc2") on ToT.  That commit was reverted in commit
>>>>> b0bb9bb6ce01 ("Revert "usb: dwc2: add bus suspend/resume for dwc2"")
>>>>> because apparently it broke the Altera SOCFPGA.
>>>>>
>>>>> With all the changes that have happened to dwc2 in the meantime, it's
>>>>> possible that the Altera SOCFPGA will just magically work with this
>>>>> change now.  ...and it would be good to get bus suspend/resume
>>>>> implemented.
>>>>>
>>>>> This change is a forward port of one that's been living in the Chrome
>>>>> OS 3.14 kernel tree.
>>>>>
>>>>> Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
>>>>> ---
>>>>> This patch was last posted at:
>>>>>
>>>>> https://urldefense.proofpoint.com/v2/url?u=https-3A__lkml.kernel.org_r_1446237173-2D15263-2D1-2Dgit-2Dsend-2Demail-2Ddianders-40chromium.org&d=DwIDAg&c=DPL6_X_6JkXFx7AXWqB0tg&r=9hPBFKCJ_nBjJhGVrrlYOeOQjP_HlVzYqrC_D7niMJI&m=MMfe-4lZePyty6F5zfQ54kiYGuJWNulyRat944LkOsc&s=nExFpAPP_0plZfO5LMG1B-mqt1vyCvE35elVcyVgs8Y&e=
>>>>>
>>>>> ...and appears to have died the death of silence.  Maybe it could get
>>>>> some bake time in linuxnext if we can't find any proactive testing?
>>>>>
>>>>> I will also freely admit that I don't know tons about the theory
>>>>> behind this patch.  I'm mostly just re-hashing the original commit
>>>>> from Kever that was reverted since:
>>>>> * Turning on partial power down on rk3288 doesn't "just work".  I
>>>>>      don't get hotplug events.  This is despite dwc2 auto-detecting that
>>>>>      we are power optimized.
>>>> What do you mean by doesn't "just work" ? It seem to me that even after
>>>> adding this patch you don't get issues fixed.
>>>> You mention that you don't get the hotplug events. Please provide dwc2
>>>> debug logs and register dumps on this issue.
>>>
>>> I mean that partial power down in the currently upstream driver
>>> doesn't work.  AKA: if I turn on partial power down in the upstream
>>> driver then hotplug events break.  I can try to provide some logs.  On
>>> what exact version of the code do you want logs?  Just your series?
>>> Just my series?  Mainline?  Some attempt at combining both series?  As
>>> I said things seem to sorta work with the combined series.  I can try
>>> to clarify if that's the series you want me to test with.  ...or I can
>>> wait for your next version?
>> As I said this patch doesn't fix the issue with hotplug. With this patch
>> or without the hotplug behaves as it was. I have tested it on our setup.
>>
>> Have you debugged your patch? Does it make any difference on your setup
>> ? Does it fix the issue with hotplug?
> 
> I think we're still not taking on the same page.
> 
> My patch makes no attempt to make partial power down mode work.  My
> patch attempts to make things work a little better when using
> DWC2_POWER_DOWN_PARAM_NONE.  There is no use testing my patch with
> partial power down as it shouldn't have any impact there.
> 
> 
>>> I am by no means an expert on dwc2, but an assumption made in my patch
>>> is that even cores that can't support partial power down can still
>>> save some amount of power when hcd_suspend is called.
>> Have you tried to debug dwc2 with power_down == DWC2_POWER_DOWN_PARAM_NONE ?
>>>
>>> Some evidence that this should be possible: looking at mainline Linux
>>> and at dwc2_port_suspend(), I see:
>>>
>>> * It is currently called even when we have DWC2_POWER_DOWN_PARAM_NONE
>> Currently (without your and my patches) (looking at mainline Linux) the
>> function dwc2_port_suspend() is called anyway because its call is issued
>> by the system. But it performs entering to suspend only in case of
>> DWC2_POWER_DOWN_PARAM_PARTIAL.
>>
>> This is not an assumption. What I am pointing out is based on debugging
>> and before making assumptions without debugging for me seems not ok.
>>
>> Currently without your patch and without my patches. In the
>> dwc2_port_suspend() it will enter to suspend only in case that
>> power_down == DWC2_POWER_DOWN_PARAM_PARTIAL. Because if you look at the
>> code more carefully you will see
>>
>>          if (hsotg->params.power_down != DWC2_POWER_DOWN_PARAM_PARTIAL)
>>                  goto skip_power_saving;
>>
>> This says if power_down is not DWC2_POWER_DOWN_PARAM_PARTIAL then skip
>> power saving.
>>
>> So but after your patch. If power_down is DWC2_POWER_DOWN_PARAM_NONE it
>> tries to suspend.
> 
> We must be looking at different code.  I'm looking at Linux's tree, AKA:
> 
> https://urldefense.proofpoint.com/v2/url?u=https-3A__git.kernel.org_pub_scm_linux_kernel_git_torvalds_linux.git_tree_drivers_usb_dwc2_hcd.c-23n3488&d=DwIFaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=9hPBFKCJ_nBjJhGVrrlYOeOQjP_HlVzYqrC_D7niMJI&m=IWkDOOGTr0q-H1piDv2KOZe_Hnrz18g6rXFx-DsTuv4&s=AHu2iOKkybliRGtIfN7cF5p070UdvUKTYJsyAKYojis&e=
Here you are looking at the old code. After that there are several of 
changes related to suspend/resume functions.

This is the link to the code with changes. Latest version of those 
functions.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/usb/dwc2/hcd.c#n4489

Your changes are sitting on that latest version of code. Not the old 
version of it.

> 
> I took a mainline kernel ("v5.1-rc7-5-g83a50840e72a") and added
> printouts in dwc2_port_suspend() next to where it set HPRT0_SUSP and
> PCGCTL_STOPPCLK in dwc2_port_suspend().
I think you did this tests on the old version of the code

I have tested the flow myself with the mainline Kernel on 
"torvalds/master" and not HPRT0_SUSP nor PCGCTL_STOPPCLK are not being set.

So here you need to review those things again.

> 
> [  454.906364] dwc2 ff540000.usb: I'm setting HPRT0_SUSP
> [  454.906367] dwc2 ff540000.usb: I'm setting PCGCTL_STOPPCLK
> 
> ...and just to confirm:
> 
> # grep '^power' /sys/kernel/debug/*.usb/params
> /sys/kernel/debug/ff540000.usb/params:power_down                    : 0
> /sys/kernel/debug/ff580000.usb/params:power_down                    : 0
> 
> So I'm really quite convinced that on mainline Linux with
> DWC2_POWER_DOWN_PARAM_NONE that dwc2_port_suspend() sets HPRT0_SUSP
> and PCGCTL_STOPPCLK.
> 
> 
>>> ...I believe that the net effect of my patch ends up doing both those
>>> same two things in hcd_suspend.  That is: when power_down is
>>> DWC2_POWER_DOWN_PARAM_NONE I believe my patch is really just doing the
>>> same thing that dwc2_port_suspend() would do in the same case.  Is
>>> that not OK?
>> No if your patch is doing the same thing as it was doing before what is
>> the purpose of the patch ?
> 
> The purpose is to make _dwc2_hcd_suspend() work more correctly in the
> case where power_down is DWC2_POWER_DOWN_PARAM_NONE >
> 
>> My testes show that your patch doesn't fix the issue related partial
>> power down.
> 
> Right.  I have been trying to say that my patch doesn't do anything at
> all for partial power down.  I am simply trying to make
> DWC2_POWER_DOWN_PARAM_NONE work more correctly.
> 
> I haven't run all the power consumption tests in quite a long time and
> I'll try to get it hooked up tomorrow to confirm that my patch really
> truly is still needed to help with power consumption.  I did confirm
> that at least there are cases where _dwc2_hcd_suspend() is called and
> my patch is what sets the important bits.
> 
> -Doug
> 


-- 
Regards,
Artur