Message-ID: <20230425-unquote-eligible-09f743d81981@wendy>
Date:   Tue, 25 Apr 2023 07:59:44 +0100
From:   Conor Dooley <conor.dooley@...rochip.com>
To:     Changhuang Liang <changhuang.liang@...rfivetech.com>
CC:     Conor Dooley <conor@...nel.org>, Rob Herring <robh+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Emil Renner Berthing <kernel@...il.dk>,
        Paul Walmsley <paul.walmsley@...ive.com>,
        Palmer Dabbelt <palmer@...belt.com>,
        Albert Ou <aou@...s.berkeley.edu>,
        Walker Chen <walker.chen@...rfivetech.com>,
        Hal Feng <hal.feng@...rfivetech.com>,
        <devicetree@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
        <linux-riscv@...ts.infradead.org>, <vkoul@...nel.org>,
        <linux-phy@...ts.infradead.org>
Subject: Re: [RESEND v2 1/6] dt-bindings: power: Add JH7110 AON PMU support

On Tue, Apr 25, 2023 at 11:41:38AM +0800, Changhuang Liang wrote:
> On 2023/4/25 0:52, Conor Dooley wrote:
> > On Thu, Apr 20, 2023 at 03:00:10PM +0800, Changhuang Liang wrote:
> >> On 2023/4/20 2:29, Conor Dooley wrote:
> >>> On Tue, Apr 18, 2023 at 08:56:41PM -0700, Changhuang Liang wrote:
> >>>> Add AON PMU for StarFive JH7110 SoC, it can be used to turn on/off DPHY
> >>>> rx/tx power switch, and it don't need the properties of reg and
> >>>> interrupts.
> >>>
> >>> Putting this here since the DT guys are more likely to see it this way..
> >>> Given how the implementation of the code driving this new
> >>> power-controller and the code driving the existing one are rather
> >>> different (you've basically re-written the entire driver in this series),
> >>> should the dphy driver implement its own power-controller?
> >>>
> >>> I know originally Changuang had tried something along those lines:
> >>> https://lore.kernel.org/linux-riscv/5dc4ddc2-9d15-ebb2-38bc-8a544ca67e0d@starfivetech.com/
> >>>
> >>> I see that that was shut down pretty much, partly due to the
> >>> non-standard property, hence this series adding the dphy power domain to
> >>> the existing driver.
> >>>
> >>> If it was done by looking up the pmu with a
> >>> of_find_compatible_node(NULL, "power-controller", "starfive,jh7110-aon-pmu")
> >>> type thing, would that make sense? Although, maybe that is not a
> >>> question for you, and this series may actually have been better entirely
> >>> bundled with the dphy series so the whole thing can be reviewed as a
> >>> unit. I've added 
> >>>
> >>> IOW, don't change this patch, or the dts patch, but move all of the
> >>> code back into the phy driver..
> >>>
> >>
> >> Maybe this way can not do that? power domain is binding before driver probe,
> >> if I use "of_find_compatible_node" it phy(DPHY rx) probe. Maybe I can only operate 
> >> this power switch in my phy(DPHY rx) driver, so the all patch of this series isn't 
> >> make sense.
> > 
> > I'm a wee bit lost here, as I unfortunately know little about how Linux
> > handles this power-domain stuff. If the DPHY tries to probe and some
> > pre-requisite does not yet exist, you can return -EPROBE_DEFER right?
> > 
> > But I don't think that's what you are asking, as using
> > of_find_compatible_node() doesn't depend on there being a driver AFAIU.
> > 
> >> In my opinion, We will also submit DPHY TX module later which use this series.
> >> Maybe this series should independent?
> > 
> > Is the DPHY tx module a different driver to the rx one?> If yes, does it have a different bit you must set in the syscon?
> > 
> 
> Yes, DPHY tx module is a different driver to the DPHY rx. And I have do a
> different bit in PATCH 1:
> 
> #define JH7110_PD_DPHY_TX	0
> #define JH7110_PD_DPHY_RX	1
> 
> also in PATCH 5:
> 
> static const struct jh71xx_domain_info jh7110_aon_power_domains[] = {
> 	[JH7110_PD_DPHY_TX] = {
> 		.name = "DPHY-TX",
> 		.bit = 30,
> 	},
> 	[JH7110_PD_DPHY_RX] = {
> 		.name = "DPHY-RX",
> 		.bit = 31,
> 	},
> };
> 
> > +CC Walker, do you have a register map for the jh7110? My TRM only says
> > what the registers are, but not the bits in them. Would make life easier
> > if I had that info.
> > 
> > I'm fine with taking this code, I just want to make sure that the soc
> > driver doing this is the right thing to do.
> > I was kinda hoping that combining with the DPHY-rx series might allow
> > the PHY folk to spot if you are doing something here with the power
> > domains that doesn't make sense.
> > 
> 
> I asked about our soc colleagues. This syscon register,
> offset 0x00:
> bit[31] ---> dphy rx power switch
> bit[30] ---> dphy tx power switch 
> other bits ---> Reserved

Okay. Unless someone explicitly disagrees, I'm fine with doing this
stand-alone from the DPHY drivers.

> >>> Sorry for not asking this sooner Changhuang,
> >>> Conor.
> >>>
> >>> (hopefully this didn't get sent twice, mutt complained of a bad email
> >>> addr during sending the first time)
> >>>
> >>
> >> I'm sorry for that, I will notice later.
> > 
> > No, this was my mail client doing things that I was unsure of. You
> > didn't do anything wrong.
> > 
> [...]
> >>>>    - Walker Chen <walker.chen@...rfivetech.com>
> >>>> +  - Changhuang Liang <changhuang.liang@...rfivetech.com>
> >>>>  
> >>>>  description: |
> >>>>    StarFive JH7110 SoC includes support for multiple power domains which can be
> >>>> @@ -17,6 +18,7 @@ properties:
> >>>>    compatible:
> >>>>      enum:
> >>>>        - starfive,jh7110-pmu
> >>>> +      - starfive,jh7110-aon-pmu
> > 
> > I was speaking to Rob about this over the weekend, he asked:
> > 'Why isn't "starfive,jh7110-aon-syscon" just the power-domain provider
> > itself?'
> 
> Maybe not, this syscon only offset "0x00" configure power switch.
> other offset configure other functions, maybe not power, so this
> "starfive,jh7110-aon-syscon" not the power-domain itself.
> 
> > Do we actually need to add a new binding for this at all?
> > 
> > Cheers,
> > Conor.
> > 
> 
> Maybe this patch do that.
> https://lore.kernel.org/all/20230414024157.53203-6-xingyu.wu@starfivetech.com/

This makes it a child-node right? I think Rob already said no to that in
and earlier revision of this series. What he meant the other day was
making the syscon itself a power domain controller, since the child node
has no meaningful properties (reg, interrupts etc).

Cheers,
Conor.

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)

Powered by blists - more mailing lists