| Message ID | 20230113-syscon-child-mfd-v1-2-0dd31b7de373@pengutronix.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Support Child MFD Cells for the i.MX6QDL Reset Controller | expand |
On 13/01/2023 18:32, Bastian Krause wrote: > The actual syscon-reboot-mode child node can be added by a board > device-tree or fixed up by the bootloader. For the child node to be > probed, the compatible needs to include simple-mfd. The binding now > specifies this, so have the SoC dtsi adhere to it. > > Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> > Signed-off-by: Bastian Krause <bst@pengutronix.de> > --- > arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi > index ff1e0173b39be..b16be39458aa6 100644 > --- a/arch/arm/boot/dts/imx6qdl.dtsi > +++ b/arch/arm/boot/dts/imx6qdl.dtsi > @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ > }; > > src: reset-controller@20d8000 { > - compatible = "fsl,imx6q-src", "fsl,imx51-src"; > + compatible = "fsl,imx6q-src", "fsl,imx51-src", > + "syscon", "simple-mfd"; You need children here. Otherwise simple-mfd does not make sense. If you expect something else to add children (I don't understand why, usually reboot capability is fixed per SoC and only sometimes extended with some other means), then this "else" will also change compatible. Best regards, Krzysztof
On 16/01/2023 09:20, Krzysztof Kozlowski wrote: > On 13/01/2023 18:32, Bastian Krause wrote: >> The actual syscon-reboot-mode child node can be added by a board >> device-tree or fixed up by the bootloader. For the child node to be >> probed, the compatible needs to include simple-mfd. The binding now >> specifies this, so have the SoC dtsi adhere to it. >> >> Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> >> Signed-off-by: Bastian Krause <bst@pengutronix.de> >> --- >> arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi >> index ff1e0173b39be..b16be39458aa6 100644 >> --- a/arch/arm/boot/dts/imx6qdl.dtsi >> +++ b/arch/arm/boot/dts/imx6qdl.dtsi >> @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ >> }; >> >> src: reset-controller@20d8000 { >> - compatible = "fsl,imx6q-src", "fsl,imx51-src"; >> + compatible = "fsl,imx6q-src", "fsl,imx51-src", >> + "syscon", "simple-mfd"; > > You need children here. Otherwise simple-mfd does not make sense. If you > expect something else to add children (I don't understand why, usually > reboot capability is fixed per SoC and only sometimes extended with some > other means), then this "else" will also change compatible. Hm, although dtbs_check would complain... so let's clarify - why you are not adding syscon-reboot now? It should be always here - either SoC supports it or it does not. Board support is other thing, but this is DTSI for SoC. Best regards, Krzysztof
Hello Krzysztof, On 16.01.23 09:20, Krzysztof Kozlowski wrote: > On 13/01/2023 18:32, Bastian Krause wrote: >> The actual syscon-reboot-mode child node can be added by a board >> device-tree or fixed up by the bootloader. For the child node to be >> probed, the compatible needs to include simple-mfd. The binding now >> specifies this, so have the SoC dtsi adhere to it. >> >> Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> >> Signed-off-by: Bastian Krause <bst@pengutronix.de> >> --- >> arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi >> index ff1e0173b39be..b16be39458aa6 100644 >> --- a/arch/arm/boot/dts/imx6qdl.dtsi >> +++ b/arch/arm/boot/dts/imx6qdl.dtsi >> @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ >> }; >> >> src: reset-controller@20d8000 { >> - compatible = "fsl,imx6q-src", "fsl,imx51-src"; >> + compatible = "fsl,imx6q-src", "fsl,imx51-src", >> + "syscon", "simple-mfd"; > > You need children here. Otherwise simple-mfd does not make sense. If you > expect something else to add children (I don't understand why, usually > reboot capability is fixed per SoC and only sometimes extended with some > other means), then this "else" will also change compatible. It's about syscon-reboot-mode, not syscon-reboot. Such modes are board- not soc-specific. Yes, a board DTS can override a compatible, but this is error prone as you can't just add a compatible, you need to hardcode the other compatibles in the SoC dtsi, which may change for good reason in future. barebox supports fixing up syscon-reboot-mode nodes defined in its own device tree to the Linux DT. While in theory, it could check if the parent node contains syscon/simple-mfd and fix it up if necessary, it sounds to me like this should be rather part of the upstream DT. Cheers, Ahmad > > Best regards, > Krzysztof > >
On 16/01/2023 09:27, Ahmad Fatoum wrote: > Hello Krzysztof, > > On 16.01.23 09:20, Krzysztof Kozlowski wrote: >> On 13/01/2023 18:32, Bastian Krause wrote: >>> The actual syscon-reboot-mode child node can be added by a board >>> device-tree or fixed up by the bootloader. For the child node to be >>> probed, the compatible needs to include simple-mfd. The binding now >>> specifies this, so have the SoC dtsi adhere to it. >>> >>> Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> >>> Signed-off-by: Bastian Krause <bst@pengutronix.de> >>> --- >>> arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- >>> 1 file changed, 2 insertions(+), 1 deletion(-) >>> >>> diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi >>> index ff1e0173b39be..b16be39458aa6 100644 >>> --- a/arch/arm/boot/dts/imx6qdl.dtsi >>> +++ b/arch/arm/boot/dts/imx6qdl.dtsi >>> @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ >>> }; >>> >>> src: reset-controller@20d8000 { >>> - compatible = "fsl,imx6q-src", "fsl,imx51-src"; >>> + compatible = "fsl,imx6q-src", "fsl,imx51-src", >>> + "syscon", "simple-mfd"; >> >> You need children here. Otherwise simple-mfd does not make sense. If you >> expect something else to add children (I don't understand why, usually >> reboot capability is fixed per SoC and only sometimes extended with some >> other means), then this "else" will also change compatible. > > It's about syscon-reboot-mode, not syscon-reboot. Such modes are board- > not soc-specific. syscon-reboot-mode is also mostly SoC specific. What exactly would differ on different boards? Register offsets of SoC component? Register values used by SoC power management unit? Anyway, the binding is then not correct - it does not allow syscon-reboot-mode child. I'll comment there. Best regards, Krzysztof
On 16.01.23 10:14, Krzysztof Kozlowski wrote: > On 16/01/2023 09:27, Ahmad Fatoum wrote: >> Hello Krzysztof, >> >> On 16.01.23 09:20, Krzysztof Kozlowski wrote: >>> On 13/01/2023 18:32, Bastian Krause wrote: >>>> The actual syscon-reboot-mode child node can be added by a board >>>> device-tree or fixed up by the bootloader. For the child node to be >>>> probed, the compatible needs to include simple-mfd. The binding now >>>> specifies this, so have the SoC dtsi adhere to it. >>>> >>>> Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> >>>> Signed-off-by: Bastian Krause <bst@pengutronix.de> >>>> --- >>>> arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- >>>> 1 file changed, 2 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi >>>> index ff1e0173b39be..b16be39458aa6 100644 >>>> --- a/arch/arm/boot/dts/imx6qdl.dtsi >>>> +++ b/arch/arm/boot/dts/imx6qdl.dtsi >>>> @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ >>>> }; >>>> >>>> src: reset-controller@20d8000 { >>>> - compatible = "fsl,imx6q-src", "fsl,imx51-src"; >>>> + compatible = "fsl,imx6q-src", "fsl,imx51-src", >>>> + "syscon", "simple-mfd"; >>> >>> You need children here. Otherwise simple-mfd does not make sense. If you >>> expect something else to add children (I don't understand why, usually >>> reboot capability is fixed per SoC and only sometimes extended with some >>> other means), then this "else" will also change compatible. >> >> It's about syscon-reboot-mode, not syscon-reboot. Such modes are board- >> not soc-specific. > > syscon-reboot-mode is also mostly SoC specific. What exactly would > differ on different boards? Register offsets of SoC component? Register > values used by SoC power management unit? The modes supported. Let's say you want a bootloader mode that drops the board's bootloader into a fastboot gadget mode. You'd add a syscon-reboot-mode pointing at one of the non-volatile registers and you would define a magic value to indicate fastboot, both in the bootloader and Linux. In theory, the reboot mode could also talk to the bootrom[1] to change the bootsource. This is also not board-agnostic, because it may not make sense to have a spinor reboot mode if your board doesn't have one. We have this scheme for STM32MP1 already and that's why I suggested Bastian to do it likewise for i.MX as he needs this functionality: https://lore.kernel.org/all/20201021102855.18026-1-a.fatoum@pengutronix.de/ https://elixir.bootlin.com/barebox/latest/source/arch/arm/dts/stm32mp151.dtsi#L44 Here's the barebox documentation for reboot modes btw: https://www.barebox.org/doc/latest/user/reboot-mode.html [1]: The i.MX bootrom samples two registers to determine boot mode override. The upstream syscon-reboot-mode binding supports only one 32-bit value and thus can be used as-is only for communicating with the "user" bootloader. > Anyway, the binding is then not correct - it does not allow > syscon-reboot-mode child. I'll comment there. > > > Best regards, > Krzysztof > >
On 16/01/2023 10:26, Ahmad Fatoum wrote: > On 16.01.23 10:14, Krzysztof Kozlowski wrote: >> On 16/01/2023 09:27, Ahmad Fatoum wrote: >>> Hello Krzysztof, >>> >>> On 16.01.23 09:20, Krzysztof Kozlowski wrote: >>>> On 13/01/2023 18:32, Bastian Krause wrote: >>>>> The actual syscon-reboot-mode child node can be added by a board >>>>> device-tree or fixed up by the bootloader. For the child node to be >>>>> probed, the compatible needs to include simple-mfd. The binding now >>>>> specifies this, so have the SoC dtsi adhere to it. >>>>> >>>>> Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> >>>>> Signed-off-by: Bastian Krause <bst@pengutronix.de> >>>>> --- >>>>> arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- >>>>> 1 file changed, 2 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi >>>>> index ff1e0173b39be..b16be39458aa6 100644 >>>>> --- a/arch/arm/boot/dts/imx6qdl.dtsi >>>>> +++ b/arch/arm/boot/dts/imx6qdl.dtsi >>>>> @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ >>>>> }; >>>>> >>>>> src: reset-controller@20d8000 { >>>>> - compatible = "fsl,imx6q-src", "fsl,imx51-src"; >>>>> + compatible = "fsl,imx6q-src", "fsl,imx51-src", >>>>> + "syscon", "simple-mfd"; >>>> >>>> You need children here. Otherwise simple-mfd does not make sense. If you >>>> expect something else to add children (I don't understand why, usually >>>> reboot capability is fixed per SoC and only sometimes extended with some >>>> other means), then this "else" will also change compatible. >>> >>> It's about syscon-reboot-mode, not syscon-reboot. Such modes are board- >>> not soc-specific. >> >> syscon-reboot-mode is also mostly SoC specific. What exactly would >> differ on different boards? Register offsets of SoC component? Register >> values used by SoC power management unit? > > The modes supported. Let's say you want a bootloader mode that drops > the board's bootloader into a fastboot gadget mode. You'd add a > syscon-reboot-mode pointing at one of the non-volatile registers and > you would define a magic value to indicate fastboot, both in the > bootloader and Linux. Bootloader and other firmware (e.g. ATF) is tightly tied to SoC, not to board. There might be differences between firmware used and OS (e.g. ChromeOS uses their own bootloader, different than Linux and Android on the same SoC), but again this is not board specific. > > In theory, the reboot mode could also talk to the bootrom[1] to change > the bootsource. This is also not board-agnostic, because it may not > make sense to have a spinor reboot mode if your board doesn't have one. > > We have this scheme for STM32MP1 already and that's why I suggested > Bastian to do it likewise for i.MX as he needs this functionality: > https://lore.kernel.org/all/20201021102855.18026-1-a.fatoum@pengutronix.de/ I don't understand why you use clearly wrong patches as examples. Bad patterns and bugs are not reason to use same approach. The binding is wrong - you do not allow syscon-reboot-mode and if you ever tested your patches, you would see the errors. > https://elixir.bootlin.com/barebox/latest/source/arch/arm/dts/stm32mp151.dtsi#L44 Whether this part is correct, tricky to say. Why these offsets are not valid for other board? Best regards, Krzysztof
Hello Krzysztof, On 16.01.23 11:15, Krzysztof Kozlowski wrote: >>>> It's about syscon-reboot-mode, not syscon-reboot. Such modes are board- >>>> not soc-specific. >>> >>> syscon-reboot-mode is also mostly SoC specific. What exactly would >>> differ on different boards? Register offsets of SoC component? Register >>> values used by SoC power management unit? >> >> The modes supported. Let's say you want a bootloader mode that drops >> the board's bootloader into a fastboot gadget mode. You'd add a >> syscon-reboot-mode pointing at one of the non-volatile registers and >> you would define a magic value to indicate fastboot, both in the >> bootloader and Linux. > > Bootloader and other firmware (e.g. ATF) is tightly tied to SoC, not to > board. There might be differences between firmware used and OS (e.g. > ChromeOS uses their own bootloader, different than Linux and Android on > the same SoC), but again this is not board specific. The bootloader probes from a board device tree and it also implements initialization, update, boot and fallback logic specific to the board and part of that is what reboot modes are supported. E.g. ST had particular reboot modes in mind (e.g. reboot into eMMC as usb mass storage gadget), but that's just a convention they chose for the platform, not something inherent to the SoC. >> In theory, the reboot mode could also talk to the bootrom[1] to change >> the bootsource. This is also not board-agnostic, because it may not >> make sense to have a spinor reboot mode if your board doesn't have one. >> >> We have this scheme for STM32MP1 already and that's why I suggested >> Bastian to do it likewise for i.MX as he needs this functionality: >> https://lore.kernel.org/all/20201021102855.18026-1-a.fatoum@pengutronix.de/ > > I don't understand why you use clearly wrong patches as examples. Bad > patterns and bugs are not reason to use same approach. I am trying to give you some context. It may be evident to you what's so clearly wrong about them, but for me it worked and I am trying to understand where you see a problem. > The binding is wrong - you do not allow syscon-reboot-mode and if you > ever tested your patches, you would see the errors. I did indeed not dump the device tree after the bootloader fixed it up and run into through the DT bindings checker. >> https://elixir.bootlin.com/barebox/latest/source/arch/arm/dts/stm32mp151.dtsi#L44 > > Whether this part is correct, tricky to say. Why these offsets are not > valid for other board? The offsets are valid, it may just not work. Also the user may choose to place the reboot mode somewhere else within the syscon if the register is unused otherwise. Still we could probably add a reboot-mode child node to the device tree with no extra modes and leave it disabled. That way boards can fill in the modes they support, enable it and use it. Does this work for you? Thanks, Ahmad > > > Best regards, > Krzysztof > >
diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi index ff1e0173b39be..b16be39458aa6 100644 --- a/arch/arm/boot/dts/imx6qdl.dtsi +++ b/arch/arm/boot/dts/imx6qdl.dtsi @@ -865,7 +865,8 @@ epit2: epit@20d4000 { /* EPIT2 */ }; src: reset-controller@20d8000 { - compatible = "fsl,imx6q-src", "fsl,imx51-src"; + compatible = "fsl,imx6q-src", "fsl,imx51-src", + "syscon", "simple-mfd"; reg = <0x020d8000 0x4000>; interrupts = <0 91 IRQ_TYPE_LEVEL_HIGH>, <0 96 IRQ_TYPE_LEVEL_HIGH>;
The actual syscon-reboot-mode child node can be added by a board device-tree or fixed up by the bootloader. For the child node to be probed, the compatible needs to include simple-mfd. The binding now specifies this, so have the SoC dtsi adhere to it. Suggested-by: Ahmad Fatoum <a.fatoum@pengutronix.de> Signed-off-by: Bastian Krause <bst@pengutronix.de> --- arch/arm/boot/dts/imx6qdl.dtsi | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)