| Message ID | 20210915120325.20248-1-heiko.thiery@gmail.com |
|---|---|
| State | Accepted |
| Commit | 9786cca4b477f2b2f9d573d474c929d87579b501 |
| Headers | show |
| Series | arm64: dts: imx8mm-kontron-n801x-som: do not allow to switch off buck2 | expand |
On 15.09.21 14:05, Michael Walle wrote: > Am 2021-09-15 14:03, schrieb Heiko Thiery: >> The buck2 output of the PMIC is the VDD core voltage of the cpu. >> Switching off this will poweroff the CPU. Add the 'regulator-always-on' >> property to avoid this. > > Mh, have this ever worked? Is there a commit which introduced a regression? Yes, this did work before, even without 'regulator-always-on'. I currently don't understand why this is needed. The regulator is referenced in the CPU nodes as 'cpu-supply'. This should be enough to not disable it as long as the CPU is up.
Hi Frieder, Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf <frieder.schrempf@kontron.de>: > > On 15.09.21 14:05, Michael Walle wrote: > > Am 2021-09-15 14:03, schrieb Heiko Thiery: > >> The buck2 output of the PMIC is the VDD core voltage of the cpu. > >> Switching off this will poweroff the CPU. Add the 'regulator-always-on' > >> property to avoid this. > > > > Mh, have this ever worked? Is there a commit which introduced a regression? > > Yes, this did work before, even without 'regulator-always-on'. I > currently don't understand why this is needed. The regulator is > referenced in the CPU nodes as 'cpu-supply'. This should be enough to > not disable it as long as the CPU is up. I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the same issue: [ 31.716031] vdd-5v: disabling [ 31.719032] rst-usb-eth2: disabling [ 31.722553] buck2: disabling While on that I tried to compare with other boards and see that they also have the cpu-voltage marked as "regulator-always-on". The only exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that has not set this property. I agree with you and don't understand why this is happening. Has anyone else an explanation? [1] https://elixir.bootlin.com/linux/latest/source/arch/arm64/boot/dts/freescale/imx8mq-librem5-devkit.dts#L319 -- Heiko
Am 17. September 2021 09:28:51 MESZ schrieb Heiko Thiery <heiko.thiery@gmail.com>: >Hi Frieder, > >Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf ><frieder.schrempf@kontron.de>: >> >> On 15.09.21 14:05, Michael Walle wrote: >> > Am 2021-09-15 14:03, schrieb Heiko Thiery: >> >> The buck2 output of the PMIC is the VDD core voltage of the cpu. >> >> Switching off this will poweroff the CPU. Add the 'regulator-always-on' >> >> property to avoid this. >> > >> > Mh, have this ever worked? Is there a commit which introduced a regression? >> >> Yes, this did work before, even without 'regulator-always-on'. I >> currently don't understand why this is needed. The regulator is >> referenced in the CPU nodes as 'cpu-supply'. This should be enough to >> not disable it as long as the CPU is up. > >I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the >same issue: > >[ 31.716031] vdd-5v: disabling >[ 31.719032] rst-usb-eth2: disabling >[ 31.722553] buck2: disabling > >While on that I tried to compare with other boards and see that they >also have the cpu-voltage marked as "regulator-always-on". The only >exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that >has not set this property. > >I agree with you and don't understand why this is happening. Has >anyone else an explanation? Do you have the CPU frequency scaling driver enabled? -michael
HI Michael, Am Fr., 17. Sept. 2021 um 10:21 Uhr schrieb Michael Walle <michael@walle.cc>: > > Am 17. September 2021 09:28:51 MESZ schrieb Heiko Thiery <heiko.thiery@gmail.com>: > >Hi Frieder, > > > >Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf > ><frieder.schrempf@kontron.de>: > >> > >> On 15.09.21 14:05, Michael Walle wrote: > >> > Am 2021-09-15 14:03, schrieb Heiko Thiery: > >> >> The buck2 output of the PMIC is the VDD core voltage of the cpu. > >> >> Switching off this will poweroff the CPU. Add the 'regulator-always-on' > >> >> property to avoid this. > >> > > >> > Mh, have this ever worked? Is there a commit which introduced a regression? > >> > >> Yes, this did work before, even without 'regulator-always-on'. I > >> currently don't understand why this is needed. The regulator is > >> referenced in the CPU nodes as 'cpu-supply'. This should be enough to > >> not disable it as long as the CPU is up. > > > >I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the > >same issue: > > > >[ 31.716031] vdd-5v: disabling > >[ 31.719032] rst-usb-eth2: disabling > >[ 31.722553] buck2: disabling > > > >While on that I tried to compare with other boards and see that they > >also have the cpu-voltage marked as "regulator-always-on". The only > >exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that > >has not set this property. > > > >I agree with you and don't understand why this is happening. Has > >anyone else an explanation? > > Do you have the CPU frequency scaling driver enabled? That was a good hint. What I now see is in my case the driver is not loaded (I boot via network). In the default kernel config the module is marked as 'm'. When switching to 'y' as expected it will be probed and the the buck2 (cpu-supply) will not be deactivated after 30 seconds. -- Heiko
Am Freitag, dem 17.09.2021 um 09:28 +0200 schrieb Heiko Thiery: > Hi Frieder, > > Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf > <frieder.schrempf@kontron.de>: > > > > On 15.09.21 14:05, Michael Walle wrote: > > > Am 2021-09-15 14:03, schrieb Heiko Thiery: > > > > The buck2 output of the PMIC is the VDD core voltage of the cpu. > > > > Switching off this will poweroff the CPU. Add the 'regulator-always-on' > > > > property to avoid this. > > > > > > Mh, have this ever worked? Is there a commit which introduced a regression? > > > > Yes, this did work before, even without 'regulator-always-on'. I > > currently don't understand why this is needed. The regulator is > > referenced in the CPU nodes as 'cpu-supply'. This should be enough to > > not disable it as long as the CPU is up. > > I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the > same issue: > > [ 31.716031] vdd-5v: disabling > [ 31.719032] rst-usb-eth2: disabling > [ 31.722553] buck2: disabling > > While on that I tried to compare with other boards and see that they > also have the cpu-voltage marked as "regulator-always-on". The only > exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that > has not set this property. > > I agree with you and don't understand why this is happening. Has > anyone else an explanation? > > [1] https://elixir.bootlin.com/linux/latest/source/arch/arm64/boot/dts/freescale/imx8mq-librem5-devkit.dts#L319 > Maybe your kernel config is missing the cpufreq driver, so you don't have a consumer of the regulator? Marking the regulator as always-on seems like the right thing to do, you don't want to depend on a consumer showing up to make sure that your CPU voltage isn't cut... Regards, Lucas
Hi Lucas, Am Fr., 17. Sept. 2021 um 13:44 Uhr schrieb Lucas Stach <l.stach@pengutronix.de>: > > Am Freitag, dem 17.09.2021 um 09:28 +0200 schrieb Heiko Thiery: > > Hi Frieder, > > > > Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf > > <frieder.schrempf@kontron.de>: > > > > > > On 15.09.21 14:05, Michael Walle wrote: > > > > Am 2021-09-15 14:03, schrieb Heiko Thiery: > > > > > The buck2 output of the PMIC is the VDD core voltage of the cpu. > > > > > Switching off this will poweroff the CPU. Add the 'regulator-always-on' > > > > > property to avoid this. > > > > > > > > Mh, have this ever worked? Is there a commit which introduced a regression? > > > > > > Yes, this did work before, even without 'regulator-always-on'. I > > > currently don't understand why this is needed. The regulator is > > > referenced in the CPU nodes as 'cpu-supply'. This should be enough to > > > not disable it as long as the CPU is up. > > > > I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the > > same issue: > > > > [ 31.716031] vdd-5v: disabling > > [ 31.719032] rst-usb-eth2: disabling > > [ 31.722553] buck2: disabling > > > > While on that I tried to compare with other boards and see that they > > also have the cpu-voltage marked as "regulator-always-on". The only > > exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that > > has not set this property. > > > > I agree with you and don't understand why this is happening. Has > > anyone else an explanation? > > > > [1] https://elixir.bootlin.com/linux/latest/source/arch/arm64/boot/dts/freescale/imx8mq-librem5-devkit.dts#L319 > > > Maybe your kernel config is missing the cpufreq driver, so you don't > have a consumer of the regulator? > > Marking the regulator as always-on seems like the right thing to do, > you don't want to depend on a consumer showing up to make sure that > your CPU voltage isn't cut... shouldn't it be that the node cpu-supply here is a consumer of the referenced voltage? -- Heiko
On 17.09.21 18:10, Heiko Thiery wrote: > Hi Lucas, > > Am Fr., 17. Sept. 2021 um 13:44 Uhr schrieb Lucas Stach > <l.stach@pengutronix.de>: >> >> Am Freitag, dem 17.09.2021 um 09:28 +0200 schrieb Heiko Thiery: >>> Hi Frieder, >>> >>> Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf >>> <frieder.schrempf@kontron.de>: >>>> >>>> On 15.09.21 14:05, Michael Walle wrote: >>>>> Am 2021-09-15 14:03, schrieb Heiko Thiery: >>>>>> The buck2 output of the PMIC is the VDD core voltage of the cpu. >>>>>> Switching off this will poweroff the CPU. Add the 'regulator-always-on' >>>>>> property to avoid this. >>>>> >>>>> Mh, have this ever worked? Is there a commit which introduced a regression? >>>> >>>> Yes, this did work before, even without 'regulator-always-on'. I >>>> currently don't understand why this is needed. The regulator is >>>> referenced in the CPU nodes as 'cpu-supply'. This should be enough to >>>> not disable it as long as the CPU is up. >>> >>> I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the >>> same issue: >>> >>> [ 31.716031] vdd-5v: disabling >>> [ 31.719032] rst-usb-eth2: disabling >>> [ 31.722553] buck2: disabling >>> >>> While on that I tried to compare with other boards and see that they >>> also have the cpu-voltage marked as "regulator-always-on". The only >>> exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that >>> has not set this property. >>> >>> I agree with you and don't understand why this is happening. Has >>> anyone else an explanation? >>> >>> [1] https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Felixir.bootlin.com%2Flinux%2Flatest%2Fsource%2Farch%2Farm64%2Fboot%2Fdts%2Ffreescale%2Fimx8mq-librem5-devkit.dts%23L319&data=04%7C01%7Cfrieder.schrempf%40kontron.de%7Cce9d266ad78a4d06721008d979f5aeed%7C8c9d3c973fd941c8a2b1646f3942daf1%7C0%7C0%7C637674918380815550%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PsTKX4MXGwvvP6WxmQ1OWc8e32YI6Nsu%2FEkcNR4V8is%3D&reserved=0 >>> >> Maybe your kernel config is missing the cpufreq driver, so you don't >> have a consumer of the regulator? >> >> Marking the regulator as always-on seems like the right thing to do, >> you don't want to depend on a consumer showing up to make sure that >> your CPU voltage isn't cut... > > shouldn't it be that the node cpu-supply here is a consumer of the > referenced voltage? Yes, but as Michael and Lucas noted, the consumer is only active if the cpufreq driver is loaded and we don't want to depend on this. In my config I always had this compiled into the kernel so I didn't notice that the always-on property is missing. I'm not sure if shutting down the regulator is allowed to save power in suspend, but that's something we could tackle later, for now: Reviewed-by: Frieder Schrempf <frieder.schrempf@kontron.de> Thanks Heiko for investigating and Michael and Lucas for helping!
Am 20. September 2021 09:31:20 MESZ schrieb Frieder Schrempf <frieder.schrempf@kontron.de>: >On 17.09.21 18:10, Heiko Thiery wrote: >> Hi Lucas, >> >> Am Fr., 17. Sept. 2021 um 13:44 Uhr schrieb Lucas Stach >> <l.stach@pengutronix.de>: >>> >>> Am Freitag, dem 17.09.2021 um 09:28 +0200 schrieb Heiko Thiery: >>>> Hi Frieder, >>>> >>>> Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf >>>> <frieder.schrempf@kontron.de>: >>>>> >>>>> On 15.09.21 14:05, Michael Walle wrote: >>>>>> Am 2021-09-15 14:03, schrieb Heiko Thiery: >>>>>>> The buck2 output of the PMIC is the VDD core voltage of the cpu. >>>>>>> Switching off this will poweroff the CPU. Add the 'regulator-always-on' >>>>>>> property to avoid this. >>>>>> >>>>>> Mh, have this ever worked? Is there a commit which introduced a regression? >>>>> >>>>> Yes, this did work before, even without 'regulator-always-on'. I >>>>> currently don't understand why this is needed. The regulator is >>>>> referenced in the CPU nodes as 'cpu-supply'. This should be enough to >>>>> not disable it as long as the CPU is up. >>>> >>>> I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the >>>> same issue: >>>> >>>> [ 31.716031] vdd-5v: disabling >>>> [ 31.719032] rst-usb-eth2: disabling >>>> [ 31.722553] buck2: disabling >>>> >>>> While on that I tried to compare with other boards and see that they >>>> also have the cpu-voltage marked as "regulator-always-on". The only >>>> exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that >>>> has not set this property. >>>> >>>> I agree with you and don't understand why this is happening. Has >>>> anyone else an explanation? >>>> >>>> [1] https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Felixir.bootlin.com%2Flinux%2Flatest%2Fsource%2Farch%2Farm64%2Fboot%2Fdts%2Ffreescale%2Fimx8mq-librem5-devkit.dts%23L319&data=04%7C01%7Cfrieder.schrempf%40kontron.de%7Cce9d266ad78a4d06721008d979f5aeed%7C8c9d3c973fd941c8a2b1646f3942daf1%7C0%7C0%7C637674918380815550%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PsTKX4MXGwvvP6WxmQ1OWc8e32YI6Nsu%2FEkcNR4V8is%3D&reserved=0 >>>> >>> Maybe your kernel config is missing the cpufreq driver, so you don't >>> have a consumer of the regulator? >>> >>> Marking the regulator as always-on seems like the right thing to do, >>> you don't want to depend on a consumer showing up to make sure that >>> your CPU voltage isn't cut... >> >> shouldn't it be that the node cpu-supply here is a consumer of the >> referenced voltage? > >Yes, but as Michael and Lucas noted, the consumer is only active if the >cpufreq driver is loaded and we don't want to depend on this. In my >config I always had this compiled into the kernel so I didn't notice >that the always-on property is missing. Hm. So any cpu core voltage has to have the "always-on" property? On any SoC? Shouldn't there be some code to prevent the disabling of the voltage if the cpu nodes have a cpu-supply phandle? -michael
Am Montag, dem 20.09.2021 um 09:43 +0200 schrieb Michael Walle: > Am 20. September 2021 09:31:20 MESZ schrieb Frieder Schrempf <frieder.schrempf@kontron.de>: > > On 17.09.21 18:10, Heiko Thiery wrote: > > > Hi Lucas, > > > > > > Am Fr., 17. Sept. 2021 um 13:44 Uhr schrieb Lucas Stach > > > <l.stach@pengutronix.de>: > > > > > > > > Am Freitag, dem 17.09.2021 um 09:28 +0200 schrieb Heiko Thiery: > > > > > Hi Frieder, > > > > > > > > > > Am Mi., 15. Sept. 2021 um 14:09 Uhr schrieb Frieder Schrempf > > > > > <frieder.schrempf@kontron.de>: > > > > > > > > > > > > On 15.09.21 14:05, Michael Walle wrote: > > > > > > > Am 2021-09-15 14:03, schrieb Heiko Thiery: > > > > > > > > The buck2 output of the PMIC is the VDD core voltage of the cpu. > > > > > > > > Switching off this will poweroff the CPU. Add the 'regulator-always-on' > > > > > > > > property to avoid this. > > > > > > > > > > > > > > Mh, have this ever worked? Is there a commit which introduced a regression? > > > > > > > > > > > > Yes, this did work before, even without 'regulator-always-on'. I > > > > > > currently don't understand why this is needed. The regulator is > > > > > > referenced in the CPU nodes as 'cpu-supply'. This should be enough to > > > > > > not disable it as long as the CPU is up. > > > > > > > > > > I rechecked that with 5.11, 5.10 and 5.9 and I see on all of them the > > > > > same issue: > > > > > > > > > > [ 31.716031] vdd-5v: disabling > > > > > [ 31.719032] rst-usb-eth2: disabling > > > > > [ 31.722553] buck2: disabling > > > > > > > > > > While on that I tried to compare with other boards and see that they > > > > > also have the cpu-voltage marked as "regulator-always-on". The only > > > > > exception in dts/freescale is in imx8mq-librem5-devkit.dts [1] that > > > > > has not set this property. > > > > > > > > > > I agree with you and don't understand why this is happening. Has > > > > > anyone else an explanation? > > > > > > > > > > [1] https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Felixir.bootlin.com%2Flinux%2Flatest%2Fsource%2Farch%2Farm64%2Fboot%2Fdts%2Ffreescale%2Fimx8mq-librem5-devkit.dts%23L319&data=04%7C01%7Cfrieder.schrempf%40kontron.de%7Cce9d266ad78a4d06721008d979f5aeed%7C8c9d3c973fd941c8a2b1646f3942daf1%7C0%7C0%7C637674918380815550%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PsTKX4MXGwvvP6WxmQ1OWc8e32YI6Nsu%2FEkcNR4V8is%3D&reserved=0 > > > > > > > > > Maybe your kernel config is missing the cpufreq driver, so you don't > > > > have a consumer of the regulator? > > > > > > > > Marking the regulator as always-on seems like the right thing to do, > > > > you don't want to depend on a consumer showing up to make sure that > > > > your CPU voltage isn't cut... > > > > > > shouldn't it be that the node cpu-supply here is a consumer of the > > > referenced voltage? > > > > Yes, but as Michael and Lucas noted, the consumer is only active if the > > cpufreq driver is loaded and we don't want to depend on this. In my > > config I always had this compiled into the kernel so I didn't notice > > that the always-on property is missing. > > Hm. So any cpu core voltage has to have the "always-on" property? On > any SoC? Shouldn't there be some code to prevent the disabling of the > voltage if the cpu nodes have a cpu-supply phandle? Yes, all regulators that must not be shut down for proper system operation must be marked as always-on. Just because it is a CPU supply doesn't make it special in any way. Otherwise where would you stop with adding special code to keep supplies enabled without a consumer? Have code to keep the DDR memory regulators on when no DDR controller driver is loaded? Code to keep 3.3V for the board level peripherals enabled? Just mark those as always-on. Regards, Lucas
On Wed, Sep 15, 2021 at 02:03:26PM +0200, Heiko Thiery wrote: > The buck2 output of the PMIC is the VDD core voltage of the cpu. > Switching off this will poweroff the CPU. Add the 'regulator-always-on' > property to avoid this. > > Fixes: 8668d8b2e67f ("arm64: dts: Add the Kontron i.MX8M Mini SoMs and baseboards") > Signed-off-by: Heiko Thiery <heiko.thiery@gmail.com> Applied, thanks!
diff --git a/arch/arm64/boot/dts/freescale/imx8mm-kontron-n801x-som.dtsi b/arch/arm64/boot/dts/freescale/imx8mm-kontron-n801x-som.dtsi index d0456daefda8..9db9b90bf2bc 100644 --- a/arch/arm64/boot/dts/freescale/imx8mm-kontron-n801x-som.dtsi +++ b/arch/arm64/boot/dts/freescale/imx8mm-kontron-n801x-som.dtsi @@ -102,6 +102,7 @@ reg_vdd_arm: BUCK2 { regulator-min-microvolt = <850000>; regulator-max-microvolt = <950000>; regulator-boot-on; + regulator-always-on; regulator-ramp-delay = <3125>; nxp,dvs-run-voltage = <950000>; nxp,dvs-standby-voltage = <850000>;
The buck2 output of the PMIC is the VDD core voltage of the cpu. Switching off this will poweroff the CPU. Add the 'regulator-always-on' property to avoid this. Fixes: 8668d8b2e67f ("arm64: dts: Add the Kontron i.MX8M Mini SoMs and baseboards") Signed-off-by: Heiko Thiery <heiko.thiery@gmail.com> --- arch/arm64/boot/dts/freescale/imx8mm-kontron-n801x-som.dtsi | 1 + 1 file changed, 1 insertion(+)