arm64: dts: qcom: hamoa/x1: Fix TODO in system power domain node
PATCH 1/1 arm64: dts: qcom: hamoa/x1: Fix TODO in system power domain node
At present, designs based on the Qualcomm X1 Hamoa reference platform have an always-on system power domain since the OF node values aren't implement.
Fix this by using the CPU C4, cluster CL5 and system DRIPS parameters from the ACPI DSDT Windows uses, together with the Low Power Idle _LPI minimum residency of 9000us and wake latency of 5000us as exit latency. Finally, assume the entry latency is the difference of these two values.
Tested on a Lenovo Yoga Slim 7x with Qualcomm X1E-80-100.
--- arch/arm64/boot/dts/qcom/hamoa.dtsi | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/hamoa.dtsi +9 -1
@@ -290,6 +290,14 @@ cluster_cl5: cluster-sleep-1 { exit-latency-us = <4000>; min-residency-us = <7000>; }; + + domain_ss3: domain-sleep-0 { + compatible = "domain-idle-state"; + arm,psci-suspend-param = <0x0200c354>; + entry-latency-us = <4000>; + exit-latency-us = <5000>; + min-residency-us = <9000>; + }; }; };
@@ -448,7 +456,7 @@ cluster_pd2: power-domain-cpu-cluster2 { system_pd: power-domain-system { #power-domain-cells = <0>; - /* TODO: system-wide idle states */ + domain-idle-states = <&domain_ss3>; }; };
Signature · 1 line
2.51.0
Quoted 1 line · Daniel J Blueman
At present, designs based on the Qualcomm X1 Hamoa reference platform have an always-on system power domain since the OF node values aren't implement.
Implemented.
Quoted 2 lines
Fix this by using the CPU C4, cluster CL5 and system DRIPS parameters from the ACPI DSDT Windows uses, together with the Low Power Idle
_LPIminimum residency of 9000us and wake latency of 5000us as exit latency. Finally, assume the entry latency is the difference of these two values.
ACPI used 5000usec as total exit latency, with this change total exit latency will be 9000usec (sum of entry + exit).
Quoted 4 lines
Tested on a Lenovo Yoga Slim 7x with Qualcomm X1E-80-100.
Fixes f33767e3cfa5 ("arm64: dts: qcom: x1e80100: Add missing system-wide PSCI power domain")
Using this fixes tag, can make the change back ported to stable kernels without dependencies and may break the GPIO IRQs.
Background:
PDC monitors the wakeup capable IRQs during system wide low power state, hitting the system low power mode can break to wake via GPIO IRQs. The system-wide idle state was not added since the wakeup capable GPIO IRQs were not configured at PDC with 602cb14e310a ("pinctrl: qcom: x1e80100: Bypass PDC wakeup parent for now").
So IMO this fixes tag should be used instead of above with the changes to configure PDC to monitor GPIO wake ups. I have these changes to configure GPIO IRQs at PDC and enable back domain_ss3 idle state in my local tree, which i plan to post this week or next.
Thanks,
Maulik
Quoted 33 lines
Signed-off-by Daniel J Blueman <[email protected]>--- arch/arm64/boot/dts/qcom/hamoa.dtsi | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)arch/arm64/boot/dts/qcom/hamoa.dtsi +9
-1@@ -290,6 +290,14 @@ cluster_cl5: cluster-sleep-1 { exit-latency-us = <4000>; min-residency-us = <7000>; }; + + domain_ss3: domain-sleep-0 { + compatible = "domain-idle-state"; + arm,psci-suspend-param = <0x0200c354>; + entry-latency-us = <4000>; + exit-latency-us = <5000>; + min-residency-us = <9000>; + }; }; };@@ -448,7 +456,7 @@ cluster_pd2: power-domain-cpu-cluster2 { system_pd: power-domain-system { #power-domain-cells = <0>; - /* TODO: system-wide idle states */ + domain-idle-states = <&domain_ss3>; }; };
Quoted 1 line · Maulik Shah (mkshah)
On 2/21/2026 4:21 PM, Daniel J Blueman wrote:
[..]
Quoted 1 line
Quoted 1 line
Fix this by using the CPU C4, cluster CL5 and system DRIPS parameters from the ACPI DSDT Windows uses, together with the Low Power Idle
_LPIminimum residency of 9000us and wake latency of 5000us as exit latency. Finally, assume the entry latency is the difference of these two values.
[..]
Quoted 5 lines
Quoted 1 line
Fixes f33767e3cfa5 ("arm64: dts: qcom: x1e80100: Add missing system-wide PSCI power domain")Using this fixes tag, can make the change back ported to stable kernels without dependencies and may break the GPIO IRQs.
Background:
Quoted 7 lines
PDC monitors the wakeup capable IRQs during system wide low power state, hitting the system low power mode can break to wake via GPIO IRQs. The system-wide idle state was not added since the wakeup capable GPIO IRQs were not configured at PDC with
602cb14e310a("pinctrl: qcom: x1e80100: Bypass PDC wakeup parent for now").So IMO this fixes tag should be used instead of above with the changes to configure PDC to monitor GPIO wake ups. I have these changes to configure GPIO IRQs at PDC and enable back
domain_ss3idle state in my local tree, which i plan to post this week or next.
On a previous episode of L-K-M-L… :)
https://lore.kernel.org/all/[email protected]/
FWIW I have just tested Konrad's patch from there that adds all three states (0x02000154, 0x02000254, and 0x0200c354):
❯ doas cat /sys/kernel/debug/pm_genpd/power-domain-system/idle_states State Time(ms) Usage Rejected Above Below S2idle S0 367 330 9 315 9 0 S1 2719 2057 553 2520 71 0 S2 0 0 1 1 0 0
As of right now I don't see any improvement in idle power consumption from just this S1 thing, compared to not adding anything and having the implicitly-added-by-code S0 state only.
It still only goes as low as 2.5W in screen-off idle, and that's with runtime PM enabled for 3 USB controllers out of 4 (enabling it on all 4 makes the system shut down), without doing that it's more like 2.75W.
---
Maulik, since you seem to be the oss.qualcomm person familiar with power management — could you please shine some light onto the mystery of how Windows achieves ~0.5W battery consumption in screen-off idle (and only slightly higher in screen-on idle) i.e. what exactly could be wasting those extra 2W under Linux? Ever since people started daily driving X series based laptops this has been an eternal topic/question in aarch64-laptops…
Is it just Windows doing something "extraordinary" like opportunistic full-system sleep (as deep as CX collapse), even with display on and in PSR?
Or are we still missing something big in Linux?
That issue with runtime-suspending all four USB controllers shutting the system down, does that mean there's some rail where USB ends up being the last load-bearing thing holding it up and we'd like to let it go down properly?
Thanks
~val
Bjorn Andersson
maintainerQuoted 2 lines · Val Packett
Quoted 1 line · Maulik Shah (mkshah)
On 2/21/2026 4:21 PM, Daniel J Blueman wrote:
[..]
Quoted 5 lines
That issue with runtime-suspending all four USB controllers shutting the system down, does that mean there's some rail where USB ends up being the last load-bearing thing holding it up and we'd like to let it go down properly?
The three USB nodes have power-domains = <&gcc GCC_USB30_*_GDSC> and required-opps = <&opp_nom>. &gcc has power-domains = < &rpmhpd CX>. This means that as long as the USB controllers are active, CX is kept at at least nominal corner.
The reason why the USB node has &opp_nom is because it's required by the 200MHz MASTER_CLK, which iirc relates to the possibility that you might have something SuperSpeed connected. (But as we flatten dwc3, we should be able to scale this based on what is actually connected)
I don't know why there's multiple pm_runtime_forbid() in the USB stack, this needs to be revisited.
Then to your actual question, yes, if you suspend the USB controllers and the result is that the system dies, then we have other resources piggybacking on the USB controllers' votes for CX.
I did some experiments on SC8280XP earlier and was able to drop the pm_runtime_forbid() - and see CX drop below NOM. Iirc more work was needed to get reasonable behavior in terms of plug-detect, so the patches got stuck in their topic branch...
Regards,
Bjorn
Quoted 4 lines
Thanks
~val
Bjorn Andersson
maintainerQuoted 20 lines · Maulik Shah (mkshah)
Quoted 1 line · Daniel J Blueman
At present, designs based on the Qualcomm X1 Hamoa reference platform have an always-on system power domain since the OF node values aren't implement.
Implemented.
Quoted 2 lines
Fix this by using the CPU C4, cluster CL5 and system DRIPS parameters from the ACPI DSDT Windows uses, together with the Low Power Idle
_LPIminimum residency of 9000us and wake latency of 5000us as exit latency. Finally, assume the entry latency is the difference of these two values.ACPI used 5000usec as total exit latency, with this change total exit latency will be 9000usec (sum of entry + exit).
Quoted 4 lines
Tested on a Lenovo Yoga Slim 7x with Qualcomm X1E-80-100.
Fixes f33767e3cfa5 ("arm64: dts: qcom: x1e80100: Add missing system-wide PSCI power domain")Using this fixes tag, can make the change back ported to stable kernels without dependencies and may break the GPIO IRQs.
Background:
Quoted 8 lines
PDC monitors the wakeup capable IRQs during system wide low power state, hitting the system low power mode can break to wake via GPIO IRQs. The system-wide idle state was not added since the wakeup capable GPIO IRQs were not configured at PDC with
602cb14e310a("pinctrl: qcom: x1e80100: Bypass PDC wakeup parent for now").So IMO this fixes tag should be used instead of above with the changes to configure PDC to monitor GPIO wake ups. I have these changes to configure GPIO IRQs at PDC and enable back
domain_ss3idle state in my local tree, which i plan to post this week or next.
No, this patch does not fix 602cb14e310a ("pinctrl: qcom: x1e80100:
Bypass PDC wakeup parent for now"), please see Documentation/process/stable-kernel-rules.rst on how to document dependencies for fixes.
It could however be argued that this doesn't fix a bug with f33767e3cfa5 ("arm64: dts: qcom: x1e80100: Add missing system-wide PSCI power domain"), but instead just implements the missing domain-idle-state (i.e. drop the Fixes).
PS. What I do expect is that 602cb14e310a ("pinctrl: qcom: x1e80100:
Bypass PDC wakeup parent for now") should have been reverted once the PDC driver gained the necessary workarounds!
Regards,
Bjorn
Quoted 37 lines
Thanks,
MaulikQuoted 33 lines
Signed-off-by Daniel J Blueman <[email protected]>--- arch/arm64/boot/dts/qcom/hamoa.dtsi | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)arch/arm64/boot/dts/qcom/hamoa.dtsi +9
-1@@ -290,6 +290,14 @@ cluster_cl5: cluster-sleep-1 { exit-latency-us = <4000>; min-residency-us = <7000>; }; + + domain_ss3: domain-sleep-0 { + compatible = "domain-idle-state"; + arm,psci-suspend-param = <0x0200c354>; + entry-latency-us = <4000>; + exit-latency-us = <5000>; + min-residency-us = <9000>; + }; }; };@@ -448,7 +456,7 @@ cluster_pd2: power-domain-cpu-cluster2 { system_pd: power-domain-system { #power-domain-cells = <0>; - /* TODO: system-wide idle states */ + domain-idle-states = <&domain_ss3>; }; };
Bjorn Andersson
maintainerQuoted 3 lines · Daniel J Blueman
At present, designs based on the Qualcomm X1 Hamoa reference platform have an always-on system power domain since the OF node values aren't implement.
Fix this by using the CPU C4, cluster CL5 and system DRIPS parameters from the ACPI DSDT Windows uses, together with the Low Power Idle
_LPIminimum residency of 9000us and wake latency of 5000us as exit latency. Finally, assume the entry latency is the difference of these two values.
Please wrap your lines to 72 characters.
Regards,
Bjorn
Quoted 40 lines
Tested on a Lenovo Yoga Slim 7x with Qualcomm X1E-80-100.
Fixes f33767e3cfa5 ("arm64: dts: qcom: x1e80100: Add missing system-wide PSCI power domain")Signed-off-by Daniel J Blueman <[email protected]>--- arch/arm64/boot/dts/qcom/hamoa.dtsi | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)arch/arm64/boot/dts/qcom/hamoa.dtsi +9
-1@@ -290,6 +290,14 @@ cluster_cl5: cluster-sleep-1 { exit-latency-us = <4000>; min-residency-us = <7000>; }; + + domain_ss3: domain-sleep-0 { + compatible = "domain-idle-state"; + arm,psci-suspend-param = <0x0200c354>; + entry-latency-us = <4000>; + exit-latency-us = <5000>; + min-residency-us = <9000>; + }; }; };@@ -448,7 +456,7 @@ cluster_pd2: power-domain-cpu-cluster2 { system_pd: power-domain-system { #power-domain-cells = <0>; - /* TODO: system-wide idle states */ + domain-idle-states = <&domain_ss3>; }; };Signature · 1 line
2.51.0