On 20.12.2024 12:27 PM, Sudeep Holla wrote: > On Thu, Dec 19, 2024 at 08:43:27PM +0100, Konrad Dybcio wrote: >> On 6.12.2024 11:21 AM, Sudeep Holla wrote: >>> On Mon, Oct 28, 2024 at 03:22:57PM +0100, Konrad Dybcio wrote: >>>> From: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx> >>>> >>>> Certain firmware implementations (such as the ones found on Qualcomm >>>> SoCs between roughly 2015 and 2023) expose an S3-like S2RAM state >>>> through the CPU_SUSPEND call, as opposed to exposing PSCIv1.0's >>>> optional PSCI_SYSTEM_SUSPEND. >>>> >>> >>> If so, can you elaborate why s2idle doesn't work as an alternative to what >>> you are hacking up here. >> >> Please see other branches of this thread >> >>> >>>> This really doesn't work well with the model where we associate all >>>> calls to CPU_SUSPEND with cpuidle. Allow specifying a single special >>>> CPU_SUSPEND suspend parameter value that is to be treated just like >>>> SYSTEM_SUSPEND from the OS's point of view. >>>> >>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx> >>>> --- >>>> Documentation/devicetree/bindings/arm/psci.yaml | 6 ++++++ >>>> 1 file changed, 6 insertions(+) >>>> >>>> diff --git a/Documentation/devicetree/bindings/arm/psci.yaml b/Documentation/devicetree/bindings/arm/psci.yaml >>>> index cbb012e217ab80c1ca88e611e7acc06c6d56fad0..a6901878697c8e1ec1cbfed62298ae3bc58f2501 100644 >>>> --- a/Documentation/devicetree/bindings/arm/psci.yaml >>>> +++ b/Documentation/devicetree/bindings/arm/psci.yaml >>>> @@ -98,6 +98,12 @@ properties: >>>> [1] Kernel documentation - ARM idle states bindings >>>> Documentation/devicetree/bindings/cpu/idle-states.yaml >>>> >>>> + arm,psci-s2ram-param: >>>> + $ref: /schemas/types.yaml#/definitions/uint32 >>>> + description: >>>> + power_state parameter denoting the S2RAM/S3-like system suspend state >>> >>> Yet another NACK as this corresponds to PSCI SYSTEM_SUSPEND and as per >>> specification it takes no such parameter. This is just misleading. >>> >> >> Yeah PSCI_SYSTEM_SUSPEND takes care of this on platforms that expose it. >> > > And those that don't advertise/expose don't get to use, simple. The spec says: "The call is equivalent to using the CPU_SUSPEND call for the deepest possible platform powerdown state." so by that logic, I'd rather call implementing PSCI_SYSTEM_SUSPEND in Linux unnecessary bloat.. >> DEN0022F.b Section 6.5. recommends that CPU_SUSPEND StateID includes >> a field for system-level power down states. This binding change only >> adds a way for DT-based platforms to associate such state with S2RAM >> suspend. >> > > Sure, just use the CPU_SUSPEND bindings that already exist. No need to > define this as some special case if it is exposed as CPU_SUSPEND idle > state. Not sure why you want to do it differently. I understand the > need to handle it different in the kernel, but I don't understand to > define the new bindings for that. Just use the existing bindings for the > idle states. Again I see no exception for this case. The bindings exist for core/cluster idle states. This whole series tries to include a system-wide suspend state that has no business being described as a cpuidle one and depends on more than just the CPUs being powered down. >> That may be a bit Linux-specific whereas bindings are supposed to be >> OS-agnostic, but since we effectively want one PSCI state for deep >> suspend regardless of the OS, I would think this kind of hint is fine. >> > > Exactly, that's the reason for not changing this into special case and > special binding for that special case created. I simply don't think it's fitting to lie about system suspend states being just CPU idle states, see above. Konrad
