On 7/4/23 11:03, Mukunda,Vijendar wrote: > On 04/07/23 13:57, Pierre-Louis Bossart wrote: >> >> On 7/4/23 09:37, Mukunda,Vijendar wrote: >>> On 04/07/23 12:41, Pierre-Louis Bossart wrote: >>>> On 7/4/23 08:36, Mukunda,Vijendar wrote: >>>>> On 03/07/23 22:48, Pierre-Louis Bossart wrote: >>>>>> On 7/3/23 17:18, Mukunda,Vijendar wrote: >>>>>>> On 03/07/23 20:15, Pierre-Louis Bossart wrote: >>>>>>>> On 7/3/23 16:46, Mukunda,Vijendar wrote: >>>>>>>>> On 03/07/23 19:50, Pierre-Louis Bossart wrote: >>>>>>>>>> On 7/3/23 15:31, Mukunda,Vijendar wrote: >>>>>>>>>>> On 03/07/23 18:30, Shuming [范書銘] wrote: >>>>>>>>>>>>>>>> During ClockStop Mode0, peripheral interrupts are disabled. >>>>>>>>>>>>>>> I can see that the interrupts are disabled in >>>>>>>>>>>>>>> rt5682_dev_system_suspend(), which is NOT a mode where the clock stop >>>>>>>>>>>>>>> is used... I don't think this commit message is correct. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> The IMPL_DEF interrupt which is used for jack detection is not >>>>>>>>>>>>>>> disabled at all during any clock stop mode, and it shouldn't >>>>>>>>>>>>>>> otherwise that would break the jack detection. >>>>>>>>>>>>>> You are right. The commit message is wrong and not clear. >>>>>>>>>>>>>> The situation is that the manager driver uses the clock stop mode0 to do >>>>>>>>>>>>> system suspension. >>>>>>>>>>>>> >>>>>>>>>>>>> No it does not. The clock stop is ONLY used for pm_runtime, never for system >>>>>>>>>>>>> suspend. We cannot go to system suspend with the link in clock-stop mode, >>>>>>>>>>>>> that will create lots of issues, that's why we perform a full pm_runtime resume >>>>>>>>>>>>> in the .prepare stage. >>>>>>>>>>>> OK, I got your point. Thanks. However, this issue reported by AMD. >>>>>>>>>>>> The AMD platform validated system level pm and runtime pm ops with the different modes. >>>>>>>>>>>> >>>>>>>>>>>> Hi Vijendar, >>>>>>>>>>>> Do you have any comments? >>>>>>>>>>> On AMD platforms, we are supporting two power modes. >>>>>>>>>>> 1) Bus reset mode >>>>>>>>>>> 2) Clock Stop Mode >>>>>>>>>>> >>>>>>>>>>> In Bus reset mode, bus will re-enumerate the peripheral devices >>>>>>>>>>> whereas in ClockStop Mode, applying ClockStop Mode0 >>>>>>>>>>> in both pm ops (runtime pm ops and system level pm ops). >>>>>>>>>>> >>>>>>>>>>> Currently, SDCA interrupts are disabled on peripheral side, when system level >>>>>>>>>>> suspend is invoked. >>>>>>>>>>> For ClockStop mode SDW manager is not receiving any jack alert when >>>>>>>>>>> SoundWire manager device is in D3 state. >>>>>>>>>> That was precisely the point of clock stop mode: a peripheral can >>>>>>>>>> restart the system even when it's in lower-power mode. >>>>>>>>>> >>>>>>>>>> If there's no means to let a peripheral restart, the only benefit is >>>>>>>>>> maybe to skip the enumeration time. That's not what the spec intended.... >>>>>>>>> As per our understanding, you are pointing to ClockStopMode1. >>>>>>>>> ClockStopMode1 requires re-enumeration as peripherals move >>>>>>>>> to unattached state. >>>>>>>>> We have cross-checked ClockStopMode0 description in spec. >>>>>>>>> It doesn't specify about peripheral device state as Unattached. >>>>>>>>> We are referring here "ClockStopMode0" only. >>>>>>>> No I was describing the difference between the 'Bus reset mode' and the >>>>>>>> 'clock stop mode' on the manager side. >>>>>>>> >>>>>>>> There's also nothing in the spec preventing the manager from doing a >>>>>>>> reset at any time, including after exiting the clock mode0 stop. >>>>>>>> >>>>>>>> >>>>>>> Partly I agree. As per our understanding, If any of the peripherals lost's sync, >>>>>>> and re-enumeration is required. >>>>>>> If continuous parity errors/bus clash conditions are reported over the link, >>>>>>> Sdw Manager bus reset sequence should be invoked. This is a different scenario. >>>>>>> Both the scenarios are asynchronous. >>>>>>> >>>>>>> Going with Spec definition for ClockStopMode0, as it's Imp defined for >>>>>>> SoundWire Manager, want to stick to Clockstop when D3 call is invoked >>>>>>> and restore the clock when D0 call is invoked for our platforms. >>>>>> The problem is that 'D3' can be used for two separate scenarios >>>>>> - S0/D3: that's pm_runtime suspend >>>>>> - Sx/D4: that's system suspend >>>>>> It's very unclear what the benefit of clock stop mode would be for the >>>>>> latter case. >>>>>> >>>>> We are pointing system level suspend scenario only. >>>>> >>>>> We have already SoundWire Power off mode - ClockStop Mode will be >>>>> applied followed by bus reset and SoundWire manager will be disabled. >>>>> On parent side, ACP soft reset will be applied and ACP parent driver can >>>>> wake for PME events for Soundwire manager. >>>>> >>>>> Sx/D4 case, want to avoid re-enumeration when multiple peripherals are connected. >>>>> For Clock Stop + Bus reset (keeping Sdw manager enabled), we can define a new power mode >>>>> on SoundWire manager side. >>>>> >>>>> Having said that, below are power modes we want to support on our side. >>>>> >>>>> 1) Idle Mode - Don't stop the clock for SoundWire manager and Soundwire manager is not >>>>> disabled during suspend callbacks. >>>>> 2) ClockStop Mode - Only apply ClockStopMode0 on Soundwire manager side and restore >>>>> the clock when resume sequence is invoked (even in case of system level suspend also) >>>>> 3) ClockStop Mode + Bus reset - Apply ClockStopMode0 when Sound wire manager entering >>>>> runtime suspend state. In system level resume case, apply clock stop exit and bus reset which >>>>> will re-enumerate all peripheral devices over the links. >>>>> 4) Power off mode: apply Clock Stop + SoundWire manager IP power off sequence. >>>>> Parent device(ACP) will be powered off. ACP can wake up for in band wake-up events. >>>>> >>>>> We expect if someone wants to go with only ClockStop mode, peripheral side also required >>>>> support should be enabled. >>>> I am not following any of this. You wrote both >>>> >>>> " >>>> For ClockStop mode SDW manager is not receiving any jack alert when >>>> SoundWire manager device is in D3 state. >>>> " >>>> >>>> and >>>> >>>> " >>>> Parent device(ACP) will be powered off. ACP can wake up for in band >>>> wake-up events. >>>> " >>>> >>>> I can't figure out how you deal with wake-ups and when exactly you >>>> expect jack detection to be functional. >>> Both power modes (case 2 & case 4) are different. >>> The current patch series is targeted to address JD events during system level >>> suspend/resume scenario, when ClockStop Power mode is selected.i.e case 2 >>> as mentioned above. We're not expecting bus reset during system level resume in >>> this power mode. >>> As SDCA interrrupts are not enabled on peripheral side during system level resume, >>> No JD events are reported. >>> >>> Test scenario will be let system level supend gets invoked. During suspend, plug/un-plug jack. >>> During system level resume, JD status should be reported correctly. >> This is completely different to the definition of 'clock stop mode'. >> What you want is to make sure the status is reported correctly when the >> system resume, not to resume the system using the clock stop exit >> mechanism at the initiative of the peripheral. > No. What we want to convey here is in system level resume for SoundWire manager, > clock stop exit sequence will be invoked. > After clock stop exit sequence completed, we expect JD alert should be reported if there is any > JD event during Sx/D4 state. > As SDCA interrupts are disabled previously when system level suspend > is invoked on peripheral side, even after finishing clock stop exit sequence, we are > no longer observing Jack alerts as SDCA interrupts are disabled. The existing code in the codec driver disables the IMPL_DEF or SDCA interrupt in the system suspend routines, no matter what happens on the manager side. see e.g. rt711_dev_system_suspend This patch really has nothing to do with clock stop mode0 proper. I think what happens is that if there is a new attachment, the sdw_initialize_slave() helper will re-enable the IMPL_DEF or SDCA interrupt, and that does not happen if there is no attachment and re-enumeration. I will be looking forward to the v2, but as I indicated earlier it might be a better idea to fix this difference in programming flows in the clock stop exit code rather in some specific codec drivers.