> -----Original Message----- > From: linux-omap-owner@xxxxxxxxxxxxxxx [mailto:linux-omap-owner@xxxxxxxxxxxxxxx] On Behalf Of Benoit > Cousson > >>>> Hi Vishwa, > >>>> > >>>> On 5/13/2010 12:12 PM, shweta gulati wrote: > >>>>> From: Vishwanath Sripathy<vishwanath.bs@xxxxxx> > >>>>> > >>>>> OMAP3430/3630 has a Silicon bug because of which SDRC is > >>>>> released from IDLE even before Core DPLL has locked. This leads > >>>>> to undefined behaviour of SDRC DLL. > >>>>> > >>>>> This patch has workaround for the same. > >>>>> > >>>>> Description of WA for 3430: > >>>>> Initialization: > >>>>> Disable DPLL3 automatic mode by default. Issue will not be faced as DPLL3 > >>>>> is always locked. > >>>>> > >>>>> Before CORE Voltage Domain (VDD2) Sleep Transition to RETENTION or OFF > >> mode: > >>>>> 1. Reduce DPLL3 M2 Frequency to get L3 running at OPP2 Frequency > >>>>> (by changing M2 Divider value). This is increasing the period duration of > >>>>> one L3 clock cycle. > >>>>> o In case of CORE is at OPP3 (166MHz@xxxxx): > >>>>> " Lower the frequency to 83MHz. > >>>>> > >>>>> o In case of CORE is at OPP2 (83MHz@xxxxx): > >>>>> " Keep the frequency as it is (83MHz). > >>>>> > >>>>> 2. Increase CORE Voltage to 1.2V. This is reducing the timing duration of > >> the > >>>>> critical path signal which will now fit to one L3 clock cycle. > >>>>> > >>>>> 3. Enable DPLL3 Automatic mode. This will ensure proper transition to > >>>>> RETENTION or OFF mode. > >>>>> > >>>>> After CORE Voltage Domain Wakeup Transition from RETENTION or OFF mode: > >>>>> 1. Disable DPLL3 Automatic mode. > >>>>> 2. Restore previous DPLL3 M2 Frequency and CORE Voltage values. > >>>>> > >>>>> Description of WA for 3630: > >>>>> Initialization: > >>>>> Disable DPLL3 automatic mode by default. Issue will not be faced as DPLL3 is > >>>> always locked. > >>>>> > >>>>> Before CORE Voltage Domain(VDD2) Sleep Transition to RETENTION or OFF > >> mode: > >>>>> 1. Reduce DPLL3 M2 Frequency to get L3 running at OPP50 Frequency > >>>>> (by changing M2 Divider value) and set VDD2 Voltage for OPP100. > >>>>> This is increasing the period duration of one L3 clock cycle and reducing > >>>>> the timing duration of the critical path signal which will now fit to one > >>>>> L3 clock cycle. > >>>>> o In case of CORE is at OPP100 (L3=200MHz, VDD2=1.1375V): > >>>>> " Lower the frequency to 100MHz. > >>>>> " Keep the voltage as it is (1.1375V). > >>>>> > >>>>> o In case of CORE is at OPP50 (L3=100MHz, VDD2=0.93V): > >>>>> " Keep the frequency as it is (100MHz). > >>>>> " Increase the voltage to 1.1375V. > >>>>> > >>>>> 2. Enable DPLL3 Automatic mode. This will ensure proper transition to > >>>>> RETENTION or OFF mode. > >>>>> > >>>>> After CORE Voltage Domain Wakeup Transition from RETENTION or OFF mode: > >>>>> 1. Disable DPLL3 Automatic mode. > >>>>> 2. Restore previous DPLL3 M2 Frequency and CORE Voltage values. > >>>>> > >>>>> Also OSWR should not be attempted if DPLL3 has locked. This should be done as > >>>> part of OSWR patch series. > >>>>> > >>>>> Patch tested on 3430SDP and 3630 ZOOM3. > >>>>> > >>>> > >>>> Do you have a more accurate description of the bug? What is the defect ID? > >>>> > >>> Defect Id is i581. > >>>> The subject is about DPLL3 lock issue, and the description is all about > >>>> the transition to CORE RET or OFF and playing with voltage... and why > >>>> OSWR is affected as well? > >>>> I'm a little bit confused by that... > >>>> > >>>> Is this bug dependent of the target power state? What about INACTIVE? > >>> The root cause of the issue is that SDRC IDLEREQ is deasserted before DPLL3 has > >> locked. Because of this DLL may/may not lock based on Process Voltage Temperature > >> conditions. The bug can occur when DPLL3 automatic transition is enabled. So DPLL3 > >> automatic transition is disabled by default and it is enabled only when system is > >> entering ret/off state (to facilitate voltage scaling). So when system is entering ret/off > >> state, WA is applied (since DPLL3 autoidle is enabled, we can possibly hit the issue; > >> hence the WA) > >> > >> Thanks, but that still not explain why this WA is considered only for > >> device transition to RET or OFF. The DPLL3 can got to idle as soon as > >> the CORE is INACTIVE. How this case is handled? > > DPLL3 auto idle is disabled when Core is in ON or INACTIVE state. So DPLL3 cannot goto idle state. > We enable DPPL3 auto idle only when Core is about to enter RET or OFF state (otherwise VDD2 will not > be able to scale the voltage down). > > Why did you prevent that mode for INACTIVE? If I remember well, during > the early measurement we did on 3430, the DPLL consumption in locked > mode is around 5mw, so you can save a good amount of power for very low > load use cases. > > Did you try to enable that in CPUIdle to check the power saving you can > achieve with that mode? I partially agree here, definitely there will be savings of the order ~5mW if DPLL3 is idled in CORE INA state. However, if this done then WA will be applied in all these states increasing the latency(there will be case were l3 is held high by some usecase whilst this c-state is hit; then 200->100->200 MHz transition might cause some latency and eventually hampering the functionality of the usecase). -Girish > > Benoit > -- > To unsubscribe from this list: send the line "unsubscribe linux-omap" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
