Current DVFS scale decision assumes voltage dependency - hence the scale decision is implemented based on voltage transition decision. This is wrong as the dependency between domains is at OPP tuple level, so no matter what the voltage level is, if we transition from a lower frequency to a higher frequency, we should scale dependent domain ahead of transitioning voltage/frequency for the current domain, and viceversa when switching to lower OPP. With Smartreflex AVS Class3 or Smartreflex AVS Class1.5, consider the following Scenario: The current OPP might be in Vnom (Vn) - if SR is disabled OR if SR has not yet made the first adjustment Vcurrent(Vc) - which can either be: Vsr_intermdiate(Vi) - if SR was interrupted in the middle of convergence Vsr(Vs) - if SR has completed adjustment. For all purposes Current OPP's Vn > Vc. However, when we consider the new OPP we are entering (assume for SR1.5, new OPP is not a calibrated opp in this discussion): Going down, we have the following cases: Case a) current_Vn > current_Vc > new_Vn Case b) current_Vn > new_Vn > current_Vc Going up, we could have: Case c) new_Vn > current_Vn > current_Vc OR we might even have a scenario case d) current_Vn > (current_Vc == new_Vn) Today's code compares current_Vc against new_Vn, and makes a decision when to scale dependent domain. This will work find in case (a) and case (c). However, in case (b), we scale dependent domain at the wrong time, in case of case (d), we dont even attempt to scale dependent domains!! This is wrong. since the comparable factor which closely represents the order of OPP is Vnom, the right thing to check for scaling dependent domains is Vnom. Hence use it instead. Change-Id: I93868743301e115fab8e33ff14317b07b81ef5d8 Signed-off-by: Nishanth Menon <nm@ti.com>