Am Freitag 13 Juni 2008 23:16:05 schrieb Andy Green: > The following series gives me 240ms resume performance to video and > userspace up; resume feels very snappy indeed now.
Great. Can't wait to test. > I didn't like the need to hold LCM in reset during suspend so I > changed it to leave LCM powered in deep sleep mode. Does this have any negative impact on power consumption during suspend? > It also fixes a resume order dependency issue with glamo mci wanting > to set power state of SD Card and therefore needing pcf50633 up. > > pcf50633 interrupt workqueue is made to enforce resume state before > it is willing to stop deferring, the weird bugs around this went away > with a timeout on the glamo mci stuff above and subsequent fixing of it. > > Individual I2C transactions in pcf50633 get wholesale moved to > single bulk autoincrenment transactions, the code is WAY simpler > after this is done to RTC as well and it provides major wins in > speed during suspend / resume. > > Current limit notifications from USB stack got migrated into pcf50633 > where they belonged in the first place, and made robust against coming > in during resume at bad times. > > There is still at least one serious issue hiding in suspend / resume > racing -- I can kill resume stone dead by just adding an mdelay() in > Glamo suspend code, it should have zero effect. > > So it's very interesting to hear any results from anyone else trying > these patches (which are on andy branch now). Will do later today. :M:
