On 8/18/20 4:10 AM, Maxime Ripard wrote: > Hi! > > On Sat, Aug 08, 2020 at 09:18:22PM -0500, Samuel Holland wrote: >> When possible, system firmware on 64-bit Allwinner platforms disables >> OSC24M during system suspend. Since this oscillator is the clock source >> for the ARM architectural timer, this causes the timer to stop counting. >> Therefore, the ARM architectural timer must not be marked as NONSTOP on >> these platforms, or the time will be wrong after system resume. >> >> Adding the arm,no-tick-in-suspend property forces the kernel to ignore >> the ARM architectural timer when calculating sleeptime; it falls back to >> reading the RTC. Note that this only affects deep suspend, not s2idle. >> >> Signed-off-by: Samuel Holland <samuel@xxxxxxxxxxxx> > > Applied, thanks! > > I assume it affects all the SoCs with a Cortex-A7 as well? Yes, they all have the same ability to turn off OSC24M. Though they don't yet have support for deep sleep. Supposedly they can also reparent OSC24M from the crystal to IOSC, but I have not got that to work, and IOSC belongs nowhere near timekeeping anyway (20%+ frequency error). Ideally, we would run some MMIO counter off of LOSC during suspend. This would be several orders of magnitude more accurate than the RTC for accounting sleeptime. However, none of the basic timer blocks appear to work when OSC24M is disabled and AHB1 is running off of LOSC; they count for about 10 cycles and then stop. The HSTIMER is the only timer block that I got working. It runs at the same frequency as AHB1, so it would only be useful for timekeeping if we reparented AHB1 to LOSC during Linux's suspend process before switching clock sources. I doubt that is workable. So the RTC is the best solution I know of for now. > Maxime Samuel
