Hi Mike, On Thu, Mar 10, 2016 at 3:01 AM, Mike Looijmans <mike.looijm...@topic.nl> wrote: > On 10-03-16 11:17, Nathan Rossi wrote: >> >> On Thu, Mar 10, 2016 at 7:29 PM, Mike Looijmans <mike.looijm...@topic.nl> >> wrote: >>> >>> After migrating to the current 4.4 kernel, I noticed that the internal >>> timer >>> no longer works correctly. >>> >>> When the CPU frequency scaling sets the CPU speed to 333MHz, the "wall >>> clock" time also slows down to half the normal rate (e.g. a "sleep 1" >>> will >>> actually take 2 seconds). >>> >>> This wasn't the case on the 4.0 kernels. >> >> >> Is this only on linux-xlnx kernels? or linux-yocto and or mainline kernels >> too? > > > It's quite easy to test on ANY kernel, just set the CPU to 333MHz, > typically: > > echo 333333 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed > > and then check how long "sleep 1" really takes. > > (Most boards default to "userspace" governor, setting the governor to > "ondemand" will also reduce the speed)
This is a known issue with Zynq. Stuff also horribly breaks if you run speedgrade 3 devices with cpufreq, as the prescaler hack goes horribly wrong for these. The issue is that the cadence ttc's clock depends on the cpu clock ... which changes ... I just turn off cpufreq on my Zynq targets since it breaks the timers for non speedgrade 1 devices and use the arm global timer instead. Cheers, Moritz -- _______________________________________________ meta-xilinx mailing list meta-xilinx@yoctoproject.org https://lists.yoctoproject.org/listinfo/meta-xilinx