Hi there,
When I am trying to use ftrace to measure cpuidle latency, I find that the timestamp is always ending with 0000, like below.
bash-2.05b# head trace
# tracer: nop
#
# TASK-PID CPU# TIMESTAMP FUNCTION
# | | | | |
<idle>-0 [000] 79.530000: cpu_idle: state=0 cpu_id=0
<idle>-0 [000] 79.530000: cpu_idle: state=4294967295 cpu_id=0 <idle>-0 [000] 79.530000: cpu_idle: state=0 cpu_id=0
<idle>-0 [000] 79.530000: cpu_idle: state=4294967295 cpu_id=0 <idle>-0 [000] 79.530000: cpu_idle: state=0 cpu_id=0
<idle>-0 [000] 79.530000: cpu_idle: state=4294967295 cpu_id=0
But it is supposed to be more accurate like:
<idle>-0 [000] 270.396210: cpu_idle: state=4294967295 cpu_id=0 <idle>-0 [000] 270.422424: cpu_idle: state=0 cpu_id=0 <idle>-0 [000] 270.427124: cpu_idle: state=4294967295 cpu_id=0 <idle>-0 [000] 270.427185: cpu_idle: state=2 cpu_id=0
Did I miss something in kernel configuration or some certain features?
Thanks Yong
Hello,
When I am trying to use ftrace to measure cpuidle latency, I find that the timestamp is always ending with 0000, like below. <idle>-0 [000] 79.530000: cpu_idle: state=0 cpu_id=0
What platform and kernel is it? 2.6.35 and Versatile Express by any chance?
Cheers!
Paweł
On Mon, Feb 21, 2011 at 6:31 PM, Pawel Moll pawel.moll@arm.com wrote:
Hello,
When I am trying to use ftrace to measure cpuidle latency, I find that the timestamp is always ending with 0000, like below. <idle>-0 [000] 79.530000: cpu_idle: state=0 cpu_id=0
What platform and kernel is it? 2.6.35 and Versatile Express by any chance?
2.6.38 on imx51 babbage board.
Thanks Yong
Cheers!
Paweł
2.6.38 on imx51 babbage board.
Then I won't be able to provide you with a simple hack, sorry...
Generally looks like the sched_clock implementation is missing so the ftrace is reverting to jiffies as the timestamp source (this was the case with mach-vexpress with kernels < 2.6.37)
Good luck!
Paweł