| |||||||||||||
A big.LITTLE scheduler updateA big.LITTLE scheduler updatePosted Jun 13, 2012 19:32 UTC (Wed) by charlesgt_arm (subscriber, #83016)Parent article: A big.LITTLE scheduler update
Regarding simulation of b/L in symmetric. Would another option be to use hrtimers? I am not that well versed there, but initial reading suggested to me that you can run them per core, and in interrupt context. If the latter is true (former helps but not essential) then it should be fairly trivial to create a driver which interrupts the cores at given frequency and then spins in the callbacks for a period of time. Ugly, but then all cycle stealing is. The driver could offer a simple interface to select period and duty cycle (spin time). The solution (assuming hrtimer is generally supported) would be generic
Cheers
Charles
(Log in to post comments)
A big.LITTLE scheduler update Posted Jun 13, 2012 23:38 UTC (Wed) by PaulMcKenney (subscriber, #9624) [Link]
Sounds like another potentially reasonable approach to me!
A big.LITTLE scheduler update Posted Jun 20, 2012 18:08 UTC (Wed) by olivier_cozette (subscriber, #84087) [Link]
The main reason we choose perf is that there is nothing to do, no kernel change, nothing new to write in user mode.
We just need to run the user mode perf on every CPU that should be a LITTLE one and that's all!
Cheers,
A big.LITTLE scheduler update Posted Jun 21, 2012 12:30 UTC (Thu) by pantoniou (subscriber, #85257) [Link]
Please take a look at this:
[PATCH] [CPUFREQ] VCPUfreq: Virtual CPU frequency driver
http://lists.linaro.org/pipermail/linaro-dev/2012-June/01...
I did try to use hrtimers at first but the API to do what was needed seems to be missing.
Anyway, using the omap timers was relatively easy, and I'm sure most recent SoCs have a multitude
|
|||||||||||||