Hi, Jack On 10/7/10, Jack Daniel <wanders.thi...@gmail.com> wrote: > Hi Wu, > > I was going through your latency tracing implementation for loongson. > Could you please assist me in understanding the following. > > 1) How is the latency tracing implemented here different from ftrace > tracer function "irqsoff" ? "irqsoff" calculates the time when > interrupts are disabled.
What do you mean "the latency tracing implemented here"? In reality, the latency tracers I have used are the ones you said above: irqsoff tracer, preemptoff tracer and preemptirqsoff tracer, wakeup tracer ... But at the time I was using them, they didn't work well for they could not provide high resolution timestamp: The unit of the timestamp resolution is only ms, so, the latency result is always >= 1000 us, which can not reflect the real situation of the latencies. Therefore, I have tried to fix it via adding the high resolution sched_clock() support. > > 2) Could you give me a reference or guide on how to implement the same > for another mips platform such as Octeon? Which kernel version are you using? For the latest kernel versions, the latency tracers should be supported very well except the high resolution sched_clock(). For Octeon, the support of high resolution sched_clock() has already been there from 2.6.33, but for the r4k MIPS, you may need to apply the patches from the patchwork of linux-mips: http://patchwork.linux-mips.org/patch/1102/ http://patchwork.linux-mips.org/patch/1104/ Regards, Wu Zhangjin -- You received this message because you are subscribed to the Google Groups "loongson-dev" group. To post to this group, send email to loongson-...@googlegroups.com. To unsubscribe from this group, send email to loongson-dev+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/loongson-dev?hl=en.
