On 9/18/19 7:38 AM, Yafang Shao wrote:
> A new perf script page-reclaim is introduced in this patch. This new script
> is used to report the page reclaim details. The possible usage of this
> script is as bellow,
> - identify latency spike caused by direct reclaim
> - whehter the latency spike is relevant with pageout
> - why is page reclaim requested, i.e. whether it is because of memory
> fragmentation
> - page reclaim efficiency
> etc
> In the future we may also enhance it to analyze the memcg reclaim.
>
> Bellow is how to use this script,
> # Record, one of the following
> $ perf record -e 'vmscan:mm_vmscan_*' ./workload
> $ perf script record page-reclaim
>
> # Report
> $ perf script report page-reclaim
>
> # Report per process latency
> $ perf script report page-reclaim -- -p
I tested it with global-dhp__pagereclaim-performance from mmtests and got what appears to be reasonable results and the output looks correct and useful. However I'm not a vm expert so I can't comment further. Hopefully someone on linux-mm can give more specific feedback.
There is one issue with Python3, see below. I didn't test with Python2.
>
> + @classmethod
> + def shrink_inactive(cls, pid, scanned, reclaimed, flags):
> + event = cls.events.get(pid)
> + if event and event.tracing():
> + # RECLAIM_WB_ANON 0x1
> + # RECLAIM_WB_FILE 0x2
> + _type = (flags & 0x2) >> 1
> + event.process_lru(lru[_type], scanned, reclaimed)
> +
> + @classmethod
> + def writepage(cls, pid, flags):
> + event = cls.events.get(pid)
> + if event and event.tracing():
> + # RECLAIM_WB_ANON 0x1
> + # RECLAIM_WB_FILE 0x2
> + # RECLAIM_WB_SYNC 0x4
> + # RECLAIM_WB_ASYNC 0x8
> + _type = (flags & 0x2) >> 1
> + _io = (flags & 0x4) >> 2
> +
> + event.process_writepage(lru[_type], sync_io[_io])
> +
> + @classmethod
Space indentation on line above. For python3 this results in:
File "tools/perf/scripts/python/page-reclaim.py", line 217
@classmethod
^
TabError: inconsistent use of tabs and spaces in indentation
> + def iterate_proc(cls):
> + if show_opt != Show.DEFAULT:
> + print("\nPer process latency (ms):")
> + print_proc_latency(latency_metric, 'pid', '[comm]')
> +
> + if show_opt == Show.VERBOSE:
> + print("%20s %s" % ('timestamp','latency(ns)'))
Thanks
Tony
