Compare dmprof GLOBAL_STATS with sum of /proc/pid/smaps

[WenSheng He]

In order to get the /proc files at the point of dmprof dump, I modify the DumpOrderedProfile function to backup them using system("cp command").

Then I write shell script to sum different groups of VSS and RSS, and compare with the GLOBAL_STATS. Details as below:

File	VSS			RSS			Formula for smaps
	smaps	dmprof	Diff	smaps	dmprof	Diff
file-exec	139920	139920	=	39428	39428	=	File mapping with x flag
file-nonexec	260140	260140	=	25504	25504	=	File mapping without x flag
anonymous	148860	148860	=	107568	107572	↑4[HWS1]	Mapping with empty Pathname
stack	136	132	↓4[HWS2]	76	76	=	Main thread stack
other	215220	215216	↓4[HWS3]	10608	10608	=	Thread stack, [vdso], [vsyscall]
total	764276	764268	↓8	183184	183188	↑4	VSS = sum of Size RSS = sum of Rss

---

 [HWS1]Maybe Reasonable wave? I check other snapshots, some have the same problem, some don't.

07667000-076db000 rw-p 00000000 00:00 0

 [HWS2]It’s really unbelievable that the smaps mapping section is same as the MMAP_LIST in dmprof, and minus result should be 132KB, why does smaps show Size as 136KB? The arithmetic  of Linux is wrong? Next I should check the kernel code.

7fffd4bd5000-7fffd4bf6000 rw-p 00000000 00:00 0                          [stack]

 [HWS3]dmprof didn’t account for [vdso]

[Dai Mikurube]

Such a 4KB ( == 1 page) difference often happens. It's very hard to get all data at the exact same time. We should accept such a small difference.

Also, small differences may happen because of limitation of API/procfs and difference of definition. Do you really need such an "exact" value? I think it's an unreasonably hard yak-shaving. I don't recommend to dive into it.

--
You received this message because you are subscribed to the Google Groups "dmprof" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dmprof+un...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/dmprof/5b13af64-a3e6-490d-9715-dff8d24302e2%40chromium.org.

--

Dai MIKURUBE

   dmik...@chromium.org

[WenSheng He]

Thank you for your advice, I accept.

在 2014年8月4日星期一UTC+8下午5时12分30秒，Dai Mikurube写道：