#systemtap on 2020-10-20 — irc logs at freenode.irclog.whitequark.org

2015-11-12 23:18 fche changed the topic of #systemtap to: http://sourceware.org/systemtap; email systemtap@sourceware.org if answers here not timely, conversations may be logged

00:00 derek0883 has quit [Ping timeout: 246 seconds]

00:21 derek0883 has joined #systemtap

00:23 mjw has quit [Quit: Leaving]

00:24 derek0883 has quit [Remote host closed the connection]

00:25 derek0883 has joined #systemtap

00:31 derek0883 has quit [Remote host closed the connection]

00:31 derek0883 has joined #systemtap

01:04 derek0883 has quit [Remote host closed the connection]

01:05 derek0883 has joined #systemtap

01:08 orivej has quit [Ping timeout: 256 seconds]

01:10 derek0883 has quit [Ping timeout: 256 seconds]

01:20 hpt has joined #systemtap

01:28 sscox has joined #systemtap

01:37 orivej has joined #systemtap

01:46 orivej has quit [Remote host closed the connection]

01:47 orivej has joined #systemtap

01:59 derek0883 has joined #systemtap

02:01 orivej has quit [Ping timeout: 256 seconds]

02:06 orivej has joined #systemtap

02:27 derek0883 has quit [Remote host closed the connection]

02:30 orivej has quit [Read error: Connection reset by peer]

02:31 orivej has joined #systemtap

02:34 derek0883 has joined #systemtap

02:39 orivej has quit [Quit: No Ping reply in 180 seconds.]

02:41 orivej has joined #systemtap

03:59 orivej has quit [Ping timeout: 260 seconds]

06:33 derek0883 has quit [Remote host closed the connection]

07:02 khaled has joined #systemtap

08:11 orivej has joined #systemtap

10:58 hpt has quit [Ping timeout: 256 seconds]

11:20 irker694 has quit [Quit: transmission timeout]

11:25 przemoc has joined #systemtap

11:25 przemoc has quit [Changing host]

11:25 przemoc has joined #systemtap

11:36 mjw has joined #systemtap

13:40 derek0883 has joined #systemtap

13:46 derek0883 has quit [Ping timeout: 260 seconds]

13:56 amerey has joined #systemtap

14:33 tromey has joined #systemtap

14:48 sscox has quit [Ping timeout: 244 seconds]

15:38 <kerneltoast> fche, good morning

15:38 <kerneltoast> i hope you slept well

15:45 derek0883 has joined #systemtap

15:53 orivej has quit [Ping timeout: 258 seconds]

16:00 <agentzh> fche: it seems __stp_tf_vma_map has inherient hash conflicts involved with many different entries for all the DSO deps for every process.

16:00 <agentzh> but that hash table is only indexed by the tasks.

16:02 <agentzh> frequent func calls to _stp_umodule_relocate() would always iterate through the heavily conflicted hash table bucket linearly to search for a hit.

16:03 <agentzh> i wonder if we could introduce a secondary hash table for each process in that hash table. the 2nd level hash table can use the ->user field (which is the _stp_module pointer value) as the keys.

16:05 <agentzh> maybe there're better ways to optimize this.

16:36 derek0883 has quit [Remote host closed the connection]

16:36 derek0883 has joined #systemtap

16:59 sscox has joined #systemtap

17:02 derek0883 has quit [Remote host closed the connection]

17:02 derek0883 has joined #systemtap

17:07 derek0883 has quit [Ping timeout: 260 seconds]

17:08 derek0883 has joined #systemtap

17:58 <fche> hi

17:59 <fche> what are we using as the lookup/hash key right now? pid?

18:03 <kerneltoast> fche, pid yeah

18:04 <kerneltoast> I changed it to the task struct pointer in my patch

18:04 <kerneltoast> since pids are not a reliable anchor to a task struct

18:05 <kerneltoast> fche, i ran the testsuite with and without my patch and collected the logs for you

18:05 <fche> the diff over the .sum file is what one'd start the analysis with there

18:05 <kerneltoast> they are attached to my gist (scroll to the bottom): https://gist.github.com/kerneltoast/8a9518f7f17b8fc1613be69a5d0719ff

18:06 <kerneltoast> Yeah it has the .sum file

18:06 <kerneltoast> I just have no idea what to look for diffing it myself, since I'm not sure which failures are normal

18:10 <kerneltoast> i tested with a 5.8 kernel btw

18:14 <kerneltoast> fche, here is the diff: https://gist.github.com/kerneltoast/8a9518f7f17b8fc1613be69a5d0719ff#file-sum-diff

18:34 <fche> ok, those diffs are mild, good to see

18:34 <fche> I'd look next at those test cases that have any conceivable relationship to the piece you modified

18:34 <fche> so something like things related to process probes

18:35 <fche> check dmesg for any diagnostics

19:17 orivej has joined #systemtap

19:49 mjw has quit [Quit: Leaving]

20:21 derek0883 has quit [Remote host closed the connection]

20:40 derek0883 has joined #systemtap

20:43 derek0883 has quit [Remote host closed the connection]

20:43 derek0883 has joined #systemtap

20:46 tromey has quit [Quit: ERC (IRC client for Emacs 27.1.50)]

21:00 root_ has joined #systemtap

21:00 <root_> if some kernel code has a global variable, how can i print it out in systemtap

21:00 root_ is now known as Guest10353

21:01 <Guest10353> e.g. a kernel module has its file module.c, in this file you have a globally defined variable

21:01 <Guest10353> my program will run some code, which has access to this variable

21:03 derek0883 has quit [Remote host closed the connection]

21:04 derek0883 has joined #systemtap

21:04 <fche> try @var("foo")

21:04 <fche> or @var("foo", "kernel") I think

21:04 <fche> though if the variable is in scope at your probe point, then just $var

21:05 <Guest10353> okay

21:05 derek0883 has quit [Read error: Connection reset by peer]

21:05 derek0883 has joined #systemtap

21:08 <Guest10353> unable to find global

21:09 <Guest10353> its basically a variable in a c file which is global, static int var;

21:10 derek0883 has quit [Remote host closed the connection]

21:11 derek0883 has joined #systemtap

21:15 <Guest10353> i got it working

21:15 Guest10353 has quit [Quit: Lost terminal]

21:36 derek0883 has quit [Ping timeout: 272 seconds]

21:46 derek0883 has joined #systemtap

21:51 <fche> ah good, how?

21:56 <kerneltoast> fche, it doesn't look like any of the failed test cases use process probes

21:57 <fche> and nothing alarming in dmesg / lockdep?

22:01 <kerneltoast> fche, dmesg is clean

22:02 <kerneltoast> i didn't test with lockdep

22:02 <kerneltoast> i figured the massive testsuite would be sufficient instead

22:03 khaled has quit [Ping timeout: 264 seconds]

22:07 khaled has joined #systemtap

22:15 <kerneltoast> fche, how comfortable do you feel about the patch?

22:15 <kerneltoast> i can throw in another cookie of course

22:31 derek0883 has quit [Remote host closed the connection]

22:39 amerey has quit [Quit: Leaving]

22:39 <fche> heh

22:40 <fche> I'd be okay in throwing it into the master branch in a tentative basis, let our buildbots chew over it

22:40 derek0883 has joined #systemtap

22:40 <fche> in the worst case we revert and look again

22:41 <fche> but I can't think of a good reason not to take this, anyone else?

22:45 derek0883 has quit [Remote host closed the connection]

22:46 <kerneltoast> it's good rcu magic

22:46 derek0883 has joined #systemtap

22:46 <kerneltoast> can't get more micro-optimized than that

22:48 <fche> hahah

22:48 <fche> ok, go for it.

22:48 <fche> if the performance effects may be user noticeable, please add a sentence blurb to the NEWS as to what people may observe

22:49 <kerneltoast> agentzh has a better grasp of where the performance difference will be noticed

22:52 derek0883 has quit [Ping timeout: 260 seconds]

22:53 <kerneltoast> fche, it did make the testsuite run faster fwiw

22:53 <fche> ah interesting, how much?

22:53 <kerneltoast> it took 40 minutes instead of 41.5 minutes

22:54 <kerneltoast> when run in parallel

22:54 <fche> 2% might just be noise but whatever

22:54 <fche> at least it's not -10% :)

22:54 <kerneltoast> it was consistent

22:54 <kerneltoast> i ran it twice

22:54 <fche> interesting

22:54 <kerneltoast> on a system without cpu scaling inconsistencies

22:54 <fche> spherical cow?

22:54 derek0883 has joined #systemtap

22:55 <kerneltoast> hah

22:55 <kerneltoast> no it's a ryzen laptop with some msr twiddling

22:56 <kerneltoast> so the cpu frequency gets pinned across all cores, plus or minus 50 MHz

22:57 <kerneltoast> the performance gain for the specific workload that stresses that file should be much greater than 2%

23:00 derek0883 has quit [Remote host closed the connection]

23:00 derek0883 has joined #systemtap

23:21 derek0883 has quit [Ping timeout: 246 seconds]

23:29 derek0883 has joined #systemtap

23:32 derek0883 has quit [Remote host closed the connection]

23:33 derek0883 has joined #systemtap

23:45 khaled has quit [Quit: Konversation terminated!]

23:46 khaled has joined #systemtap

23:47 khaled has quit [Remote host closed the connection]

23:49 khaled has joined #systemtap

23:49 khaled has quit [Client Quit]

23:50 khaled has joined #systemtap

23:51 khaled has quit [Remote host closed the connection]

23:53 khaled has joined #systemtap