15:00 <lorddoskias> is it possible to silence systemtap when it finds module version mismatches (due to multiple compilations)?

15:04 <fche> lorddoskias, like the buildid mismatch warnings?

15:05 <lorddoskias> more like: ERROR: module version mismatch (#49 SMP Wed Feb 10 15:37:04 EET 2016 vs #48 SMP Wed Feb 10 12:02:37 EET 2016), release 3.12.51-clouder3

15:05 <lorddoskias> so yeah, the build number is incremented

15:06 <fche> we try to be very careful about assuming binary compatibility between kernel headers and running binaries

15:06 <fche> if we're running on a different version of the kernel, we have no guarantee that the same kernel configuration is used e.g.

15:07 <lorddoskias> yeah, but is there a way to override this?

15:07 <lorddoskias> a la guru mode?

15:08 <lorddoskias> (tried it to no avail)

15:08 <fche> hm, perhaps we could relax that check with -g mode, or perhaps a -D flag (maybe there is one already)

15:08 <lorddoskias> because what i did now is modify the source of the module and recompile, which naturally bumped up the build number but fundamentally nothing changed on the core kernel side

15:11 <fche> there is a series of checks that stap emits, not just that one, and there's currently no knob to override them

15:12 <fche> see translate.cxx c_unparser::emit_module_init

15:12 <lorddoskias> i as jsut reading this file

15:12 <fche> there is a -DSTP_NO_BUILDID_CHECK that disables build-id hash checking

15:12 <fche> we could consider another -DSTP_NO_VERREL_CHECK or something to disable the other stuff in translate.cxx

15:13 <lorddoskias> so i'm reading this line: o->newline(1) << "_stp_error (\"module release mismatch (%s vs %s)\", "

15:13 <lorddoskias> << "release, "

15:13 <lorddoskias> << lex_cast_qstring (session->kernel_release)

15:13 <lorddoskias> << ");";

15:13 <fche> it'd be only a few lines of code.

15:13 <lorddoskias> but doesn't this output the actual error

15:13 <lorddoskias> where does the error checking happen?

15:13 <lorddoskias> o is the translator output

15:14 <lorddoskias> aaaaah, i se this genereates code, ok :D

15:14 <fche> yup

15:14 <fche> so it could generate a #ifndef STP_NO_VERREL_CHECK and matching #endif into the output stream

15:14 <lorddoskias> yep

15:15 <lorddoskias> let my try and do that :)

15:18 <lorddoskias> fche: but would such a define be accessible from within translate.cxx )

15:18 <lorddoskias> ebcause the buildid check one is used in a different file which is evaluated at runtime or something?

15:18 <fche> you emit the #ifndef as a string

15:18 <lorddoskias> right

15:19 <fche> then when you -run- the resulting stap binary, use stap -DSTP_NO_VERREL_CHECK ... to disable the wrapped code

15:19 <lorddoskias> yep, i see

15:22 <fche> (one little complication can be that the emitted code should be -Wall -clean, with or without that -D activated)

15:24 <lorddoskias> http://pastebin.com/GLi27z0v

15:24 <fche> you might want to move that a little wider

15:24 <fche> suppress the ->version check just afterwards

15:24 <fche> and some other stuff just before IIRC

15:25 <lorddoskias> okay, because with this one it didn't work

15:25 <lorddoskias> though the check which emits that particular error message ought to be disabled

15:25 <fche> module release vs. module version !!

15:25 <fche> anyway I'd put the #ifndef line at translate.cxx line ... 1731ish

15:25 <fche> and the #endif around ... 1765ish

15:26 <fche> hm make that 1730 for the former, outside the { }

15:26 <fche> dunno, whatever works syntactically, hiding the const char * release / version decls too

15:26 <lorddoskias> yeah

15:26 <lorddoskias> put it exactly on those lines just before you said it :D

15:28 <lorddoskias> In function ‘systemtap_module_init’:

15:28 <lorddoskias> /tmp/stapJDDurI/stap_125b0a3c53515242c8448b419e9ba101_1351_src.c:508: error: no return statement in function returning non-void

15:28 <lorddoskias> hmz, commenting the wholoe code block just breaks it

15:28 <lorddoskias> and it seems systemtap doesn't preserve the tmp files

15:29 <lorddoskias> okay, testing with -k and then examining the code :)

15:29 <fche> (or stap -p3)

15:29 <fche> you'll have this working in no time

15:31 <lorddoskias> hm, the generated code for systemtap_kernel_module_init does look really broken :D http://pastebin.com/Ssv8qAnU

15:33 <fche> what a lovely no-op

15:33 <lorddoskias> okay, did it :)

15:33 <lorddoskias> so the verrel dif check has to be inside the code block :D

15:36 <lorddoskias> fche: so if i send a patch would it be accepted :D ?

15:37 <fche> why not.

15:37 <fche> please make sure it's tested with and without that -D flag

15:38 <fche> (an explicit test suite item would be wonderful too)

15:39 <fche> (btw curious, for git commit name purposes, would you prefer Lord or Nikolay ?)

15:41 <lorddoskias> Nikolay :D

15:42 <lorddoskias> i still haven't had the chance to test the security scenario we discussed at FOSDEM

15:42 <lorddoskias> ;\

15:43 <lorddoskias> but i think this won't be an issue because when stap is running in recorder mode, e.g. someone with a client can connect to an already running probe this would require access to the descriptor object on the hdd, right?

15:43 <lorddoskias> (i have only skimmed the flight recorder documentation)

16:05 <fche> yeah, the file descriptors are under /proc/systemtap somewhere

16:12 <lorddoskias> hm, i see no /proc/systemtap though when i have run a module as flight recorder

16:14 <fche> could be /sys/kernel/debug/systemtap/ instead

16:15 <lorddoskias> yep, it's there :)

16:15 <lorddoskias> thanks

16:16 <lorddoskias> righto, so if anyone wants to connect they need to have access to the debugfs in order to acquire the descriptor?

16:19 <fche> and the kernel's sysfs file permissions widgetry would have to let them in

18:48 <irker366> systemtap: mcermak systemtap.git:refs/heads/master * release-2.9-274-g24ef0e8 / : Add more czech localized manpages. http://tinyurl.com/gqt5w9x

18:48 <irker366> systemtap: mcermak systemtap.git:refs/heads/master * release-2.9-275-g7173001 / configure doc/Makefile.in java/Makefile.in man/cs/Makefile.in: Autoreconf following commit 24ef0e886 (automake-1.15-4.fc23). http://tinyurl.com/hmkxwag

18:50 <irker366> systemtap: dsmith systemtap.git:refs/heads/master * release-2.9-274-gde6d99c / testsuite/systemtap.examples/memory/hw_watch_addr.meta testsuite/systemtap.examples/memory/hw_watch_addr.tcl testsuite/systemtap.examples/memory/hw_watch_sym.meta testsuite/systemtap.examples/memory/hw_watch_sym.tcl: Better test for hardware breakpoint support in the examples. http://tinyurl.com/hgo5lv3

18:50 <irker366> systemtap: dsmith systemtap.git:refs/heads/master * release-2.9-277-gadbeb02 / : Merge branch 'master' of ssh://sourceware.org/git/systemtap http://tinyurl.com/jmosjjz

19:07 <irker366> systemtap: dsmith systemtap.git:refs/heads/master * release-2.9-278-gdc6bf1e / staprun/relay.c: Fix PR19560 by improving the relay code. http://tinyurl.com/zegz7ok