00:56 <irker474> systemtap: yichun systemtap.git:master * release-4.2-65-g6de9136a2 / tapset/uconversions.stp: uconversions.stp: worked around kernel's pointer value hashing.

19:30 <agentzh> fche: it'll be great if you can have a look at this: https://sourceware.org/bugzilla/show_bug.cgi?id=25851

19:30 <agentzh> not sure how to debug this.

19:30 <agentzh> thanks!

19:30 <fche> did you see the [man error::dwarf] man page?

19:31 <fche> sometimes these things are out of our hands

19:31 <fche> i.e., mere tapset magic sometimes cannot workaround compiler optimization artifacts

19:31 <fche> sometimes it can

19:33 <agentzh> yeah, i was aware of that. wondering if it'll be better to simply use the register directly for specific arch.

19:33 <fche> the point is the register is not identified in dwarf

19:33 <agentzh> not sure if vfs_write itself will be inlined at all.

19:33 <fche> so we don't know which one (or where else)

19:34 <agentzh> sure. i'm talking about the tapset probe vfs.write. not dwarf.

19:36 <agentzh> i also ran into something similar while probing some glibc functions. using the x64 abi regs solved the problem.

19:37 <fche> when you probe at abi function entry points, the parameters should be available

19:37 <fche> but an inlined function cannot be an abi function entry point :)

19:42 <agentzh> yes, i was aware of that.

19:43 <agentzh> also, what do you think of this issue? https://sourceware.org/bugzilla/show_bug.cgi?id=25852

19:43 <fche> hm yeah, that variable should be in the context

19:44 <fche> may find other similar things in the context already, conditionalized on some #ifdef kind of thing

19:52 <agentzh> okay, i will make the change.

19:52 <agentzh> also, i filed this PR to record the print_backtrace() inaccuracy i reported earlier: https://sourceware.org/bugzilla/show_bug.cgi?id=25853

19:52 <agentzh> this time, with a oneliner to demonstrate the issue.