10:11 <lorddoskias> does systemtap support the ORC unwinder

10:11 <lorddoskias> or should i be using FP ?

11:45 <mjw> ehe?

11:45 <mjw> Pass 5: run failed. [man error::pass5]

11:45 <mjw> WARNING: /home/mark/build/systemtap-obj/staprun/run-staprun exited with status: 127

11:45 <mjw> /var/tmp/scl7OqDQa: line 8: -P: command not found

11:48 <mjw> ok, ignore that, apparently I configured in srcdir by accident causing weirdness in the generated files.

11:53 <mjw> Or something... how odd. I keep getting that. But that /var/tmp file disappears before I can inspect it...

11:54 <mjw> If only we had a tool to catch processes creating files... :)

12:17 <mjw> gah, no idea what this is. It only happens if I do an in-build-tree ./run-stap, if I install it and run stap from the install dir it works fine.

12:17 <mjw> weird

13:33 <irker788> systemtap: mark systemtap.git:refs/heads/master * release-3.2-178-g17ee540 / runtime/unwind.c: Make kernel DWARF unwinder work with ksalr. http://tinyurl.com/y9dhu5px

13:33 <mjw> pfff, that took all day. For a oneliner... embarrassing.

13:53 <fche> ship it

13:54 <fche> why is that specific to the kernel btw?

13:54 <fche> I'd imagine the exact same relocation calculation is appropriate for module/sections too

13:54 <fche> mjw^

13:54 <mjw> fche, because the kernel is "special" :)

13:55 <mjw> It is not really relocatable, but then it is, so it looks somewhat like an executable, but really is a shared object, except that it isn't.

13:55 <mjw> So for all other types we already do the right thing.

13:56 <fche> how do we do the right thing for modules already?

13:58 <mjw> fche, at the end of that same function.

13:59 <fche> ah. too bad we have all these offset-like things we must or must not add differently for each type of target

13:59 <mjw> Now that you say it, maybe we could fall-through. But I am a little afraid to do that now.

14:00 <mjw> I did test the module thingy. So it does work for that as it is now.

14:03 <mjw> fche, yeah, so for modules it is even weirder, because they can have multiple .text sections and the kernel itself acts as "dynamic loader" possibly laying out the sections randomly. For user space we don't need to keep track of sections at all, it is just one module.

14:03 <mjw> normally the linker takes care of all that. placing all hot and cold sections together, etc. determining the layout of the ELF module. But the kernel... it is special :)

17:10 <tonyj> What are the current systemtap "rules" on build-ids? Originally we had common build-ids for vmlinuz, vmlinux and vmlinux.debug. We broke this (vmlinuz was different) and started getting errors from stap like "RROR: Build-id mismatch: "kernel" vs. "vmlinux-3.0.34-0.7-default.debug" byte 0 (0x78 vs 0xa6) address 0xffffffff8144d128 rc 0"

17:11 <tonyj> We've managed to break it again (vmlinux and /sys/kernel/notes no longer match vmlinux/vmlinux.debug) but stap no longer seems to care.

17:11 <tonyj> I notice on Fedora27 that the /boot/vmlinuz and /boot/vmlinuz+debug do not have the same build-id.

17:37 <jistone> tonyj, I think vmlinuz+debug is a distinct kernel build, with various CONFIG_DEBUG options set

17:38 <jistone> (afaik stap still verifies build-ids, but I'm not directly involved these days)

17:53 <tonyj> jistone: I know before, if /sys/notes/kernel build-id didn't match vmlinux.debug stap would complain. it seems no longer, it doesn't match for us and it's not complaining, I'll have to look at the code.

21:59 <fche> tonyj, stap -DDEBUG_SYMBOLS should get a little deeper into what it's trying to do

22:00 <fche> but the buildids in question aren't fetched from /sys/notes per se but rather from dereferencing kernel-side addresses computed at -p3 time

22:00 <fche> but yeah it should still check

22:00 <fche> please test git stap, some kaslr-related patches just went in, and might be relevant

22:49 <irker990> systemtap: amerey systemtap.git:refs/heads/master * release-3.2-179-gfa4f499 / runtime/stp_string.c: PR22991: Fix invalid truncation of quoted strings http://tinyurl.com/y7cn7adr

22:58 <irker990> systemtap: amerey systemtap.git:refs/heads/master * release-3.2-180-gd0e7f51 / runtime/stp_string.c: runtime/stp_string.c: Remove unused variable. http://tinyurl.com/y9la2xcd