12:03 <mjw> When adding a new runtime/linux/autoconf-*.c file in buildrun, how do you "test" it?

12:03 <mjw> I can see it fails, because my build variable doesn't get defined, but why?

12:10 <mjw> hmmm, somehow you'll have to be able to recreate the module build environment

12:10 <mjw> but how?

12:12 <mjw> So, I do get a /tmp/stapxxx dir with -k that seems to have most build infra

12:12 <mjw> but how to invoke it so that I can see the output of CHECK_BUILD for my new autoconf file?

12:16 <mjw> right, so removing the /dev/null redirect and removing the generated stapconf_ header file I can see it fail when running make

12:17 <mjw> but just running make, everything fails

12:17 <mjw> that doesn't seem to really setup the kernel module build environment.

12:17 <mjw> hmmm

12:19 <mjw> any pointers how to setup this environment?

12:27 <mjw> Finally!

12:27 <mjw> /home/mark/src/systemtap/runtime/linux/autoconf-unwind-stack-trace.c:4:6: error: function declaration isn’t a prototype [-Werror=strict-prototypes]

12:27 <mjw> void unwind_stack_trace ()

12:27 <mjw> doh

12:30 <mjw> OK, so what you do is the completely cargo-culted: rm stapconf_*.h; make -C /lib/modules/`uname -r`/build M=`pwd` SRCDIR=`pwd` clean; make -C /lib/modules/`uname -r`/build M=`pwd` SRCDIR=`pwd`

12:30 <mjw> No idea why. But it gives me something I can debug

12:31 <mjw> (O, plus editing that Makefile to remove the /dev/null redirect above of course)

12:48 <mjw> O... now my branch is 2 months out of sync. woops.

12:49 <mjw> wow, it rebased cleanly. Nice!

12:49 <mjw> (Or maybe not, that would mean I am the only one every hacking on this part of the code...)

12:49 <fche> stap --poison-cache ?

12:50 <mjw> fche, what, I have a perfectly fine working tree, and you are poking the hornest nest? :)

12:51 <fche> poke poke

12:51 <fche> anyway that is the easiest way of forcing regeneration of auxiliary files like config headers etc.

12:51 <fche> during pass 2/4

12:51 <mjw> yes, thanks.

12:52 <mjw> I'll run some tests and then try again on a different setup with an older kernel.

12:52 <mjw> But for now I have a new kernel unwind fallback that seems to work nicely.

12:53 <fche> how does it do it?

12:53 <mjw> The DWARF unwinder is still broken, but at least the fallback seems ok. So we get at least some output. And if the kernel is build with frame pointers it actually seems good.

12:53 <fche> ok. is it only the newest kaslr that's preventing dwarf from working too?

12:54 <mjw> yes, I believe so. Still looking at why we don't get things correct.

12:54 <fche> great

12:55 <mjw> The DWARF unwinder is a little wonky in that it creates the unwind search table "offline" (but apparently already relocated against _stext), but then gets used at runtime (where _stext is gone somewhere else?)

12:55 <mjw> I don't fully understand it yet.

12:56 <mjw> So I thought I first get the fallback unwinder working, so we at least have something.

12:56 <fche> ok

12:56 <mjw> https://code.wildebeest.org/git/user/mjw/systemtap/commit/?h=kernel_unwind_fallback

12:56 <fche> yeah we need to relocate runtime addresses against _stext ... but I thought we had done that for years. something must have broken fairly recently

12:56 <mjw> First lunch, then testing that on some older setups.

12:56 <fche> thanks

12:58 <mjw> fche, we do, but like I said the unwind code is "weird" and the linux kernel itself is already relocatable. Apparently we relocate the unwind table addresses "too early" (when collecting them from disk, not at runtime).

12:58 <fche> we need to do both

12:58 <mjw> The unwind tables come from the vmlinux debug, not .eh_frame...

12:58 <fche> to subtract out the link-time base address

12:58 <fche> t

12:58 <fche> then add in the run-time base address

12:59 <mjw> right. something like that.

12:59 <mjw> But without some lunch in my belly I am not going to understand which base gets added/subtracted from which address where and when :)

13:07 <fche> (after lunch, see also what we do with any such address, like a kprobes probe, or a symbol table dump)

14:45 <irker885> systemtap: mark systemtap.git:refs/heads/master * release-3.2-171-g553b6df / buildrun.cxx runtime/linux/autoconf-unwind-stack-trace.c runtime/stack.c runtime/unwind.c runtime/unwind/unwind.h: linux runtime: Add support for new kernel unwind fallback. http://tinyurl.com/y7tjhr7p

15:27 <irker885> systemtap: dsmith systemtap.git:refs/heads/master * release-3.2-172-g1ba4be1 / httpd/docker/fedora_install_package.py: Fix httpd/docker/fedora_install_package.py to correctly handle file paths. http://tinyurl.com/y9c57fxj

15:43 <irker885> systemtap: smakarov systemtap.git:refs/heads/master * release-3.2-173-gdb9c9d6 / runtime/sym.c runtime/transport/symbols.c: RHBZ1566422 - fix spurious Build-id mismatch when probing reinserted kernel module http://tinyurl.com/yapvvzyf

16:04 <irker885> systemtap: dsmith systemtap.git:refs/heads/master * release-3.2-174-g2e9e800 / httpd/backends.cxx httpd/docker/centos.json httpd/docker/fedora.json: Change the http docker backend to use the docker image as a sysroot. http://tinyurl.com/yb3c72xx

16:51 <irker885> systemtap: mark systemtap.git:refs/heads/master * release-3.2-175-g0b7f8e4 / loc2stap.cxx: loc2stap.c: Give proper error message for unhandled DW_OP_GNU_parameter_ref. http://tinyurl.com/yckncntc

20:08 <irker885> systemtap: smakarov systemtap.git:refs/heads/master * release-3.2-176-gf579ca5 / elaborate.cxx staptree.cxx staptree.h testsuite/runok/regex_noelide.stp: Revert "PR22193: don't elide regex match '=~' operator since it's now side-effecting" http://tinyurl.com/ydclhuas

20:40 <irker885> systemtap: dsmith systemtap.git:refs/heads/master * release-3.2-177-gf9bca69 / testsuite/lib/http_server.exp: Document remote server testing in testsuite/lib/http_server.exp. http://tinyurl.com/ydbxybr4