18:18 <agentzh> fche: never mind, we did more stress testing, and enabling STP_UTRACE_USE_TASK_WORK_QUEUE would lead to kernel lockup (100% CPU).

18:19 <agentzh> seems like we need another way to fix the process.begin vs vma map callback running order.

19:44 <agentzh> fche: maybe combine the vma map callback with the process.begin callback to form a single utrace engine callback?

19:44 <agentzh> so that we can always make sure they run in the correct order when attached to the target process?

20:10 <agentzh> fche: it'll be great if you can provide some suggestions on this one: https://sourceware.org/bugzilla/show_bug.cgi?id=26397

20:12 <fche> hmmmm

20:12 <agentzh> it took me a lot of time debugging this and finally realizing they are no longer cputime_t...

20:13 <fche> I'd work by identifying the kernel commit that changed the type of those fields

20:13 <agentzh> the results are hugly larger.

20:13 <fche> if it's a declaration change, a runtime/autoconf*.c file may be able to tell the before & after apart (is sizeof() different? signedness?)

20:13 <fche> if it's the same compile-time data type, then maybe just a KERNEL_VERSION based test in the tapset/embedded-c code could adjust scaling accordingly

20:14 <agentzh> okay, i'll dig it up.

20:14 <agentzh> thanks for the suggestion.

20:15 <fche> righto

20:16 <agentzh> i found it was introduced by commit 5613fda9a503c in mainline kernel.

20:16 <agentzh> not sure if redhat would backport it to its older kernels.

20:16 <agentzh> or has backported

20:16 <fche> might happen, so that's why autoconf* based tests are used

20:17 <agentzh> i don't see any data type changes in that commit. alas.

20:17 <agentzh> it was introduced in kernel 4.10.

20:18 <agentzh> v4.10-rc6-79-g5613fda9a503

20:19 <fche> cputime_t -> u64

20:20 <agentzh> so we declare this function again in the autotest program?

20:20 <fche> not necessarily

20:20 <fche> but -some- function that is compilable iff the kernel commit is in effect

20:21 <fche> (or not compilable iff ...) -- either way

20:22 <agentzh> hmm

20:23 <agentzh> i'll try it out

20:26 <agentzh> i remember stap's autoconf uses -Wall to compile so it should be easier.

20:26 <fche> yup

20:26 <agentzh> -Wall and -Werr

20:26 <agentzh> *-Werror

20:28 <fche> but yeah cputime_t itself has gone away some years ago generally

20:28 <fche> but anyway see what you can come up with

20:29 <fche> and it seems it was frequently just a typedef to u64 anyway

22:11 <agentzh> yeah