10:51 <irker327> systemtap: alizhang systemtap.git:pr13838/fpsupport * release-4.4-43-g3f01e0af4 / runtime/loc2c-runtime.h: PR13838: update aarch64 pt_reg section, added macro definition for SIMD FP register V0-V7, (reg64-reg71).

20:45 <kerneltoast> fche, happy friday

20:45 <kerneltoast> now I'm here to make your Friday unhappy

20:46 <fche> MOOHAHAHAHAHAHAHAHAH

20:46 <fche> abuse of power

20:47 <fche> ok so where were we

20:48 <kerneltoast> yes, it was about your vehicle's extended warranty

20:48 <kerneltoast> the utrace gigapatch is ready to ship

20:48 <kerneltoast> i need to work on splitting it up now, but you may as well take a careful read through the big commit

20:49 <kerneltoast> https://gist.github.com/kerneltoast/43983b672946f3f1df3aadfbef45dafa

20:49 <kerneltoast> notable things:

20:49 <kerneltoast> utrace structs now pin their task structs

20:50 <kerneltoast> bandwagon rcu read locks are gone

20:50 <kerneltoast> utrace_report_death() no longer has a buggy in_atomic() check

20:52 <kerneltoast> utrace->utrace_flags was shrunk to make room for a refcount. it's fine because there are way fewer than 32 utrace flags

20:52 <kerneltoast> utrace structs can no longer be used after free by various task workers

20:53 <kerneltoast> and they can't be used after free at all now in fact

20:54 <kerneltoast> any dangling utrace_engine structs are now always freed when the utrace struct is freed

20:54 <kerneltoast> that about sums it up i think

20:56 <kerneltoast> that patch has withstood a day of heavy fuzzing that would kill stap master very quickly

20:57 <kerneltoast> and it passes our test suite

21:06 <kerneltoast> it also cooked me a nice breakfast and filed my taxes

21:15 <fche> ok, will sit down over the weekend, not look it too deeply, but let it gnaw away at my sense of humanity

21:15 <kerneltoast> don't wanna push today and let the buildbots cry?

21:21 <irker327> systemtap: scox systemtap.git:master * release-4.4-89-g1ef2edcd3 / dwflpp.cxx dwflpp.h testsuite/systemtap.base/enum.cxx testsuite/systemtap.base/enum.exp testsuite/systemtap.base/enum.stp: Add support for accessing enumerators

21:46 <fche> kerneltoast, will have to take a look at it later tonight

21:46 <kerneltoast> mmkay

21:46 <kerneltoast> i'll give you split up commits in a little bit

21:47 <fche> yupper

22:38 <kerneltoast> fche, https://gist.github.com/kerneltoast/10e6ffdecac6510818324c4b7835fd9d

22:38 <kerneltoast> lemme know if you think it can be split more

22:41 <fche> reading

22:43 <kerneltoast> I'll make the refcount patch's commit message nicer later

22:48 <fche> noting still some rcu usage, e.g. in utrace_exit()

22:48 <fche> are those still needed?

22:50 <fche> also, did you have a chance to run the normal stap testsuite also, or just your haircutter ?

22:53 <fche> in utrace_finish_stop(), what's the purpose of lock then unlock of utrace->lock?

22:53 <fche> meh, previous code does the same thing

23:02 <kerneltoast> fche, yeah some locks can be trimmed i think, like in utrace cleanup since no more engines will be added

23:03 <kerneltoast> i haven't run this on the normal testsuite

23:03 <kerneltoast> I'll do that when i get back to muh laptop

23:04 <kerneltoast> rcu is needed to iterate on the hashlist

23:06 <kerneltoast> and i too have no idea what that lock/unlock is for

23:06 <fche> guess it makes sure no one else had it, at least briefly

23:07 <kerneltoast> it might not be needed with refcounts

23:07 <kerneltoast> unless there's some synchronization requirement

23:39 <kerneltoast> okey back at laptop

23:40 <kerneltoast> fche, the rcu read lock in utrace_exit() is probably not needed but i was paranoid

23:42 <kerneltoast> i think the justification i gave myself was it's needed if a grace period hadn't elapsed between the time of the last utrace free and iterating through the hashlist in utrace_exit()

23:44 <kerneltoast> oh yeah i kept the utrace->lock in utrace_cleanup() in case an engine could be in use while freeing the utrace struct

23:45 <kerneltoast> if that's possible, then we need to listen to this comment: /* FIXME: hmm, should this be utrace_engine_put()? */