23:01 <kerneltoast> fche, bonjour

23:01 <kerneltoast> how uneasy does this patch make you feel? https://gist.github.com/kerneltoast/54c49d6cdf121d50d99ec7fab396ed27

23:11 <fche> could the comment better spell out the lifespan of the relevant objects in the union?

23:11 <fche> like ... "rcu is only used in functions x y z AFTER an event takes place"

23:11 <fche> "attached/attaching are only used in function a b c BEFORE this event"

23:12 <fche> and some way to feel sure that the transition of that event is atomic, i.e., there is no moment when both conditions are true in some sort of inatomic way

23:16 <kerneltoast> hmm i could also make those objects' lifetimes more strict wrt RCU and easier to audit

23:28 <kerneltoast> fche, oh no

23:28 <kerneltoast> i found a bug in my first stap commit

23:28 <fche> probably there was one in my first one too

23:38 <irker249> systemtap: sultan systemtap.git:master * release-4.4-62-g23ef716aa / runtime/stp_utrace.c: stp_utrace: remove kmem cache usage

23:38 <irker249> systemtap: sultan systemtap.git:master * release-4.4-63-gd5c4be8aa / runtime/stp_utrace.c: stp_utrace: add missing rcu_read_unlock() in get_utrace_lock() error path

23:38 <kerneltoast> CRAP

23:38 <kerneltoast> CRAP

23:38 <kerneltoast> NOOOOOOOOOOOO

23:38 <kerneltoast> AAAAAAAAAAAAAAH

23:38 <kerneltoast> FCHE HELP ME

23:38 <fche> with

23:38 <kerneltoast> I MESSED UP

23:38 <kerneltoast> I DIDN'T MEAN TO PUSH THE KMEM CACHE COMMIT

23:39 <kerneltoast> AAAAAAAAAAAAAA

23:39 <fche> ok so you revert it and push

23:39 <fche> bad toast

23:39 <fche> no butter

23:39 <irker249> systemtap: sultan systemtap.git:master * release-4.4-64-g3b858b727 / runtime/stp_utrace.c: Revert "stp_utrace: remove kmem cache usage"

23:40 <kerneltoast> the message in "stp_utrace: add missing rcu_read_unlock() in get_utrace_lock() error path" didn't lie

23:45 <kerneltoast> bad day to be a toast

23:53 <kerneltoast> fche, so RCU isn't even used to strictly guarantee a utrace struct's lifetime, as seen in task_utrace_struct()

23:53 <kerneltoast> task_utrace_struct() returns a utrace struct which is then used outside the RCU lock

23:54 <kerneltoast> this was a mistake on my part when i added the RCU stuff; i just made use of the existing assumptions about utrace object lifetime when i did that

23:54 <kerneltoast> *wasn't a mistake

23:54 <kerneltoast> man i am failing hard today

23:57 <kerneltoast> so, since we aren't even relying on RCU to keep utrace structs safe, we can assume that the struct members i'm reusing for the RCU callback are not used in a way that leans on RCU's lifetime guarantees after the utrace struct is marked dead

23:57 <kerneltoast> i hope that makes sense

23:57 <fche> or how about this

23:57 <fche> if we're not SURE,

23:57 <fche> leave the code the way it is, waste the hundred or whatever bytes, who cares