03:28:56 i can ping matt in the morning since he's probably done the most work in idmap, but yeah, it looks like it needs a lock or atomic add 18:24:28 if a panic happens on lwp0 and you drop into kmdb, will that set t_preempt on lwp0's kthread_t ? 18:27:26 Good question! 18:28:20 I don't believe so 18:30:22 i'm tripping ASSERT(curthread->t_preempt > 0), yet ::print kthread_t t_preempt shows '\001' 18:30:33 which is making me scratch my head... 18:39:01 It will. 18:39:34 https://github.com/illumos/illumos-gate/blob/master/usr/src/uts/common/os/panic.c#L230 18:45:17 rmustacc: TIL '++' isn't assignment for cscope 18:47:09 Huh! That seems like an oversight haha 18:47:35 I also wonder if it would catch things like *(some_ptr + 16) = 19:03:26 aha!.. ok... so i'm not losing it (at least because of that :P) 19:53:26 btw, if anyone's tracking what would be needed to maintain compatibility with Solaris, https://blogs.oracle.com/solaris/post/whats-new-in-the-oracle-solaris-11481-cbe-release lists a bunch of things added in the past 3 years, including some libc functions/system calls and new arguments to various commands 20:02:06 I don't think anyone is, beyond when things seem like good ideas 20:03:55 so you're not going to start working on "ld -z ctf" right away? 20:04:50 heh.. had to pick -g for ptree? :) 20:05:59 it is nice list. 20:15:07 alanc: we've talked about -z ctf for years, it's stuck on todo list length :\ 20:29:10 and just for ptribble, among the nearly 10k words in that blog is one occurence of the word "tribblix" 8-) 20:29:45 alanc: I did notice, and thanks for the mention :-) 20:30:05 thanks for making the packages for us to point to 8-) 20:53:48 alanc: Is Solaris version 3 the version Oracle sent to GNU? 20:54:22 No, which is why we labeled it as Solaris version 3 to distinguish from the GNU one 20:54:46 So if I'm not careful and I end up with illumos version 3 then I'm really in trouble. 20:55:11 ours is closer to FreeBSD CTF version 3 20:55:18 Appreciate the info. 20:55:28 Oh, I thought they had adopted the GNU version at some point. 20:55:46 alanc: did you happen to add parameter names? :) 20:56:15 since andyf added types to stack traces, they've seemed like they'd be neat 20:56:52 I don't think so, it was mainly about the 16->32 bit changes 20:57:17 I'm not sure we've hit those (except probably in userland in pkgsrc) 20:57:19 but we deliver a man page now to document the format if you want to check 20:58:08 yeah, we hit them trying to add CTF to stuff from our userland consolidation (i.e. all the FOSS) 21:00:43 Appreciate the hands up and the context as always Alan. 21:01:10 I think FreeBSD version 3 is a subset of the GNU version, but that's third or fourth hand information as I'm not part of the project team 21:49:27 sommerfeld: torching the cache.db re-sets things to a broken state. That is certainly time saving! Unfortunately, the major "only if the ephemeral cache is new" is the `flag` parameter to `allocids()` (via `sidsys`), which seems entirely reasonable. It looks like it's not the easy kind, or at least, is pretty well hidden. 21:51:48 the flag paramater gets us to the kernel eph allocator, in cred.c which, thankfully, is at least a little atomic 21:57:17 the flag parameter is probably a red herring. 2nd run should find the id in cache.db rather than generating a new ephemeral id. 22:02:59 Yeah, I was just chasing it through to confirm it didn't do anything weird. It does exactly what you'd think, flushes the cache in the kernel and starts afresh