hrm.. is there a way to set the signal sent to a parent when a whild process dies? (e.g. equiv of freebsd procctl(P_PID, -, PROC_DEATHSIG_CTL, ..) or linux prctl(PR_SET_PDEATHSIG, xxx) ?

maybe something with contracts?

err i have that backwards.. want child to get signal when parent dies

as a variant: register via atexit() a function to send a signal to the children

jbk: contract should work though I believe you'd get the notification via a fd rather than a signal. A child could also potentially poll with getppid()

hi all, i have no perms for review. i'm interested in why my pers was removed

*permissions

igork: you asked that question about two days ago - did you see jbk's response to you? (have you tried signing in again?)

sommerfeld: yes, i saw, and i tried his proposed solution - it is not working

andyf: have you perhaps seen this kind of ena panic before? paste.ec/raw/Zy0f-Jek#znVpMTR+Fgus9ot5tHWFXodlhViZqvj-HY/rKeQnGPZ

i saw some removed of my +1 to richlowe changes with sparc updates by jclulow and after that i have no perms for review on tsoome changes too

jbk: thanks for your proposal but it is not working me for review, i have no permissions and i have no ideas - why ?

daleg - yes illumos 16373 (fenix).

BUG 16373: ena network interface occasionally hangs (In Progress)

↳ illumos.org/issues/16373

That is the same - just you ran something like `dladm` or something else that tried to send an admin command after it hung.

I have implemented reset for the driver, which lets it get over that bump but don't know the root cause yet.

andyf: thanks. alright. I have some crash dumps, but have not hit upon a way to reliably reproduce this. It seems to happen in spates - a few panics and reboots due to this, followed by a long period of it not occurring.

Here's the change that adds reset, and fixes some other bugs - code.illumos.org/c/illumos-gate/+/3367?usp=search

→ CODE REVIEW 3367: 16391 ena driver could support reset 16392 ena driver async event queue stalls (NEW) | illumos.org/issues/16391

I could spin a hotfix if you're on omnios. I'm working on finding the root cause but no breakthrough yet.

thanks. I'll try those in a local branch and try some more to get this to happen. It's on bsros, so I'll just do my own local application of the patch and try it.

I have never managed to reproduce it myself in AWS, but some CI runners are hitting it fairly regularly. That patch is getting them to run to completion at least while I work on it.

daleg - what's the instance type? In case it's relevant - I'm mostly testing on m7a.medium

yeah I've been trying to get our nitro story in shape and hit upon this using an m5.large. All seems quite fine otherwise.

Happy Patch Tuesday everyone! illumos.topicbox.com/groups/develop…7435d13a1-Mb16ac836483418b3ce160d2c

If anyone here has a modern (Skylake-era or later) Atom processor running any illumos, please contact me.

i'm thinking this is the software trying to use something it shouldn't, but warning on R_AMD64_COPY: file <shlib> <sym>: relocation bound to a symbol with STV_PROTECTED visibility

if i'm wrong, is there something to fix?

(specifically, this is trying to link against pkgsrc libcrypto)

datapoint: I only see one such error in my bulk builds, and it's with ncmpcpp trying to link against libicuuc.so

annoyingly, i think it's the openssl macros itself that are responsible (still digging)

ex: ASN1_TIME_it

[illumos-gate] 16311 ps: Inconsistant formatting of options in usage output -- rigzba21 <jonathan.velando01⊙gc>

is that macro part of the published openssl API/ABI? (they've been tightening up on that lately..)

that i'm not sure

though doing some more digging, i guess it's maybe non-fatal (makefiles don't make that clear)

but at least now i have a hopefully working tpm2 binary for testing

so what on earth is nfsfind?

and why is it removed nfs.*?

removing*

hmmm, is that necessary? Is there a way to kill it. This particular file system is pretty dense

lots of files

those .nfs files are leftovers, so it is good to have cleanup. it is run via root cron, if you do not have nfs service, you can disable it.

KungFuJesus: unix semantics require that you be able to continue doing I/O to an unlinked file that you have open. NFS doesn't allow for that so NFS clients don't unlink open files -- they rename them, instead. And unlink them later, but later may not happen if the client crashes.