Kernel panic advice

[webb]

What are good steps to take in case of kernel panics? System has used OpenZFS on OS X without incident for the past month, and now two OpenZFS on OS X kernel panics today. Both times I was listening to music (first in iTunes, second in Spotify) while building pytorch via MacPorts, but none of that (except iTunes, which was in use the first time) used any zfs datasets, and these are very typical activities for me which have never produced kernel panics before.

macOS 10.13.6
Mac Pro 5,1
OpenZFS on OS X 1.9.4

What I've done so far:

Code: Select all

sudo sysctl kstat.zfs.darwin.tunable.zfs_recover=1 # found this in forums for a similar issue
suzo zpool scrub <my pool> # completed in 1 minute without output

Are these reasonable actions to take? Any other suggestions?

I've attached both panic logs. The first does not appear to have anything to do with OpenZFS on OS X, whereas the second does:

Code: Select all

      Kernel Extensions in backtrace:
         net.lundman.spl(1.9.4)[3FF65C3D-2197-371C-8888-0CA18513E410]@0xffffff7f91bca000->0xffffff7f92dbefff
         net.lundman.zfs(1.9.4)[1967216C-271B-3B09-BEAB-F884ACB0F245]@0xffffff7f92def000->0xffffff7f931a9fff
            dependency: com.apple.iokit.IOStorageFamily(2.1)[F27A8A2A-6662-3608-83BD-415037509E01]@0xffffff7f92dbf000
            dependency: net.lundman.spl(1.9.4)[3FF65C3D-2197-371C-8888-0CA18513E410]@0xffffff7f91bca000

BSD process name corresponding to current thread: mdworker

But it would be a strange coincidence for them both to occur so close to each other in time without a causal connection. Pytorch build was taking near 100% CPU (all cores), so that's a reasonable guess as to the cause, but ideally that should just slow the system, not cause a kernel panic.

PS: /opt, where MacPorts writes all of its files, is already excluded from Spotlight indexing, and is also on the boot apfs disk, not a zfs dataset.

[lundman]

panic(cpu 8 caller 0xffffff8010764041): "TLB invalidation IPI timeout, unresponsive CPU bitmap: 0x4000, NMIPI acks: 0x0, now: 0x0, deadline: 2329469885883"@/BuildRoot/Library/Caches/com.apple.xbs/Sources/xnu/xnu-4570.71.82.8/osfmk/x86_64/pmap.c:2798

Those tend not be software, but rather hardware. CPU, GPU or similar.

[webb]

The backplane died a week after I posted this
(very likely from the GPU demanding too much power from the backplane PCIe power sockets)