Read errors on stripe cause hang/crash.

[zandr]

memtest86 passed with no issues. I needed to move ahead with rebuilding the array for my Photo library, but I'll take these two flaky disks and try to reproduce this on another box.

With the replacement disks, I've replicated the dataset back, and am most of the way through a scrub. Fingers Crossed.

[lundman]

MCA Error Report ***\nCPU Machine Check Architecture Error Dump

That's the first time I've seen that, and appears to only talk about hardware, not software, so possibly ZFS isn't even involved - it certainly does
not contain a list of kexts like we are used to seeing.

[zandr]

so possibly ZFS isn't even involved

Other than the "completely repeatable by importing the zpool and letting the scrub complete" bit, yeah.

- it certainly does not contain a list of kexts like we are used to seeing.

Because it isn't the same kind of crash. On the stripe, it would just lock up the machine without a panic. On the mirror, it managed to throw this.

I'll be able to do some additional testing this week. Need to get some work done using that hardware first.

[lundman]

You could try disabling all the assembler versions (sysctl, setting the 4? to default/generic) in case it is something with saving registers that go wrong.

[4ever6]

You could try disabling all the assembler versions (sysctl, setting the 4? to default/generic) in case it is something with saving registers that go wrong.

Curious about this - which sysctls are you referring to?

[lundman]

There are four: zfs_vdev_raidz_impl icp_gcm_impl icp_aes_impl zfs_fletcher_4_impl

You set them to "generic" and hmm, standard? Whatever the most benign sounding one is

[jawbroken]

I have had two of these same crashes in the last month or so with the same enclosure. I recently upgraded to the latest Catalina versions, after lagging behind a bit. It's suspicious that we all have the same hardware, but given the generic error message it could be any combination of hardware, OS, or OpenZFS.