Recent content by shallax

I've done this 5 times now, going for a 6th. I have a feeling this has worked around the issue. If I understand correctly, io_uring is only marginally faster than using native, so if I'm okay with losing a tiny bit of performance, I guess this fixes it for me.

Tried three times so far with 3 VMs running the cryptsetup stuff in parallel. It's not reproduced the issue yet. I might do this one more time or so as I'm adding quite a bit of mileage to my SSDs when I do this ;)

I'll try this now and report back today.

Sure, unfortunately you probably won't like the answer as it's a fairly bespoke setup based on a Gentoo kernel (Linux xxxx 5.10.61-gentoo #2 SMP Tue Sep 28 15:36:05 BST 2021 x86_64 Common KVM processor GenuineIntel GNU/Linux) I'm actually in an homemade initramfs that contains statically...

To try and get you some crashes, I've set up 3 VMs on which I run cryptsetup-reencrypt in parallel. I've had a segfault in one so far, but the stack trace isn't so useful (though it does appear to be within code handling IO). See attached.

That's all I can get for now.

Another one (see attached) Anecdotally, it seems like the crashes are more likely to occur if more than one VM is doing a high amount of IO at a time.

Interesting, different crash this time: Sep 28 10:17:24 node QEMU[171446]: kvm: ../softmmu/physmem.c:3193: address_space_unmap: Assertion `mr != NULL' failed. root@node:~/crash# cat gdb.txt Continuing. Thread 1 "kvm" received signal SIGABRT, Aborted. __GI_raise (sig=sig@entry=6) at...

I'm on it, chief! Giving it my best with about 10 VMs to encrypt. I'll let you know how it goes.

Just thought I'd add that I'm also receiving this issue with a Linux guest when running cryptsetup-reencrypt to encrypt my disks. It doesn't always happen, though. However, I have had it occur on multiple nodes (running AMD Ryzen CPUs - one with ECC RAM, so I'm hoping it's a software issue...