Re: hunting memory corruption bug in 6.18.x

[Nikola Ciprich <nikola.ciprich@linuxbox.cz>]

> Well we initially thought that nested virtualization was not involved.
> 
> It it indeed is in the picture then it might have been 55780d8a1dcc
> which unknowingly fixed this.
> 
> Are you able to test that commit on the 6.18 kernel?

Sure! And I have more good news: after applying 55780d8a1dcc
to 6.18.56, I was able to perform more than 30 migrations without
a glitch. After booting back into 6.18.56 without it, both VMs
crashed within 2-3 migrations. Furthermore, some memory apparently
got corrupted on one of the hosts, since various commands started
crashing too:

[  303.466091] traps: systemd-logind[3085] trap invalid opcode ip:7f7b5667e616 sp:7ffe1afdec38 error:0 in libsystemd-shared-252.so[7e616,7f7b5666d000+22e000]
[  303.493810] traps: systemd-coredum[11269] trap invalid opcode ip:7fc6bf67f614 sp:7ffc075da958 error:0 in libsystemd-shared-252.so[7f614,7fc6bf66d000+22e000]
[  377.528955] traps: systemd[5607] trap invalid opcode ip:7f791447e616 sp:7ffe83e61cf8 error:0 in libsystemd-shared-252.so[7e616,7f791446d000+22e000]
[  377.556993] traps: systemd-coredum[12080] trap invalid opcode ip:7f97ce07f614 sp:7ffe28b56a88 error:0 in libsystemd-shared-252.so[7f614,7f97ce06d000+22e000]
[ 3746.173043] traps: hostnamectl[45846] trap invalid opcode ip:7fe00cc7f614 sp:7fff89916e48 error:0 in libsystemd-shared-252.so[7f614,7fe00cc6d000+22e000]
[ 3746.202305] traps: systemd-coredum[45847] trap invalid opcode ip:7f7fe9e7f614 sp:7ffea227cfa8 error:0 in libsystemd-shared-252.so[7f614,7f7fe9e6d000+22e000]
[ 3750.581940] traps: systemctl[45903] trap invalid opcode ip:7fbb8967f614 sp:7fff733e5748 error:0 in libsystemd-shared-252.so[7f614,7fbb8966d000+22e000]
[ 3750.612651] traps: systemd-coredum[45904] trap invalid opcode ip:7f90d347f614 sp:7ffdc7f55148 error:0 in libsystemd-shared-252.so[7f614,7f90d346d000+22e000]

I'm not sure if this is of any use, but I collected a vmcore of
this state by intentionally crashing the host via /proc/sysrq-trigger.

Anyway, I'll give 55780d8a1dcc some more migration rounds, but it looks
promising.

I'll report soon

Cheers, nik