Lockdep warnings on kexec (virtio_blk, hrtimers)

[David Woodhouse]

On Thu, 2024-12-12 at 19:04 +0100, Thomas Gleixner wrote:
> 
> I tried to reproduce, but failed. Do you have a simple reproducer
> recipe?

Build current master (231825b2e1ff here). The config I'm using is at
http://david.woodhou.se/config-x86-kjump-irqs although I don't think
there's anything special other than CONFIG_KEXEC_JUMP and enough
lockdep to trigger the complaints.

Build the test case from
https://git.kernel.org/torvalds/c/07fa619f2a40c (which at some point
I'll clean up and add to the kernel selftests).

Then just run it repeatedly. Ten or twenty invocations should trigger
it at least once.

I've been running in qemu with the test case shoved into an initrd for
faster testing, but it works just as well done manually. If it matters,
the QEMU command line on my Haswell box is

 qemu-system-x86_64 -accel kvm,kernel-irqchip=split -display none \
   -serial mon:stdio -kernel arch/x86/boot/bzImage -smp 2 -m 2g \
   -append "console=ttyS0 root=/dev/vda1 no_console_suspend earlyprintk=serial" \
   -drive file=/var/lib/libvirt/images/fedora.qcow2,if=virtio \
   -cpu host --no-reboot -nic user,model=virtio 

Probably the only important part of that is the no_console_suspend.

A normal invocation looks like this:

[root at localhost ~]# ./loadret
[   83.908243] Freezing user space processes
[   83.910587] Freezing user space processes completed (elapsed 0.001 seconds)
[   83.911614] OOM killer disabled.
[   83.923550] ata2: found unknown device (class 0)
[   83.938623] Disabling non-boot CPUs ...
[   83.962083] smpboot: CPU 1 is now offline
[   83.963022] crash hp: kexec_trylock() failed, kdump image may be inaccurate
B[   83.983498] Enabling non-boot CPUs ...
[   83.984410] crash hp: kexec_trylock() failed, kdump image may be inaccurate
[   83.985792] smpboot: Booting Node 0 Processor 1 APIC 0x1
[   83.989823] CPU1 is up
[   83.999270] virtio_blk virtio1: 2/0/0 default/read/poll queues
[   84.004705] OOM killer enabled.
[   84.005299] Restarting tasks ... done.
Success


But sometimes it looks like this...

[root at localhost ~]# ./loadret
[   84.618610] Freezing user space processes
[   84.621427] Freezing user space processes completed (elapsed 0.001 seconds)
[   84.622510] OOM killer disabled.
[   84.628613] ata2: found unknown device (class 0)
[   84.652758] Disabling non-boot CPUs ...
[   84.672223] smpboot: CPU 1 is now offline
[   84.673091] crash hp: kexec_trylock() failed, kdump image may be inaccurate
B[   84.682946] ------------[ cut here ]------------
[   84.683584] Interrupts enabled after irqrouter_resume+0x0/0x50

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5965 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/kexec/attachments/20241212/48b0dd1b/attachment.p7s>