From: Vit Kabele <redacted> Sent: Monday, June 20, 2022 6:56 AM

On Thu, Jun 16, 2022 at 05:03:16PM +0200, Vitaly Kuznetsov wrote:

quoted

Vit Kabele [off-list ref] writes:

quoted

Nevertheless, I am not sure about following:

1/ The vmbus_initiate_unload function is called within the panic handler
even when the vmbus initialization does not finish (there might be no
vmbus at all). This should probably not be problem because the vmbus
unload function always checks for current connection state and does
nothing when this is "DISCONNECTED". For better readability, it might be
better to add separate panic notifier for vmbus and crash reporting.

2/ Wouldn't it be better to extract the whole reporting capability out
of the vmbus module, so that it stays present in the kernel even when
the vmbus module is possibly unloaded?

IMHO yes but as you mention hyperv_panic_event() currently does to
things:
1) Initiates VMBus unload
2) Reports panic to the hypervisor

I think untangling them moving the later to arch/x86/hyper-v (and
arch/arm64/hyperv/) makes sense.

Ok, I will send the complete patch soon.

Vit --

FYI, there's a large patch series [1] that proposed some reorganization
of the panic notifiers across the Linux kernel.  Patch 16 of the series
splits the Hyper-V panic notifier into two along the lines that you
suggest.  In addition to the patch itself, the comments and follow-on
discussion are relevant to changes you propose.  See my responses
throughout the series.

The author of the series is planning a v2, but he's out for a few weeks
so there will be a delay. [2]

Michael

[1] https://lore.kernel.org/linux-hyperv/20220427224924.592546-1-gpiccoli@igalia.com/T/#t (local)
[2] https://lore.kernel.org/linux-hyperv/20220427224924.592546-1-gpiccoli@igalia.com/T/#m3c190913bcb6f66e3ace792b4e6f2236839d4fa7 (local)