Re: [PATCH v10 00/21] Add ACPI CPER firmware first error injection on ARM emulation

[Mauro Carvalho Chehab <mchehab+huawei@kernel.org>]

Em Tue, 24 Sep 2024 15:14:29 +0200
Igor Mammedov <imammedo@redhat.com> escreveu:

> > 1) preparation patches:
...
> > 69850f550f99 acpi/generic_event_device: add an APEI error device  
> this one doesn't belong to clean ups, I think.
> Lets move this to #3 part

Ok.

> > The migration logic will require some time, and I don't want to bother
> > with the cleanup stuff while doing it. So, perhaps while I'm doing it,
> > you could review/merge the cleanups.
> > 
> > We can do the same for each of the 4 above series of patches, as it
> > makes review simpler as there will be less patches to look into on
> > each series.
> > 
> > Would it work for you?  
> 
> other than nit above, LGTM
> 

Ok, sent a PR with the first set (cleanups) at:

	https://lore.kernel.org/qemu-devel/cover.1727236561.git.mchehab+huawei@kernel.org/

You can see the full series at:

	https://gitlab.com/mchehab_kernel/qemu/-/commits/qemu_submission_v11b?ref_type=heads

It works fine, except for the migration part that I'm still working with.

For the migration, there are how two functions at ghes.c:

The one compatible with current behavior (up to version 9.1):
	https://gitlab.com/mchehab_kernel/qemu/-/blob/qemu_submission_v11b/hw/acpi/ghes.c?ref_type=heads#L411

And the new one using offsets calculated from HEST (newer versions):
	https://gitlab.com/mchehab_kernel/qemu/-/blob/qemu_submission_v11b/hw/acpi/ghes.c?ref_type=heads#L437 

With that, the migration logic can decide what function should be
called (currently, it is just checking if hest_addr_le is zero, but
I guess I'll need to change it to match some variable added by the
migration path.

Also, in preparation for the migration tests, I created a separate 
branch at:

	https://gitlab.com/mchehab_kernel/qemu/-/commits/ghes_on_v9.1.0?ref_type=heads

which contains the same patches on the top of 9.1, except for
the HEST ones. It also contains a hack to use ACPI_GHES_NOTIFY_GPIO
instead of ACPI_GHES_NOTIFY_SEA.

With that, we have a way to use the same error injection logic
on both 9.1 and upstream, hopefully being enough to test if migration
works.

Thanks,
Mauro