Re: [PATCH RFC 0/5] Change ghes driver to use HEST-based offsets

[Mauro Carvalho Chehab]

Em Wed, 13 Nov 2024 07:54:18 +0100
Mauro Carvalho Chehab <mchehab+huawei@kernel.org> escreveu:

> Em Wed, 2 Oct 2024 15:45:34 +0200
> Igor Mammedov <imammedo@redhat.com> escreveu:
> 
> > On Tue,  1 Oct 2024 13:42:45 +0200
> > Mauro Carvalho Chehab <mchehab+huawei@kernel.org> wrote:
> >   
> > > This RFC series was part of the previous PR to add generic error injection
> > > support on GHES.
> > > 
> > > It contains only the changes of the math used to calculate offsets at
> > > HEST table and hardware_error firmware file.
> > > 
> > > The first patch adds a new firmware file to store HEST address.
> > > The second patch makes use of it.
> > > The remaining ones add migration support.
> > > 
> > > PS.: I'm sending this as a RFC as using the proceudure defined at the
> > > pseudo-migration of:
> > > 
> > >   https://www.linux-kvm.org/page/Migration
> > > 
> > > Didn't work. I tried to use two different QEMU versions to check a
> > > real life case and also to use just one QEMU and trying to load a
> > > virt-9.1 state on a virt-9.2 machine. 
> > > 
> > > For instance, trying to restore a virt-9.1 state on virt-9.2 gave me
> > > this error:
> > > 
> > >   (qemu) qemu: Machine type received is 'virt-9.1' and local is 'virt-9.2'
> > >   qemu: load of migration failed: Invalid argument    
> > 
> > that's expected (idea is to keep machine type (virt-X) ABI stable so
> > it would work the same way on old and new QEMU)
> > migration is meant to move VM of the same machine type to a new/another 
> > QEMU instance.  
> 
> I found a couple of issues and, after the fixes, it can successfully
> migrate both virt-9.1 and virt-9.2 machines. 
> 
> > 
> > i.e try migrate 
> > 
> > qemu-9.1 -M virt-9.1  => qemu-9.2 -M virt-9.1
> > and vice-versa
> > migration should succeed and memory error injection should still work
> > the old way in both instances (I don't recall anymore how to simulate SEA,
> > perhaps original author left a description how to do that somewhere on 
> > mail-list).  
> 
> Those work as well, but I had to pass -cpu cortex-a57 to both 9.1
> and 9.2, as using -cpu max caused qemu to refuse loading the guest.
> 
> I tested with both:
> 
>       qemu-9.1 -M virt-9.1 -cpu cortex-a57 => qemu-9.2 -M virt-9.1 -cpu 
> cortex-a57
>       qemu-9.2 -M virt-9.1 -cpu cortex-a57 => qemu-9.1 -M virt-9.1 -cpu 
> cortex-a57

Forgot to mention, but I modified qemu-9.1 to use GPIO instead of SEA, as
it is a lot easier to do the tests using the error injection logic.
Also, I don't know how to test SEA errors.

> I'll address your other comments to the series and post a new version
> today.
> 
> Thanks,
> Mauro



Thanks,
Mauro