[RFC,0/3] Attempt to add GHES for x86

Message

Mauro Carvalho Chehab March 4, 2025, 1:30 p.m. UTC

Hi Igor,

This patch series comes after:
   https://lore.kernel.org/qemu-devel/[e-mail]/

I'm basically trying to add support for error injection for x86,
without success so far.

On x86, the notification mechanism is different: it is via QMP.
Yet, from what I saw on Linux implementation, it works on
a similar way to GED. So, I ended implementing a notification
via GED.

HEST table build seems to be working OK on it, and the
error injection notification for x86 is called. Yet, OSPM is not
receiving any notifications.

Could you help me figuring out what are the missing bits?

PS.: there are some things at the code that require polishing,
plus there are some extra printf() there to help debugging.
Finally, we would need to add x86 CPU error event at the
ghes script, but I'd like to have at least something that the 
OSPM receives before improving it further.

It can be tested with:
   $ ./scripts/ghes_inject.py arm

(Ok, this would be generating an ARM processor event , but
I guess Linux would at least mark the event as read, even if
it doesn't recognize it)

Mauro Carvalho Chehab (3):
  acpi/ghes: move use_hest_addr out of acpi_build_hest()
  hw/i186: add support for HEST table with SCI
  Add a GED device for RAS notification

 hw/acpi/ghes.c           | 16 ++++-----
 hw/arm/virt-acpi-build.c | 12 ++++---
 hw/i386/Kconfig          |  1 +
 hw/i386/acpi-build.c     | 75 ++++++++++++++++++++++++++++++++++++++++
 hw/i386/pc.c             | 41 ++++++++++++++++++++++
 include/hw/acpi/ghes.h   | 25 +++++++-------
 include/hw/i386/pc.h     |  5 +++
 include/hw/i386/x86.h    |  2 ++
 8 files changed, 153 insertions(+), 24 deletions(-)

Comments

Igor Mammedov April 4, 2025, 12:19 p.m. UTC | #1

On Tue,  4 Mar 2025 14:30:55 +0100
Mauro Carvalho Chehab <[e-mail]> wrote:

> Hi Igor,
> 
> This patch series comes after:
>    https://lore.kernel.org/qemu-devel/[e-mail]/
> 
> I'm basically trying to add support for error injection for x86,
> without success so far.
> 
> On x86, the notification mechanism is different: it is via QMP.
> Yet, from what I saw on Linux implementation, it works on
> a similar way to GED. So, I ended implementing a notification
> via GED.
> 
> HEST table build seems to be working OK on it, and the
> error injection notification for x86 is called. Yet, OSPM is not
> receiving any notifications.

microvm already uses GED, so GHES support with GED is fine there  

However ps/q35 have already acpi controller, and I'd rather
add GHES support there instead of adding extra acpi device.
and use GPE as event delivery mechanism (like it's done for
hotplug event delivery)

> Could you help me figuring out what are the missing bits?
> 
> PS.: there are some things at the code that require polishing,
> plus there are some extra printf() there to help debugging.

use tracing instead of printfs

> Finally, we would need to add x86 CPU error event at the
> ghes script, but I'd like to have at least something that the 
> OSPM receives before improving it further.
> 
> It can be tested with:
>    $ ./scripts/ghes_inject.py arm
> 
> (Ok, this would be generating an ARM processor event , but
> I guess Linux would at least mark the event as read, even if
> it doesn't recognize it)
> 
> Mauro Carvalho Chehab (3):
>   acpi/ghes: move use_hest_addr out of acpi_build_hest()
>   hw/i186: add support for HEST table with SCI
>   Add a GED device for RAS notification
> 
>  hw/acpi/ghes.c           | 16 ++++-----
>  hw/arm/virt-acpi-build.c | 12 ++++---
>  hw/i386/Kconfig          |  1 +
>  hw/i386/acpi-build.c     | 75 ++++++++++++++++++++++++++++++++++++++++
>  hw/i386/pc.c             | 41 ++++++++++++++++++++++
>  include/hw/acpi/ghes.h   | 25 +++++++-------
>  include/hw/i386/pc.h     |  5 +++
>  include/hw/i386/x86.h    |  2 ++
>  8 files changed, 153 insertions(+), 24 deletions(-)
>