Distros have started using the 'scv' instructions (glibc 2.33) which
relies on the LPCR AIL bits. Unfortunately, the LPCR of hot-plugged
CPUs is not synchronized with the rest of machine and it breaks the
guest OS.

Fix that by using the first CPU to set the LPCR value of all hot-plugged
CPUs.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
---
 hw/ppc/spapr.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index c23bcc449071..e463c2570c7a 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -3890,6 +3890,8 @@ static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
         for (i = 0; i < cc->nr_threads; i++) {
             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
                            &error_abort);
+            ppc_store_lpcr(core->threads[i],
+                           POWERPC_CPU(first_cpu)->env.spr[SPR_LPCR]);
         }
     }
 
-- 
2.31.1

On Mon, 24 May 2021 13:41:32 +0200
Cédric Le Goater <clg@kaod.org> wrote:

> Distros have started using the 'scv' instructions (glibc 2.33) which
> relies on the LPCR AIL bits. Unfortunately, the LPCR of hot-plugged
> CPUs is not synchronized with the rest of machine and it breaks the
> guest OS.
> 
> Fix that by using the first CPU to set the LPCR value of all hot-plugged
> CPUs.
> 
> Signed-off-by: Cédric Le Goater <clg@kaod.org>
> ---

LGTM

Reviewed-by: Greg Kurz <groug@kaod.org>

>  hw/ppc/spapr.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index c23bcc449071..e463c2570c7a 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -3890,6 +3890,8 @@ static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
>          for (i = 0; i < cc->nr_threads; i++) {
>              ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
>                             &error_abort);
> +            ppc_store_lpcr(core->threads[i],
> +                           POWERPC_CPU(first_cpu)->env.spr[SPR_LPCR]);
>          }
>      }
>  

On Mon, May 24, 2021 at 01:41:32PM +0200, Cédric Le Goater wrote:
> Distros have started using the 'scv' instructions (glibc 2.33) which
> relies on the LPCR AIL bits. Unfortunately, the LPCR of hot-plugged
> CPUs is not synchronized with the rest of machine and it breaks the
> guest OS.
> 
> Fix that by using the first CPU to set the LPCR value of all hot-plugged
> CPUs.
> 
> Signed-off-by: Cédric Le Goater <clg@kaod.org>

I'm assuming this is obsoleted by Nick Piggin's rework of LPCR
initialization.  This patch does fix a real bug, but it leaves LPCR
initialization a bit of a mess.

> ---
>  hw/ppc/spapr.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index c23bcc449071..e463c2570c7a 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -3890,6 +3890,8 @@ static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
>          for (i = 0; i < cc->nr_threads; i++) {
>              ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
>                             &error_abort);
> +            ppc_store_lpcr(core->threads[i],
> +                           POWERPC_CPU(first_cpu)->env.spr[SPR_LPCR]);
>          }
>      }
>  

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson