From: Benjamin Herrenschmidt <benh@kernel.crashing.org>

It's very easy for the CPU specific has_work() implementation
and the logic in ppc_hw_interrupt() to be subtly out of sync.

This can occasionally allow a CPU to wakeup from a PM state
and resume executing past the PM instruction when it should
resume at the 0x100 vector.

This detects if it happens and aborts, making it a lot easier
to catch such bugs when testing rather than chasing obscure
guest misbehaviour.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
---
 target/ppc/excp_helper.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/target/ppc/excp_helper.c b/target/ppc/excp_helper.c
index 37546bb0f0fe..1a2f469a5fa2 100644
--- a/target/ppc/excp_helper.c
+++ b/target/ppc/excp_helper.c
@@ -878,6 +878,22 @@ static void ppc_hw_interrupt(CPUPPCState *env)
             return;
         }
     }
+
+    if (env->resume_as_sreset) {
+        /*
+         * This is a bug ! It means that has_work took us out of halt without
+         * anything to deliver while in a PM state that requires getting
+         * out via a 0x100
+         *
+         * This means we will incorrectly execute past the power management
+         * instruction instead of triggering a reset.
+         *
+         * It generally means a discrepancy between the wakup conditions in the
+         * processor has_work implementation and the logic in this function.
+         */
+        cpu_abort(CPU(ppc_env_get_cpu(env)),
+                  "Wakeup from PM state but interrupt Undelivered");
+    }
 }
 
 void ppc_cpu_do_system_reset(CPUState *cs)
-- 
2.20.1

On Mon, Jan 28, 2019 at 10:46:21AM +0100, Cédric Le Goater wrote:
> From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> 
> It's very easy for the CPU specific has_work() implementation
> and the logic in ppc_hw_interrupt() to be subtly out of sync.
> 
> This can occasionally allow a CPU to wakeup from a PM state
> and resume executing past the PM instruction when it should
> resume at the 0x100 vector.
> 
> This detects if it happens and aborts, making it a lot easier
> to catch such bugs when testing rather than chasing obscure
> guest misbehaviour.
> 
> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Signed-off-by: Cédric Le Goater <clg@kaod.org>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
>  target/ppc/excp_helper.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/target/ppc/excp_helper.c b/target/ppc/excp_helper.c
> index 37546bb0f0fe..1a2f469a5fa2 100644
> --- a/target/ppc/excp_helper.c
> +++ b/target/ppc/excp_helper.c
> @@ -878,6 +878,22 @@ static void ppc_hw_interrupt(CPUPPCState *env)
>              return;
>          }
>      }
> +
> +    if (env->resume_as_sreset) {
> +        /*
> +         * This is a bug ! It means that has_work took us out of halt without
> +         * anything to deliver while in a PM state that requires getting
> +         * out via a 0x100
> +         *
> +         * This means we will incorrectly execute past the power management
> +         * instruction instead of triggering a reset.
> +         *
> +         * It generally means a discrepancy between the wakup conditions in the
> +         * processor has_work implementation and the logic in this function.
> +         */
> +        cpu_abort(CPU(ppc_env_get_cpu(env)),
> +                  "Wakeup from PM state but interrupt Undelivered");
> +    }
>  }
>  
>  void ppc_cpu_do_system_reset(CPUState *cs)

-- 
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