[Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[James Hogan]

When a KVM internal error occurs QEMU dumps the CPU state, however it
doesn't synchronise the state from KVM first so the dumped state is out
of date. Add the synchronisation calls before the dump in both locations
(which is used depends on whether the arch says to stop or not).

Note that x86 does a sync in its kvm_arch_stop_on_emulation_error()
function so at least for emulation errors is unaffected.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org
---
 kvm-all.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kvm-all.c b/kvm-all.c
index b788fcd..803141a 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -1546,6 +1546,7 @@ static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
     if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
         fprintf(stderr, "emulation failure\n");
         if (!kvm_arch_stop_on_emulation_error(cpu)) {
+            kvm_cpu_synchronize_state(cpu);
             cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
             return EXCP_INTERRUPT;
         }
@@ -1701,6 +1702,7 @@ int kvm_cpu_exec(CPUState *cpu)
     } while (ret == 0);
 
     if (ret < 0) {
+        kvm_cpu_synchronize_state(cpu);
         cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
         vm_stop(RUN_STATE_INTERNAL_ERROR);
     }
-- 
1.8.1.2

Re: [Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[Gleb Natapov]

On Fri, Aug 23, 2013 at 01:26:00PM +0100, James Hogan wrote:
> When a KVM internal error occurs QEMU dumps the CPU state, however it
> doesn't synchronise the state from KVM first so the dumped state is out
> of date. Add the synchronisation calls before the dump in both locations
> (which is used depends on whether the arch says to stop or not).
> 
x86_cpu_dump_state() calls cpu_synchronize_state() already.

> Note that x86 does a sync in its kvm_arch_stop_on_emulation_error()
> function so at least for emulation errors is unaffected.
> 
> Signed-off-by: James Hogan <james.hogan@imgtec.com>
> Cc: Gleb Natapov <gleb@redhat.com>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: kvm@vger.kernel.org
> ---
>  kvm-all.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kvm-all.c b/kvm-all.c
> index b788fcd..803141a 100644
> --- a/kvm-all.c
> +++ b/kvm-all.c
> @@ -1546,6 +1546,7 @@ static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
>      if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
>          fprintf(stderr, "emulation failure\n");
>          if (!kvm_arch_stop_on_emulation_error(cpu)) {
> +            kvm_cpu_synchronize_state(cpu);
>              cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
>              return EXCP_INTERRUPT;
>          }
> @@ -1701,6 +1702,7 @@ int kvm_cpu_exec(CPUState *cpu)
>      } while (ret == 0);
>  
>      if (ret < 0) {
> +        kvm_cpu_synchronize_state(cpu);
>          cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
>          vm_stop(RUN_STATE_INTERNAL_ERROR);
>      }
> -- 
> 1.8.1.2
> 

--
			Gleb.

Re: [Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[James Hogan]

On 23/08/13 13:58, Gleb Natapov wrote:
> On Fri, Aug 23, 2013 at 01:26:00PM +0100, James Hogan wrote:
>> When a KVM internal error occurs QEMU dumps the CPU state, however it
>> doesn't synchronise the state from KVM first so the dumped state is out
>> of date. Add the synchronisation calls before the dump in both locations
>> (which is used depends on whether the arch says to stop or not).
>>
> x86_cpu_dump_state() calls cpu_synchronize_state() already.

Ah yes, thanks. I hadn't noticed that.

Out of the arches that support KVM only x86 and ppc call it. arm, mips
(qemu support not upstream yet), and s390 don't. s390 never seems to
emit that exit code, and arm only does so for unsupported exceptions
(which should never happen).

I'll fix in mips_cpu_dump_state() instead.

Cheers
James

Re: [Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[Gleb Natapov]

On Fri, Aug 23, 2013 at 02:41:13PM +0100, James Hogan wrote:
> On 23/08/13 13:58, Gleb Natapov wrote:
> > On Fri, Aug 23, 2013 at 01:26:00PM +0100, James Hogan wrote:
> >> When a KVM internal error occurs QEMU dumps the CPU state, however it
> >> doesn't synchronise the state from KVM first so the dumped state is out
> >> of date. Add the synchronisation calls before the dump in both locations
> >> (which is used depends on whether the arch says to stop or not).
> >>
> > x86_cpu_dump_state() calls cpu_synchronize_state() already.
> 
> Ah yes, thanks. I hadn't noticed that.
> 
> Out of the arches that support KVM only x86 and ppc call it. arm, mips
> (qemu support not upstream yet), and s390 don't. s390 never seems to
> emit that exit code, and arm only does so for unsupported exceptions
> (which should never happen).
> 
> I'll fix in mips_cpu_dump_state() instead.
> 
Moving cpu_synchronize_state() up to cpu_dump_state() would be better.

--
			Gleb.

Re: [Qemu-devel] [PATCH] kvm: sync cpu state on internal error before dump

[Andreas Färber]

Am 24.08.2013 12:37, schrieb Gleb Natapov:
> On Fri, Aug 23, 2013 at 02:41:13PM +0100, James Hogan wrote:
>> On 23/08/13 13:58, Gleb Natapov wrote:
>>> On Fri, Aug 23, 2013 at 01:26:00PM +0100, James Hogan wrote:
>>>> When a KVM internal error occurs QEMU dumps the CPU state, however it
>>>> doesn't synchronise the state from KVM first so the dumped state is out
>>>> of date. Add the synchronisation calls before the dump in both locations
>>>> (which is used depends on whether the arch says to stop or not).
>>>>
>>> x86_cpu_dump_state() calls cpu_synchronize_state() already.
>>
>> Ah yes, thanks. I hadn't noticed that.
>>
>> Out of the arches that support KVM only x86 and ppc call it. arm, mips
>> (qemu support not upstream yet), and s390 don't. s390 never seems to
>> emit that exit code, and arm only does so for unsupported exceptions
>> (which should never happen).
>>
>> I'll fix in mips_cpu_dump_state() instead.
>>
> Moving cpu_synchronize_state() up to cpu_dump_state() would be better.

Yes, please. I did not review the hooks themselves much, just avoided
global functions.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg