[Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Paolo Bonzini]

io-error is for block device errors; it should always be preceded
by a BLOCK_IO_ERROR event.  I think vfio wants to use
RUN_STATE_INTERNAL_ERROR instead.

Cc: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/misc/vfio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/hw/misc/vfio.c b/hw/misc/vfio.c
index 7437c2e..9d71d97 100644
--- a/hw/misc/vfio.c
+++ b/hw/misc/vfio.c
@@ -3978,7 +3978,7 @@ static void vfio_err_notifier_handler(void *opaque)
                  __func__, vdev->host.domain, vdev->host.bus,
                  vdev->host.slot, vdev->host.function);
 
-    vm_stop(RUN_STATE_IO_ERROR);
+    vm_stop(RUN_STATE_INTERNAL_ERROR);
 }
 
 /*
-- 
1.8.3.1

Re: [Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Alex Williamson]

On Fri, 2014-06-27 at 16:32 +0200, Paolo Bonzini wrote:
> io-error is for block device errors; it should always be preceded
> by a BLOCK_IO_ERROR event.

Where does this requirement come from?  I only see a loose association
of IO_ERROR to disk in libvirt and none in QEMU.

>   I think vfio wants to use
> RUN_STATE_INTERNAL_ERROR instead.

But that seems to put us into an "unknown" paused state in libvirt.  An
I/O error still seems like a more appropriate state.  I'm not sure this
qualifies as a trivial patch, it is small, but subtle.  Thanks,

Alex

> 
> Cc: Alex Williamson <alex.williamson@redhat.com>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  hw/misc/vfio.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/hw/misc/vfio.c b/hw/misc/vfio.c
> index 7437c2e..9d71d97 100644
> --- a/hw/misc/vfio.c
> +++ b/hw/misc/vfio.c
> @@ -3978,7 +3978,7 @@ static void vfio_err_notifier_handler(void *opaque)
>                   __func__, vdev->host.domain, vdev->host.bus,
>                   vdev->host.slot, vdev->host.function);
>  
> -    vm_stop(RUN_STATE_IO_ERROR);
> +    vm_stop(RUN_STATE_INTERNAL_ERROR);
>  }
>  
>  /*

Re: [Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Paolo Bonzini]

----- Messaggio originale -----
> Da: "Alex Williamson" <alex.williamson@redhat.com>
> A: "Paolo Bonzini" <pbonzini@redhat.com>
> Cc: qemu-devel@nongnu.org, qemu-trivial@nongnu.org
> Inviato: Venerdì, 27 giugno 2014 18:34:59
> Oggetto: Re: [PATCH for 2.1] vfio: use correct runstate
> 
> On Fri, 2014-06-27 at 16:32 +0200, Paolo Bonzini wrote:
> > io-error is for block device errors; it should always be preceded
> > by a BLOCK_IO_ERROR event.
> 
> Where does this requirement come from?  I only see a loose association
> of IO_ERROR to disk in libvirt and none in QEMU.

See the RunState enum in qapi-schema.json:

##
# @RunState
#
# An enumeration of VM run states.
#
# ...
#
# @internal-error: An internal error that prevents further guest execution
# has occurred
#
# @io-error: the last IOP has failed and the device is configured to pause
# on I/O errors
#
# @paused: guest has been paused via the 'stop' command

The point of io-error is that management can look at block devices, see if
any have an error reported, and then resume execution (see documentation of
rerror=stop and werror=stop/enospc).  This is counter to the intentions you
have in vfio.

> > I think vfio wants to use
> > RUN_STATE_INTERNAL_ERROR instead.
> 
> But that seems to put us into an "unknown" paused state in libvirt.

I think paused is incorrect, because (unlike RUN_STATE_IO_ERROR), you cannot
resume from RUN_STATE_INTERNAL_ERROR except with a reset.  QEMU enforces that,
and this matches the error you are reporting:

    error_report("%s(%04x:%02x:%02x.%x) Unrecoverable error detected.  "
                 "Please collect any data possible and then kill the guest",
                 __func__, vdev->host.domain, vdev->host.bus,
                 vdev->host.slot, vdev->host.function);

libvirt has a crashed state, I think that's what libvirt should call the
internal-error runstate.  IIRC on Xen you get to crashed when the processor
raises an error on vmentry, for example.

Libvirt only knows about crashed/unknown, but one could add crashed/internal-error
too.

Paolo

Re: [Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Alex Williamson]

On Fri, 2014-06-27 at 13:51 -0400, Paolo Bonzini wrote:
> ----- Messaggio originale -----
> > Da: "Alex Williamson" <alex.williamson@redhat.com>
> > A: "Paolo Bonzini" <pbonzini@redhat.com>
> > Cc: qemu-devel@nongnu.org, qemu-trivial@nongnu.org
> > Inviato: Venerdì, 27 giugno 2014 18:34:59
> > Oggetto: Re: [PATCH for 2.1] vfio: use correct runstate
> > 
> > On Fri, 2014-06-27 at 16:32 +0200, Paolo Bonzini wrote:
> > > io-error is for block device errors; it should always be preceded
> > > by a BLOCK_IO_ERROR event.
> > 
> > Where does this requirement come from?  I only see a loose association
> > of IO_ERROR to disk in libvirt and none in QEMU.
> 
> See the RunState enum in qapi-schema.json:
> 
> ##
> # @RunState
> #
> # An enumeration of VM run states.
> #
> # ...
> #
> # @internal-error: An internal error that prevents further guest execution
> # has occurred
> #
> # @io-error: the last IOP has failed and the device is configured to pause
> # on I/O errors
> #
> # @paused: guest has been paused via the 'stop' command
> 
> The point of io-error is that management can look at block devices, see if
> any have an error reported, and then resume execution (see documentation of
> rerror=stop and werror=stop/enospc).  This is counter to the intentions you
> have in vfio.
> 
> > > I think vfio wants to use
> > > RUN_STATE_INTERNAL_ERROR instead.
> > 
> > But that seems to put us into an "unknown" paused state in libvirt.
> 
> I think paused is incorrect, because (unlike RUN_STATE_IO_ERROR), you cannot
> resume from RUN_STATE_INTERNAL_ERROR except with a reset.  QEMU enforces that,
> and this matches the error you are reporting:
> 
>     error_report("%s(%04x:%02x:%02x.%x) Unrecoverable error detected.  "
>                  "Please collect any data possible and then kill the guest",
>                  __func__, vdev->host.domain, vdev->host.bus,
>                  vdev->host.slot, vdev->host.function);
> 
> libvirt has a crashed state, I think that's what libvirt should call the
> internal-error runstate.  IIRC on Xen you get to crashed when the processor
> raises an error on vmentry, for example.
> 
> Libvirt only knows about crashed/unknown, but one could add crashed/internal-error
> too.

Ok, I'll go with it.  I'll include this in the pull request I intend to
do on Monday.  Thanks,

Alex

Re: [Qemu-devel] [PATCH for 2.1] vfio: use correct runstate

[Eric Blake]

[-- Attachment #1: Type: text/plain, Size: 2439 bytes --]

[adding libvirt]

On 06/27/2014 11:51 AM, Paolo Bonzini wrote:
> ----- Messaggio originale -----
>> Da: "Alex Williamson" <alex.williamson@redhat.com>
>> A: "Paolo Bonzini" <pbonzini@redhat.com>
>> Cc: qemu-devel@nongnu.org, qemu-trivial@nongnu.org
>> Inviato: Venerdì, 27 giugno 2014 18:34:59
>> Oggetto: Re: [PATCH for 2.1] vfio: use correct runstate
>>
>> On Fri, 2014-06-27 at 16:32 +0200, Paolo Bonzini wrote:
>>> io-error is for block device errors; it should always be preceded
>>> by a BLOCK_IO_ERROR event.
>>
>> Where does this requirement come from?  I only see a loose association
>> of IO_ERROR to disk in libvirt and none in QEMU.
> 
> See the RunState enum in qapi-schema.json:
> 
> ##
> # @RunState
> #
> # An enumeration of VM run states.
> #
> # ...
> #
> # @internal-error: An internal error that prevents further guest execution
> # has occurred
> #
> # @io-error: the last IOP has failed and the device is configured to pause
> # on I/O errors
> #
> # @paused: guest has been paused via the 'stop' command
> 
> The point of io-error is that management can look at block devices, see if
> any have an error reported, and then resume execution (see documentation of
> rerror=stop and werror=stop/enospc).  This is counter to the intentions you
> have in vfio.
> 
>>> I think vfio wants to use
>>> RUN_STATE_INTERNAL_ERROR instead.
>>
>> But that seems to put us into an "unknown" paused state in libvirt.
> 
> I think paused is incorrect, because (unlike RUN_STATE_IO_ERROR), you cannot
> resume from RUN_STATE_INTERNAL_ERROR except with a reset.  QEMU enforces that,
> and this matches the error you are reporting:
> 
>     error_report("%s(%04x:%02x:%02x.%x) Unrecoverable error detected.  "
>                  "Please collect any data possible and then kill the guest",
>                  __func__, vdev->host.domain, vdev->host.bus,
>                  vdev->host.slot, vdev->host.function);
> 
> libvirt has a crashed state, I think that's what libvirt should call the
> internal-error runstate.  IIRC on Xen you get to crashed when the processor
> raises an error on vmentry, for example.
> 
> Libvirt only knows about crashed/unknown, but one could add crashed/internal-error
> too.

Yes, we probably need to teach libvirt about this state.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]