Re: [Qemu-devel] [PATCH 0/7] virtio: avoid inappropriate QEMU termination

[Greg Kurz]

On Thu, 22 Sep 2016 07:38:29 +0000
"Gonglei (Arei)" <address@hidden> wrote:

> Hi Greg,
> 
> 
> 
> > -----Original Message-----
> > From: Greg Kurz [mailto:address@hidden
> > Sent: Thursday, September 22, 2016 3:22 PM
> > To: Gonglei (Arei)
> > Cc: address@hidden; Kevin Wolf; Michael S. Tsirkin; Jason Wang; Max
> > Reitz; Aneesh Kumar K.V; Stefan Hajnoczi; Cornelia Huck; Paolo Bonzini
> > Subject: Re: [Qemu-devel] [PATCH 0/7] virtio: avoid inappropriate QEMU
> > termination
> > 
> > On Thu, 22 Sep 2016 06:55:43 +0000
> > "Gonglei (Arei)" <address@hidden> wrote:
> >   
> > > > -----Original Message-----
> > > > From: Greg Kurz [mailto:address@hidden
> > > > Sent: Thursday, September 22, 2016 2:43 PM
> > > > To: Gonglei (Arei)
> > > > Cc: address@hidden; Kevin Wolf; Michael S. Tsirkin; Jason Wang;  
> > Max  
> > > > Reitz; Aneesh Kumar K.V; Stefan Hajnoczi; Cornelia Huck; Paolo Bonzini
> > > > Subject: Re: [Qemu-devel] [PATCH 0/7] virtio: avoid inappropriate QEMU
> > > > termination
> > > >
> > > > On Thu, 22 Sep 2016 09:19:49 +0800
> > > > Gonglei <address@hidden> wrote:
> > > >  
> > > > > On 2016/9/21 21:13, Greg Kurz wrote:  
> > > > > > This series is a follow up to Stefan's work to eradicate most calls 
> > > > > > to
> > > > > > exit() we currently have in the virtio code.
> > > > > >
> > > > > > It addresses all exit() call sites in the blk, net and scsi device 
> > > > > > code,
> > > > > > where the error is about a missing or malformed in/out header sent 
> > > > > > by
> > > > > > the guest. They are converted to use virtio_error() and stop any  
> > processing,  
> > > > > > instead of exiting.
> > > > > >  
> > > > > Actually if you just stop procesing when encounter a missing in/out 
> > > > > header  
> > > > but  
> > > > > send a interrupt to the guest, the guest maybe be stuck.  
> > > > virtio_net_handle_ctrl()
> > > >
> > > > The virtio_error() function sets the device status to 
> > > > DEVICE_NEEDS_RESET  
> > and  
> > > > does send a device configuration change interrupt to the guest, so it 
> > > > can  
> > take  
> > > > appropriate action (i.e. reset the device).
> > > >  
> 
> I saw virtio_error() only handle the virtio 1.0 device, the legacy virtio 
> device may
> still stuck, am I right?
> 

The DEVICE_NEEDS_RESET bit was introduced by the virtio 1.0 specification.

http://docs.oasis-open.org/virtio/virtio/v1.0/cs04/virtio-v1.0-cs04.html#x1-100001

Yes, the legacy device will stay stuck since the broken flag is set.

Cheers.

--
Greg

> +void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, 
> ...)
> +{
> +    va_list ap;
> +
> +    va_start(ap, fmt);
> +    error_vreport(fmt, ap);
> +    va_end(ap);
> +
> +    vdev->broken = true;
> +
> +    if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
> +        virtio_set_status(vdev, vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET);
> +        virtio_notify_config(vdev);
> +    }
> +}
> 
> 
> Regards,
> -Gonglei
> 
> > > That's appropriate. Where is realization of virtio_error() ?
> > > I'm sure I missed something.
> > >  
> > 
> > Sorry for that... Michael already "lectured" me about not providing these
> > details. He is right indeed :)
> > 
> > 
> > This is work in progress by Stefan. The latest version of the patchset (v5) 
> > was
> > posted yesterday:
> > 
> > <address@hidden>
> > 
> > The virtio_error() function itself is in patch 2/9:
> > 
> > <address@hidden>
> > 
> > Cheers.
> > 
> > --
> > Greg
> >   
> > > Regards,
> > > -Gonglei
> > >  
> > > > Maybe I should have mentioned that in the changelog...
> > > >
> > > > Cheers.
> > > >
> > > > --
> > > > Greg
> > > >  
> > > > > is an example, the guest frontend driver infinite loop to wait the  
> > interrupt's  
> > > > coming.  
> > > > > The guest can't work anymore though you don't exit the Qemu process .
> > > > >
> > > > > Regards,
> > > > > -Gonglei
> > > > >  
> > > > > > The remaining call sites are related to a host misconfiguration or a
> > > > > > migration stream issue.
> > > > > >
> > > > > > The 9P code currently calls assert() instead of exit(), but it also 
> > > > > > about
> > > > > > malformed or missing headers, so it gets converted the same way.
> > > > > >
> > > > > > Next work will be to check all assert() call sites in the device 
> > > > > > code, in
> > > > > > case some of them actually refer to a bug in the guest, and should 
> > > > > > be
> > > > > > converted to use virtio_error() as well.
> > > > > >
> > > > > > ---
> > > > > >
> > > > > > Greg Kurz (7):
> > > > > >       virtio-9p: handle handle_9p_output() error
> > > > > >       virtio-blk: handle virtio_blk_handle_request() errors
> > > > > >       virtio-net: handle virtio_net_handle_ctrl() error
> > > > > >       virtio-net: handle virtio_net_receive() errors
> > > > > >       virtio-net: handle virtio_net_flush_tx() errors
> > > > > >       virtio-scsi: convert virtio_scsi_bad_req() to use 
> > > > > > virtio_error()
> > > > > >       virtio-scsi: handle virtio_scsi_set_config() error
> > > > > >
> > > > > >
> > > > > >  hw/9pfs/virtio-9p-device.c |   14 ++++++++++--
> > > > > >  hw/block/virtio-blk.c      |   27 +++++++++++++++--------
> > > > > >  hw/net/virtio-net.c        |   51  
> > > > +++++++++++++++++++++++++-------------------  
> > > > > >  hw/scsi/virtio-scsi.c      |   21 ++++++++++--------
> > > > > >  4 files changed, 70 insertions(+), 43 deletions(-)
> > > > > >
> > > > > > --
> > > > > > Greg
> > > > > >
> > > > > >
> > > > > >
> > > > > >  
> > > > >  
> > >  
>