Re: [PATCH] device-assignment: Register as un-migratable

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 2586 bytes --]

Am 15.11.2010 21:25, Alex Williamson wrote:
> On Mon, 2010-11-15 at 21:05 +0100, Jan Kiszka wrote:
>> Am 15.11.2010 20:41, Alex Williamson wrote:
>>> Use register_device_unmigratable() to declare ourselves as
>>> non-migratable.
>>>
>>> Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
>>> ---
>>>
>>>  hw/device-assignment.c |   15 +++++++++++++++
>>>  1 files changed, 15 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/hw/device-assignment.c b/hw/device-assignment.c
>>> index bde231d..cd93941 100644
>>> --- a/hw/device-assignment.c
>>> +++ b/hw/device-assignment.c
>>> @@ -1434,6 +1434,13 @@ static void assigned_dev_unregister_msix_mmio(AssignedDevice *dev)
>>>      dev->msix_table_page = NULL;
>>>  }
>>>  
>>> +/* This should never get called, but we're required to create a save_state
>>> + * handler or else the no_migrate flag will never be checked. */
>>> +static void assigned_save(QEMUFile* f, void *opaque)
>>> +{
>>> +    abort();
>>> +}
>>> +
>>>  static int assigned_initfn(struct PCIDevice *pci_dev)
>>>  {
>>>      AssignedDevice *dev = DO_UPCAST(AssignedDevice, dev, pci_dev);
>>> @@ -1490,6 +1497,13 @@ static int assigned_initfn(struct PCIDevice *pci_dev)
>>>  
>>>      assigned_dev_load_option_rom(dev);
>>>      QLIST_INSERT_HEAD(&devs, dev, next);
>>> +
>>> +    /* Assigned devices are not migratable, register a save
>>> +     * state entry so that we can mark it unmigratable. */
>>> +    register_savevm(&dev->dev.qdev, "pci-assign", 0, 0,
>>> +                    assigned_save, NULL, dev);
>>> +    register_device_unmigratable(&dev->dev.qdev, "pci-assign", dev);
>>> +
>>
>> Isn't this expressible via some VMStateDescription? If not, that should
>> be changed first.
> 
> Nope, save state handlers aren't allowed to fail.  I tried to fix it:
> 
> http://lists.nongnu.org/archive/html/qemu-devel/2010-11/msg00417.html
> 
> (you can find more discussion in other branches of that subject)  I've
> succumbed to not getting that series in, so now I'm just trying to use
> the code as it exists.  Thanks,

Hmm, didn't get why you need that series for the purpose of no_migration
declaration. My point is:

struct VMStateDescription {
    const char *name;
    int version_id;
    int minimum_version_id;
    int minimum_version_id_old;
    int no_migrate; /* or 'flags' */
    ...

so that you can specify an empty vmstate with that flag set, and you do
not need to register/unregister things via to-be-deprecated service
calls. Or am I missing some subtle detail?

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 259 bytes --]