[libvirt] [PATCH 1/2] vfio/mdev: add version field as mandatory attribute for mdev device

Wed May 15 11:18:46 UTC 2019

On 4/30/19 11:29 AM, Cornelia Huck wrote:
> On Wed, 24 Apr 2019 04:15:58 -0400
> Yan Zhao <yan.y.zhao at intel.com> wrote:
> 
>> On Wed, Apr 24, 2019 at 03:56:24PM +0800, Cornelia Huck wrote:
>>> On Tue, 23 Apr 2019 23:10:37 -0400
>>> Yan Zhao <yan.y.zhao at intel.com> wrote:
>>>    
>>>> On Tue, Apr 23, 2019 at 05:59:32PM +0800, Cornelia Huck wrote:
>>>>> On Fri, 19 Apr 2019 04:35:04 -0400
>>>>> Yan Zhao <yan.y.zhao at intel.com> wrote:
> 
>>>>>> @@ -225,6 +228,8 @@ Directories and files under the sysfs for Each Physical Device
>>>>>>     [<type-id>], device_api, and available_instances are mandatory attributes
>>>>>>     that should be provided by vendor driver.
>>>>>>
>>>>>> +  version is a mandatory attribute if a mdev device supports live migration.
>>>>>
>>>>> What about "An mdev device wishing to support live migration must
>>>>> provide the version attribute."?
>>>> yes, I just want to keep consistent with the line above it
>>>> " [<type-id>], device_api, and available_instances are mandatory attributes
>>>>    that should be provided by vendor driver."
>>>> what about below one?
>>>>    "version is a mandatory attribute if a mdev device wishing to support live
>>>>    migration."
>>>
>>> My point is that an attribute is not mandatory if it can be left out :)
>>> (I'm not a native speaker, though; maybe this makes perfect sense
>>> after all?)
>>>
>>> Maybe "version is a required attribute if live migration is supported
>>> for an mdev device"?

My two cents: This is the best of the suggestions

Tony Krowiak

>>>    
>> you are right, "mandatory" may bring some confusion.
>> Maybe
>> "vendor driver must provide version attribute for an mdev device wishing to
>> support live migration." ?
>> based on your first version :)
> 
> "The vendor driver must provide the version attribute for any mdev
> device it wishes to support live migration for." ?
> 
>>
>>>>
>>>>    
>>>>>> +
>>>>>>   * [<type-id>]
>>>>>>
>>>>>>     The [<type-id>] name is created by adding the device driver string as a prefix
>>>>>> @@ -246,6 +251,35 @@ Directories and files under the sysfs for Each Physical Device
>>>>>>     This attribute should show the number of devices of type <type-id> that can be
>>>>>>     created.
>>>>>>
>>>>>> +* version
>>>>>> +
>>>>>> +  This attribute is rw. It is used to check whether two devices are compatible
>>>>>> +  for live migration. If this attribute is missing, then the corresponding mdev
>>>>>> +  device is regarded as not supporting live migration.
>>>>>> +
>>>>>> +  It consists of two parts: common part and vendor proprietary part.
>>>>>> +  common part: 32 bit. lower 16 bits is vendor id and higher 16 bits identifies
>>>>>> +               device type. e.g., for pci device, it is
>>>>>> +               "pci vendor id" | (VFIO_DEVICE_FLAGS_PCI << 16).
>>>>>> +  vendor proprietary part: this part is varied in length. vendor driver can
>>>>>> +               specify any string to identify a device.
>>>>>> +
>>>>>> +  When reading this attribute, it should show device version string of the device
>>>>>> +  of type <type-id>. If a device does not support live migration, it should
>>>>>> +  return errno.
>>>>>> +  When writing a string to this attribute, it returns errno for incompatibility
>>>>>> +  or returns written string length in compatibility case. If a device does not
>>>>>> +  support live migration, it always returns errno.
>>>>>
>>>>> I'm not sure whether a device that does not support live migration
>>>>> should expose this attribute in the first place. Or is that to cover
>>>>> cases where a driver supports live migration only for some of the
>>>>> devices it supports?
>>>> yes, driver returning error code is to cover the cases where only part of devices it
>>>> supports can be migrated.
>>>>
>>>>    
>>>>> Also, I'm not sure if a string that has to be parsed is a good idea...
>>>>> is this 'version' attribute supposed to convey some human-readable
>>>>> information as well? The procedure you describe for compatibility
>>>>> checking does the checking within the vendor driver which I would
>>>>> expect to have a table/rules for that anyway.
>>>> right. if a vendor driver has the confidence to migrate between devices of
>>>> diffent platform or mdev types, it can maintain a compatibility table for that
>>>> purpose. That's the reason why we would leave the compatibility check to vendor
>>>> driver. vendor driver can freely choose its own complicated way to decide
>>>> which device is migratable to which device.
>>>
>>> I think there are two scenarios here:
>>> - Migrating between different device types, which is unlikely to work,
>>>    except in special cases.
>>> - Migrating between different versions of the same device type, which
>>>    may work for some drivers/devices (and at least migrating to a newer
>>>    version looks quite reasonable).
>>>
>>> But both should be something that is decided by the individual driver;
>>> I hope we don't want to support migration between different drivers :-O
>>>
>>> Can we make this a driver-defined format?
>>>   
>> yes, this is indeed driver-defined format.
>> Actually we define it into two parts: common part and vendor proprietary part.
>> common part: 32 bit. lower 16 bits is vendor id and higher 16 bits
>>               identifies device type. e.g., for pci device, it is
>>               "pci vendor id" | (VFIO_DEVICE_FLAGS_PCI << 16).
>> vendor proprietary part: this part is varied in length. vendor driver can
>>               specify any string to identify a device.
>>
>> vendor proprietary part is defined by vendor driver. vendor driver can
>> define any format it wishes to use. Also it is its own responsibility to
>> ensure backward compatibility if it wants to update format definition in this
>> part.
>>
>> So user space only needs to get source side's version string, and asks
>> target side whether the two are compatible. The decision maker is the
>> vendor driver:)
> 
> If I followed the discussion correctly, I think you plan to drop this
> format, don't you? I'd be happy if a vendor driver can use a simple
> number without any prefixes if it so chooses.
> 
> I also like the idea of renaming this "migration_version" so that it is
> clear we're dealing with versioning of the migration capability (and
> not a version of the device or so).
> 
> --
> libvir-list mailing list
> libvir-list at redhat.com
> https://www.redhat.com/mailman/listinfo/libvir-list
>