Re: [PATCH v4 1/2] vhost: Add callback and private data for vhost PMD

[Tetsuya Mukawa <mukawa@igel.co.jp>]

On 2015/11/19 15:31, Yuanhan Liu wrote:
> On Thu, Nov 19, 2015 at 02:58:56PM +0900, Tetsuya Mukawa wrote:
>> On 2015/11/19 14:45, Yuanhan Liu wrote:
>>> On Thu, Nov 19, 2015 at 02:14:13PM +0900, Tetsuya Mukawa wrote:
>>>> On 2015/11/19 12:33, Yuanhan Liu wrote:
>>>>> On Thu, Nov 19, 2015 at 12:13:38PM +0900, Tetsuya Mukawa wrote:
>>>>>> On 2015/11/19 11:18, Yuanhan Liu wrote:
>>>>>>> On Thu, Nov 19, 2015 at 11:03:50AM +0900, Tetsuya Mukawa wrote:
>>>>>>>> On 2015/11/17 22:29, Yuanhan Liu wrote:
>>>>>>>>> On Fri, Nov 13, 2015 at 02:20:30PM +0900, Tetsuya Mukawa wrote:
>>>>>>>>>> These variables are needed to be able to manage one of virtio devices
>>>>>>>>>> using both vhost library APIs and vhost PMD.
>>>>>>>>>> For example, if vhost PMD uses current callback handler and private data
>>>>>>>>>> provided by vhost library, A DPDK application that links vhost library
>>>>>>>>>> cannot use some of vhost library APIs.
>>>>>>>>> Can you be more specific about this?
>>>>>>>>>
>>>>>>>>> 	--yliu
>>>>>>>> How about like below?
>>>>>>>>
>>>>>>>> commit log:
>>>>>>>> Currently, when virtio device is created and destroyed, vhost library
>>>>>>>> will call one of callback handlers.
>>>>>>>> The vhost PMD need to use this pair of callback handlers to know which
>>>>>>>> virtio devices are connected actually.
>>>>>>>> Because we can register only one pair of callbacks to vhost library, if
>>>>>>>> the PMD use it, DPDK applications
>>>>>>>> cannot have a way to know the events.
>>>>>>> Will (and why) the two co-exist at same time?
>>>>>> Yes it is. Sure, I will describe below in commit log.
>>>>>>
>>>>>> Because we cannot map some of vhost library APIs to ethdev APIs, in some
>>>>>> cases, we still
>>>>>> need to use vhost library APIs for a port created by the vhost PMD. One
>>>>>> of example is
>>>>>> rte_vhost_enable_guest_notification().
>>>>> I don't get it why it has something to do with a standalone PMD callback.
>>>>> And if you don't call rte_vhost_enable_guest_notification() inside vhost
>>>>> PMD, where else can you call that? I mean, you can't start vhost-pmd
>>>>> and vhost-swithc in the mean time, right?
>>>> No it's not true, even after connecting to virtio-net device, you can
>>>> change the flag.
>>>> It's just a hint for virtio-net driver, and it will be used while queuing.
>>>> (We may be able to change the flag, even while sending or receiving packets)
>>>>
>>>>> And, pmd callback and the old notify callback will not exist at same
>>>>> time in one case, right? If so, why is that needed?
>>>>>
>>>>> BTW, if it's a MUST, would you provide a specific example?
>>>> Actually, this patch is not a MUST.
>>>>
>>>> But still the users need callback handlers to know when virtio-net
>>>> device is connected or disconnected.
>>>> This is because the user can call rte_vhost_enable_guest_notification()
>>>> only while connection is established.
>>> What does "the user" mean? Is there a second user of vhost lib besides
>>> vhost PMD, that he has to interact with those connected devices? If so,
>>> how?
>> Sorry, my English is wrong.
>> Not a second user.
>>
>> For example, If DPDK application has a port created by vhost PMD, then
>> needs to call rte_vhost_enable_guest_notification() to the port.
> So, you are mixing the usage of vhost PMD and vhost lib in a DPDK
> application? Say,

Yes, that is my intention.
Using ethdev(PMD) APIs and some library specific APIs to a same port is
used in bonding PMD also.

Thanks,
Tetsuya

>         DPDK application
>                start_vhost_pmd
>                       rte_vhost_driver_pmd_callback_register
>                rte_vhost_driver_callback_register
>
> I know little about PMD, and not quite sure it's a good combo.
>
> Huawei, comments?
>
> 	--yliu
>
>> DPDK application needs to know when virtio-net device is connected or
>> disconnected, because the function is only valid while connecting.
>> But without callback handler, DPDK application cannot know it.
>>
>> This is what I wanted to explain.
>>
>> Thanks,
>> Tetsuya
>>
>>> 	--yliu
>>>> Probably we can use link status changed callback of the PMD for this
>>>> purpose.
>>>> (The vhost PMD will notice DPDK application using link status callback)
>>>>
>>>> But I am not sure whether we need to implement link status changed
>>>> callback for this purpose.
>>>> While processing this callback handler, the users will only calls vhost
>>>> library APIs that ethdev API cannot map, or store some variables related
>>>> with vhost library.
>>>> If so, this callback handler itself is specific for using vhost library.
>>>> And it may be ok that callback itself is implemented as one of vhost
>>>> library APIs.
>>>>
>>>> Tetsuya
>>>>
>>>>> 	--yliu
>>>>>> Thanks,
>>>>>> Tetsuya
>>>>>>
>>>>>>
>>>>>>> 	--yliu
>>>>>>>
>>>>>>>> This may break legacy DPDK
>>>>>>>> applications that uses vhost library.
>>>>>>>> To prevent it, this patch adds one more pair of callbacks to vhost
>>>>>>>> library especially for the vhost PMD.
>>>>>>>> With the patch, legacy applications can use the vhost PMD even if they
>>>>>>>> need additional specific handling
>>>>>>>> for virtio device creation and destruction.
>>>>>>>> For example, legacy application can call
>>>>>>>> rte_vhost_enable_guest_notification() in callbacks to change setting.
>>>>>>>>
>>>>>>>> Tetsuya