[dpdk-dev] vhost: add virtio configuration space access socket messages

[Liu, Changpeng]

> -----Original Message-----
> From: Maxime Coquelin [mailto:maxime.coquelin at redhat.com]
> Sent: Wednesday, February 27, 2019 5:12 PM
> To: Liu, Changpeng <changpeng.liu at intel.com>; Ilya Maximets
> <i.maximets at samsung.com>; dev at dpdk.org
> Cc: Stojaczyk, Dariusz <dariusz.stojaczyk at intel.com>; Bie, Tiwei
> <tiwei.bie at intel.com>; Wang, Zhihong <zhihong.wang at intel.com>; Jason Wang
> <jasowang at redhat.com>
> Subject: Re: vhost: add virtio configuration space access socket messages
> 
> 
> 
> On 2/27/19 2:31 AM, Liu, Changpeng wrote:
> >
> >
> >> -----Original Message-----
> >> From: Maxime Coquelin [mailto:maxime.coquelin at redhat.com]
> >> Sent: Tuesday, February 26, 2019 8:32 PM
> >> To: Ilya Maximets <i.maximets at samsung.com>; Liu, Changpeng
> >> <changpeng.liu at intel.com>; dev at dpdk.org
> >> Cc: Stojaczyk, Dariusz <dariusz.stojaczyk at intel.com>; Bie, Tiwei
> >> <tiwei.bie at intel.com>; Wang, Zhihong <zhihong.wang at intel.com>; Jason
> Wang
> >> <jasowang at redhat.com>
> >> Subject: Re: vhost: add virtio configuration space access socket messages
> >>
> >>
> >>
> >> On 2/26/19 9:42 AM, Ilya Maximets wrote:
> >>> On 26.02.2019 11:13, Liu, Changpeng wrote:
> >>>>
> >>>>
> >>>>> -----Original Message-----
> >>>>> From: Ilya Maximets [mailto:i.maximets at samsung.com]
> >>>>> Sent: Tuesday, February 26, 2019 3:39 PM
> >>>>> To: Liu, Changpeng <changpeng.liu at intel.com>; dev at dpdk.org
> >>>>> Cc: Stojaczyk, Dariusz <dariusz.stojaczyk at intel.com>;
> >>>>> maxime.coquelin at redhat.com; Bie, Tiwei <tiwei.bie at intel.com>; Wang,
> >>>>> Zhihong <zhihong.wang at intel.com>; Jason Wang <jasowang at redhat.com>
> >>>>> Subject: Re: vhost: add virtio configuration space access socket messages
> >>>>>
> >>>>> On 26.02.2019 10:01, Liu, Changpeng wrote:
> >>>>>>
> >>>>>>
> >>>>>>> -----Original Message-----
> >>>>>>> From: Ilya Maximets [mailto:i.maximets at samsung.com]
> >>>>>>> Sent: Monday, February 25, 2019 9:20 PM
> >>>>>>> To: Liu, Changpeng <changpeng.liu at intel.com>; dev at dpdk.org
> >>>>>>> Cc: Stojaczyk, Dariusz <dariusz.stojaczyk at intel.com>;
> >>>>>>> maxime.coquelin at redhat.com; Bie, Tiwei <tiwei.bie at intel.com>; Wang,
> >>>>>>> Zhihong <zhihong.wang at intel.com>; Jason Wang
> <jasowang at redhat.com>
> >>>>>>> Subject: Re: vhost: add virtio configuration space access socket
> messages
> >>>>>>>
> >>>>>>> On 25.02.2019 10:51, Changpeng Liu wrote:
> >>>>>>>> This patch adds new vhost user messages GET_CONFIG and
> SET_CONFIG
> >>>>>>>> used to get/set virtio device's PCI configuration space.
> >>>>>>>
> >>>>>>> Beside the fact that some additional description and reasoning required,
> >>>>>>> I do not see the usage of this feature. You're defining the flag
> >>>>>>> VHOST_USER_PROTOCOL_F_CONFIG, but it's never used. So, none of
> >> dpdk
> >>>>> vhost
> >>>>>>> backends (vdpa, vhost-user) will use this feature.
> >>>>>>> You, probably, missed adding it to VHOST_USER_PROTOCOL_FEATURES
> or
> >>>>>>> VDPA_SUPPORTED_PROTOCOL_FEATURES.
> >>>>>>>
> >>>>>>>   From the other side, current implementation forces application to
> >> properly
> >>>>>>> implement the get/set_config callbacks. Otherwise, receiving of the
> >> messages
> >>>>>>> will result in RTE_VHOST_MSG_RESULT_ERR and subsequent vhost
> >>>>>>> disconnection.
> >>>>>>> This looks strange, because supported protocol features normally
> enabled
> >> by
> >>>>>>> default. Am I misunderstood something ?
> >>>>>> QEMU will not send the messages if VHOST_USER_PROTOCOL_F_CONFIG
> >>>>> wasn't enabled.
> >>>>>
> >>>>> So, you're going to enable it only by explicit call to
> >>>>> 'rte_vhost_driver_set_features' ?
> >>>>>
> >>>>> In this case I'm assuming that you're implementing your own vhost
> backend.
> >>>>> But why you're not using 'dev->extern_ops' and corresponding
> >> 'pre_msg_handle'
> >>>>> or 'post_msg_handle' to handle your GET/SET_CONFIG messages like it
> does
> >>>>> 'vhost_crypto' backend ?
> >>>> The patch was developed one year ago, while DPDK didn't have external
> ops.
> >>>
> >>> So, maybe it's time to reconsider the implementation.
> >>
> >> +1
> > Okay, I will only add related messages definition in this patch and remove the
> > Callbacks.
> 
> What we need to do is fix vhost lib so that you don't have anything to
> do in dpdk to add support for the new requests.
> 
> So we need to fix the few bits in vhost_user_msg_handler() I mentionned
> yesterday. And I also notice we are missing
> rte_vhost_driver_set_protocol_feature(), so that you can advertise
> VHOST_USER_PROTOCOL_F_CONFIG support by your external backend.
Sounds good to me, I will not post a new version and wait for the new API.
> 
> I'll try to cook the dpdk patch today.
> 
> Cheers,
> Maxime
> 
> >>
> >>>> The get_config/set_config was defined for all the virtio devices, so I think it
> >> makes
> >>>> more sense adding here.
> >>>
> >>> VHOST_USER_*_CRYPTO_SESSION messages are defined for all the virtio
> >> devices
> >>> too, however they makes sense for vhost_crypto backend only. These
> >> messages
> >>> (GET/SET_CONFIG) makes sense only when callbacks (get/set_config) are
> >>> implemented, so IMHO it's better to implement their handlers along with
> the
> >>> callbacks, i.e. inside the implementation of your vhost backend.
> >>>
> >>> Maxime, Tiwei, what do you think ?
> >>
> >> I would prefer it to be implemented in SPDK directly as a pre_handler
> >> callback, as I don't foresee a need for it for other backends, and it
> >> would avoid breaking the API.
> >>
> >> It would imply fixing the beginning of vhost_user_msg_handler() to
> >> accept requests > VHOST_USER_MAX and add necessary check before doing
> >> the debug logs.
> >>
> >> With above change we would also be able to remove VHOST_CRYPTO
> requests
> >> from vhost_user.c, and we could then work on moving vhost-net bits
> >> out of this file too.
> >>
> >> Regards,
> >> Maxime
> >