mbox series

[v7,0/5] vhost: handle memory hotplug for async vhost

Message ID 20210722040907.20468-1-cheng1.jiang@intel.com (mailing list archive)
Headers show
Series vhost: handle memory hotplug for async vhost | expand

Message

Jiang, Cheng1 July 22, 2021, 4:09 a.m. UTC
When the guest memory is hotplugged, the vhost application which
enables DMA acceleration must stop DMA transfers before the vhost
re-maps the guest memory.

This patch set is to provide an unsafe API to drain inflight pkts
which are submitted to DMA engine in vhost async data path, and
notify the vhost application of stopping DMA transfers. And enable it
in vhost example.

v7:
 * rebased on the latest codes
 * improved commit log
v6:
 * removed unnecessary args for the new API
 * improved variable names and function names
 * added enable notification in set_mem_table
 * fixed vhost example queue clear process
v5:
 * added fixes in 'vhost: fix async vhost ops return type'
 * improved git log, variable names and logs
v4:
 * rebased on the latest codes
v3:
 * added a patch to fix async ops return type
 * fixed async ops fail handler
 * updated the doc
v2:
 * changed the patch structure

Cheng Jiang (4):
  vhost: fix async vhost ops return type
  vhost: add unsafe API to clear packets in async vhost
  examples/vhost: handle memory hotplug for async vhost
  doc: update doc for queue clear API in vhost lib

Jiayu Hu (1):
  vhost: handle memory hotplug for async vhost

 doc/guides/prog_guide/vhost_lib.rst    |   5 +
 doc/guides/rel_notes/release_21_08.rst |   5 +
 examples/vhost/ioat.c                  |   4 +-
 examples/vhost/ioat.h                  |   4 +-
 examples/vhost/main.c                  |  55 ++++++++-
 examples/vhost/main.h                  |   1 +
 lib/vhost/rte_vhost_async.h            |  30 ++++-
 lib/vhost/version.map                  |   1 +
 lib/vhost/vhost_user.c                 |  16 +++
 lib/vhost/virtio_net.c                 | 152 ++++++++++++++++++++-----
 10 files changed, 232 insertions(+), 41 deletions(-)

--
2.29.2

Comments

Xia, Chenbo July 22, 2021, 5:07 a.m. UTC | #1
> -----Original Message-----
> From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> Sent: Thursday, July 22, 2021 12:09 PM
> To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: dev@dpdk.org; Hu, Jiayu <jiayu.hu@intel.com>; Yang, YvonneX
> <yvonnex.yang@intel.com>; Jiang, Cheng1 <cheng1.jiang@intel.com>
> Subject: [PATCH v7 0/5] vhost: handle memory hotplug for async vhost
> 
> When the guest memory is hotplugged, the vhost application which
> enables DMA acceleration must stop DMA transfers before the vhost
> re-maps the guest memory.
> 
> This patch set is to provide an unsafe API to drain inflight pkts
> which are submitted to DMA engine in vhost async data path, and
> notify the vhost application of stopping DMA transfers. And enable it
> in vhost example.
> 
> v7:
>  * rebased on the latest codes
>  * improved commit log
> v6:
>  * removed unnecessary args for the new API
>  * improved variable names and function names
>  * added enable notification in set_mem_table
>  * fixed vhost example queue clear process
> v5:
>  * added fixes in 'vhost: fix async vhost ops return type'
>  * improved git log, variable names and logs
> v4:
>  * rebased on the latest codes
> v3:
>  * added a patch to fix async ops return type
>  * fixed async ops fail handler
>  * updated the doc
> v2:
>  * changed the patch structure
> 
> Cheng Jiang (4):
>   vhost: fix async vhost ops return type
>   vhost: add unsafe API to clear packets in async vhost
>   examples/vhost: handle memory hotplug for async vhost
>   doc: update doc for queue clear API in vhost lib
> 
> Jiayu Hu (1):
>   vhost: handle memory hotplug for async vhost
> 
>  doc/guides/prog_guide/vhost_lib.rst    |   5 +
>  doc/guides/rel_notes/release_21_08.rst |   5 +
>  examples/vhost/ioat.c                  |   4 +-
>  examples/vhost/ioat.h                  |   4 +-
>  examples/vhost/main.c                  |  55 ++++++++-
>  examples/vhost/main.h                  |   1 +
>  lib/vhost/rte_vhost_async.h            |  30 ++++-
>  lib/vhost/version.map                  |   1 +
>  lib/vhost/vhost_user.c                 |  16 +++
>  lib/vhost/virtio_net.c                 | 152 ++++++++++++++++++++-----
>  10 files changed, 232 insertions(+), 41 deletions(-)
> 
> --
> 2.29.2

Series applied to next-virtio/main. Thanks
Thomas Monjalon July 22, 2021, 4:12 p.m. UTC | #2
22/07/2021 07:07, Xia, Chenbo:
> From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > When the guest memory is hotplugged, the vhost application which
> > enables DMA acceleration must stop DMA transfers before the vhost
> > re-maps the guest memory.
> > 
> > This patch set is to provide an unsafe API to drain inflight pkts
> > which are submitted to DMA engine in vhost async data path, and
> > notify the vhost application of stopping DMA transfers. And enable it
> > in vhost example.
> 
> Series applied to next-virtio/main. Thanks

I cannot pull this series in main branch.

There is a compilation error seen on Arm cross-compilation:

examples/vhost/main.c:1493:51: error: assignment to 'int32_t (*)(int,  uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka 'int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned int)'} from incompatible pointer type 'uint32_t (*)(int,  uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka 'unsigned int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned int)'} [-Werror=incompatible-pointer-types]
 1493 |                         channel_ops.transfer_data = ioat_transfer_data_cb;
      |                                                   ^

Other comments about the last patch:
- it is updating doc out of the original patch doing the code changes
- there is not even a reference to the code patch (Fixes: line)
- the addition in the release notes is not sorted

Last question while at it, why having the API documentation
in the vhost guide (rst file)?
Doxygen is not enough to describe the functions?
Xia, Chenbo July 23, 2021, 5:06 a.m. UTC | #3
Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon <thomas@monjalon.net>
> Sent: Friday, July 23, 2021 12:13 AM
> To: Jiang, Cheng1 <cheng1.jiang@intel.com>; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: maxime.coquelin@redhat.com; dev@dpdk.org; Hu, Jiayu <jiayu.hu@intel.com>;
> Yang, YvonneX <yvonnex.yang@intel.com>; david.marchand@redhat.com
> Subject: Re: [dpdk-dev] [PATCH v7 0/5] vhost: handle memory hotplug for async
> vhost
> 
> 22/07/2021 07:07, Xia, Chenbo:
> > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > > When the guest memory is hotplugged, the vhost application which
> > > enables DMA acceleration must stop DMA transfers before the vhost
> > > re-maps the guest memory.
> > >
> > > This patch set is to provide an unsafe API to drain inflight pkts
> > > which are submitted to DMA engine in vhost async data path, and
> > > notify the vhost application of stopping DMA transfers. And enable it
> > > in vhost example.
> >
> > Series applied to next-virtio/main. Thanks
> 
> I cannot pull this series in main branch.
> 
> There is a compilation error seen on Arm cross-compilation:
> 
> examples/vhost/main.c:1493:51: error: assignment to 'int32_t (*)(int,
> uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *,
> uint16_t)' {aka 'int (*)(int,  short unsigned int,  struct
> rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned int)'}
> from incompatible pointer type 'uint32_t (*)(int,  uint16_t,  struct
> rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka
> 'unsigned int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *,
> struct rte_vhost_async_status *, short unsigned int)'} [-Werror=incompatible-
> pointer-types]
>  1493 |                         channel_ops.transfer_data =
> ioat_transfer_data_cb;
>       |                                                   ^

I see. @Cheng, please fix it in new version.

> 
> Other comments about the last patch:
> - it is updating doc out of the original patch doing the code changes
> - there is not even a reference to the code patch (Fixes: line)

I think the doc patch could be combined with the code patch in the same series.
But personally, sometimes I am not very clear when doc patch should be split.
For example, in this case we can combine as the update in release note is related
only to the code patch. What if it's related to multiple patch? Should we split or
add doc changes to every related patches? Just a bit confused. Maybe you can give
me some general guidance so that we will be on the same page.

> - the addition in the release notes is not sorted

Not very clear on this. The change is put in the bottom. Is there any sorting
rules?

> 
> Last question while at it, why having the API documentation
> in the vhost guide (rst file)?
> Doxygen is not enough to describe the functions?

Good point. To be honest, I have not thought about it :P

I think it could be moved to the doxygen later (maybe in another patch). The only
concern of mine is some API description in the vhost guide is a bit long.

@Maxime What do you think?

Thanks,
Chenbo

>
Thomas Monjalon July 23, 2021, 7:25 a.m. UTC | #4
23/07/2021 07:06, Xia, Chenbo:
> From: Thomas Monjalon <thomas@monjalon.net>
> > 22/07/2021 07:07, Xia, Chenbo:
> > > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > > > When the guest memory is hotplugged, the vhost application which
> > > > enables DMA acceleration must stop DMA transfers before the vhost
> > > > re-maps the guest memory.
> > > >
> > > > This patch set is to provide an unsafe API to drain inflight pkts
> > > > which are submitted to DMA engine in vhost async data path, and
> > > > notify the vhost application of stopping DMA transfers. And enable it
> > > > in vhost example.
> > >
> > > Series applied to next-virtio/main. Thanks
> > 
> > I cannot pull this series in main branch.
> > 
> > There is a compilation error seen on Arm cross-compilation:
> > 
> > examples/vhost/main.c:1493:51: error: assignment to 'int32_t (*)(int,
> > uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *,
> > uint16_t)' {aka 'int (*)(int,  short unsigned int,  struct
> > rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned int)'}
> > from incompatible pointer type 'uint32_t (*)(int,  uint16_t,  struct
> > rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka
> > 'unsigned int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *,
> > struct rte_vhost_async_status *, short unsigned int)'} [-Werror=incompatible-
> > pointer-types]
> >  1493 |                         channel_ops.transfer_data =
> > ioat_transfer_data_cb;
> >       |                                                   ^
> 
> I see. @Cheng, please fix it in new version.
> 
> > 
> > Other comments about the last patch:
> > - it is updating doc out of the original patch doing the code changes
> > - there is not even a reference to the code patch (Fixes: line)
> 
> I think the doc patch could be combined with the code patch in the same series.
> But personally, sometimes I am not very clear when doc patch should be split.
> For example, in this case we can combine as the update in release note is related
> only to the code patch. What if it's related to multiple patch? Should we split or
> add doc changes to every related patches? Just a bit confused. Maybe you can give
> me some general guidance so that we will be on the same page.

The doc must be updated in each patch.
Sometimes, the same line is updated to add a word related to the patch.

> > - the addition in the release notes is not sorted
> 
> Not very clear on this. The change is put in the bottom. Is there any sorting
> rules?

Read the comment at the beginning of the section, it explains
how things must be sorted:

     Suggested order in release notes items:
     * Core libs (EAL, mempool, ring, mbuf, buses)
     * Device abstraction libs and PMDs (ordered alphabetically by vendor name)
       - ethdev (lib, PMDs)
       - cryptodev (lib, PMDs)
       - eventdev (lib, PMDs)
       - etc
     * Other libs
     * Apps, Examples, Tools (if significant)

vhost is usually at the end of ethdev PMDs.

> > Last question while at it, why having the API documentation
> > in the vhost guide (rst file)?
> > Doxygen is not enough to describe the functions?
> 
> Good point. To be honest, I have not thought about it :P
> 
> I think it could be moved to the doxygen later (maybe in another patch). The only
> concern of mine is some API description in the vhost guide is a bit long.

So you can improve doxygen and remove this part of the guide.
The guide should be an overview, a tutorial and an internal design reference.

> @Maxime What do you think?
Xia, Chenbo July 23, 2021, 7:34 a.m. UTC | #5
Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon <thomas@monjalon.net>
> Sent: Friday, July 23, 2021 3:25 PM
> To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: Jiang, Cheng1 <cheng1.jiang@intel.com>; dev@dpdk.org; Hu, Jiayu
> <jiayu.hu@intel.com>; Yang, YvonneX <yvonnex.yang@intel.com>;
> david.marchand@redhat.com; Yigit, Ferruh <ferruh.yigit@intel.com>
> Subject: Re: [dpdk-dev] [PATCH v7 0/5] vhost: handle memory hotplug for async
> vhost
> 
> 23/07/2021 07:06, Xia, Chenbo:
> > From: Thomas Monjalon <thomas@monjalon.net>
> > > 22/07/2021 07:07, Xia, Chenbo:
> > > > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > > > > When the guest memory is hotplugged, the vhost application which
> > > > > enables DMA acceleration must stop DMA transfers before the vhost
> > > > > re-maps the guest memory.
> > > > >
> > > > > This patch set is to provide an unsafe API to drain inflight pkts
> > > > > which are submitted to DMA engine in vhost async data path, and
> > > > > notify the vhost application of stopping DMA transfers. And enable it
> > > > > in vhost example.
> > > >
> > > > Series applied to next-virtio/main. Thanks
> > >
> > > I cannot pull this series in main branch.
> > >
> > > There is a compilation error seen on Arm cross-compilation:
> > >
> > > examples/vhost/main.c:1493:51: error: assignment to 'int32_t (*)(int,
> > > uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *,
> > > uint16_t)' {aka 'int (*)(int,  short unsigned int,  struct
> > > rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned
> int)'}
> > > from incompatible pointer type 'uint32_t (*)(int,  uint16_t,  struct
> > > rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka
> > > 'unsigned int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *,
> > > struct rte_vhost_async_status *, short unsigned int)'} [-
> Werror=incompatible-
> > > pointer-types]
> > >  1493 |                         channel_ops.transfer_data =
> > > ioat_transfer_data_cb;
> > >       |                                                   ^
> >
> > I see. @Cheng, please fix it in new version.
> >
> > >
> > > Other comments about the last patch:
> > > - it is updating doc out of the original patch doing the code changes
> > > - there is not even a reference to the code patch (Fixes: line)
> >
> > I think the doc patch could be combined with the code patch in the same
> series.
> > But personally, sometimes I am not very clear when doc patch should be split.
> > For example, in this case we can combine as the update in release note is
> related
> > only to the code patch. What if it's related to multiple patch? Should we
> split or
> > add doc changes to every related patches? Just a bit confused. Maybe you can
> give
> > me some general guidance so that we will be on the same page.
> 
> The doc must be updated in each patch.
> Sometimes, the same line is updated to add a word related to the patch.

Thanks for the guidance!

> 
> > > - the addition in the release notes is not sorted
> >
> > Not very clear on this. The change is put in the bottom. Is there any
> sorting
> > rules?
> 
> Read the comment at the beginning of the section, it explains
> how things must be sorted:
> 
>      Suggested order in release notes items:
>      * Core libs (EAL, mempool, ring, mbuf, buses)
>      * Device abstraction libs and PMDs (ordered alphabetically by vendor name)
>        - ethdev (lib, PMDs)
>        - cryptodev (lib, PMDs)
>        - eventdev (lib, PMDs)
>        - etc
>      * Other libs
>      * Apps, Examples, Tools (if significant)
> 
> vhost is usually at the end of ethdev PMDs.

Oops.. I should notice it..

> 
> > > Last question while at it, why having the API documentation
> > > in the vhost guide (rst file)?
> > > Doxygen is not enough to describe the functions?
> >
> > Good point. To be honest, I have not thought about it :P
> >
> > I think it could be moved to the doxygen later (maybe in another patch). The
> only
> > concern of mine is some API description in the vhost guide is a bit long.
> 
> So you can improve doxygen and remove this part of the guide.
> The guide should be an overview, a tutorial and an internal design reference.

Make sense to me. For this patch, I suggest to keep the api doc in vhost guide.
Then I will send a patch to move them all if we all agree on this.

Thanks,
Chenbo

> 
> > @Maxime What do you think?
> 
>
Thomas Monjalon July 23, 2021, 7:39 a.m. UTC | #6
23/07/2021 09:34, Xia, Chenbo:
> From: Thomas Monjalon <thomas@monjalon.net>
> > 23/07/2021 07:06, Xia, Chenbo:
> > > From: Thomas Monjalon <thomas@monjalon.net>
> > > > 22/07/2021 07:07, Xia, Chenbo:
> > > > > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > > > > > When the guest memory is hotplugged, the vhost application which
> > > > > > enables DMA acceleration must stop DMA transfers before the vhost
> > > > > > re-maps the guest memory.
> > > > > >
> > > > > > This patch set is to provide an unsafe API to drain inflight pkts
> > > > > > which are submitted to DMA engine in vhost async data path, and
> > > > > > notify the vhost application of stopping DMA transfers. And enable it
> > > > > > in vhost example.
> > > > >
> > > > > Series applied to next-virtio/main. Thanks
> > > >
> > > > I cannot pull this series in main branch.
> > > >
> > > > There is a compilation error seen on Arm cross-compilation:
> > > >
> > > > examples/vhost/main.c:1493:51: error: assignment to 'int32_t (*)(int,
> > > > uint16_t,  struct rte_vhost_async_desc *, struct rte_vhost_async_status *,
> > > > uint16_t)' {aka 'int (*)(int,  short unsigned int,  struct
> > > > rte_vhost_async_desc *, struct rte_vhost_async_status *, short unsigned
> > int)'}
> > > > from incompatible pointer type 'uint32_t (*)(int,  uint16_t,  struct
> > > > rte_vhost_async_desc *, struct rte_vhost_async_status *, uint16_t)' {aka
> > > > 'unsigned int (*)(int,  short unsigned int,  struct rte_vhost_async_desc *,
> > > > struct rte_vhost_async_status *, short unsigned int)'} [-
> > Werror=incompatible-
> > > > pointer-types]
> > > >  1493 |                         channel_ops.transfer_data =
> > > > ioat_transfer_data_cb;
> > > >       |                                                   ^
> > >
> > > I see. @Cheng, please fix it in new version.
> > >
> > > >
> > > > Other comments about the last patch:
> > > > - it is updating doc out of the original patch doing the code changes
> > > > - there is not even a reference to the code patch (Fixes: line)
> > >
> > > I think the doc patch could be combined with the code patch in the same
> > series.
> > > But personally, sometimes I am not very clear when doc patch should be split.
> > > For example, in this case we can combine as the update in release note is
> > related
> > > only to the code patch. What if it's related to multiple patch? Should we
> > split or
> > > add doc changes to every related patches? Just a bit confused. Maybe you can
> > give
> > > me some general guidance so that we will be on the same page.
> > 
> > The doc must be updated in each patch.
> > Sometimes, the same line is updated to add a word related to the patch.
> 
> Thanks for the guidance!
> 
> > 
> > > > - the addition in the release notes is not sorted
> > >
> > > Not very clear on this. The change is put in the bottom. Is there any
> > sorting
> > > rules?
> > 
> > Read the comment at the beginning of the section, it explains
> > how things must be sorted:
> > 
> >      Suggested order in release notes items:
> >      * Core libs (EAL, mempool, ring, mbuf, buses)
> >      * Device abstraction libs and PMDs (ordered alphabetically by vendor name)
> >        - ethdev (lib, PMDs)
> >        - cryptodev (lib, PMDs)
> >        - eventdev (lib, PMDs)
> >        - etc
> >      * Other libs
> >      * Apps, Examples, Tools (if significant)
> > 
> > vhost is usually at the end of ethdev PMDs.
> 
> Oops.. I should notice it..
> 
> > 
> > > > Last question while at it, why having the API documentation
> > > > in the vhost guide (rst file)?
> > > > Doxygen is not enough to describe the functions?
> > >
> > > Good point. To be honest, I have not thought about it :P
> > >
> > > I think it could be moved to the doxygen later (maybe in another patch). The
> > only
> > > concern of mine is some API description in the vhost guide is a bit long.
> > 
> > So you can improve doxygen and remove this part of the guide.
> > The guide should be an overview, a tutorial and an internal design reference.
> 
> Make sense to me. For this patch, I suggest to keep the api doc in vhost guide.

Yes of course, don't change everything for this patch :)

> Then I will send a patch to move them all if we all agree on this.

Thank you.
Xia, Chenbo July 23, 2021, 8:03 a.m. UTC | #7
Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon <thomas@monjalon.net>
> Sent: Friday, July 23, 2021 3:40 PM
> To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: Jiang, Cheng1 <cheng1.jiang@intel.com>; dev@dpdk.org; Hu, Jiayu
> <jiayu.hu@intel.com>; Yang, YvonneX <yvonnex.yang@intel.com>;
> david.marchand@redhat.com; Yigit, Ferruh <ferruh.yigit@intel.com>
> Subject: Re: [dpdk-dev] [PATCH v7 0/5] vhost: handle memory hotplug for async
> vhost
> 
> 23/07/2021 09:34, Xia, Chenbo:
> > From: Thomas Monjalon <thomas@monjalon.net>
> > > 23/07/2021 07:06, Xia, Chenbo:
> > > > From: Thomas Monjalon <thomas@monjalon.net>
> > > > > 22/07/2021 07:07, Xia, Chenbo:
> > > > > > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > > > > > > When the guest memory is hotplugged, the vhost application which
> > > > > > > enables DMA acceleration must stop DMA transfers before the vhost
> > > > > > > re-maps the guest memory.
> > > > > > >

[...]

> > > Read the comment at the beginning of the section, it explains
> > > how things must be sorted:
> > >
> > >      Suggested order in release notes items:
> > >      * Core libs (EAL, mempool, ring, mbuf, buses)
> > >      * Device abstraction libs and PMDs (ordered alphabetically by vendor
> name)
> > >        - ethdev (lib, PMDs)
> > >        - cryptodev (lib, PMDs)
> > >        - eventdev (lib, PMDs)
> > >        - etc
> > >      * Other libs
> > >      * Apps, Examples, Tools (if significant)
> > >
> > > vhost is usually at the end of ethdev PMDs.
> >
> > Oops.. I should notice it..

I want to make clear for this. The release note is a vhost lib update which belongs
to '* Other libs '. Since there're no app/example/tool change, it just happens to be
the last item..

Thanks,
Chenbo
Thomas Monjalon July 23, 2021, 8:57 a.m. UTC | #8
23/07/2021 10:03, Xia, Chenbo:
> From: Thomas Monjalon <thomas@monjalon.net>
> > 23/07/2021 09:34, Xia, Chenbo:
> > > From: Thomas Monjalon <thomas@monjalon.net>
> > > > 23/07/2021 07:06, Xia, Chenbo:
> > > > > From: Thomas Monjalon <thomas@monjalon.net>
> > > > Read the comment at the beginning of the section, it explains
> > > > how things must be sorted:
> > > >
> > > >      Suggested order in release notes items:
> > > >      * Core libs (EAL, mempool, ring, mbuf, buses)
> > > >      * Device abstraction libs and PMDs (ordered alphabetically by vendor
> > name)
> > > >        - ethdev (lib, PMDs)
> > > >        - cryptodev (lib, PMDs)
> > > >        - eventdev (lib, PMDs)
> > > >        - etc
> > > >      * Other libs
> > > >      * Apps, Examples, Tools (if significant)
> > > >
> > > > vhost is usually at the end of ethdev PMDs.
> > >
> > > Oops.. I should notice it..
> 
> I want to make clear for this. The release note is a vhost lib update which belongs
> to '* Other libs '. Since there're no app/example/tool change, it just happens to be
> the last item..

vhost is an exception.
Most of the time it is about networking vhost,
so it fits better at the end of ethdev lib and PMDs in my opinion.
In the case there are net+crypto changes,
it can still be between ethdev and crypto.