| Message ID | 20240620002741.1029936-1-kuba@kernel.org (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 4558645d139c01d5635cac7e3528a3d19bbb6526 |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next,v2] docs: net: document guidance of implementing the SR-IOV NDOs | expand |
On 6/19/24 5:27 PM, Jakub Kicinski wrote: > New drivers were prevented from adding ndo_set_vf_* callbacks > over the last few years. This was expected to result in broader > switchdev adoption, but seems to have had little effect. > > Based on recent netdev meeting there is broad support for allowing > adding those ops. > > There is a problem with the current API supporting a limited number > of VFs (100+, which is less than some modern HW supports). > We can try to solve it by adding similar functionality on devlink > ports, but that'd be another API variation to maintain. > So a netlink attribute reshuffling is a more likely outcome. > > Document the guidance, make it clear that the API is frozen. > > Signed-off-by: Jakub Kicinski <kuba@kernel.org> > --- > v2: > - edits from Randy > v1: https://lore.kernel.org/all/20240618192818.554646-1-kuba@kernel.org/ > > CC: corbet@lwn.net > CC: rdunlap@infradead.org > CC: linux-doc@vger.kernel.org > --- > Documentation/networking/index.rst | 1 + > Documentation/networking/sriov.rst | 25 +++++++++++++++++++++++++ > 2 files changed, 26 insertions(+) > create mode 100644 Documentation/networking/sriov.rst Reviewed-by: Randy Dunlap <rdunlap@infradead.org> Thanks.
On Wed, Jun 19, 2024 at 05:27:41PM -0700, Jakub Kicinski wrote: > New drivers were prevented from adding ndo_set_vf_* callbacks > over the last few years. This was expected to result in broader > switchdev adoption, but seems to have had little effect. > > Based on recent netdev meeting there is broad support for allowing > adding those ops. > > There is a problem with the current API supporting a limited number > of VFs (100+, which is less than some modern HW supports). > We can try to solve it by adding similar functionality on devlink > ports, but that'd be another API variation to maintain. > So a netlink attribute reshuffling is a more likely outcome. > > Document the guidance, make it clear that the API is frozen. > > Signed-off-by: Jakub Kicinski <kuba@kernel.org> Thanks Jakub, This is as was discussed at the public netdev call the other day [1]. And a change I supported in that meeting. Reviewed-by: Simon Horman <horms@kernel.org> [1] https://lore.kernel.org/netdev/20240617083955.1039b2e3@kernel.org/
Hello: This patch was applied to netdev/net-next.git (main) by David S. Miller <davem@davemloft.net>: On Wed, 19 Jun 2024 17:27:41 -0700 you wrote: > New drivers were prevented from adding ndo_set_vf_* callbacks > over the last few years. This was expected to result in broader > switchdev adoption, but seems to have had little effect. > > Based on recent netdev meeting there is broad support for allowing > adding those ops. > > [...] Here is the summary with links: - [net-next,v2] docs: net: document guidance of implementing the SR-IOV NDOs https://git.kernel.org/netdev/net-next/c/4558645d139c You are awesome, thank you!
diff --git a/Documentation/networking/index.rst b/Documentation/networking/index.rst index a6443851a142..b4b2a002f183 100644 --- a/Documentation/networking/index.rst +++ b/Documentation/networking/index.rst @@ -105,6 +105,7 @@ Refer to :ref:`netdev-FAQ` for a guide on netdev development process specifics. seg6-sysctl skbuff smc-sysctl + sriov statistics strparser switchdev diff --git a/Documentation/networking/sriov.rst b/Documentation/networking/sriov.rst new file mode 100644 index 000000000000..5deb4ff3154f --- /dev/null +++ b/Documentation/networking/sriov.rst @@ -0,0 +1,25 @@ +.. SPDX-License-Identifier: GPL-2.0 + +=============== +NIC SR-IOV APIs +=============== + +Modern NICs are strongly encouraged to focus on implementing the ``switchdev`` +model (see :ref:`switchdev`) to configure forwarding and security of SR-IOV +functionality. + +Legacy API +========== + +The old SR-IOV API is implemented in ``rtnetlink`` Netlink family as part of +the ``RTM_GETLINK`` and ``RTM_SETLINK`` commands. On the driver side +it consists of a number of ``ndo_set_vf_*`` and ``ndo_get_vf_*`` callbacks. + +Since the legacy APIs do not integrate well with the rest of the stack +the API is considered frozen; no new functionality or extensions +will be accepted. New drivers should not implement the uncommon callbacks; +namely the following callbacks are off limits: + + - ``ndo_get_vf_port`` + - ``ndo_set_vf_port`` + - ``ndo_set_vf_rss_query_en``
New drivers were prevented from adding ndo_set_vf_* callbacks over the last few years. This was expected to result in broader switchdev adoption, but seems to have had little effect. Based on recent netdev meeting there is broad support for allowing adding those ops. There is a problem with the current API supporting a limited number of VFs (100+, which is less than some modern HW supports). We can try to solve it by adding similar functionality on devlink ports, but that'd be another API variation to maintain. So a netlink attribute reshuffling is a more likely outcome. Document the guidance, make it clear that the API is frozen. Signed-off-by: Jakub Kicinski <kuba@kernel.org> --- v2: - edits from Randy v1: https://lore.kernel.org/all/20240618192818.554646-1-kuba@kernel.org/ CC: corbet@lwn.net CC: rdunlap@infradead.org CC: linux-doc@vger.kernel.org --- Documentation/networking/index.rst | 1 + Documentation/networking/sriov.rst | 25 +++++++++++++++++++++++++ 2 files changed, 26 insertions(+) create mode 100644 Documentation/networking/sriov.rst