[dpdk-dev] [PATCH v3] ethdev: add isolated mode to flow API

Adrien Mazarguil adrien.mazarguil at 6wind.com
Wed Jun 14 15:35:29 CEST 2017


On Wed, Jun 14, 2017 at 04:01:46PM +0300, Andrew Rybchenko wrote:
> On 06/14/2017 03:45 PM, Adrien Mazarguil wrote:
> >Isolated mode can be requested by applications on individual ports to avoid
> >ingress traffic outside of the flow rules they define.
> >
> >Besides making ingress more deterministic, it allows PMDs to safely reuse
> >resources otherwise assigned to handle the remaining traffic, such as
> >global RSS configuration settings, VLAN filters, MAC address entries,
> >legacy filter API rules and so on in order to expand the set of possible
> >flow rule types.
> >
> >To minimize code complexity, PMDs implementing this mode may provide
> >partial (or even no) support for flow rules when not enabled (e.g. no
> >priorities, no RSS action). Applications written to use the flow API are
> >therefore encouraged to enable it.
> >
> >Once effective, leaving isolated mode may not be possible depending on PMD
> >implementation.
> >
> >Signed-off-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com>
> >Acked-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com>
> >
> >---
> >
> >v3:
> >- Rebased on next-net/master. Note this patch depends on
> >   commit c0688ef1eded ("net/igb: parse flow API n-tuple filter") due to a
> >   necessary fix in igb's rte_flow_ops definition to avoid a compilation
> >   issue.
> >
> >v2:
> >- Rebased on master.
> >---
> >  app/test-pmd/cmdline.c                      |  4 ++
> >  app/test-pmd/cmdline_flow.c                 | 49 ++++++++++++++++++++-
> >  app/test-pmd/config.c                       | 16 +++++++
> >  app/test-pmd/testpmd.h                      |  1 +
> >  doc/guides/prog_guide/rte_flow.rst          | 56 ++++++++++++++++++++++++
> >  doc/guides/testpmd_app_ug/testpmd_funcs.rst | 48 +++++++++++++++++++-
> >  drivers/net/e1000/igb_flow.c                |  9 ++--
> >  drivers/net/ixgbe/ixgbe_flow.c              |  9 ++--
> >  lib/librte_ether/rte_ether_version.map      |  7 +++
> >  lib/librte_ether/rte_flow.c                 | 18 ++++++++
> >  lib/librte_ether/rte_flow.h                 | 33 ++++++++++++++
> >  lib/librte_ether/rte_flow_driver.h          |  5 +++
> >  12 files changed, 242 insertions(+), 13 deletions(-)
> 
> <snip>
> 
> >diff --git a/doc/guides/prog_guide/rte_flow.rst b/doc/guides/prog_guide/rte_flow.rst
> >index b587ba9..699d2b2 100644
> >--- a/doc/guides/prog_guide/rte_flow.rst
> >+++ b/doc/guides/prog_guide/rte_flow.rst
> >@@ -1517,6 +1517,62 @@ Return values:
> >  - 0 on success, a negative errno value otherwise and ``rte_errno`` is set.
> >+Isolated mode
> >+-------------
> >+
> >+The general expectation for ingress traffic is that flow rules process it
> >+first; the remaining unmatched or pass-through traffic usually ends up in a
> >+queue (with or without RSS, locally or in some sub-device instance)
> >+depending on the global configuration settings of a port.
> >+
> >+While fine from a compatibility standpoint, this approach makes drivers more
> >+complex as they have to check for possible side effects outside of this API
> >+when creating or destroying flow rules. It results in a more limited set of
> >+available rule types due to the way device resources are assigned (e.g. no
> >+support for the RSS action even on capable hardware).
> >+
> >+Given that nonspecific traffic can be handled by flow rules as well,
> >+isolated mode is a means for applications to tell a driver that ingress on
> >+the underlying port must be injected from the defined flow rules only; that
> >+no default traffic is expected outside those rules.
> >+
> >+This has the following benefits:
> >+
> >+- Applications get finer-grained control over the kind of traffic they want
> >+  to receive (no traffic by default).
> >+
> >+- More importantly they control at what point nonspecific traffic is handled
> >+  relative to other flow rules, by adjusting priority levels.
> >+
> >+- Drivers can assign more hardware resources to flow rules and expand the
> >+  set of supported rule types.
> >+
> >+Because toggling isolated mode may cause profound changes to the ingress
> >+processing path of a driver, it may not be possible to leave it once
> >+entered. Likewise, existing flow rules or global configuration settings may
> >+prevent a driver from entering isolated mode.
> >+
> >+Applications relying on this mode are therefore encouraged to toggle it as
> >+soon as possible after device initialization, ideally before the first call
> >+to ``rte_eth_dev_configure()`` to avoid possible failures due to conflicting
> >+settings.
> >+
> 
> I think it would be useful to highlight how isolated mode coexists with
> promiscuous
> and all-multicast. What is the expected behaviour of the functions which
> toggle
> promiscuous and all-multicast mode if isolated mode is enabled? These
> functions
> return void right now, so it is impossible to return error. What should
> rte_eth_promiscuous_get() and  rte_eth_allmulticast_get() return?

They can technically return nothing/anything as long as they have no effect
on received traffic, as described.

Modifying existing wrappers that currently return void instead of an error
is outside the scope of this patch and requires ABI breakage. This can be
done later when the need arises.

For mlx4/mlx5, we plan to expose a different set of rte_eth_dev_ops
depending on whether isolated mode is toggled. When enabled, the
allmulti/promisc/MAC/VLAN/etc callbacks would be NULL for instance, and the
associated ethdev wrappers would automatically return an error where
applicable.

-- 
Adrien Mazarguil
6WIND


More information about the dev mailing list