Date:	Fri, 26 Feb 2016 11:29:52 +0100
From:	Jiri Pirko <jiri@...nulli.us>
To:	John Fastabend <john.fastabend@...il.com>
Cc:	daniel@...earbox.net, simon.horman@...ronome.com,
	netdev@...r.kernel.org, alexei.starovoitov@...il.com,
	davem@...emloft.net, jhs@...atatu.com
Subject: Re: [net-next PATCH v2 3/3] net: sched: cls_u32 add bit to specify
 software only rules

Fri, Feb 26, 2016 at 12:20:45AM CET, john.fastabend@...il.com wrote:
>In the initial implementation the only way to stop a rule from being
>inserted into the hardware table was via the device feature flag.
>However this doesn't work well when working on an end host system
>where packets are expect to hit both the hardware and software
>datapaths.
>
>For example we can imagine a rule that will match an IP address and
>increment a field. If we install this rule in both hardware and
>software we may increment the field twice. To date we have only
>added support for the drop action so we have been able to ignore
>these cases. But as we extend the action support we will hit this
>example plus more such cases. Arguably these are not even corner
>cases in many working systems these cases will be common.
>
>To avoid forcing the driver to always abort (i.e. the above example)
>this patch adds a flag to add a rule in software only. A careful
>user can use this flag to build software and hardware datapaths
>that work together. One example we have found particularly useful
>is to use hardware resources to set the skb->mark on the skb when
>the match may be expensive to run in software but a mark lookup
>in a hash table is cheap. The idea here is hardware can do in one
>lookup what the u32 classifier may need to traverse multiple lists
>and hash tables to compute. The flag is only passed down on inserts
>on deletion to avoid stale references in hardware we always try
>to remove a rule if it exists.
>
>The flags field is part of the classifier specific options. Although
>it is tempting to lift this into the generic structure doing this
>proves difficult do to how the tc netlink attributes are implemented
>along with how the dump/change routines are called. There is also
>precedence for putting seemingly generic pieces in the specific
>classifier options such as TCA_U32_POLICE, TCA_U32_ACT, etc. So
>although not ideal I've left FLAGS in the u32 options as well as it
>simplifies the code greatly and user space has already learned how
>to manage these bits ala 'tc' tool.
>
>Another thing if trying to update a rule we require the flags to
>be unchanged. This is to force user space, software u32 and
>the hardware u32 to keep in sync. Thanks to Simon Horman for
>catching this case.
>
>Signed-off-by: John Fastabend <john.r.fastabend@...el.com>
>---
> include/net/pkt_cls.h        |   13 +++++++++++--
> include/uapi/linux/pkt_cls.h |    1 +
> net/sched/cls_u32.c          |   37 +++++++++++++++++++++++++++----------
> 3 files changed, 39 insertions(+), 12 deletions(-)
>
>diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
>index 6096e96..42dc412 100644
>--- a/include/net/pkt_cls.h
>+++ b/include/net/pkt_cls.h
>@@ -392,12 +392,21 @@ struct tc_cls_u32_offload {
> 	};
> };
> 
>-static inline bool tc_should_offload(struct net_device *dev)
>+/* tca flags definitions */
>+#define TCA_CLS_FLAGS_SOFTWARE 1

I'm sorry, the flag name is misleading to me.
We have by default, both SW and HW.

Now this flag should say: "do not push to HW".

In future, there will be another flag saying: "do not push to SW".

So I suggest what I already suggested before:

TCA_CLS_FLAGS_SKIP_HW for this one and
TCA_CLS_FLAGS_SKIP_KERNEL for the future one.

Sounds sane?

Powered by blists - more mailing lists