Re: [RFC] net/sched/em_canid: Ematch rule to match CAN frames according to their CAN IDs

[Eric Dumazet <eric.dumazet@gmail.com>]

On Tue, 2012-06-12 at 15:48 +0200, Rostislav Lisovy wrote:
> em_canid is an ematch capable of classifying CAN frames according to
> their CAN IDs.
> 
> This RFC/Patch contains a reworked classifier initially posted in
> http://www.spinics.net/lists/netdev/msg200114.html
> The functionality is the same however there is almost 50% reduction
> in the source code length.
> 
> There is a slight difference between this ematch and other available
> ematches. Other ematches implement only a simple match operation and
> are meant to be combined with logic conjunctions (e.g. AND, OR).
> Our ematch makes it possible to use up to 32 rules in single
> 'configuration statement' (with OR semantics). This allows us to take
> the advantage of the bit field data-structure in the implementation of
> the match function.
> 
> Example: canid(sff 0x123 eff 0x124 sff 0x230:0x7f0)
> This ematch would match CAN SFF frames with the following IDs:
> 0x123, 0x230--0x23f or EFF frame with ID 0x124.
> 
> Signed-off-by: Rostislav Lisovy <lisovy@gmail.com>
> ---
>  include/linux/can.h     |    3 +
>  include/linux/pkt_cls.h |    5 +-
>  net/sched/Kconfig       |   10 +++
>  net/sched/Makefile      |    1 +
>  net/sched/em_canid.c    |  217 +++++++++++++++++++++++++++++++++++++++++++++++
>  5 files changed, 234 insertions(+), 2 deletions(-)
>  create mode 100644 net/sched/em_canid.c
> 
> diff --git a/include/linux/can.h b/include/linux/can.h
> index 9a19bcb..08d1610 100644
> --- a/include/linux/can.h
> +++ b/include/linux/can.h
> @@ -38,6 +38,9 @@
>   */
>  typedef __u32 canid_t;
>  
> +#define CAN_SFF_ID_BITS		11
> +#define CAN_EFF_ID_BITS		29
> +

> +struct canid_match {
> +	struct can_filter rules_raw[EM_CAN_RULES_SIZE]; /* Raw rules copied
> +		from netlink message; Used for sending information to
> +		userspace (when 'tc filter show' is invoked) AND when
> +		matching EFF frames*/
> +	DECLARE_BITMAP(match_sff, (1 << CAN_SFF_ID_BITS)); /* For each SFF CAN
> +		ID (11 bit) there is one record in this	bitfield */
> +	int rules_count;
> +	int eff_rules_count;
> +	int sff_rules_count;
> +
> +	struct rcu_head rcu;
> +};

The size of kmalloc() blob to hold this structure is 4 Mbytes

This is a huge cost, and unlikely to succeed but shortly after boot...

(this happens to be the largest possible kmalloc() allocation by the way
on x86)