On Fri, May 19, 2017 at 5:58 PM, Amritha Nambiar
<amritha.namb...@intel.com> wrote:
> This patch introduces a new hardware offload mode in mqprio
> which makes full use of the mqprio options, the TCs, the
> queue configurations and the bandwidth rates for the TCs.
> This is achieved by setting the value 2 for the "hw" option.
> This new offload mode supports new attributes for traffic
> class such as minimum and maximum values for bandwidth rate limits.
>
> Introduces a new datastructure 'tc_mqprio_qopt_offload' for offloading
> mqprio queue options and use this to be shared between the kernel and
> device driver. This contains a copy of the exisiting datastructure
> for mqprio queue options. This new datastructure can be extended when
> adding new attributes for traffic class such as bandwidth rate limits. The
> existing datastructure for mqprio queue options will be shared between the
> kernel and userspace.
>
> This patch enables configuring additional attributes associated
> with a traffic class such as minimum and maximum bandwidth
> rates and can be offloaded to the hardware in the new offload mode.
> The min and max limits for bandwidth rates are provided
> by the user along with the the TCs and the queue configurations
> when creating the mqprio qdisc.
>
> Example:
> # tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
> queues 4@0 4@4 min_rate 0Mbit 0Mbit max_rate 55Mbit 60Mbit hw 2
>
> To dump the bandwidth rates:
>
> # tc qdisc show dev eth0
> qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
> queues:(0:3) (4:7)
> min rates:0bit 0bit
> max rates:55Mbit 60Mbit
>
> Signed-off-by: Amritha Nambiar <amritha.namb...@intel.com>
> ---
> include/linux/netdevice.h | 2
> include/net/pkt_cls.h | 7 ++
> include/uapi/linux/pkt_sched.h | 13 +++
> net/sched/sch_mqprio.c | 169
> +++++++++++++++++++++++++++++++++++++---
> 4 files changed, 180 insertions(+), 11 deletions(-)
>
[...]
> diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c
> index 0a4cf27..6457ec9 100644
> --- a/net/sched/sch_mqprio.c
> +++ b/net/sched/sch_mqprio.c
> @@ -18,10 +18,13 @@
> #include <net/netlink.h>
> #include <net/pkt_sched.h>
> #include <net/sch_generic.h>
> +#include <net/pkt_cls.h>
>
> struct mqprio_sched {
> struct Qdisc **qdiscs;
> int hw_offload;
> + u32 flags;
> + u64 min_rate[TC_QOPT_MAX_QUEUE], max_rate[TC_QOPT_MAX_QUEUE];
> };
>
> static void mqprio_destroy(struct Qdisc *sch)
> @@ -39,10 +42,21 @@ static void mqprio_destroy(struct Qdisc *sch)
> }
>
> if (priv->hw_offload && dev->netdev_ops->ndo_setup_tc) {
> - struct tc_mqprio_qopt offload = { 0 };
> - struct tc_to_netdev tc = { .type = TC_SETUP_MQPRIO,
> - { .mqprio = &offload } };
> + struct tc_mqprio_qopt_offload offload = { 0 };
So this is currently throwing a warning when I pull these patches and
build this with gcc 6.2. I think the correct setup is "offload = {{ 0
}}" in order to indicate that we are initializing the inner structure
and all other data to 0.
> + struct tc_to_netdev tc = { 0 };
>
> + switch (priv->hw_offload) {
> + case TC_MQPRIO_HW_OFFLOAD_TCS:
> + tc.type = TC_SETUP_MQPRIO;
> + tc.mqprio = &offload.qopt;
> + break;
> + case TC_MQPRIO_HW_OFFLOAD:
> + tc.type = TC_SETUP_MQPRIO_EXT;
> + tc.mqprio_qopt = &offload;
> + break;
> + default:
> + return;
> + }
> dev->netdev_ops->ndo_setup_tc(dev, sch->handle, 0, &tc);
> } else {
> netdev_set_num_tc(dev, 0);
