[PATCH net] net/sched: cls_flower: Set egress_dev mark when calling into the HW driver

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress
Hardware device') made sure (when fl_hw_replace_filter is called) to put
the egress_dev mark on persisent structure instance. Hence, following calls
into the HW driver for stats and deletion will note it and act accordingly.

With commit de4784ca030f this property is lost and hence when called,
the HW driver failes to operate (stats, delete) on the offloaded flow.

Fix it by setting the egress_dev flag whenever the ingress device is
different from the hw device since this is exactly the condition under
which we're calling into the HW driver through the egress port net-device.

Fixes: de4784ca030f ('net: sched: get rid of struct tc_to_netdev')
Signed-off-by: Or Gerlitz <redacted>
Signed-off-by: Roi Dayan <redacted>
---

Hi Dave, the bug was introduced in 4.14-rc1 but later the related
code was changed in net-next, hence the fix must not go to net-next, Or.

 net/sched/cls_flower.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/sched/cls_flower.c b/net/sched/cls_flower.c
index d230cb4..b480d7c 100644
--- a/net/sched/cls_flower.c
+++ b/net/sched/cls_flower.c

@@ -234,6 +234,7 @@ static void fl_hw_destroy_filter(struct tcf_proto *tp, struct cls_fl_filter *f)
 	tc_cls_common_offload_init(&cls_flower.common, tp);
 	cls_flower.command = TC_CLSFLOWER_DESTROY;
 	cls_flower.cookie = (unsigned long) f;
+	cls_flower.egress_dev = f->hw_dev != tp->q->dev_queue->dev;
 
 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSFLOWER, &cls_flower);
 }

@@ -289,6 +290,7 @@ static void fl_hw_update_stats(struct tcf_proto *tp, struct cls_fl_filter *f)
 	cls_flower.command = TC_CLSFLOWER_STATS;
 	cls_flower.cookie = (unsigned long) f;
 	cls_flower.exts = &f->exts;
+	cls_flower.egress_dev = f->hw_dev != tp->q->dev_queue->dev;
 
 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSFLOWER,
 				      &cls_flower);

-- 
2.3.7

Mon, Oct 16, 2017 at 11:19:48AM CEST, ogerlitz@mellanox.com wrote:

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress
Hardware device') made sure (when fl_hw_replace_filter is called) to put
the egress_dev mark on persisent structure instance. Hence, following calls
into the HW driver for stats and deletion will note it and act accordingly.

With commit de4784ca030f this property is lost and hence when called,
the HW driver failes to operate (stats, delete) on the offloaded flow.

Fix it by setting the egress_dev flag whenever the ingress device is
different from the hw device since this is exactly the condition under
which we're calling into the HW driver through the egress port net-device.

Fixes: de4784ca030f ('net: sched: get rid of struct tc_to_netdev')
Signed-off-by: Or Gerlitz <redacted>
Signed-off-by: Roi Dayan <redacted>

Acked-by: Jiri Pirko <redacted>

From: Or Gerlitz <redacted>
Date: Mon, 16 Oct 2017 12:19:48 +0300

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress
Hardware device') made sure (when fl_hw_replace_filter is called) to put
the egress_dev mark on persisent structure instance. Hence, following calls
into the HW driver for stats and deletion will note it and act accordingly.

With commit de4784ca030f this property is lost and hence when called,
the HW driver failes to operate (stats, delete) on the offloaded flow.

Fix it by setting the egress_dev flag whenever the ingress device is
different from the hw device since this is exactly the condition under
which we're calling into the HW driver through the egress port net-device.

Fixes: de4784ca030f ('net: sched: get rid of struct tc_to_netdev')
Signed-off-by: Or Gerlitz <redacted>
Signed-off-by: Roi Dayan <redacted>
---

Hi Dave, the bug was introduced in 4.14-rc1 but later the related
code was changed in net-next, hence the fix must not go to net-next, Or.

Ok, applied to 'net' and I'll watch out for this next time I merge into
net-next.

Thanks.

Hello!

On 10/16/2017 12:19 PM, Or Gerlitz wrote:

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress

    At least 12 digits are needed.

Hardware device') made sure (when fl_hw_replace_filter is called) to put
the egress_dev mark on persisent structure instance. Hence, following calls

    Persistent.

into the HW driver for stats and deletion will note it and act accordingly.

With commit de4784ca030f this property is lost and hence when called,
the HW driver failes to operate (stats, delete) on the offloaded flow.

    Fails.

Fix it by setting the egress_dev flag whenever the ingress device is
different from the hw device since this is exactly the condition under
which we're calling into the HW driver through the egress port net-device.

Fixes: de4784ca030f ('net: sched: get rid of struct tc_to_netdev')
Signed-off-by: Or Gerlitz <redacted>
Signed-off-by: Roi Dayan <redacted>

[...]

MBR, Sergei

On Tue, Oct 17, 2017 at 2:29 PM, Sergei Shtylyov
[off-list ref] wrote:

Hello!

On 10/16/2017 12:19 PM, Or Gerlitz wrote:

quoted

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress

   At least 12 digits are needed.

Also on such a spot (e.g not on the Fixes line)?

quoted

Hardware device') made sure (when fl_hw_replace_filter is called) to put
the egress_dev mark on persisent structure instance. Hence, following
calls

   Persistent.

right

quoted

into the HW driver for stats and deletion will note it and act
accordingly.

quoted

With commit de4784ca030f this property is lost and hence when called,
the HW driver failes to operate (stats, delete) on the offloaded flow.

   Fails.

correct

This is applied, will take more care next time

Hello!

On 10/18/2017 10:14 AM, Or Gerlitz wrote:

quoted

quoted

Commit 7091d8c '(net/sched: cls_flower: Add offload support using egress

    At least 12 digits are needed.

Also on such a spot (e.g not on the Fixes line)?

    Yes.

[...]

MBR, Sergei