Re: [dpdk-dev] [PATCH] net/mlx5: fix E-Switch flow without port item

[Yongseok Koh <yskoh@mellanox.com>]

> On Apr 25, 2019, at 12:02 AM, Ori Kam <orika@mellanox.com> wrote:
> 
> Hi Yongseok,
> PSB,
> 
> Ori
> 
>> -----Original Message-----
>> From: Yongseok Koh
>> Sent: Wednesday, April 24, 2019 12:03 AM
>> To: Ori Kam <orika@mellanox.com>
>> Cc: Shahaf Shuler <shahafs@mellanox.com>; Matan Azrad
>> <matan@mellanox.com>; dev@dpdk.org
>> Subject: Re: [PATCH] net/mlx5: fix E-Switch flow without port item
>> 
>> On Tue, Apr 23, 2019 at 11:19:16AM +0000, Ori Kam wrote:
>>> When creating a flow rule without the port_id pattern item, always the
>>> PF was selected.
>>> 
>>> This commit fixes this issue, if no port_id pattern item is available
>>> then we use the port that the flow was created on as source port.
>>> 
>>> Fixes: 822fb3195348 ("net/mlx5: add port id item to Direct Verbs")
>>> 
>>> Signed-off-by: Ori Kam <orika@mellanox.com>
>>> ---
>>> drivers/net/mlx5/mlx5_flow_dv.c | 30 +++++++++++++++---------------
>>> 1 file changed, 15 insertions(+), 15 deletions(-)
>>> 
>>> diff --git a/drivers/net/mlx5/mlx5_flow_dv.c
>> b/drivers/net/mlx5/mlx5_flow_dv.c
>>> index c2a2fc6..d17adbe 100644
>>> --- a/drivers/net/mlx5/mlx5_flow_dv.c
>>> +++ b/drivers/net/mlx5/mlx5_flow_dv.c
>>> @@ -3623,6 +3623,8 @@ struct field_modify_info modify_tcp[] = {
>>> 	union flow_dv_attr flow_attr = { .attr = 0 };
>>> 	struct mlx5_flow_dv_tag_resource tag_resource;
>>> 	uint32_t modify_action_position = UINT32_MAX;
>>> +	void *match_mask = matcher.mask.buf;
>>> +	void *match_value = dev_flow->dv.value.buf;
>>> 
>>> 	flow->group = attr->group;
>>> 	if (attr->transfer)
>>> @@ -3895,23 +3897,8 @@ struct field_modify_info modify_tcp[] = {
>>> 	}
>>> 	dev_flow->dv.actions_n = actions_n;
>>> 	flow->actions = action_flags;
>>> -	if (attr->ingress && !attr->transfer &&
>>> -	    (priv->representor || priv->master)) {
>>> -		/* It was validated - we support unidirection flows only. */
>>> -		assert(!attr->egress);
>>> -		/*
>>> -		 * Add matching on source vport index only
>>> -		 * for ingress rules in E-Switch configurations.
>>> -		 */
>>> -		flow_dv_translate_item_source_vport(matcher.mask.buf,
>>> -						    dev_flow->dv.value.buf,
>>> -						    priv->vport_id,
>>> -						    0xffff);
>>> -	}
>>> 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
>>> 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
>>> -		void *match_mask = matcher.mask.buf;
>>> -		void *match_value = dev_flow->dv.value.buf;
>>> 
>>> 		switch (items->type) {
>>> 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
>>> @@ -4018,6 +4005,19 @@ struct field_modify_info modify_tcp[] = {
>>> 		}
>>> 		item_flags |= last_item;
>>> 	}
>>> +	if (((attr->ingress && !attr->transfer) ||
>>> +	     (attr->transfer && !(item_flags & MLX5_FLOW_ITEM_PORT_ID)))
>> &&
>>> +	    (priv->representor || priv->master)) {
>> 
>> From the validations, I could figure out
>> - Either ingress (I) or egress (E) must be specified
>> - Transfer (T) can't be egress
> 0> - Port ID (P) is valid only if transfer (T) is specified.
>> 
>> (!T and I) or (T and !P)
>> = (I - T) + (T - P)
>> = I - P
>> 
>> So, this condition is equivalent to
>> 	if (attr->ingress && !!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
>> 	    (priv->representor || priv->master)) {
>> 		...
>> 	}
>> 
>> Right?

Right that was my typo.
Thanks,
Yongseok

>> 
> 
> You are right that we correnlty only support ingress rules for E-Switch, I want to keep it open if in future we
> will support also egress for E-Switch rules, but I guess we can update it when it will be relevant.
> Regarding the if you wrote there should be only one ! not 2 since this code is relevant only if the user
> didn't specified port_id.
> 
> Am I right?
> 
>> If agreed, please add comment properly.
>> 
>>> +		/* It was validated - we support unidirection flows only. */
>>> +		assert(!attr->egress);
>> 
>> This comment and assert are there to mention ingress and egress are
>> exclusive.
>> Is it still relevant? Did you also test the patch with enabling DEBUG?
>> 
> 
> I will remove this code.
> 
>>> +		/*
>>> +		 * Add matching on source vport index only
>>> +		 * for ingress rules in E-Switch configurations.
>>> +		 */
>> 
>> Please make this comment appropriate as well.
>> 
> 
> This comment is correct, due to the second part of the if (E-Switch mode is enabled, never mind if 
> it is E-Switch rule or Nic rule), but I will remove this comment and add it as part of the if updated comment.
> 
>> Thanks,
>> Yongseok
>> 
>>> +		if (flow_dv_translate_item_port_id(dev, match_mask,
>>> +						   match_value, NULL))
>>> +			return -rte_errno;
>>> +	}
>>> 	assert(!flow_dv_check_valid_spec(matcher.mask.buf,
>>> 					 dev_flow->dv.value.buf));
>>> 	dev_flow->layers = item_flags;
>>> --
>>> 1.8.3.1