2022-04-20 07:22:40

by Bo Jiao

Subject: [PATCH] mt76: mt7915: report qos_ctl without ACK policy

From: Bo Jiao <[email protected]>

there is no need to report to the mac80211 layer the ACK policy which may
cause rx out-of-order issue, because the hardware rmac module already handle it

Signed-off-by: Bo Jiao <[email protected]>
---
drivers/net/wireless/mediatek/mt76/mt7915/mac.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index b47aea6..5080280 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -675,6 +675,11 @@ mt7915_mac_fill_rx(struct mt7915_dev *dev, struct sk_buff *skb)

fc = cpu_to_le16(FIELD_GET(MT_RXD6_FRAME_CONTROL, v0));
qos_ctl = FIELD_GET(MT_RXD8_QOS_CTL, v2);
+ /*
+ * there is no need to report the ACK policy to the
+ * mac80211 layer, because hw has already processed it.
+ */
+ qos_ctl &= ~IEEE80211_QOS_CTL_ACK_POLICY_MASK;
seq_ctrl = FIELD_GET(MT_RXD8_SEQ_CTRL, v2);

rxd += 4;
--
2.18.0

2022-04-21 00:39:35

by Felix Fietkau

Subject: Re: [PATCH] mt76: mt7915: report qos_ctl without ACK policy

On 20.04.22 06:03, Bo Jiao wrote:
> From: Bo Jiao <[email protected]>
>
> there is no need to report to the mac80211 layer the ACK policy which may
> cause rx out-of-order issue, because the hardware rmac module already handle it
>
> Signed-off-by: Bo Jiao <[email protected]>
> ---
> drivers/net/wireless/mediatek/mt76/mt7915/mac.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
> index b47aea6..5080280 100644
> --- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
> +++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
> @@ -675,6 +675,11 @@ mt7915_mac_fill_rx(struct mt7915_dev *dev, struct sk_buff *skb)
>
> fc = cpu_to_le16(FIELD_GET(MT_RXD6_FRAME_CONTROL, v0));
> qos_ctl = FIELD_GET(MT_RXD8_QOS_CTL, v2);
> + /*
> + * there is no need to report the ACK policy to the
> + * mac80211 layer, because hw has already processed it.
> + */
> + qos_ctl &= ~IEEE80211_QOS_CTL_ACK_POLICY_MASK;
I don't understand this change at all. First of all, the qos_ctl field
isn't really reported to the mac80211 layer, it's used in mt76 internally.
The rx reorder code uses it to detect no-ack policy in order to avoid
reordering for it.
Since you're masking out the ack policy, the value will be
IEEE80211_QOS_CTL_ACK_POLICY_NORMAL, which ends up forcibly enabling rx
reordering, even when the policy indicates no-ack.
How does this help, and what part does the rmac module already handle?

- Felix