* [PATCH 0/2] wifi: mt76: fix NAT performance degradation on MT799x
@ 2026-05-10 10:29 Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 1/2] wifi: mt76: connac: add MT7991A (0x7991) to is_mt7996() Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 2/2] wifi: mt76: mt7996: always fill txwi for 802.3 packets Dmitry Gomzyakov
0 siblings, 2 replies; 3+ messages in thread
From: Dmitry Gomzyakov @ 2026-05-10 10:29 UTC (permalink / raw)
To: linux-wireless; +Cc: Dmitry Gomzyakov
From: Dmitry Gomzyakov <nicerok11@gmail.com>
This series fixes severe IPv4 NAT download performance degradation
(30-40 Mbps limit) observed on MT799x platforms (MT7987A+MT7991A,
MT7992, MT7996).
Patch 1 adds MT7991A (0x7991) to is_mt7996() to ensure correct
chip-specific settings are used.
Patch 2 restores host-driven TXWI filling for all packets, fixing
firmware misinterpretation of NAT-modified IPv4 packets.
Dmitry Gomzyakov (2):
wifi: mt76: connac: add MT7991A (0x7991) to is_mt7996()
wifi: mt76: mt7996: always fill txwi for 802.3 packets
mt76_connac.h | 3 ++-
mt7996/mac.c | 12 +++++-------
2 files changed, 7 insertions(+), 8 deletions(-)
--
2.54.0
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH 1/2] wifi: mt76: connac: add MT7991A (0x7991) to is_mt7996()
2026-05-10 10:29 [PATCH 0/2] wifi: mt76: fix NAT performance degradation on MT799x Dmitry Gomzyakov
@ 2026-05-10 10:29 ` Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 2/2] wifi: mt76: mt7996: always fill txwi for 802.3 packets Dmitry Gomzyakov
1 sibling, 0 replies; 3+ messages in thread
From: Dmitry Gomzyakov @ 2026-05-10 10:29 UTC (permalink / raw)
To: linux-wireless; +Cc: Dmitry Gomzyakov
From: Dmitry Gomzyakov <nicerok11@gmail.com>
The MT7991A chipset uses PCI device ID 0x7991 (MT7996_DEVICE_ID_2),
but is_mt7996() only checks for 0x7990. This causes MT7991A devices
to use incorrect chip-specific settings, such as:
- MSDU_CNT_V2 instead of MSDU_CNT in TX descriptors
- Wrong WTBL BMC size (32 instead of 64)
- Incorrect prefetch depth for MCU queues
Fixes: 63a4edf1cb74 ("wifi: mt76: mt7996: add macros for pci decive id")
Signed-off-by: Dmitry Gomzyakov <nicerok11@gmail.com>
---
mt76_connac.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mt76_connac.h b/mt76_connac.h
index 51423c77..db07f3dc 100644
--- a/mt76_connac.h
+++ b/mt76_connac.h
@@ -230,7 +230,8 @@ static inline bool is_mt798x(struct mt76_dev *dev)
static inline bool is_mt7996(struct mt76_dev *dev)
{
- return mt76_chip(dev) == 0x7990;
+ u16 chip = mt76_chip(dev);
+ return chip == 0x7990 || chip == 0x7991;
}
static inline bool is_mt7992(struct mt76_dev *dev)
--
2.54.0
^ permalink raw reply related [flat|nested] 3+ messages in thread
* [PATCH 2/2] wifi: mt76: mt7996: always fill txwi for 802.3 packets
2026-05-10 10:29 [PATCH 0/2] wifi: mt76: fix NAT performance degradation on MT799x Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 1/2] wifi: mt76: connac: add MT7991A (0x7991) to is_mt7996() Dmitry Gomzyakov
@ 2026-05-10 10:29 ` Dmitry Gomzyakov
1 sibling, 0 replies; 3+ messages in thread
From: Dmitry Gomzyakov @ 2026-05-10 10:29 UTC (permalink / raw)
To: linux-wireless; +Cc: Dmitry Gomzyakov
From: Dmitry Gomzyakov <nicerok11@gmail.com>
Commit 55218e636092 reverted host-driven TXD filling to rely on
Software Defined Offload (SDO). However, SDO cannot correctly parse
IPv4 packets modified by NAT (changed TTL and recalculated IP/TCP
checksums), leaving TXWI mostly zero-filled.
This causes the firmware to misinterpret or drop packets,
resulting in severe throughput degradation (30-40 Mbps limit on
WiFi IPv4 download through NAT).
Always fill TXWI via mt7996_mac_write_txwi() and always set
MT_CT_INFO_APPLY_TXD, matching the behavior of mt7915 and the
proprietary driver.
Fixes: 55218e636092 ("Revert \"wifi: mt76: mt7996: fill txd by host driver\"")
Signed-off-by: Dmitry Gomzyakov <nicerok11@gmail.com>
---
mt7996/mac.c | 12 +++++-------
1 file changed, 5 insertions(+), 7 deletions(-)
diff --git a/mt7996/mac.c b/mt7996/mac.c
index 7d454e21..a219b063 100644
--- a/mt7996/mac.c
+++ b/mt7996/mac.c
@@ -1093,10 +1093,9 @@ int mt7996_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
pid = mt76_tx_status_skb_add(mdev, wcid, tx_info->skb);
memset(txwi_ptr, 0, MT_TXD_SIZE);
- /* Transmit non qos data by 802.11 header and need to fill txd by host*/
- if (!is_8023 || pid >= MT_PACKET_ID_FIRST)
- mt7996_mac_write_txwi(dev, txwi_ptr, tx_info->skb, wcid, key,
- pid, qid, 0);
+ /* Always fill txwi for firmware to avoid incorrect header processing */
+ mt7996_mac_write_txwi(dev, txwi_ptr, tx_info->skb, wcid, key,
+ pid, qid, 0);
/* MT7996 and MT7992 require driver to provide the MAC TXP for AddBA
* req
@@ -1145,9 +1144,8 @@ int mt7996_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
txp->fw.nbuf = nbuf;
txp->fw.flags = cpu_to_le16(MT_CT_INFO_FROM_HOST);
-
- if (!is_8023 || pid >= MT_PACKET_ID_FIRST)
- txp->fw.flags |= cpu_to_le16(MT_CT_INFO_APPLY_TXD);
+ /* Always apply TXD to ensure correct firmware processing */
+ txp->fw.flags |= cpu_to_le16(MT_CT_INFO_APPLY_TXD);
if (!key)
txp->fw.flags |= cpu_to_le16(MT_CT_INFO_NONE_CIPHER_FRAME);
--
2.54.0
^ permalink raw reply related [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-05-10 10:29 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-05-10 10:29 [PATCH 0/2] wifi: mt76: fix NAT performance degradation on MT799x Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 1/2] wifi: mt76: connac: add MT7991A (0x7991) to is_mt7996() Dmitry Gomzyakov
2026-05-10 10:29 ` [PATCH 2/2] wifi: mt76: mt7996: always fill txwi for 802.3 packets Dmitry Gomzyakov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox