From a9b56c1678204b1f67af27517a95ddfaa8c99618 Mon Sep 17 00:00:00 2001
From: Bo Jiao <Bo.Jiao@mediatek.com>
Date: Mon, 6 Nov 2023 11:10:10 +0800
Subject: [PATCH 1026/1044] mtk: wifi: mt76: try more times when send message
 timeout.

Signed-off-by: Bo Jiao <Bo.Jiao@mediatek.com>
---
 dma.c        |  7 ++++--
 mcu.c        | 65 ++++++++++++++++++++++++++++++++++++----------------
 mt7996/mac.c | 49 ++++++++++-----------------------------
 3 files changed, 62 insertions(+), 59 deletions(-)

diff --git a/dma.c b/dma.c
index 4cf2afdf..e4758e80 100644
--- a/dma.c
+++ b/dma.c
@@ -504,9 +504,12 @@ mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, struct mt76_queue *q,
 {
 	struct mt76_queue_buf buf = {};
 	dma_addr_t addr;
+	int ret = -ENOMEM;
 
-	if (test_bit(MT76_MCU_RESET, &dev->phy.state))
+	if (test_bit(MT76_MCU_RESET, &dev->phy.state)) {
+		ret = -EAGAIN;
 		goto error;
+	}
 
 	if (q->queued + 1 >= q->ndesc - 1)
 		goto error;
@@ -528,7 +531,7 @@ mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, struct mt76_queue *q,
 
 error:
 	dev_kfree_skb(skb);
-	return -ENOMEM;
+	return ret;
 }
 
 static int
diff --git a/mcu.c b/mcu.c
index fa4b0544..2926f715 100644
--- a/mcu.c
+++ b/mcu.c
@@ -4,6 +4,7 @@
  */
 
 #include "mt76.h"
+#include "mt76_connac.h"
 #include <linux/moduleparam.h>
 
 struct sk_buff *
@@ -74,35 +75,59 @@ int mt76_mcu_skb_send_and_get_msg(struct mt76_dev *dev, struct sk_buff *skb,
 				  int cmd, bool wait_resp,
 				  struct sk_buff **ret_skb)
 {
+#define MT76_MSG_MAX_RETRY_CNT 3
 	unsigned long expires;
-	int ret, seq;
+	int ret, seq, retry_cnt;
+	struct sk_buff *skb_tmp;
+	bool retry = wait_resp && is_mt7996(dev);
 
 	if (ret_skb)
 		*ret_skb = NULL;
 
 	mutex_lock(&dev->mcu.mutex);
-
-	ret = dev->mcu_ops->mcu_skb_send_msg(dev, skb, cmd, &seq);
-	if (ret < 0)
-		goto out;
-
-	if (!wait_resp) {
-		ret = 0;
-		goto out;
+	retry_cnt = retry ? MT76_MSG_MAX_RETRY_CNT : 1;
+	while (retry_cnt) {
+		skb_tmp = mt76_mcu_msg_alloc(dev, skb->data, skb->len);
+		if (!skb_tmp)
+			goto out;
+
+		if (retry && retry_cnt < MT76_MSG_MAX_RETRY_CNT) {
+			if (test_bit(MT76_MCU_RESET, &dev->phy.state))
+				usleep_range(200000, 500000);
+			dev_err(dev->dev, "send message %08x timeout, try again.\n", cmd);
+		}
+
+		ret = dev->mcu_ops->mcu_skb_send_msg(dev, skb_tmp, cmd, &seq);
+		if (ret < 0 && ret != -EAGAIN)
+			goto out;
+
+		if (!wait_resp) {
+			ret = 0;
+			goto out;
+		}
+
+		expires = jiffies + dev->mcu.timeout;
+
+		do {
+			skb_tmp = mt76_mcu_get_response(dev, expires);
+			ret = dev->mcu_ops->mcu_parse_response(dev, cmd, skb_tmp, seq);
+			if (ret == -ETIMEDOUT)
+				break;
+
+			if (!ret && ret_skb)
+				*ret_skb = skb_tmp;
+			else
+				dev_kfree_skb(skb_tmp);
+
+			if (ret != -EAGAIN)
+				goto out;
+		} while (ret == -EAGAIN);
+
+		retry_cnt--;
 	}
 
-	expires = jiffies + dev->mcu.timeout;
-
-	do {
-		skb = mt76_mcu_get_response(dev, expires);
-		ret = dev->mcu_ops->mcu_parse_response(dev, cmd, skb, seq);
-		if (!ret && ret_skb)
-			*ret_skb = skb;
-		else
-			dev_kfree_skb(skb);
-	} while (ret == -EAGAIN);
-
 out:
+	dev_kfree_skb(skb);
 	mutex_unlock(&dev->mcu.mutex);
 
 	return ret;
diff --git a/mt7996/mac.c b/mt7996/mac.c
index b3762728..7efe7db5 100644
--- a/mt7996/mac.c
+++ b/mt7996/mac.c
@@ -1663,18 +1663,6 @@ mt7996_mac_restart(struct mt7996_dev *dev)
 			mt76_wr(dev, MT_PCIE1_MAC_INT_ENABLE, 0x0);
 	}
 
-	set_bit(MT76_RESET, &dev->mphy.state);
-	set_bit(MT76_MCU_RESET, &dev->mphy.state);
-	wake_up(&dev->mt76.mcu.wait);
-	if (phy2) {
-		set_bit(MT76_RESET, &phy2->mt76->state);
-		set_bit(MT76_MCU_RESET, &phy2->mt76->state);
-	}
-	if (phy3) {
-		set_bit(MT76_RESET, &phy3->mt76->state);
-		set_bit(MT76_MCU_RESET, &phy3->mt76->state);
-	}
-
 	/* lock/unlock all queues to ensure that no tx is pending */
 	mt76_txq_schedule_all(&dev->mphy);
 	if (phy2)
@@ -1788,13 +1776,24 @@ mt7996_mac_full_reset(struct mt7996_dev *dev)
 	phy3 = mt7996_phy3(dev);
 	dev->recovery.hw_full_reset = true;
 
-	wake_up(&dev->mt76.mcu.wait);
 	ieee80211_stop_queues(mt76_hw(dev));
 	if (phy2)
 		ieee80211_stop_queues(phy2->mt76->hw);
 	if (phy3)
 		ieee80211_stop_queues(phy3->mt76->hw);
 
+	set_bit(MT76_RESET, &dev->mphy.state);
+	set_bit(MT76_MCU_RESET, &dev->mphy.state);
+	wake_up(&dev->mt76.mcu.wait);
+	if (phy2) {
+		set_bit(MT76_RESET, &phy2->mt76->state);
+		set_bit(MT76_MCU_RESET, &phy2->mt76->state);
+	}
+	if (phy3) {
+		set_bit(MT76_RESET, &phy3->mt76->state);
+		set_bit(MT76_MCU_RESET, &phy3->mt76->state);
+	}
+
 	cancel_work_sync(&dev->wed_rro.work);
 	cancel_delayed_work_sync(&dev->mphy.mac_work);
 	if (phy2)
@@ -1897,16 +1896,6 @@ void mt7996_mac_reset_work(struct work_struct *work)
 	set_bit(MT76_MCU_RESET, &dev->mphy.state);
 	wake_up(&dev->mt76.mcu.wait);
 
-	cancel_work_sync(&dev->wed_rro.work);
-	cancel_delayed_work_sync(&dev->mphy.mac_work);
-	if (phy2) {
-		set_bit(MT76_RESET, &phy2->mt76->state);
-		cancel_delayed_work_sync(&phy2->mt76->mac_work);
-	}
-	if (phy3) {
-		set_bit(MT76_RESET, &phy3->mt76->state);
-		cancel_delayed_work_sync(&phy3->mt76->mac_work);
-	}
 	mt76_worker_disable(&dev->mt76.tx_worker);
 	mt76_for_each_q_rx(&dev->mt76, i) {
 		if (mtk_wed_device_active(&dev->mt76.mmio.wed) &&
@@ -1917,8 +1906,6 @@ void mt7996_mac_reset_work(struct work_struct *work)
 	}
 	napi_disable(&dev->mt76.tx_napi);
 
-	mutex_lock(&dev->mt76.mutex);
-
 	mt76_wr(dev, MT_MCU_INT_EVENT, MT_MCU_INT_EVENT_DMA_STOPPED);
 
 	if (mt7996_wait_reset_state(dev, MT_MCU_CMD_RESET_DONE)) {
@@ -1991,20 +1978,8 @@ void mt7996_mac_reset_work(struct work_struct *work)
 	if (phy3)
 		ieee80211_wake_queues(phy3->mt76->hw);
 
-	mutex_unlock(&dev->mt76.mutex);
-
 	mt7996_update_beacons(dev);
 
-	ieee80211_queue_delayed_work(mt76_hw(dev), &dev->mphy.mac_work,
-				     MT7996_WATCHDOG_TIME);
-	if (phy2)
-		ieee80211_queue_delayed_work(phy2->mt76->hw,
-					     &phy2->mt76->mac_work,
-					     MT7996_WATCHDOG_TIME);
-	if (phy3)
-		ieee80211_queue_delayed_work(phy3->mt76->hw,
-					     &phy3->mt76->mac_work,
-					     MT7996_WATCHDOG_TIME);
 	dev_info(dev->mt76.dev,"\n%s L1 SER recovery completed.",
 		 wiphy_name(dev->mt76.hw->wiphy));
 }
-- 
2.18.0