recipes-wifi/linux-mt76/files/patches-3.x/0015-wifi-mt76-mt7996-add-L0.5-system-error-recovery-supp.patch

From 0809af5eefeebb30854486e1e928c06ed0a667d6 Mon Sep 17 00:00:00 2001
From: Bo Jiao <Bo.Jiao@mediatek.com>
Date: Mon, 13 Feb 2023 18:00:25 +0800
Subject: [PATCH 15/29] wifi: mt76: mt7996: add L0.5 system error recovery
 support

Signed-off-by: Bo Jiao <Bo.Jiao@mediatek.com>
---
 mt7996/debugfs.c | 155 +++++++++++++++++++++++---
 mt7996/dma.c     |  65 +++++++++++
 mt7996/init.c    |   9 +-
 mt7996/mac.c     | 285 +++++++++++++++++++++++++++++++++++++++--------
 mt7996/main.c    |  18 ++-
 mt7996/mcu.c     |  22 ++--
 mt7996/mcu.h     |  28 +++--
 mt7996/mmio.c    |   7 +-
 mt7996/mt7996.h  |  17 ++-
 mt7996/regs.h    |  36 +++++-
 10 files changed, 542 insertions(+), 100 deletions(-)

diff --git a/mt7996/debugfs.c b/mt7996/debugfs.c
index 9c5e9ac..f2c46a5 100644
--- a/mt7996/debugfs.c
+++ b/mt7996/debugfs.c
@@ -48,12 +48,12 @@ DEFINE_DEBUGFS_ATTRIBUTE(fops_implicit_txbf, mt7996_implicit_txbf_get,
 
 /* test knob of system error recovery */
 static ssize_t
-mt7996_fw_ser_set(struct file *file, const char __user *user_buf,
-		  size_t count, loff_t *ppos)
+mt7996_sys_recovery_set(struct file *file, const char __user *user_buf,
+			size_t count, loff_t *ppos)
 {
 	struct mt7996_phy *phy = file->private_data;
 	struct mt7996_dev *dev = phy->dev;
-	u8 band_idx = phy->mt76->band_idx;
+	bool band = phy->mt76->band_idx;
 	char buf[16];
 	int ret = 0;
 	u16 val;
@@ -73,17 +73,49 @@ mt7996_fw_ser_set(struct file *file, const char __user *user_buf,
 		return -EINVAL;
 
 	switch (val) {
-	case SER_SET_RECOVER_L1:
-	case SER_SET_RECOVER_L2:
-	case SER_SET_RECOVER_L3_RX_ABORT:
-	case SER_SET_RECOVER_L3_TX_ABORT:
-	case SER_SET_RECOVER_L3_TX_DISABLE:
-	case SER_SET_RECOVER_L3_BF:
-		ret = mt7996_mcu_set_ser(dev, SER_ENABLE, BIT(val), band_idx);
+	/*
+	 * 0: grab firmware current SER state.
+	 * 1: trigger & enable system error L1 recovery.
+	 * 2: trigger & enable system error L2 recovery.
+	 * 3: trigger & enable system error L3 rx abort.
+	 * 4: trigger & enable system error L3 tx abort.
+	 * 5: trigger & enable system error L3 tx disable.
+	 * 6: trigger & enable system error L3 bf recovery.
+	 * 7: trigger & enable system error L4 MDP recovery.
+	 * 8: trigger & enable system error full recovery.
+	 * 9: trigger firmware crash.
+	 */
+	case UNI_CMD_SER_QUERY:
+		ret = mt7996_mcu_set_ser(dev, UNI_CMD_SER_QUERY, 0, band);
+		break;
+	case UNI_CMD_SER_SET_RECOVER_L1:
+	case UNI_CMD_SER_SET_RECOVER_L2:
+	case UNI_CMD_SER_SET_RECOVER_L3_RX_ABORT:
+	case UNI_CMD_SER_SET_RECOVER_L3_TX_ABORT:
+	case UNI_CMD_SER_SET_RECOVER_L3_TX_DISABLE:
+	case UNI_CMD_SER_SET_RECOVER_L3_BF:
+	case UNI_CMD_SER_SET_RECOVER_L4_MDP:
+		ret = mt7996_mcu_set_ser(dev, UNI_CMD_SER_SET, BIT(val), band);
+		if (ret)
+			return ret;
+
+		ret = mt7996_mcu_set_ser(dev, UNI_CMD_SER_TRIGGER, val, band);
+		break;
+
+	/* enable full chip reset */
+	case UNI_CMD_SER_SET_RECOVER_FULL:
+		mt76_set(dev, MT_WFDMA0_MCU_HOST_INT_ENA, MT_MCU_CMD_WDT_MASK);
 		if (ret)
 			return ret;
 
-		ret = mt7996_mcu_set_ser(dev, SER_RECOVER, val, band_idx);
+		dev->recovery.state |= MT_MCU_CMD_WDT_MASK;
+		mt7996_reset(dev);
+		break;
+
+	/* WARNING: trigger firmware crash */
+	case UNI_CMD_SER_SET_SYSTEM_ASSERT:
+		mt76_wr(dev, MT_MCU_WM_CIRQ_EINT_MASK_CLR_ADDR, BIT(18));
+		mt76_wr(dev, MT_MCU_WM_CIRQ_EINT_SOFT_ADDR, BIT(18));
 		break;
 	default:
 		break;
@@ -92,9 +124,101 @@ mt7996_fw_ser_set(struct file *file, const char __user *user_buf,
 	return ret ? ret : count;
 }
 
-static const struct file_operations mt7996_fw_ser_ops = {
-	.write = mt7996_fw_ser_set,
-	/* TODO: ser read */
+static ssize_t
+mt7996_sys_recovery_get(struct file *file, char __user *user_buf,
+			size_t count, loff_t *ppos)
+{
+	struct mt7996_phy *phy = file->private_data;
+	struct mt7996_dev *dev = phy->dev;
+	char *buff;
+	int desc = 0;
+	ssize_t ret;
+	static const size_t bufsz = 1024;
+
+	buff = kmalloc(bufsz, GFP_KERNEL);
+	if (!buff)
+		return -ENOMEM;
+
+	/* HELP */
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "Please echo the correct value ...\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "0: grab firmware transient SER state\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "1: trigger system error L1 recovery\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "2: trigger system error L2 recovery\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "3: trigger system error L3 rx abort\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "4: trigger system error L3 tx abort\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "5: trigger system error L3 tx disable\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "6: trigger system error L3 bf recovery\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "7: trigger system error L4 MDP recovery\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "8: trigger system error full recovery\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "9: trigger firmware crash\n");
+
+	/* SER statistics */
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "\nlet's dump firmware SER statistics...\n");
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_STATUS        = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_SER_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_PLE_ERR       = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_PLE_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_PLE_ERR_1     = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_PLE1_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_PLE_ERR_AMSDU = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_PLE_AMSDU_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_PSE_ERR       = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_PSE_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_PSE_ERR_1     = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_PSE1_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR6_B0 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR6_BN0_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR6_B1 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR6_BN1_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR6_B2 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR6_BN2_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR7_B0 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR7_BN0_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR7_B1 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR7_BN1_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_LMAC_WISR7_B2 = 0x%08x\n",
+			  mt76_rr(dev, MT_SWDEF_LAMC_WISR7_BN2_STATS));
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "::E  R , SER_WFDMA_ERR = 0x%08x\n",
+			  mt76_rr(dev, WF_SWDEF_WFDMA_STATUS_ADDR));
+
+	desc += scnprintf(buff + desc, bufsz - desc,
+			  "\nSYS_RESET_COUNT: WM %d, WA %d\n",
+			  dev->recovery.wm_reset_count,
+			  dev->recovery.wa_reset_count);
+
+	ret = simple_read_from_buffer(user_buf, count, ppos, buff, desc);
+	kfree(buff);
+	return ret;
+}
+
+static const struct file_operations mt7996_sys_recovery_ops = {
+	.write = mt7996_sys_recovery_set,
+	.read = mt7996_sys_recovery_get,
 	.open = simple_open,
 	.llseek = default_llseek,
 };
@@ -674,6 +798,8 @@ int mt7996_init_debugfs(struct mt7996_phy *phy)
 	debugfs_create_file("xmit-queues", 0400, dir, phy,
 			    &mt7996_xmit_queues_fops);
 	debugfs_create_file("tx_stats", 0400, dir, phy, &mt7996_tx_stats_fops);
+	debugfs_create_file("sys_recovery", 0600, dir, phy,
+			    &mt7996_sys_recovery_ops);
 	debugfs_create_file("fw_debug_wm", 0600, dir, dev, &fops_fw_debug_wm);
 	debugfs_create_file("fw_debug_wa", 0600, dir, dev, &fops_fw_debug_wa);
 	debugfs_create_file("fw_debug_bin", 0600, dir, dev, &fops_fw_debug_bin);
@@ -684,7 +810,6 @@ int mt7996_init_debugfs(struct mt7996_phy *phy)
 			    &fops_implicit_txbf);
 	debugfs_create_devm_seqfile(dev->mt76.dev, "twt_stats", dir,
 				    mt7996_twt_stats);
-	debugfs_create_file("fw_ser", 0600, dir, phy, &mt7996_fw_ser_ops);
 	debugfs_create_file("rf_regval", 0600, dir, dev, &fops_rf_regval);
 
 	if (phy->mt76->cap.has_5ghz) {
diff --git a/mt7996/dma.c b/mt7996/dma.c
index c09fe42..18ea758 100644
--- a/mt7996/dma.c
+++ b/mt7996/dma.c
@@ -352,6 +352,71 @@ int mt7996_dma_init(struct mt7996_dev *dev)
 	return 0;
 }
 
+void mt7996_dma_reset(struct mt7996_dev *dev, bool force)
+{
+	struct mt76_phy *phy2 = dev->mt76.phys[MT_BAND1];
+	struct mt76_phy *phy3 = dev->mt76.phys[MT_BAND2];
+	u32 hif1_ofs = MT_WFDMA0_PCIE1(0) - MT_WFDMA0(0);
+	int i;
+
+	mt76_clear(dev, MT_WFDMA0_GLO_CFG,
+		   MT_WFDMA0_GLO_CFG_TX_DMA_EN |
+		   MT_WFDMA0_GLO_CFG_RX_DMA_EN);
+
+	if (dev->hif2)
+		mt76_clear(dev, MT_WFDMA0_GLO_CFG + hif1_ofs,
+			   MT_WFDMA0_GLO_CFG_TX_DMA_EN |
+			   MT_WFDMA0_GLO_CFG_RX_DMA_EN);
+
+	usleep_range(1000, 2000);
+
+	for (i = 0; i < __MT_TXQ_MAX; i++) {
+		mt76_queue_tx_cleanup(dev, dev->mphy.q_tx[i], true);
+		if (phy2)
+			mt76_queue_tx_cleanup(dev, phy2->q_tx[i], true);
+		if (phy3)
+			mt76_queue_tx_cleanup(dev, phy3->q_tx[i], true);
+	}
+
+	for (i = 0; i < __MT_MCUQ_MAX; i++)
+		mt76_queue_tx_cleanup(dev, dev->mt76.q_mcu[i], true);
+
+	mt76_for_each_q_rx(&dev->mt76, i)
+		//mt76_queue_rx_reset(dev, i);
+		mt76_queue_rx_cleanup(dev, &dev->mt76.q_rx[i]);
+
+	mt76_tx_status_check(&dev->mt76, true);
+
+	/* reset wfsys */
+	if (force)
+		mt7996_wfsys_reset(dev);
+
+	mt7996_dma_disable(dev, force);
+
+	/* reset hw queues */
+	for (i = 0; i < __MT_TXQ_MAX; i++) {
+		mt76_queue_reset(dev, dev->mphy.q_tx[i]);
+		if (phy2)
+			mt76_queue_reset(dev, phy2->q_tx[i]);
+		if (phy3)
+			mt76_queue_reset(dev, phy3->q_tx[i]);
+	}
+
+	for (i = 0; i < __MT_MCUQ_MAX; i++)
+		mt76_queue_reset(dev, dev->mt76.q_mcu[i]);
+
+	mt76_for_each_q_rx(&dev->mt76, i) {
+		mt76_queue_reset(dev, &dev->mt76.q_rx[i]);
+	}
+
+	mt76_tx_status_check(&dev->mt76, true);
+
+	mt76_for_each_q_rx(&dev->mt76, i)
+		mt76_queue_rx_reset(dev, i);
+
+	mt7996_dma_enable(dev);
+}
+
 void mt7996_dma_cleanup(struct mt7996_dev *dev)
 {
 	mt7996_dma_disable(dev, true);
diff --git a/mt7996/init.c b/mt7996/init.c
index 7350194..479b2ce 100644
--- a/mt7996/init.c
+++ b/mt7996/init.c
@@ -278,8 +278,7 @@ static void mt7996_led_set_brightness(struct led_classdev *led_cdev,
 		mt7996_led_set_config(led_cdev, 0xff, 0);
 }
 
-static void
-mt7996_init_txpower(struct mt7996_dev *dev,
+void mt7996_init_txpower(struct mt7996_dev *dev,
 		    struct ieee80211_supported_band *sband)
 {
 	int i, nss = hweight8(dev->mphy.antenna_mask);
@@ -434,7 +433,7 @@ mt7996_mac_init_band(struct mt7996_dev *dev, u8 band)
 	mt76_rmw(dev, MT_WTBLOFF_RSCR(band), mask, set);
 }
 
-static void mt7996_mac_init(struct mt7996_dev *dev)
+void mt7996_mac_init(struct mt7996_dev *dev)
 {
 #define HIF_TXD_V2_1	4
 	int i;
@@ -468,7 +467,7 @@ static void mt7996_mac_init(struct mt7996_dev *dev)
 		mt7996_mac_init_band(dev, i);
 }
 
-static int mt7996_txbf_init(struct mt7996_dev *dev)
+int mt7996_txbf_init(struct mt7996_dev *dev)
 {
 	int ret;
 
@@ -1080,6 +1079,8 @@ int mt7996_register_device(struct mt7996_dev *dev)
 	if (ret)
 		return ret;
 
+	dev->recovery.hw_init_done = true;
+
 	return mt7996_init_debugfs(&dev->phy);
 }
 
diff --git a/mt7996/mac.c b/mt7996/mac.c
index 8dc3a62..4c0c8f1 100644
--- a/mt7996/mac.c
+++ b/mt7996/mac.c
@@ -1724,7 +1724,7 @@ mt7996_wait_reset_state(struct mt7996_dev *dev, u32 state)
 	bool ret;
 
 	ret = wait_event_timeout(dev->reset_wait,
-				 (READ_ONCE(dev->reset_state) & state),
+				 (READ_ONCE(dev->recovery.state) & state),
 				 MT7996_RESET_TIMEOUT);
 
 	WARN(!ret, "Timeout waiting for MCU reset state %x\n", state);
@@ -1773,68 +1773,208 @@ mt7996_update_beacons(struct mt7996_dev *dev)
 					    mt7996_update_vif_beacon, phy3->hw);
 }
 
-static void
-mt7996_dma_reset(struct mt7996_dev *dev)
+void mt7996_tx_token_put(struct mt7996_dev *dev)
 {
-	struct mt76_phy *phy2 = dev->mt76.phys[MT_BAND1];
-	struct mt76_phy *phy3 = dev->mt76.phys[MT_BAND2];
-	u32 hif1_ofs = MT_WFDMA0_PCIE1(0) - MT_WFDMA0(0);
-	int i;
+	struct mt76_txwi_cache *txwi;
+	int id;
 
-	mt76_clear(dev, MT_WFDMA0_GLO_CFG,
-		   MT_WFDMA0_GLO_CFG_TX_DMA_EN |
-		   MT_WFDMA0_GLO_CFG_RX_DMA_EN);
+	spin_lock_bh(&dev->mt76.token_lock);
+	idr_for_each_entry(&dev->mt76.token, txwi, id) {
+		mt7996_txwi_free(dev, txwi, NULL, NULL);
+		dev->mt76.token_count--;
+	}
+	spin_unlock_bh(&dev->mt76.token_lock);
+	idr_destroy(&dev->mt76.token);
+}
 
-	if (dev->hif2)
-		mt76_clear(dev, MT_WFDMA0_GLO_CFG + hif1_ofs,
-			   MT_WFDMA0_GLO_CFG_TX_DMA_EN |
-			   MT_WFDMA0_GLO_CFG_RX_DMA_EN);
 
-	usleep_range(1000, 2000);
+static int
+mt7996_mac_restart(struct mt7996_dev *dev)
+{
+	struct mt7996_phy *phy2, *phy3;
+	struct mt76_dev *mdev = &dev->mt76;
+	int i, ret;
 
-	for (i = 0; i < __MT_TXQ_MAX; i++) {
-		mt76_queue_tx_cleanup(dev, dev->mphy.q_tx[i], true);
-		if (phy2)
-			mt76_queue_tx_cleanup(dev, phy2->q_tx[i], true);
-		if (phy3)
-			mt76_queue_tx_cleanup(dev, phy3->q_tx[i], true);
+	phy2 = mt7996_phy2(dev);
+	phy3 = mt7996_phy3(dev);
+
+	if (dev->hif2) {
+		mt76_wr(dev, MT_INT1_MASK_CSR, 0x0);
+		mt76_wr(dev, MT_INT1_SOURCE_CSR, ~0);
 	}
 
-	for (i = 0; i < __MT_MCUQ_MAX; i++)
-		mt76_queue_tx_cleanup(dev, dev->mt76.q_mcu[i], true);
+	if (dev_is_pci(mdev->dev)) {
+		mt76_wr(dev, MT_PCIE_MAC_INT_ENABLE, 0x0);
+		if (dev->hif2)
+			mt76_wr(dev, MT_PCIE1_MAC_INT_ENABLE, 0x0);
+	}
 
-	mt76_for_each_q_rx(&dev->mt76, i)
-		mt76_queue_rx_reset(dev, i);
+	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);
+	}
 
-	mt76_tx_status_check(&dev->mt76, true);
+	/* lock/unlock all queues to ensure that no tx is pending */
+	mt76_txq_schedule_all(&dev->mphy);
+	if (phy2)
+		mt76_txq_schedule_all(phy2->mt76);
+	if (phy3)
+		mt76_txq_schedule_all(phy3->mt76);
+
+	/* disable all tx/rx napi */
+	mt76_worker_disable(&dev->mt76.tx_worker);
+	mt76_for_each_q_rx(mdev, i) {
+		if (mdev->q_rx[i].ndesc)
+			napi_disable(&dev->mt76.napi[i]);
+	}
+	napi_disable(&dev->mt76.tx_napi);
 
-	/* re-init prefetch settings after reset */
-	mt7996_dma_prefetch(dev);
+	/* token reinit */
+	mt7996_tx_token_put(dev);
+	idr_init(&dev->mt76.token);
 
-	mt76_set(dev, MT_WFDMA0_GLO_CFG,
-		 MT_WFDMA0_GLO_CFG_TX_DMA_EN | MT_WFDMA0_GLO_CFG_RX_DMA_EN);
+	mt7996_dma_reset(dev, true);
 
-	if (dev->hif2)
-		mt76_set(dev, MT_WFDMA0_GLO_CFG + hif1_ofs,
-			 MT_WFDMA0_GLO_CFG_TX_DMA_EN |
-			 MT_WFDMA0_GLO_CFG_RX_DMA_EN);
+	local_bh_disable();
+	mt76_for_each_q_rx(mdev, i) {
+		if (mdev->q_rx[i].ndesc) {
+			napi_enable(&dev->mt76.napi[i]);
+			napi_schedule(&dev->mt76.napi[i]);
+		}
+	}
+	local_bh_enable();
+	clear_bit(MT76_MCU_RESET, &dev->mphy.state);
+	clear_bit(MT76_STATE_MCU_RUNNING, &dev->mphy.state);
+
+	mt76_wr(dev, MT_INT_MASK_CSR, dev->mt76.mmio.irqmask);
+	mt76_wr(dev, MT_INT_SOURCE_CSR, ~0);
+	if (dev->hif2) {
+		mt76_wr(dev, MT_INT1_MASK_CSR, dev->mt76.mmio.irqmask);
+		mt76_wr(dev, MT_INT1_SOURCE_CSR, ~0);
+	}
+	if (dev_is_pci(mdev->dev)) {
+		mt76_wr(dev, MT_PCIE_MAC_INT_ENABLE, 0xff);
+		if (dev->hif2)
+			mt76_wr(dev, MT_PCIE1_MAC_INT_ENABLE, 0xff);
+	}
+
+	/* load firmware */
+	ret = mt7996_mcu_init_firmware(dev);
+	if (ret)
+		goto out;
+
+	/* set the necessary init items */
+	ret = mt7996_mcu_set_eeprom(dev);
+	if (ret)
+		goto out;
+
+	mt7996_mac_init(dev);
+	mt7996_init_txpower(dev, &dev->mphy.sband_2g.sband);
+	mt7996_init_txpower(dev, &dev->mphy.sband_5g.sband);
+	mt7996_init_txpower(dev, &dev->mphy.sband_6g.sband);
+	ret = mt7996_txbf_init(dev);
+
+	if (test_bit(MT76_STATE_RUNNING, &dev->mphy.state)) {
+		ret = mt7996_run(dev->mphy.hw);
+		if (ret)
+			goto out;
+	}
+
+	if (phy2 && test_bit(MT76_STATE_RUNNING, &phy2->mt76->state)) {
+		ret = mt7996_run(phy2->mt76->hw);
+		if (ret)
+			goto out;
+	}
+
+	if (phy3 && test_bit(MT76_STATE_RUNNING, &phy3->mt76->state)) {
+		ret = mt7996_run(phy3->mt76->hw);
+		if (ret)
+			goto out;
+	}
+
+out:
+	/* reset done */
+	clear_bit(MT76_RESET, &dev->mphy.state);
+	if (phy2)
+		clear_bit(MT76_RESET, &phy2->mt76->state);
+	if (phy3)
+		clear_bit(MT76_RESET, &phy3->mt76->state);
+
+	local_bh_disable();
+	napi_enable(&dev->mt76.tx_napi);
+	napi_schedule(&dev->mt76.tx_napi);
+	local_bh_enable();
+
+	mt76_worker_enable(&dev->mt76.tx_worker);
+	return ret;
 }
 
-void mt7996_tx_token_put(struct mt7996_dev *dev)
+static void
+mt7996_mac_full_reset(struct mt7996_dev *dev)
 {
-	struct mt76_txwi_cache *txwi;
-	int id;
+	struct mt7996_phy *phy2, *phy3;
+	int i;
 
-	spin_lock_bh(&dev->mt76.token_lock);
-	idr_for_each_entry(&dev->mt76.token, txwi, id) {
-		mt7996_txwi_free(dev, txwi, NULL, NULL);
-		dev->mt76.token_count--;
+	phy2 = mt7996_phy2(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);
+
+	cancel_delayed_work_sync(&dev->mphy.mac_work);
+	if (phy2)
+		cancel_delayed_work_sync(&phy2->mt76->mac_work);
+	if (phy3)
+		cancel_delayed_work_sync(&phy3->mt76->mac_work);
+
+	mutex_lock(&dev->mt76.mutex);
+	for (i = 0; i < 10; i++) {
+		if (!mt7996_mac_restart(dev))
+			break;
 	}
-	spin_unlock_bh(&dev->mt76.token_lock);
-	idr_destroy(&dev->mt76.token);
+	mutex_unlock(&dev->mt76.mutex);
+
+	if (i == 10)
+		dev_err(dev->mt76.dev, "chip full reset failed\n");
+
+	ieee80211_restart_hw(mt76_hw(dev));
+	if (phy2)
+		ieee80211_restart_hw(phy2->mt76->hw);
+	if (phy3)
+		ieee80211_restart_hw(phy3->mt76->hw);
+
+	ieee80211_wake_queues(mt76_hw(dev));
+	if (phy2)
+		ieee80211_wake_queues(phy2->mt76->hw);
+	if (phy3)
+		ieee80211_wake_queues(phy3->mt76->hw);
+
+	dev->recovery.hw_full_reset = false;
+	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);
 }
 
-/* system error recovery */
 void mt7996_mac_reset_work(struct work_struct *work)
 {
 	struct mt7996_phy *phy2, *phy3;
@@ -1845,9 +1985,36 @@ void mt7996_mac_reset_work(struct work_struct *work)
 	phy2 = mt7996_phy2(dev);
 	phy3 = mt7996_phy3(dev);
 
-	if (!(READ_ONCE(dev->reset_state) & MT_MCU_CMD_STOP_DMA))
+	/* chip full reset */
+	if (dev->recovery.restart) {
+		/* disable WA/WM WDT */
+		mt76_clear(dev, MT_WFDMA0_MCU_HOST_INT_ENA,
+			   MT_MCU_CMD_WDT_MASK);
+
+		if (READ_ONCE(dev->recovery.state) & MT_MCU_CMD_WA_WDT)
+			dev->recovery.wa_reset_count++;
+		else
+			dev->recovery.wm_reset_count++;
+
+		mt7996_mac_full_reset(dev);
+
+		/* enable mcu irq */
+		mt7996_irq_enable(dev, MT_INT_MCU_CMD);
+		mt7996_irq_disable(dev, 0);
+
+		/* enable WA/WM WDT */
+		mt76_set(dev, MT_WFDMA0_MCU_HOST_INT_ENA, MT_MCU_CMD_WDT_MASK);
+
+		dev->recovery.state = MT_MCU_CMD_NORMAL_STATE;
+		dev->recovery.restart = false;
 		return;
+	}
 
+	if (!(READ_ONCE(dev->recovery.state) & MT_MCU_CMD_STOP_DMA))
+		return;
+
+	dev_info(dev->mt76.dev,"\n%s L1 SER recovery start.",
+		 wiphy_name(dev->mt76.hw->wiphy));
 	ieee80211_stop_queues(mt76_hw(dev));
 	if (phy2)
 		ieee80211_stop_queues(phy2->mt76->hw);
@@ -1876,7 +2043,7 @@ void mt7996_mac_reset_work(struct work_struct *work)
 	mt76_wr(dev, MT_MCU_INT_EVENT, MT_MCU_INT_EVENT_DMA_STOPPED);
 
 	if (mt7996_wait_reset_state(dev, MT_MCU_CMD_RESET_DONE)) {
-		mt7996_dma_reset(dev);
+		mt7996_dma_reset(dev, false);
 
 		mt7996_tx_token_put(dev);
 		idr_init(&dev->mt76.token);
@@ -1931,6 +2098,32 @@ void mt7996_mac_reset_work(struct work_struct *work)
 		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));
+}
+
+void mt7996_reset(struct mt7996_dev *dev)
+{
+	if (!dev->recovery.hw_init_done)
+		return;
+
+	if (dev->recovery.hw_full_reset)
+		return;
+
+	/* wm/wa exception: do full recovery */
+	if (READ_ONCE(dev->recovery.state) & MT_MCU_CMD_WDT_MASK) {
+		dev->recovery.restart = true;
+		dev_info(dev->mt76.dev,
+			 "%s indicated firmware crash, attempting recovery\n",
+			 wiphy_name(dev->mt76.hw->wiphy));
+
+		mt7996_irq_disable(dev, MT_INT_MCU_CMD);
+		queue_work(dev->mt76.wq, &dev->reset_work);
+		return;
+	}
+
+	queue_work(dev->mt76.wq, &dev->reset_work);
+	wake_up(&dev->reset_wait);
 }
 
 void mt7996_mac_update_stats(struct mt7996_phy *phy)
diff --git a/mt7996/main.c b/mt7996/main.c
index d8d578c..cb0e0d3 100644
--- a/mt7996/main.c
+++ b/mt7996/main.c
@@ -22,17 +22,13 @@ static bool mt7996_dev_running(struct mt7996_dev *dev)
 	return phy && test_bit(MT76_STATE_RUNNING, &phy->mt76->state);
 }
 
-static int mt7996_start(struct ieee80211_hw *hw)
+int mt7996_run(struct ieee80211_hw *hw)
 {
 	struct mt7996_dev *dev = mt7996_hw_dev(hw);
 	struct mt7996_phy *phy = mt7996_hw_phy(hw);
 	bool running;
 	int ret;
 
-	flush_work(&dev->init_work);
-
-	mutex_lock(&dev->mt76.mutex);
-
 	running = mt7996_dev_running(dev);
 	if (!running) {
 		ret = mt7996_mcu_set_hdr_trans(dev, true);
@@ -71,6 +67,18 @@ static int mt7996_start(struct ieee80211_hw *hw)
 		mt7996_mac_reset_counters(phy);
 
 out:
+	return ret;
+}
+
+static int mt7996_start(struct ieee80211_hw *hw)
+{
+	struct mt7996_dev *dev = mt7996_hw_dev(hw);
+	int ret;
+
+	flush_work(&dev->init_work);
+
+	mutex_lock(&dev->mt76.mutex);
+	ret = mt7996_run(hw);
 	mutex_unlock(&dev->mt76.mutex);
 
 	return ret;
diff --git a/mt7996/mcu.c b/mt7996/mcu.c
index 0dbe2e0..a949ff7 100644
--- a/mt7996/mcu.c
+++ b/mt7996/mcu.c
@@ -2629,17 +2629,10 @@ mt7996_mcu_init_rx_airtime(struct mt7996_dev *dev)
 				     MCU_WM_UNI_CMD(VOW), true);
 }
 
-int mt7996_mcu_init(struct mt7996_dev *dev)
+int mt7996_mcu_init_firmware(struct mt7996_dev *dev)
 {
-	static const struct mt76_mcu_ops mt7996_mcu_ops = {
-		.headroom = sizeof(struct mt76_connac2_mcu_txd), /* reuse */
-		.mcu_skb_send_msg = mt7996_mcu_send_message,
-		.mcu_parse_response = mt7996_mcu_parse_response,
-	};
 	int ret;
 
-	dev->mt76.mcu_ops = &mt7996_mcu_ops;
-
 	/* force firmware operation mode into normal state,
 	 * which should be set before firmware download stage.
 	 */
@@ -2680,6 +2673,19 @@ int mt7996_mcu_init(struct mt7996_dev *dev)
 				 MCU_WA_PARAM_RED, 0, 0);
 }
 
+int mt7996_mcu_init(struct mt7996_dev *dev)
+{
+	static const struct mt76_mcu_ops mt7996_mcu_ops = {
+		.headroom = sizeof(struct mt76_connac2_mcu_txd), /* reuse */
+		.mcu_skb_send_msg = mt7996_mcu_send_message,
+		.mcu_parse_response = mt7996_mcu_parse_response,
+	};
+
+	dev->mt76.mcu_ops = &mt7996_mcu_ops;
+
+	return mt7996_mcu_init_firmware(dev);
+}
+
 void mt7996_mcu_exit(struct mt7996_dev *dev)
 {
 	mt7996_mcu_restart(&dev->mt76);
diff --git a/mt7996/mcu.h b/mt7996/mcu.h
index ad66a1f..778deed 100644
--- a/mt7996/mcu.h
+++ b/mt7996/mcu.h
@@ -692,23 +692,21 @@ enum {
 };
 
 enum {
-	UNI_CMD_SER_QUERY = 0x0,
-	UNI_CMD_SER_SET = 0x2,
-	UNI_CMD_SER_TRIGGER = 0x3,
-};
-
-enum {
-	SER_QUERY,
+	UNI_CMD_SER_QUERY,
 	/* recovery */
-	SER_SET_RECOVER_L1,
-	SER_SET_RECOVER_L2,
-	SER_SET_RECOVER_L3_RX_ABORT,
-	SER_SET_RECOVER_L3_TX_ABORT,
-	SER_SET_RECOVER_L3_TX_DISABLE,
-	SER_SET_RECOVER_L3_BF,
+	UNI_CMD_SER_SET_RECOVER_L1,
+	UNI_CMD_SER_SET_RECOVER_L2,
+	UNI_CMD_SER_SET_RECOVER_L3_RX_ABORT,
+	UNI_CMD_SER_SET_RECOVER_L3_TX_ABORT,
+	UNI_CMD_SER_SET_RECOVER_L3_TX_DISABLE,
+	UNI_CMD_SER_SET_RECOVER_L3_BF,
+	UNI_CMD_SER_SET_RECOVER_L4_MDP,
+	UNI_CMD_SER_SET_RECOVER_FULL,
+	UNI_CMD_SER_SET_SYSTEM_ASSERT,
 	/* action */
-	SER_ENABLE = 2,
-	SER_RECOVER
+	UNI_CMD_SER_ENABLE = 1,
+	UNI_CMD_SER_SET,
+	UNI_CMD_SER_TRIGGER
 };
 
 enum {
diff --git a/mt7996/mmio.c b/mt7996/mmio.c
index 6610cc4..0e11f39 100644
--- a/mt7996/mmio.c
+++ b/mt7996/mmio.c
@@ -289,10 +289,9 @@ static void mt7996_irq_tasklet(struct tasklet_struct *t)
 		u32 val = mt76_rr(dev, MT_MCU_CMD);
 
 		mt76_wr(dev, MT_MCU_CMD, val);
-		if (val & MT_MCU_CMD_ERROR_MASK) {
-			dev->reset_state = val;
-			ieee80211_queue_work(mt76_hw(dev), &dev->reset_work);
-			wake_up(&dev->reset_wait);
+		if (val & (MT_MCU_CMD_ERROR_MASK | MT_MCU_CMD_WDT_MASK)) {
+			dev->recovery.state = val;
+			mt7996_reset(dev);
 		}
 	}
 }
diff --git a/mt7996/mt7996.h b/mt7996/mt7996.h
index 25b20fa..923e6fc 100644
--- a/mt7996/mt7996.h
+++ b/mt7996/mt7996.h
@@ -268,7 +268,14 @@ struct mt7996_dev {
 	struct work_struct rc_work;
 	struct work_struct reset_work;
 	wait_queue_head_t reset_wait;
-	u32 reset_state;
+	struct {
+		u32 state;
+		u32 wa_reset_count;
+		u32 wm_reset_count;
+		bool hw_full_reset:1;
+		bool hw_init_done:1;
+		bool restart:1;
+	} recovery;
 
 	struct list_head sta_rc_list;
 	struct list_head sta_poll_list;
@@ -401,9 +408,16 @@ int mt7996_eeprom_get_target_power(struct mt7996_dev *dev,
 				   struct ieee80211_channel *chan);
 s8 mt7996_eeprom_get_power_delta(struct mt7996_dev *dev, int band);
 int mt7996_dma_init(struct mt7996_dev *dev);
+void mt7996_dma_reset(struct mt7996_dev *dev, bool force);
 void mt7996_dma_prefetch(struct mt7996_dev *dev);
 void mt7996_dma_cleanup(struct mt7996_dev *dev);
+void mt7996_init_txpower(struct mt7996_dev *dev,
+		    struct ieee80211_supported_band *sband);
+int mt7996_txbf_init(struct mt7996_dev *dev);
+void mt7996_reset(struct mt7996_dev *dev);
+int mt7996_run(struct ieee80211_hw *hw);
 int mt7996_mcu_init(struct mt7996_dev *dev);
+int mt7996_mcu_init_firmware(struct mt7996_dev *dev);
 int mt7996_mcu_twt_agrt_update(struct mt7996_dev *dev,
 			       struct mt7996_vif *mvif,
 			       struct mt7996_twt_flow *flow,
@@ -496,6 +510,7 @@ static inline void mt7996_irq_disable(struct mt7996_dev *dev, u32 mask)
 		mt76_set_irq_mask(&dev->mt76, MT_INT_MASK_CSR, mask, 0);
 }
 
+void mt7996_mac_init(struct mt7996_dev *dev);
 u32 mt7996_mac_wtbl_lmac_addr(struct mt7996_dev *dev, u16 wcid, u8 dw);
 bool mt7996_mac_wtbl_update(struct mt7996_dev *dev, int idx, u32 mask);
 void mt7996_mac_reset_counters(struct mt7996_phy *phy);
diff --git a/mt7996/regs.h b/mt7996/regs.h
index 7a28cae..0775ca5 100644
--- a/mt7996/regs.h
+++ b/mt7996/regs.h
@@ -317,6 +317,8 @@ enum base_rev {
 #define MT_WFDMA0_RX_INT_PCIE_SEL		MT_WFDMA0(0x154)
 #define MT_WFDMA0_RX_INT_SEL_RING3		BIT(3)
 
+#define MT_WFDMA0_MCU_HOST_INT_ENA		MT_WFDMA0(0x1f4)
+
 #define MT_WFDMA0_GLO_CFG			MT_WFDMA0(0x208)
 #define MT_WFDMA0_GLO_CFG_TX_DMA_EN		BIT(0)
 #define MT_WFDMA0_GLO_CFG_RX_DMA_EN		BIT(2)
@@ -444,6 +446,10 @@ enum base_rev {
 #define MT_MCU_CMD_NORMAL_STATE			BIT(5)
 #define MT_MCU_CMD_ERROR_MASK			GENMASK(5, 1)
 
+#define MT_MCU_CMD_WA_WDT			BIT(31)
+#define MT_MCU_CMD_WM_WDT			BIT(30)
+#define MT_MCU_CMD_WDT_MASK			GENMASK(31, 30)
+
 /* l1/l2 remap */
 #define MT_HIF_REMAP_L1				0x155024
 #define MT_HIF_REMAP_L1_MASK			GENMASK(31, 16)
@@ -468,8 +474,27 @@ enum base_rev {
 #define MT_INFRA_MCU_END			0x7c3fffff
 
 /* FW MODE SYNC */
-#define MT_SWDEF_MODE				0x9143c
+#define MT_SWDEF_BASE				0x00401400
+
+#define MT_SWDEF(ofs)				(MT_SWDEF_BASE + (ofs))
+#define MT_SWDEF_MODE				MT_SWDEF(0x3c)
 #define MT_SWDEF_NORMAL_MODE			0
+#define MT_SWDEF_ICAP_MODE			1
+#define MT_SWDEF_SPECTRUM_MODE			2
+
+#define MT_SWDEF_SER_STATS			MT_SWDEF(0x040)
+#define MT_SWDEF_PLE_STATS			MT_SWDEF(0x044)
+#define MT_SWDEF_PLE1_STATS			MT_SWDEF(0x048)
+#define MT_SWDEF_PLE_AMSDU_STATS		MT_SWDEF(0x04C)
+#define MT_SWDEF_PSE_STATS			MT_SWDEF(0x050)
+#define MT_SWDEF_PSE1_STATS			MT_SWDEF(0x054)
+#define MT_SWDEF_LAMC_WISR6_BN0_STATS		MT_SWDEF(0x058)
+#define MT_SWDEF_LAMC_WISR6_BN1_STATS		MT_SWDEF(0x05C)
+#define MT_SWDEF_LAMC_WISR6_BN2_STATS		MT_SWDEF(0x060)
+#define MT_SWDEF_LAMC_WISR7_BN0_STATS		MT_SWDEF(0x064)
+#define MT_SWDEF_LAMC_WISR7_BN1_STATS		MT_SWDEF(0x068)
+#define MT_SWDEF_LAMC_WISR7_BN2_STATS		MT_SWDEF(0x06C)
+#define WF_SWDEF_WFDMA_STATUS_ADDR		MT_SWDEF(0x090)
 
 /* LED */
 #define MT_LED_TOP_BASE				0x18013000
@@ -506,7 +531,7 @@ enum base_rev {
 #define MT_TOP_MISC_FW_STATE			GENMASK(2, 0)
 
 #define MT_HW_REV				0x70010204
-#define MT_WF_SUBSYS_RST			0x70002600
+#define MT_WF_SUBSYS_RST			0x70028600
 
 /* PCIE MAC */
 #define MT_PCIE_MAC_BASE			0x74030000
@@ -539,4 +564,11 @@ enum base_rev {
 #define MT_WF_PHYRX_CSD_BAND_RXTD12_IRPI_SW_CLR_ONLY	BIT(18)
 #define MT_WF_PHYRX_CSD_BAND_RXTD12_IRPI_SW_CLR		BIT(29)
 
+#define MT_MCU_WM_CIRQ_BASE			0x89010000
+#define MT_MCU_WM_CIRQ(ofs)			(MT_MCU_WM_CIRQ_BASE + (ofs))
+#define MT_MCU_WM_CIRQ_IRQ_MASK_CLR_ADDR	MT_MCU_WM_CIRQ(0x80)
+#define MT_MCU_WM_CIRQ_IRQ_SOFT_ADDR		MT_MCU_WM_CIRQ(0xc0)
+#define MT_MCU_WM_CIRQ_EINT_MASK_CLR_ADDR	MT_MCU_WM_CIRQ(0x108)
+#define MT_MCU_WM_CIRQ_EINT_SOFT_ADDR		MT_MCU_WM_CIRQ(0x118)
+
 #endif
-- 
2.18.0