wifi: mt76: mt7915: enable coredump support

Host triggered and catastrophic event triggered firmware core dumping
for basic firmware issues triage, including state reporting, task/irq
info, function calltrace and MCU memory dump.
Signed-off-by: Ryder Lee <ryder.lee@mediatek.com>
Signed-off-by: Felix Fietkau <nbd@nbd.name>

drivers/net/wireless/mediatek/mt76/mt7915/Kconfig

@@ -2,6 +2,7 @@
 config MT7915E
 	tristate "MediaTek MT7915E (PCIe) support"
 	select MT76_CONNAC_LIB
+	select WANT_DEV_COREDUMP
 	depends on MAC80211
 	depends on PCI
 	select RELAY

drivers/net/wireless/mediatek/mt76/mt7915/Makefile

@@ -7,3 +7,4 @@ mt7915e-y := pci.o init.o dma.o eeprom.o main.o mcu.o mac.o \
 
 mt7915e-$(CONFIG_NL80211_TESTMODE) += testmode.o
 mt7915e-$(CONFIG_MT7986_WMAC) += soc.o
+mt7915e-$(CONFIG_DEV_COREDUMP) += coredump.o

drivers/net/wireless/mediatek/mt76/mt7915/coredump.h

+/* SPDX-License-Identifier: ISC */
+/* Copyright (C) 2022 MediaTek Inc. */
+
+#ifndef _COREDUMP_H_
+#define _COREDUMP_H_
+
+#include "mt7915.h"
+
+struct trace {
+	u32 id;
+	u32 timestamp;
+};
+
+struct mt7915_coredump {
+	char magic[16];
+
+	u32 len;
+
+	guid_t guid;
+
+	/* time-of-day stamp */
+	u64 tv_sec;
+	/* time-of-day stamp, nano-seconds */
+	u64 tv_nsec;
+	/* kernel version */
+	char kernel[64];
+	/* firmware version */
+	char fw_ver[ETHTOOL_FWVERS_LEN];
+
+	u32 device_id;
+
+	/* exception state */
+	char fw_state[12];
+
+	u32 last_msg_id;
+	u32 eint_info_idx;
+	u32 irq_info_idx;
+	u32 sched_info_idx;
+
+	/* schedule info */
+	char trace_sched[32];
+	struct {
+		struct trace t;
+		u32 pc;
+	} sched[60];
+
+	/* irq info */
+	char trace_irq[32];
+	struct trace irq[60];
+
+	/* task queue status */
+	char task_qid[32];
+	struct {
+		u32 read;
+		u32 write;
+	} taskq[2];
+
+	/* task stack info */
+	char task_info[32];
+	struct {
+		u32 start;
+		u32 end;
+		u32 size;
+	} taski[2];
+
+	/* firmware context */
+	char fw_context[24];
+	struct {
+		u32 idx;
+		u32 handler;
+	} context;
+
+	/* link registers calltrace */
+	u32 call_stack[16];
+
+	/* memory content */
+	u8 data[];
+} __packed;
+
+struct mt7915_coredump_mem {
+	u32 len;
+	u8 data[];
+} __packed;
+
+struct mt7915_mem_hdr {
+	u32 start;
+	u32 len;
+	u8 data[];
+};
+
+struct mt7915_mem_region {
+	u32 start;
+	size_t len;
+
+	const char *name;
+};
+
+#ifdef CONFIG_DEV_COREDUMP
+
+const struct mt7915_mem_region *
+mt7915_coredump_get_mem_layout(struct mt7915_dev *dev, u32 *num);
+struct mt7915_crash_data *mt7915_coredump_new(struct mt7915_dev *dev);
+int mt7915_coredump_submit(struct mt7915_dev *dev);
+int mt7915_coredump_register(struct mt7915_dev *dev);
+void mt7915_coredump_unregister(struct mt7915_dev *dev);
+
+#else /* CONFIG_DEV_COREDUMP */
+
+static inline const struct mt7915_mem_region *
+mt7915_coredump_get_mem_layout(struct mt7915_dev *dev, u32 *num)
+{
+	return NULL;
+}
+
+static inline int mt7915_coredump_submit(struct mt7915_dev *dev)
+{
+	return 0;
+}
+
+static inline struct mt7915_crash_data *mt7915_coredump_new(struct mt7915_dev *dev)
+{
+	return NULL;
+}
+
+static inline int mt7915_coredump_register(struct mt7915_dev *dev)
+{
+	return 0;
+}
+
+static inline void mt7915_coredump_unregister(struct mt7915_dev *dev)
+{
+}
+
+#endif /* CONFIG_DEV_COREDUMP */
+
+#endif /* _COREDUMP_H_ */

drivers/net/wireless/mediatek/mt76/mt7915/init.c

@@ -8,6 +8,7 @@
 #include "mt7915.h"
 #include "mac.h"
 #include "mcu.h"
+#include "coredump.h"
 #include "eeprom.h"
 
 static const struct ieee80211_iface_limit if_limits[] = {
@@ -1100,6 +1101,8 @@ int mt7915_register_device(struct mt7915_dev *dev)
 
 	init_waitqueue_head(&dev->reset_wait);
 	INIT_WORK(&dev->reset_work, mt7915_mac_reset_work);
+	INIT_WORK(&dev->dump_work, mt7915_mac_dump_work);
+	mutex_init(&dev->dump_mutex);
 
 	dev->dbdc_support = mt7915_band_config(dev);
 
@@ -1142,7 +1145,13 @@ int mt7915_register_device(struct mt7915_dev *dev)
 
 	dev->recovery.hw_init_done = true;
 
-	mt7915_init_debugfs(&dev->phy);
+	ret = mt7915_init_debugfs(&dev->phy);
+	if (ret)
+		goto unreg_thermal;
+
+	ret = mt7915_coredump_register(dev);
+	if (ret)
+		goto unreg_thermal;
 
 	return 0;
 
@@ -1161,6 +1170,7 @@ int mt7915_register_device(struct mt7915_dev *dev)
 void mt7915_unregister_device(struct mt7915_dev *dev)
 {
 	mt7915_unregister_ext_phy(dev);
+	mt7915_coredump_unregister(dev);
 	mt7915_unregister_thermal(&dev->phy);
 	mt76_unregister_device(&dev->mt76);
 	mt7915_stop_hardware(dev);

drivers/net/wireless/mediatek/mt76/mt7915/mac.c

@@ -3,6 +3,7 @@
 
 #include <linux/etherdevice.h>
 #include <linux/timekeeping.h>
+#include "coredump.h"
 #include "mt7915.h"
 #include "../dma.h"
 #include "mac.h"
@@ -1620,6 +1621,76 @@ void mt7915_mac_reset_work(struct work_struct *work)
 					     MT7915_WATCHDOG_TIME);
 }
 
+/* firmware coredump */
+void mt7915_mac_dump_work(struct work_struct *work)
+{
+	const struct mt7915_mem_region *mem_region;
+	struct mt7915_crash_data *crash_data;
+	struct mt7915_dev *dev;
+	struct mt7915_mem_hdr *hdr;
+	size_t buf_len;
+	int i;
+	u32 num;
+	u8 *buf;
+
+	dev = container_of(work, struct mt7915_dev, dump_work);
+
+	mutex_lock(&dev->dump_mutex);
+
+	crash_data = mt7915_coredump_new(dev);
+	if (!crash_data) {
+		mutex_unlock(&dev->dump_mutex);
+		goto skip_coredump;
+	}
+
+	mem_region = mt7915_coredump_get_mem_layout(dev, &num);
+	if (!mem_region || !crash_data->memdump_buf_len) {
+		mutex_unlock(&dev->dump_mutex);
+		goto skip_memdump;
+	}
+
+	buf = crash_data->memdump_buf;
+	buf_len = crash_data->memdump_buf_len;
+
+	/* dumping memory content... */
+	memset(buf, 0, buf_len);
+	for (i = 0; i < num; i++) {
+		if (mem_region->len > buf_len) {
+			dev_warn(dev->mt76.dev, "%s len %lu is too large\n",
+				 mem_region->name,
+				 (unsigned long)mem_region->len);
+			break;
+		}
+
+		/* reserve space for the header */
+		hdr = (void *)buf;
+		buf += sizeof(*hdr);
+		buf_len -= sizeof(*hdr);
+
+		mt7915_memcpy_fromio(dev, buf, mem_region->start,
+				     mem_region->len);
+
+		hdr->start = mem_region->start;
+		hdr->len = mem_region->len;
+
+		if (!mem_region->len)
+			/* note: the header remains, just with zero length */
+			break;
+
+		buf += mem_region->len;
+		buf_len -= mem_region->len;
+
+		mem_region++;
+	}
+
+	mutex_unlock(&dev->dump_mutex);
+
+skip_memdump:
+	mt7915_coredump_submit(dev);
+skip_coredump:
+	queue_work(dev->mt76.wq, &dev->reset_work);
+}
+
 void mt7915_reset(struct mt7915_dev *dev)
 {
 	if (!dev->recovery.hw_init_done)
@@ -1636,7 +1707,7 @@ void mt7915_reset(struct mt7915_dev *dev)
 			 wiphy_name(dev->mt76.hw->wiphy));
 
 		mt7915_irq_disable(dev, MT_INT_MCU_CMD);
-		queue_work(dev->mt76.wq, &dev->reset_work);
+		queue_work(dev->mt76.wq, &dev->dump_work);
 		return;
 	}
 

drivers/net/wireless/mediatek/mt76/mt7915/mmio.c

@@ -25,7 +25,22 @@ static const u32 mt7915_reg[] = {
 	[WFDMA_EXT_CSR_ADDR]	= 0xd7000,
 	[CBTOP1_PHY_END]	= 0x77ffffff,
 	[INFRA_MCU_ADDR_END]	= 0x7c3fffff,
-	[FW_EXCEPTION_ADDR]	= 0x219848,
+	[FW_ASSERT_STAT_ADDR]	= 0x219848,
+	[FW_EXCEPT_TYPE_ADDR]	= 0x21987c,
+	[FW_EXCEPT_COUNT_ADDR]	= 0x219848,
+	[FW_CIRQ_COUNT_ADDR]	= 0x216f94,
+	[FW_CIRQ_IDX_ADDR]	= 0x216ef8,
+	[FW_CIRQ_LISR_ADDR]	= 0x2170ac,
+	[FW_TASK_ID_ADDR]	= 0x216f90,
+	[FW_TASK_IDX_ADDR]	= 0x216f9c,
+	[FW_TASK_QID1_ADDR]	= 0x219680,
+	[FW_TASK_QID2_ADDR]	= 0x219760,
+	[FW_TASK_START_ADDR]	= 0x219558,
+	[FW_TASK_END_ADDR]	= 0x219554,
+	[FW_TASK_SIZE_ADDR]	= 0x219560,
+	[FW_LAST_MSG_ID_ADDR]	= 0x216f70,
+	[FW_EINT_INFO_ADDR]	= 0x219818,
+	[FW_SCHED_INFO_ADDR]	= 0x219828,
 	[SWDEF_BASE_ADDR]	= 0x41f200,
 	[TXQ_WED_RING_BASE]	= 0xd7300,
 	[RXQ_WED_RING_BASE]	= 0xd7410,
@@ -43,7 +58,22 @@ static const u32 mt7916_reg[] = {
 	[WFDMA_EXT_CSR_ADDR]	= 0xd7000,
 	[CBTOP1_PHY_END]	= 0x7fffffff,
 	[INFRA_MCU_ADDR_END]	= 0x7c085fff,
-	[FW_EXCEPTION_ADDR]	= 0x022050bc,
+	[FW_ASSERT_STAT_ADDR]	= 0x02204c14,
+	[FW_EXCEPT_TYPE_ADDR]	= 0x022051a4,
+	[FW_EXCEPT_COUNT_ADDR]	= 0x022050bc,
+	[FW_CIRQ_COUNT_ADDR]	= 0x022001ac,
+	[FW_CIRQ_IDX_ADDR]	= 0x02204f84,
+	[FW_CIRQ_LISR_ADDR]	= 0x022050d0,
+	[FW_TASK_ID_ADDR]	= 0x0220406c,
+	[FW_TASK_IDX_ADDR]	= 0x0220500c,
+	[FW_TASK_QID1_ADDR]	= 0x022028c8,
+	[FW_TASK_QID2_ADDR]	= 0x02202a38,
+	[FW_TASK_START_ADDR]	= 0x0220286c,
+	[FW_TASK_END_ADDR]	= 0x02202870,
+	[FW_TASK_SIZE_ADDR]	= 0x02202878,
+	[FW_LAST_MSG_ID_ADDR]	= 0x02204fe8,
+	[FW_EINT_INFO_ADDR]	= 0x0220525c,
+	[FW_SCHED_INFO_ADDR]	= 0x0220516c,
 	[SWDEF_BASE_ADDR]	= 0x411400,
 	[TXQ_WED_RING_BASE]	= 0xd7300,
 	[RXQ_WED_RING_BASE]	= 0xd7410,
@@ -61,7 +91,22 @@ static const u32 mt7986_reg[] = {
 	[WFDMA_EXT_CSR_ADDR]	= 0x27000,
 	[CBTOP1_PHY_END]	= 0x7fffffff,
 	[INFRA_MCU_ADDR_END]	= 0x7c085fff,
-	[FW_EXCEPTION_ADDR]	= 0x02204ffc,
+	[FW_ASSERT_STAT_ADDR]	= 0x02204b54,
+	[FW_EXCEPT_TYPE_ADDR]	= 0x022050dc,
+	[FW_EXCEPT_COUNT_ADDR]	= 0x02204ffc,
+	[FW_CIRQ_COUNT_ADDR]	= 0x022001ac,
+	[FW_CIRQ_IDX_ADDR]	= 0x02204ec4,
+	[FW_CIRQ_LISR_ADDR]	= 0x02205010,
+	[FW_TASK_ID_ADDR]	= 0x02204fac,
+	[FW_TASK_IDX_ADDR]	= 0x02204f4c,
+	[FW_TASK_QID1_ADDR]	= 0x02202814,
+	[FW_TASK_QID2_ADDR]	= 0x02202984,
+	[FW_TASK_START_ADDR]	= 0x022027b8,
+	[FW_TASK_END_ADDR]	= 0x022027bc,
+	[FW_TASK_SIZE_ADDR]	= 0x022027c4,
+	[FW_LAST_MSG_ID_ADDR]	= 0x02204f28,
+	[FW_EINT_INFO_ADDR]	= 0x02205194,
+	[FW_SCHED_INFO_ADDR]	= 0x022051a4,
 	[SWDEF_BASE_ADDR]	= 0x411400,
 	[TXQ_WED_RING_BASE]	= 0x24420,
 	[RXQ_WED_RING_BASE]	= 0x24520,
@@ -457,6 +502,14 @@ static u32 __mt7915_reg_addr(struct mt7915_dev *dev, u32 addr)
 	return mt7915_reg_map_l2(dev, addr);
 }
 
+void mt7915_memcpy_fromio(struct mt7915_dev *dev, void *buf, u32 offset,
+			  size_t len)
+{
+	u32 addr = __mt7915_reg_addr(dev, offset);
+
+	memcpy_fromio(buf, dev->mt76.mmio.regs + addr, len);
+}
+
 static u32 mt7915_rr(struct mt76_dev *mdev, u32 offset)
 {
 	struct mt7915_dev *dev = container_of(mdev, struct mt7915_dev, mt76);

drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h

@@ -224,6 +224,15 @@ struct mib_stats {
 	u32 tx_amsdu_cnt;
 };
 
+/* crash-dump */
+struct mt7915_crash_data {
+	guid_t guid;
+	struct timespec64 timestamp;
+
+	u8 *memdump_buf;
+	size_t memdump_buf_len;
+};
+
 struct mt7915_hif {
 	struct list_head list;
 
@@ -305,6 +314,7 @@ struct mt7915_dev {
 
 	struct work_struct init_work;
 	struct work_struct rc_work;
+	struct work_struct dump_work;
 	struct work_struct reset_work;
 	wait_queue_head_t reset_wait;
 
@@ -317,6 +327,14 @@ struct mt7915_dev {
 		bool restart:1;
 	} recovery;
 
+	/* protects coredump data */
+	struct mutex dump_mutex;
+#ifdef CONFIG_DEV_COREDUMP
+	struct {
+		struct mt7915_crash_data *crash_data;
+	} coredump;
+#endif
+
 	struct list_head sta_rc_list;
 	struct list_head sta_poll_list;
 	struct list_head twt_list;
@@ -562,6 +580,9 @@ static inline void mt7915_irq_disable(struct mt7915_dev *dev, u32 mask)
 		mt76_set_irq_mask(&dev->mt76, MT_INT_MASK_CSR, mask, 0);
 }
 
+void mt7915_memcpy_fromio(struct mt7915_dev *dev, void *buf, u32 offset,
+			  size_t len);
+
 void mt7915_mac_init(struct mt7915_dev *dev);
 u32 mt7915_mac_wtbl_lmac_addr(struct mt7915_dev *dev, u16 wcid, u8 dw);
 bool mt7915_mac_wtbl_update(struct mt7915_dev *dev, int idx, u32 mask);
@@ -581,6 +602,7 @@ void mt7915_mac_sta_remove(struct mt76_dev *mdev, struct ieee80211_vif *vif,
 			   struct ieee80211_sta *sta);
 void mt7915_mac_work(struct work_struct *work);
 void mt7915_mac_reset_work(struct work_struct *work);
+void mt7915_mac_dump_work(struct work_struct *work);
 void mt7915_mac_sta_rc_work(struct work_struct *work);
 void mt7915_mac_update_stats(struct mt7915_phy *phy);
 void mt7915_mac_twt_teardown_flow(struct mt7915_dev *dev,

drivers/net/wireless/mediatek/mt76/mt7915/regs.h

@@ -24,7 +24,22 @@ enum reg_rev {
 	WFDMA_EXT_CSR_ADDR,
 	CBTOP1_PHY_END,
 	INFRA_MCU_ADDR_END,
-	FW_EXCEPTION_ADDR,
+	FW_ASSERT_STAT_ADDR,
+	FW_EXCEPT_TYPE_ADDR,
+	FW_EXCEPT_COUNT_ADDR,
+	FW_CIRQ_COUNT_ADDR,
+	FW_CIRQ_IDX_ADDR,
+	FW_CIRQ_LISR_ADDR,
+	FW_TASK_ID_ADDR,
+	FW_TASK_IDX_ADDR,
+	FW_TASK_QID1_ADDR,
+	FW_TASK_QID2_ADDR,
+	FW_TASK_START_ADDR,
+	FW_TASK_END_ADDR,
+	FW_TASK_SIZE_ADDR,
+	FW_LAST_MSG_ID_ADDR,
+	FW_EINT_INFO_ADDR,
+	FW_SCHED_INFO_ADDR,
 	SWDEF_BASE_ADDR,
 	TXQ_WED_RING_BASE,
 	RXQ_WED_RING_BASE,
@@ -979,7 +994,22 @@ enum offs_rev {
 #define MT_ADIE_TYPE_MASK		BIT(1)
 
 /* FW MODE SYNC */
-#define MT_FW_EXCEPTION			__REG(FW_EXCEPTION_ADDR)
+#define MT_FW_ASSERT_STAT		__REG(FW_ASSERT_STAT_ADDR)
+#define MT_FW_EXCEPT_TYPE		__REG(FW_EXCEPT_TYPE_ADDR)
+#define MT_FW_EXCEPT_COUNT		__REG(FW_EXCEPT_COUNT_ADDR)
+#define MT_FW_CIRQ_COUNT		__REG(FW_CIRQ_COUNT_ADDR)
+#define MT_FW_CIRQ_IDX			__REG(FW_CIRQ_IDX_ADDR)
+#define MT_FW_CIRQ_LISR			__REG(FW_CIRQ_LISR_ADDR)
+#define MT_FW_TASK_ID			__REG(FW_TASK_ID_ADDR)
+#define MT_FW_TASK_IDX			__REG(FW_TASK_IDX_ADDR)
+#define MT_FW_TASK_QID1			__REG(FW_TASK_QID1_ADDR)
+#define MT_FW_TASK_QID2			__REG(FW_TASK_QID2_ADDR)
+#define MT_FW_TASK_START		__REG(FW_TASK_START_ADDR)
+#define MT_FW_TASK_END			__REG(FW_TASK_END_ADDR)
+#define MT_FW_TASK_SIZE			__REG(FW_TASK_SIZE_ADDR)
+#define MT_FW_LAST_MSG_ID		__REG(FW_LAST_MSG_ID_ADDR)
+#define MT_FW_EINT_INFO			__REG(FW_EINT_INFO_ADDR)
+#define MT_FW_SCHED_INFO		__REG(FW_SCHED_INFO_ADDR)
 
 #define MT_SWDEF_BASE			__REG(SWDEF_BASE_ADDR)