Merge branch 'bnxt_en-Error-recovery-improvements'

Michael Chan says:

====================
bnxt_en: Error recovery improvements.

This series contains some improvements for error recovery.  The main
changes are:

1. Keep better track of the health register mappings with the
"status_reliable" flag.

2. Don't wait for firmware responses if firmware is not healthy.

3. Better retry logic of the first firmware message.

4. Set the proper flag early to let the RDMA driver know that firmware
reset has been detected.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/broadcom/bnxt/bnxt.c

@@ -4470,7 +4470,7 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 	writel(1, bp->bar0 + doorbell_offset);
 
 	if (!pci_is_enabled(bp->pdev))
-		return 0;
+		return -ENODEV;
 
 	if (!timeout)
 		timeout = DFLT_HWRM_CMD_TIMEOUT;
@@ -4500,12 +4500,15 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 			if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
 				return -EBUSY;
 			/* on first few passes, just barely sleep */
-			if (i < HWRM_SHORT_TIMEOUT_COUNTER)
+			if (i < HWRM_SHORT_TIMEOUT_COUNTER) {
 				usleep_range(HWRM_SHORT_MIN_TIMEOUT,
 					     HWRM_SHORT_MAX_TIMEOUT);
-			else
+			} else {
+				if (HWRM_WAIT_MUST_ABORT(bp, req))
+					break;
 				usleep_range(HWRM_MIN_TIMEOUT,
 					     HWRM_MAX_TIMEOUT);
+			}
 		}
 
 		if (bp->hwrm_intr_seq_id != (u16)~seq_id) {
@@ -4530,15 +4533,19 @@ static int bnxt_hwrm_do_send_msg(struct bnxt *bp, void *msg, u32 msg_len,
 			if (len)
 				break;
 			/* on first few passes, just barely sleep */
-			if (i < HWRM_SHORT_TIMEOUT_COUNTER)
+			if (i < HWRM_SHORT_TIMEOUT_COUNTER) {
 				usleep_range(HWRM_SHORT_MIN_TIMEOUT,
 					     HWRM_SHORT_MAX_TIMEOUT);
-			else
+			} else {
+				if (HWRM_WAIT_MUST_ABORT(bp, req))
+					goto timeout_abort;
 				usleep_range(HWRM_MIN_TIMEOUT,
 					     HWRM_MAX_TIMEOUT);
+			}
 		}
 
 		if (i >= tmo_count) {
+timeout_abort:
 			if (!silent)
 				netdev_err(bp->dev, "Error (timeout: %d) msg {0x%x 0x%x} len:%d\n",
 					   HWRM_TOTAL_TIMEOUT(i),
@@ -7540,6 +7547,32 @@ static void __bnxt_map_fw_health_reg(struct bnxt *bp, u32 reg)
 					 BNXT_FW_HEALTH_WIN_MAP_OFF);
 }
 
+bool bnxt_is_fw_healthy(struct bnxt *bp)
+{
+	if (bp->fw_health && bp->fw_health->status_reliable) {
+		u32 fw_status;
+
+		fw_status = bnxt_fw_health_readl(bp, BNXT_FW_HEALTH_REG);
+		if (fw_status && !BNXT_FW_IS_HEALTHY(fw_status))
+			return false;
+	}
+
+	return true;
+}
+
+static void bnxt_inv_fw_health_reg(struct bnxt *bp)
+{
+	struct bnxt_fw_health *fw_health = bp->fw_health;
+	u32 reg_type;
+
+	if (!fw_health || !fw_health->status_reliable)
+		return;
+
+	reg_type = BNXT_FW_HEALTH_REG_TYPE(fw_health->regs[BNXT_FW_HEALTH_REG]);
+	if (reg_type == BNXT_FW_HEALTH_REG_TYPE_GRC)
+		fw_health->status_reliable = false;
+}
+
 static void bnxt_try_map_fw_health_reg(struct bnxt *bp)
 {
 	void __iomem *hs;
@@ -7547,6 +7580,9 @@ static void bnxt_try_map_fw_health_reg(struct bnxt *bp)
 	u32 reg_type;
 	u32 sig;
 
+	if (bp->fw_health)
+		bp->fw_health->status_reliable = false;
+
 	__bnxt_map_fw_health_reg(bp, HCOMM_STATUS_STRUCT_LOC);
 	hs = bp->bar0 + BNXT_FW_HEALTH_WIN_OFF(HCOMM_STATUS_STRUCT_LOC);
 
@@ -7558,11 +7594,9 @@ static void bnxt_try_map_fw_health_reg(struct bnxt *bp)
 					     BNXT_FW_HEALTH_WIN_BASE +
 					     BNXT_GRC_REG_CHIP_NUM);
 		}
-		if (!BNXT_CHIP_P5(bp)) {
-			if (bp->fw_health)
-				bp->fw_health->status_reliable = false;
+		if (!BNXT_CHIP_P5(bp))
 			return;
-		}
+
 		status_loc = BNXT_GRC_REG_STATUS_P5 |
 			     BNXT_FW_HEALTH_REG_TYPE_BAR0;
 	} else {
@@ -7592,6 +7626,7 @@ static int bnxt_map_fw_health_regs(struct bnxt *bp)
 	u32 reg_base = 0xffffffff;
 	int i;
 
+	bp->fw_health->status_reliable = false;
 	/* Only pre-map the monitoring GRC registers using window 3 */
 	for (i = 0; i < 4; i++) {
 		u32 reg = fw_health->regs[i];
@@ -7604,6 +7639,7 @@ static int bnxt_map_fw_health_regs(struct bnxt *bp)
 			return -ERANGE;
 		fw_health->mapped_regs[i] = BNXT_FW_HEALTH_WIN_OFF(reg);
 	}
+	bp->fw_health->status_reliable = true;
 	if (reg_base == 0xffffffff)
 		return 0;
 
@@ -9494,9 +9530,10 @@ static int bnxt_try_recover_fw(struct bnxt *bp)
 
 		mutex_lock(&bp->hwrm_cmd_lock);
 		do {
-			rc = __bnxt_hwrm_ver_get(bp, true);
 			sts = bnxt_fw_health_readl(bp, BNXT_FW_HEALTH_REG);
-			if (!sts || !BNXT_FW_IS_BOOTING(sts))
+			rc = __bnxt_hwrm_ver_get(bp, true);
+			if (!sts || (!BNXT_FW_IS_BOOTING(sts) &&
+				     !BNXT_FW_IS_RECOVERING(sts)))
 				break;
 			retry++;
 		} while (rc == -EBUSY && retry < BNXT_FW_RETRY);
@@ -9556,13 +9593,17 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 	if (rc)
 		return rc;
 
-	if (!up)
+	if (!up) {
+		bnxt_inv_fw_health_reg(bp);
 		return 0;
+	}
 
 	if (flags & FUNC_DRV_IF_CHANGE_RESP_FLAGS_RESC_CHANGE)
 		resc_reinit = true;
 	if (flags & FUNC_DRV_IF_CHANGE_RESP_FLAGS_HOT_FW_RESET_DONE)
 		fw_reset = true;
+	else if (bp->fw_health && !bp->fw_health->status_reliable)
+		bnxt_try_map_fw_health_reg(bp);
 
 	if (test_bit(BNXT_STATE_IN_FW_RESET, &bp->state) && !fw_reset) {
 		netdev_err(bp->dev, "RESET_DONE not set during FW reset.\n");
@@ -9571,6 +9612,7 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 	}
 	if (resc_reinit || fw_reset) {
 		if (fw_reset) {
+			set_bit(BNXT_STATE_FW_RESET_DET, &bp->state);
 			if (!test_bit(BNXT_STATE_IN_FW_RESET, &bp->state))
 				bnxt_ulp_stop(bp);
 			bnxt_free_ctx_mem(bp);
@@ -9579,21 +9621,25 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 			bnxt_dcb_free(bp);
 			rc = bnxt_fw_init_one(bp);
 			if (rc) {
+				clear_bit(BNXT_STATE_FW_RESET_DET, &bp->state);
 				set_bit(BNXT_STATE_ABORT_ERR, &bp->state);
 				return rc;
 			}
 			bnxt_clear_int_mode(bp);
 			rc = bnxt_init_int_mode(bp);
 			if (rc) {
+				clear_bit(BNXT_STATE_FW_RESET_DET, &bp->state);
 				netdev_err(bp->dev, "init int mode failed\n");
 				return rc;
 			}
-			set_bit(BNXT_STATE_FW_RESET_DET, &bp->state);
 		}
 		if (BNXT_NEW_RM(bp)) {
 			struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
 
 			rc = bnxt_hwrm_func_resc_qcaps(bp, true);
+			if (rc)
+				netdev_err(bp->dev, "resc_qcaps failed\n");
+
 			hw_resc->resv_cp_rings = 0;
 			hw_resc->resv_stat_ctxs = 0;
 			hw_resc->resv_irqs = 0;
@@ -9607,7 +9653,7 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up)
 			}
 		}
 	}
-	return 0;
+	return rc;
 }
 
 static int bnxt_hwrm_port_led_qcaps(struct bnxt *bp)
@@ -11640,7 +11686,7 @@ static void bnxt_reset_all(struct bnxt *bp)
 		req.selfrst_status = FW_RESET_REQ_SELFRST_STATUS_SELFRSTASAP;
 		req.flags = FW_RESET_REQ_FLAGS_RESET_GRACEFUL;
 		rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
-		if (rc)
+		if (rc != -ENODEV)
 			netdev_warn(bp->dev, "Unable to reset FW rc=%d\n", rc);
 	}
 	bp->fw_reset_timestamp = jiffies;
@@ -11723,28 +11769,20 @@ static void bnxt_fw_reset_task(struct work_struct *work)
 		bnxt_queue_fw_reset_work(bp, bp->fw_reset_min_dsecs * HZ / 10);
 		return;
 	case BNXT_FW_RESET_STATE_ENABLE_DEV:
-		if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state)) {
-			u32 val;
-
-			if (!bp->fw_reset_min_dsecs) {
-				u16 val;
-
-				pci_read_config_word(bp->pdev, PCI_SUBSYSTEM_ID,
-						     &val);
-				if (val == 0xffff) {
-					if (bnxt_fw_reset_timeout(bp)) {
-						netdev_err(bp->dev, "Firmware reset aborted, PCI config space invalid\n");
-						goto fw_reset_abort;
-					}
-					bnxt_queue_fw_reset_work(bp, HZ / 1000);
-					return;
+		bnxt_inv_fw_health_reg(bp);
+		if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state) &&
+		    !bp->fw_reset_min_dsecs) {
+			u16 val;
+
+			pci_read_config_word(bp->pdev, PCI_SUBSYSTEM_ID, &val);
+			if (val == 0xffff) {
+				if (bnxt_fw_reset_timeout(bp)) {
+					netdev_err(bp->dev, "Firmware reset aborted, PCI config space invalid\n");
+					goto fw_reset_abort;
 				}
+				bnxt_queue_fw_reset_work(bp, HZ / 1000);
+				return;
 			}
-			val = bnxt_fw_health_readl(bp,
-						   BNXT_FW_RESET_INPROG_REG);
-			if (val)
-				netdev_warn(bp->dev, "FW reset inprog %x after min wait time.\n",
-					    val);
 		}
 		clear_bit(BNXT_STATE_FW_FATAL_COND, &bp->state);
 		if (pci_enable_device(bp->pdev)) {

drivers/net/ethernet/broadcom/bnxt/bnxt.h

@@ -671,6 +671,10 @@ struct nqe_cn {
 #define HWRM_MIN_TIMEOUT		25
 #define HWRM_MAX_TIMEOUT		40
 
+#define HWRM_WAIT_MUST_ABORT(bp, req)					\
+	(le16_to_cpu((req)->req_type) != HWRM_VER_GET &&		\
+	 !bnxt_is_fw_healthy(bp))
+
 #define HWRM_TOTAL_TIMEOUT(n)	(((n) <= HWRM_SHORT_TIMEOUT_COUNTER) ?	\
 	((n) * HWRM_SHORT_MIN_TIMEOUT) :				\
 	(HWRM_SHORT_TIMEOUT_COUNTER * HWRM_SHORT_MIN_TIMEOUT +		\
@@ -1560,6 +1564,7 @@ struct bnxt_fw_reporter_ctx {
 #define BNXT_FW_STATUS_HEALTH_MSK	0xffff
 #define BNXT_FW_STATUS_HEALTHY		0x8000
 #define BNXT_FW_STATUS_SHUTDOWN		0x100000
+#define BNXT_FW_STATUS_RECOVERING	0x400000
 
 #define BNXT_FW_IS_HEALTHY(sts)		(((sts) & BNXT_FW_STATUS_HEALTH_MSK) ==\
 					 BNXT_FW_STATUS_HEALTHY)
@@ -1570,6 +1575,9 @@ struct bnxt_fw_reporter_ctx {
 #define BNXT_FW_IS_ERR(sts)		(((sts) & BNXT_FW_STATUS_HEALTH_MSK) > \
 					 BNXT_FW_STATUS_HEALTHY)
 
+#define BNXT_FW_IS_RECOVERING(sts)	(BNXT_FW_IS_ERR(sts) &&		       \
+					 ((sts) & BNXT_FW_STATUS_RECOVERING))
+
 #define BNXT_FW_RETRY			5
 #define BNXT_FW_IF_RETRY		10
 
@@ -2228,6 +2236,7 @@ int bnxt_hwrm_set_link_setting(struct bnxt *, bool, bool);
 int bnxt_hwrm_alloc_wol_fltr(struct bnxt *bp);
 int bnxt_hwrm_free_wol_fltr(struct bnxt *bp);
 int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp, bool all);
+bool bnxt_is_fw_healthy(struct bnxt *bp);
 int bnxt_hwrm_fw_set_time(struct bnxt *);
 int bnxt_open_nic(struct bnxt *, bool, bool);
 int bnxt_half_open_nic(struct bnxt *bp);