[dpdk-dev] [PATCH v3 10/12] net/bnxt: fix firmware fatal error handling

[Ajit Khaparde]

From: Kalesh AP <kalesh-anakkur.purayil at broadcom.com>

During some fatal firmware error conditions, the PCI config space
register 0x2e which normally contains the subsystem ID will become
0xffff. This register will revert back to the normal value after
the chip has completed core reset. If we detect this condition,
we can poll this config register immediately for the value to revert.
Because we use config read cycles to poll this register, there is no
possibility of Master Abort if we happen to read it during core reset.
This speeds up recovery significantly as we don't have to wait for the
conservative min_time before polling to see if the firmware has come
out of reset. As soon as this register changes value we can proceed
to re-initialize the device.

Fixes: df6cd7c1f73a ("net/bnxt: handle reset notify async event from FW")
Cc: stable at dpdk.org

Signed-off-by: Kalesh AP <kalesh-anakkur.purayil at broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur at broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde at broadcom.com>
---
 drivers/net/bnxt/bnxt_ethdev.c | 56 ++++++++++++++++++++++++++++++++--
 drivers/net/bnxt/bnxt_util.h   |  2 ++
 2 files changed, 56 insertions(+), 2 deletions(-)

diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 9e0ec46403..67ff800da5 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -3743,6 +3743,32 @@ static void bnxt_dev_cleanup(struct bnxt *bp)
 	bnxt_uninit_resources(bp, true);
 }
 
+static int
+bnxt_check_fw_reset_done(struct bnxt *bp)
+{
+	int timeout = bp->fw_reset_max_msecs;
+	uint16_t val = 0;
+	int rc;
+
+	do {
+		rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
+		if (rc < 0) {
+			PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
+			return rc;
+		}
+		if (val != 0xffff)
+			break;
+		rte_delay_ms(1);
+	} while (timeout--);
+
+	if (val == 0xffff) {
+		PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
+		return -1;
+	}
+
+	return 0;
+}
+
 static int bnxt_restore_vlan_filters(struct bnxt *bp)
 {
 	struct rte_eth_dev *dev = bp->eth_dev;
@@ -3840,6 +3866,13 @@ static void bnxt_dev_recover(void *arg)
 	int rc = 0;
 
 	pthread_mutex_lock(&bp->err_recovery_lock);
+
+	if (!bp->fw_reset_min_msecs) {
+		rc = bnxt_check_fw_reset_done(bp);
+		if (rc)
+			goto err;
+	}
+
 	/* Clear Error flag so that device re-init should happen */
 	bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
 
@@ -3891,14 +3924,33 @@ static void bnxt_dev_recover(void *arg)
 void bnxt_dev_reset_and_resume(void *arg)
 {
 	struct bnxt *bp = arg;
+	uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
+	uint16_t val = 0;
 	int rc;
 
 	bnxt_dev_cleanup(bp);
 
 	bnxt_wait_for_device_shutdown(bp);
 
-	rc = rte_eal_alarm_set(US_PER_MS * bp->fw_reset_min_msecs,
-			       bnxt_dev_recover, (void *)bp);
+	/* During some fatal firmware error conditions, the PCI config space
+	 * register 0x2e which normally contains the subsystem ID will become
+	 * 0xffff. This register will revert back to the normal value after
+	 * the chip has completed core reset. If we detect this condition,
+	 * we can poll this config register immediately for the value to revert.
+	 */
+	if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
+		rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
+		if (rc < 0) {
+			PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
+			return;
+		}
+		if (val == 0xffff) {
+			bp->fw_reset_min_msecs = 0;
+			us = 1;
+		}
+	}
+
+	rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
 	if (rc)
 		PMD_DRV_LOG(ERR, "Error setting recovery alarm");
 }
diff --git a/drivers/net/bnxt/bnxt_util.h b/drivers/net/bnxt/bnxt_util.h
index 8de55e1038..64e97eed15 100644
--- a/drivers/net/bnxt/bnxt_util.h
+++ b/drivers/net/bnxt/bnxt_util.h
@@ -10,6 +10,8 @@
 #define BIT(n)	(1UL << (n))
 #endif /* BIT */
 
+#define PCI_SUBSYSTEM_ID_OFFSET	0x2e
+
 int bnxt_check_zero_bytes(const uint8_t *bytes, int len);
 void bnxt_eth_hw_addr_random(uint8_t *mac_addr);
 
-- 
2.21.1 (Apple Git-122.3)