diff mbox series

[v2] net/failsafe: link_update request crashing at boot

Message ID 20211021214215.1633-1-vipul.ashri@oracle.com (mailing list archive)
State New
Delegated to: Ferruh Yigit
Headers show
Series [v2] net/failsafe: link_update request crashing at boot | expand

Checks

Context Check Description
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-x86_64-compile-testing success Testing PASS
ci/iol-x86_64-unit-testing success Testing PASS
ci/intel-Testing success Testing PASS
ci/Intel-compilation success Compilation OK
ci/iol-aarch64-compile-testing success Testing PASS
ci/iol-broadcom-Performance success Performance Testing PASS
ci/github-robot: build success github build: passed
ci/iol-broadcom-Functional success Functional Testing PASS
ci/checkpatch success coding style OK

Commit Message

Vipul Ashri Oct. 21, 2021, 9:42 p.m. UTC
From: Vipul Ashri <vipul.ashri@oracle.com>

failsafe crashed while sending early link_update request during
boot time initialization.
Based on debugging we found failsafe device was good but sub-
devices were progressing towards initialization and SUBOPS macro
where expanding macro gives [partial_dev]->dev_ops->link_update()
execution of which triggered crash because dev_ops==0. similar
crash seen at failsafe_eth_dev_close()

Failsafe driver need a separate check for subdevices similar to
"RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);" which is
called to almost every eth_dev function.

Fixes: a46f8d5 ("net/failsafe: add fail-safe PMD")
Cc: stable@dpdk.org
Signed-off-by: Vipul Ashri <vipul.ashri@oracle.com>
---
 drivers/net/failsafe/failsafe_ops.c     | 45 +++++++++++++++++++++++--
 drivers/net/failsafe/failsafe_private.h |  6 ++++
 2 files changed, 49 insertions(+), 2 deletions(-)

Comments

Ferruh Yigit Nov. 22, 2021, 9:36 a.m. UTC | #1
On 10/21/2021 10:42 PM, vipul.ashri@oracle.com wrote:
> From: Vipul Ashri <vipul.ashri@oracle.com>
> 
> failsafe crashed while sending early link_update request during
> boot time initialization.
> Based on debugging we found failsafe device was good but sub-
> devices were progressing towards initialization and SUBOPS macro
> where expanding macro gives [partial_dev]->dev_ops->link_update()
> execution of which triggered crash because dev_ops==0. similar
> crash seen at failsafe_eth_dev_close()
> 
> Failsafe driver need a separate check for subdevices similar to
> "RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);" which is
> called to almost every eth_dev function.
> 
> Fixes: a46f8d5 ("net/failsafe: add fail-safe PMD")
> Cc: stable@dpdk.org
> Signed-off-by: Vipul Ashri <vipul.ashri@oracle.com>


Hi Gaetan,

Can you please review this patch?

Thanks,
ferruh
Gaëtan Rivet Nov. 22, 2021, 10:23 a.m. UTC | #2
On Thu, Oct 21, 2021, at 23:42, vipul.ashri@oracle.com wrote:
> From: Vipul Ashri <vipul.ashri@oracle.com>
>
> failsafe crashed while sending early link_update request during
> boot time initialization.
> Based on debugging we found failsafe device was good but sub-
> devices were progressing towards initialization and SUBOPS macro
> where expanding macro gives [partial_dev]->dev_ops->link_update()
> execution of which triggered crash because dev_ops==0. similar
> crash seen at failsafe_eth_dev_close()
>
> Failsafe driver need a separate check for subdevices similar to
> "RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);" which is
> called to almost every eth_dev function.
>
> Fixes: a46f8d5 ("net/failsafe: add fail-safe PMD")
> Cc: stable@dpdk.org
> Signed-off-by: Vipul Ashri <vipul.ashri@oracle.com>

Hello Vipul,

I'm sorry for the delay, I missed your fix on the mailing list.

IIUC, the issue is that failsafe finished init and received an ethdev
operation call, but one of its sub-device, although marked DEV_ACTIVE,
has its eth_dev->dev_ops field NULL.

It is really surprising to me, because there aren't many ways for a sub-device
to become DEV_ACTIVE.

The only two ways are

  * by executing 'fs_dev_configure()', which will first execute
    rte_eth_dev_configure() on the sub-device, and on error would
    stop *without* setting DEV_ACTIVE.
    rte_eth_dev_configure() will itself execute
    RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV), so it would
    return negative errno and fs_dev_configure() would abort.

  * by executing 'fs_dev_remove()' and the sub-device was 'DEV_STARTED'
    to begin with, then it is retrograded to DEV_ACTIVE once stopped.

So I don't understand yet how it is possible for a sub-device to become DEV_ACTIVE
while its eth_dev->dev_ops are NULL. It seems more like a bug, memory corruption or
just an unexpected execution pattern.

Could describe in more detail the execution?
In particular, setting the EAL log-level to debug with the option:
' --log-level pmd.net.failsafe:debug '
for example while using testpmd or your DPDK app.
It should show ethdev level accesses to the sub-devices, and error values.

Best regards,
diff mbox series

Patch

diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
index 29de39910c..8e128b9802 100644
--- a/drivers/net/failsafe/failsafe_ops.c
+++ b/drivers/net/failsafe/failsafe_ops.c
@@ -371,7 +371,8 @@  fs_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 		close(rxq->event_fd);
 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 		if (ETH(sdev)->data->rx_queues != NULL &&
-		    ETH(sdev)->data->rx_queues[rxq->qid] != NULL)
+			ETH(sdev)->data->rx_queues[rxq->qid] != NULL &&
+			SUBDEV_VALID_PORTID(sdev))
 			SUBOPS(sdev, rx_queue_release)(ETH(sdev), rxq->qid);
 	}
 	dev->data->rx_queues[rxq->qid] = NULL;
@@ -405,6 +406,12 @@  fs_rx_queue_setup(struct rte_eth_dev *dev,
 	fs_lock(dev, 0);
 	if (rx_conf->rx_deferred_start) {
 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
+			if (!SUBDEV_VALID_PORTID(sdev)) {
+				ERROR("%s: Invalid sub-device port_id=%u\n",
+					__func__, PORT_ID(sdev));
+				fs_unlock(dev, 0);
+				return -ENODEV;
+			}
 			if (SUBOPS(sdev, rx_queue_start) == NULL) {
 				ERROR("Rx queue deferred start is not "
 					"supported for subdevice %d", i);
@@ -548,7 +555,8 @@  fs_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 	fs_lock(dev, 0);
 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 		if (ETH(sdev)->data->tx_queues != NULL &&
-		    ETH(sdev)->data->tx_queues[txq->qid] != NULL)
+			ETH(sdev)->data->tx_queues[txq->qid] != NULL &&
+			SUBDEV_VALID_PORTID(sdev))
 			SUBOPS(sdev, tx_queue_release)(ETH(sdev), txq->qid);
 	}
 	dev->data->tx_queues[txq->qid] = NULL;
@@ -571,6 +579,12 @@  fs_tx_queue_setup(struct rte_eth_dev *dev,
 	fs_lock(dev, 0);
 	if (tx_conf->tx_deferred_start) {
 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
+			if (!SUBDEV_VALID_PORTID(sdev)) {
+				ERROR("%s: Invalid sub-device port_id=%u\n",
+					__func__, PORT_ID(sdev));
+				fs_unlock(dev, 0);
+				return -ENODEV;
+			}
 			if (SUBOPS(sdev, tx_queue_start) == NULL) {
 				ERROR("Tx queue deferred start is not "
 					"supported for subdevice %d", i);
@@ -645,6 +659,12 @@  failsafe_eth_dev_close(struct rte_eth_dev *dev)
 	fs_lock(dev, 0);
 	failsafe_hotplug_alarm_cancel(dev);
 	if (PRIV(dev)->state == DEV_STARTED) {
+		if (!rte_eth_dev_is_valid_port(dev->data->port_id)) {
+			ERROR("%s: Invalid sub-device port_id=%u\n",
+				__func__, dev->data->port_id);
+			fs_unlock(dev, 0);
+			return -ENODEV;
+		}
 		ret = dev->dev_ops->dev_stop(dev);
 		if (ret != 0) {
 			fs_unlock(dev, 0);
@@ -827,6 +847,12 @@  fs_link_update(struct rte_eth_dev *dev,
 
 	fs_lock(dev, 0);
 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
+		if (!SUBDEV_VALID_PORTID(sdev)) {
+			ERROR("%s: Invalid Sub-device port_id=%u\n",
+				__func__, PORT_ID(sdev));
+			fs_unlock(dev, 0);
+			return -ENODEV;
+		}
 		DEBUG("Calling link_update on sub_device %d", i);
 		ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 		if (ret && ret != -1 && sdev->remove == 0 &&
@@ -1249,6 +1275,15 @@  fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 		goto unlock;
 	}
 	edev = ETH(sdev);
+
+	if (!SUBDEV_VALID_PORTID(sdev)) {
+		ERROR("%s: Invalid TX_SUBDEV port_id=%u\n",
+			__func__, PORT_ID(sdev));
+		rte_errno = -ENODEV;
+		ret = NULL;
+		goto unlock;
+	}
+
 	/* ENOTSUP: counts as no supported ptypes */
 	if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
 		ret = NULL;
@@ -1324,6 +1359,12 @@  fs_flow_ctrl_get(struct rte_eth_dev *dev,
 		ret = 0;
 		goto unlock;
 	}
+	if (!SUBDEV_VALID_PORTID(sdev)) {
+		ERROR("%s: Invalid TX_SUBDEV port_id=%u\n",
+			__func__, PORT_ID(sdev));
+		ret = -ENODEV;
+		goto unlock;
+	}
 	if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
 		ret = -ENOTSUP;
 		goto unlock;
diff --git a/drivers/net/failsafe/failsafe_private.h b/drivers/net/failsafe/failsafe_private.h
index cd39d103c6..0227060bcb 100644
--- a/drivers/net/failsafe/failsafe_private.h
+++ b/drivers/net/failsafe/failsafe_private.h
@@ -308,6 +308,12 @@  extern int failsafe_mac_from_arg;
 	 : (PRIV(dev)->subs[PRIV(dev)->subs_tx].state < DEV_PROBED ? NULL \
 	 : &PRIV(dev)->subs[PRIV(dev)->subs_tx]))
 
+/**
+ * check for fail-safe sub-device valid port
+ */
+#define SUBDEV_VALID_PORTID(s) \
+	rte_eth_dev_is_valid_port(PORT_ID(s))
+
 /**
  * s:   (struct sub_device *)
  * ops: (struct eth_dev_ops) member