Skip to contentSkip to conversation

vmxnet3: prevent div-by-zero panic when ring resizing uninitialized dev

PATCH In review Bhavesh Davda 2 participants 6 messages Mar 6, 2013 – Mar 7, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vmxnet3: prevent div-by-zero panic when ring resizing uninitialized dev

Linux is free to call ethtool ops as soon as a netdev exists when probe finishes. However, we only allocate vmxnet3 tx/rx queues and initialize the rx_buf_per_pkt field in struct vmxnet3_adapter when the interface is opened (UP).

---
 drivers/net/vmxnet3/vmxnet3_drv.c     |    3 ++-
 drivers/net/vmxnet3/vmxnet3_ethtool.c |    6 ++++++
 drivers/net/vmxnet3/vmxnet3_int.h     |    4 ++--
 3 files changed, 10 insertions(+), 3 deletions(-)
drivers/net/vmxnet3/vmxnet3_drv.c +2 -1
index 4aad350..bfc5898 100644
@@ -2922,7 +2922,7 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 				    (int)num_online_cpus());
 	else
 #endif
-		num_rx_queues = 1;
+	num_rx_queues = 1;
 	num_rx_queues = rounddown_pow_of_two(num_rx_queues);
 
 	if (enable_mq)
@@ -2958,6 +2958,7 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 
 	adapter->num_rx_queues = num_rx_queues;
 	adapter->num_tx_queues = num_tx_queues;
+	adapter->rx_buf_per_pkt = 1;
 
 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
drivers/net/vmxnet3/vmxnet3_ethtool.c +6 -0
index a0feb17..0a4fc9e 100644
@@ -472,6 +472,12 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 						VMXNET3_RX_RING_MAX_SIZE)
 		return -EINVAL;
 
+	/* if adapter not yet initialized, do nothing */
+	if (adapter->rx_buf_per_pkt == 0) {
+		netdev_info(netdev,
+				"adapter not completely initialized, ring size cannot be changed yet\n");
+		return -EOPNOTSUPP;
+	}
 
 	/* round it up to a multiple of VMXNET3_RING_SIZE_ALIGN */
 	new_tx_ring_size = (param->tx_pending + VMXNET3_RING_SIZE_MASK) &
drivers/net/vmxnet3/vmxnet3_int.h +2 -2
index 3198384..3541814 100644
@@ -70,10 +70,10 @@
 /*
  * Version numbers
  */
-#define VMXNET3_DRIVER_VERSION_STRING   "1.1.29.0-k"
+#define VMXNET3_DRIVER_VERSION_STRING   "1.1.30.0-k"
 
 /* a 32-bit int, each byte encode a verion number in VMXNET3_DRIVER_VERSION */
-#define VMXNET3_DRIVER_VERSION_NUM      0x01011D00
+#define VMXNET3_DRIVER_VERSION_NUM      0x01011E00
 
 #if defined(CONFIG_PCI_MSI)
 	/* RSS only makes sense if MSI-X is supported. */
Signature · 1 line
1.7.1
maintainer
#1 lore raw

From: Bhavesh Davda <[email protected]>
Date: Wed, 6 Mar 2013 09:46:24 -0800

Quoted 18 lines

Linux is free to call ethtool ops as soon as a netdev exists when probe finishes. However, we only allocate vmxnet3 tx/rx queues and initialize the rx_buf_per_pkt field in struct vmxnet3_adapter when the interface is opened (UP).

---
 drivers/net/vmxnet3/vmxnet3_drv.c     |    3 ++-
 drivers/net/vmxnet3/vmxnet3_ethtool.c |    6 ++++++
 drivers/net/vmxnet3/vmxnet3_int.h     |    4 ++--
 3 files changed, 10 insertions(+), 3 deletions(-)
drivers/net/vmxnet3/vmxnet3_drv.c +0 -0
index 4aad350..bfc5898 100644

drivers/net/vmxnet3/vmxnet3_drv.c vmxnet3_probe_device(struct pci_dev *pdev, · line 2925

Quoted 5 lines
drivers/net/vmxnet3/vmxnet3_drv.c +1 -1
@@ -2922,7 +2922,7 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 				    (int)num_online_cpus());
 	else
 #endif
-		num_rx_queues = 1;
+	num_rx_queues = 1;

Why are you un-indenting this else branch? Please don't do that.

#2 lore raw
Quoted 6 lines
Quoted 6 lines
---
 drivers/net/vmxnet3/vmxnet3_drv.c     |    3 ++-
 drivers/net/vmxnet3/vmxnet3_ethtool.c |    6 ++++++
 drivers/net/vmxnet3/vmxnet3_int.h     |    4 ++--
 3 files changed, 10 insertions(+), 3 deletions(-)
Quoted 14 lines

drivers/net/vmxnet3/vmxnet3_drv.c vmxnet3_probe_device(struct pci_dev *pdev, · line 2925

Quoted 11 lines
drivers/net/vmxnet3/vmxnet3_drv.c +1 -1
index 4aad350..bfc5898 100644
@@ -2922,7 +2922,7 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 				    (int)num_online_cpus());
 	else
 #endif
-		num_rx_queues = 1;
+	num_rx_queues = 1;

Why are you un-indenting this else branch? Please don't do that

Just to annoy you :) [Sorry, couldn't resist.]

I'm sorry I didn't notice I was breaking the indent of the else branch. For some reason checkpatch didn't complain.

While at it, I'm also changing the netdev_info to netdev_err in vmxnet3_ethtool.c.

You didn't explicitly ask for it, so I wasn't sure if I should repost the patch. I'm going to repost it any way and leave it up to you on which patch you want to apply.

Thanks.

- Bhavesh

#3 lore raw

PATCH 1/1 vmxnet3: prevent div-by-zero panic when ring resizing uninitialized dev

In reply to Bhavesh Davda

Linux is free to call ethtool ops as soon as a netdev exists when probe finishes. However, we only allocate vmxnet3 tx/rx queues and initialize the rx_buf_per_pkt field in struct vmxnet3_adapter when the interface is opened (UP).

---
 drivers/net/vmxnet3/vmxnet3_drv.c     |    1 +
 drivers/net/vmxnet3/vmxnet3_ethtool.c |    6 ++++++
 drivers/net/vmxnet3/vmxnet3_int.h     |    4 ++--
 3 files changed, 9 insertions(+), 2 deletions(-)
drivers/net/vmxnet3/vmxnet3_drv.c +1 -0
index 4aad350..eae7a03 100644
@@ -2958,6 +2958,7 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 
 	adapter->num_rx_queues = num_rx_queues;
 	adapter->num_tx_queues = num_tx_queues;
+	adapter->rx_buf_per_pkt = 1;
 
 	size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
 	size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
drivers/net/vmxnet3/vmxnet3_ethtool.c +6 -0
index a0feb17..63a1243 100644
@@ -472,6 +472,12 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 						VMXNET3_RX_RING_MAX_SIZE)
 		return -EINVAL;
 
+	/* if adapter not yet initialized, do nothing */
+	if (adapter->rx_buf_per_pkt == 0) {
+		netdev_err(netdev, "adapter not completely initialized, "
+			   "ring size cannot be changed yet\n");
+		return -EOPNOTSUPP;
+	}
 
 	/* round it up to a multiple of VMXNET3_RING_SIZE_ALIGN */
 	new_tx_ring_size = (param->tx_pending + VMXNET3_RING_SIZE_MASK) &
drivers/net/vmxnet3/vmxnet3_int.h +2 -2
index 3198384..3541814 100644
@@ -70,10 +70,10 @@
 /*
  * Version numbers
  */
-#define VMXNET3_DRIVER_VERSION_STRING   "1.1.29.0-k"
+#define VMXNET3_DRIVER_VERSION_STRING   "1.1.30.0-k"
 
 /* a 32-bit int, each byte encode a verion number in VMXNET3_DRIVER_VERSION */
-#define VMXNET3_DRIVER_VERSION_NUM      0x01011D00
+#define VMXNET3_DRIVER_VERSION_NUM      0x01011E00
 
 #if defined(CONFIG_PCI_MSI)
 	/* RSS only makes sense if MSI-X is supported. */
Signature · 1 line
1.7.1
maintainer
#4 lore raw

From: Bhavesh Davda <[email protected]>
Date: Wed, 6 Mar 2013 14:04:53 -0800

Quoted 7 lines

Linux is free to call ethtool ops as soon as a netdev exists when probe finishes. However, we only allocate vmxnet3 tx/rx queues and initialize the rx_buf_per_pkt field in struct vmxnet3_adapter when the interface is opened (UP).

Maybe this is more appropriate for 'net' rather than 'net-next'?

maintainer
#5 lore raw
In reply to Bhavesh Davda

From: Bhavesh Davda <[email protected]>
Date: Wed, 6 Mar 2013 14:04:53 -0800

Quoted 7 lines

Linux is free to call ethtool ops as soon as a netdev exists when probe finishes. However, we only allocate vmxnet3 tx/rx queues and initialize the rx_buf_per_pkt field in struct vmxnet3_adapter when the interface is opened (UP).

As discussed yesterday, I'm applying this to 'net', thanks.