hinic: fix rewaking txq after netif_tx_disable
Subsystems (3)
PATCH 1/1 hinic: fix rewaking txq after netif_tx_disable
When calling hinic_close in hinic_set_channels, all queues are stopped after netif_tx_disable, but some queue may be rewaken in free_tx_poll by mistake while drv is handling tx irq. If one queue is rewaken core may call hinic_xmit_frame to send pkt after netif_tx_disable within a short time which may results in accessing memory that has been already freed in hinic_close. So we call napi_disable before netif_tx_disable in hinic_close to fix this bug.
---
V0~V1:
- call napi_disable before netif_tx_disable instead of judging whether the netdev is in down state before waking txq in free_tx_poll to fix this bug
.../net/ethernet/huawei/hinic/hinic_main.c | 24 +++++++++++++++++++ drivers/net/ethernet/huawei/hinic/hinic_tx.c | 20 ++++------------ 2 files changed, 28 insertions(+), 16 deletions(-)
drivers/net/ethernet/huawei/hinic/hinic_main.c +24 -0
@@ -174,6 +174,24 @@ static int create_txqs(struct hinic_dev *nic_dev)return err; } +static void enable_txqs_napi (struct hinic_dev *nic_dev) +{ +int num_txqs =hinic_hwdev_num_qps (nic_dev->hwdev); +int i; + +for (i =0 ; i < num_txqs; i++) +napi_enable (&nic_dev->txqs[i].napi); +} + +static void disable_txqs_napi (struct hinic_dev *nic_dev) +{ +int num_txqs =hinic_hwdev_num_qps (nic_dev->hwdev); +int i; + +for (i =0 ; i < num_txqs; i++) +napi_disable (&nic_dev->txqs[i].napi); +} +/** * free_txqs - Free the Logical Tx Queues of specific NIC device * @nic_dev: the specific NIC device
@@ -400,6 +418,8 @@ int hinic_open(struct net_device *netdev) goto err_create_txqs; } +enable_txqs_napi (nic_dev); + err =create_rxqs (nic_dev);if (err) { netif_err(nic_dev, drv, netdev,
@@ -484,6 +504,7 @@ int hinic_open(struct net_device *netdev) } err_create_rxqs: +disable_txqs_napi (nic_dev);free_txqs (nic_dev); err_create_txqs:
@@ -497,6 +518,9 @@ int hinic_close(struct net_device *netdev)struct hinic_dev *nic_dev =netdev_priv (netdev);unsigned int flags; +/* Disable txq napi firstly to aviod rewaking txq in free_tx_poll */ +disable_txqs_napi (nic_dev); +down (&nic_dev->mgmt_lock); flags = nic_dev->flags;
drivers/net/ethernet/huawei/hinic/hinic_tx.c +4 -16
@@ -745,18 +745,6 @@ static int free_tx_poll(struct napi_struct *napi, int budget)return budget; } -static void tx_napi_add (struct hinic_txq *txq,int weight) -{ -netif_napi_add (txq->netdev, &txq->napi, free_tx_poll, weight); -napi_enable (&txq->napi); -} - -static void tx_napi_del (struct hinic_txq *txq) -{ -napi_disable (&txq->napi); -netif_napi_del (&txq->napi); -} -static irqreturn_t tx_irq (int irq,void *data) {struct hinic_txq *txq = data;
@@ -790,7 +778,7 @@ static int tx_request_irq(struct hinic_txq *txq) qp = container_of(sq, struct hinic_qp, sq); -tx_napi_add (txq, nic_dev->tx_weight); +netif_napi_add (txq->netdev, &txq->napi, free_tx_poll, nic_dev->tx_weight); hinic_hwdev_msix_set(nic_dev->hwdev, sq->msix_entry, TX_IRQ_NO_PENDING, TX_IRQ_NO_COALESC,
@@ -807,14 +795,14 @@ static int tx_request_irq(struct hinic_txq *txq)if (err) {netif_err (nic_dev, drv, txq->netdev,"Failed to set TX interrupt coalescing attribute\n" ); -tx_napi_del (txq); +netif_napi_del (&txq->napi);return err; } err =request_irq (sq->irq, tx_irq,0 , txq->irq_name, txq);if (err) {dev_err (&pdev->dev,"Failed to request Tx irq\n" ); -tx_napi_del (txq); +netif_napi_del (&txq->napi);return err; }
@@ -826,7 +814,7 @@ static void tx_free_irq(struct hinic_txq *txq)struct hinic_sq *sq = txq->sq;free_irq (sq->irq, txq); -tx_napi_del (txq); +netif_napi_del (&txq->napi); }/**
Signature · 1 line
2.17.1
Quoted 10 lines · On Thu, 10 Sep 2020 22:04:40 +0800 Luo bin wrote:
When calling
hinic_closeinhinic_set_channels, all queues are stopped afternetif_tx_disable, but some queue may be rewaken infree_tx_pollby mistake while drv is handling tx irq. If one queue is rewaken core may callhinic_xmit_frameto send pkt afternetif_tx_disablewithin a short time which may results in accessing memory that has been already freed inhinic_close. So we callnapi_disablebeforenetif_tx_disableinhinic_closeto fix this bug.Fixes 2eed5a8b614b ("hinic: add set_channels ethtool_ops support")Signed-off-by Luo bin <[email protected]>
David Miller
maintainerFrom: Luo bin <[email protected]>
Date: Thu, 10 Sep 2020 22:04:40 +0800
Quoted 8 lines
When calling
hinic_closeinhinic_set_channels, all queues are stopped afternetif_tx_disable, but some queue may be rewaken infree_tx_pollby mistake while drv is handling tx irq. If one queue is rewaken core may callhinic_xmit_frameto send pkt afternetif_tx_disablewithin a short time which may results in accessing memory that has been already freed inhinic_close. So we callnapi_disablebeforenetif_tx_disableinhinic_closeto fix this bug.
Quoted 7 lines
Fixes 2eed5a8b614b ("hinic: add set_channels ethtool_ops support")Signed-off-by Luo bin <[email protected]>---
V0~V1:
- callnapi_disablebeforenetif_tx_disableinstead of judging whether the netdev is in down state before waking txq infree_tx_pollto fix this bug
Applied and queued up for -stable, thank you.