ixgbe: only process one ixgbe_watchdog_task at a time.
Subsystems (2)
PATCH v2 1/1 ixgbe: only process one ixgbe_watchdog_task at a time.
From: John Fastabend <[email protected]>
Processing multiple ixgbe_watchdog_task calls may cause the link_up variable and IXGBE_FLAG_NEED_LINK_UPDATE flag to be set incorrectly. In the worse case this is causing the netif_carrier_off to be called inappropriately which results in an interface that can't be brought up.
Although schedule_work() will only schedule the task if it is not already on the work queue the WORK_STRUCT_PENDING bits are cleared just before calling the work function. This allows WORK_STRUCT_PENDING to be cleared, the work function to start and meanwhile schedule another task.
This patch adds a mutex to the watchdog task. This bug is actualized by changing DCB settings or doing extended cable pull or reset tests.
---
drivers/net/ixgbe/ixgbe.h | 17 ++++++++--------- drivers/net/ixgbe/ixgbe_main.c | 13 +++++++++---- 2 files changed, 17 insertions(+), 13 deletions(-)
drivers/net/ixgbe/ixgbe.h +8 -9
@@ -342,15 +342,14 @@ struct ixgbe_adapter {#define IXGBE_FLAG_VMDQ_ENABLED (u32)(1 << 19)#define IXGBE_FLAG_FAN_FAIL_CAPABLE (u32)(1 << 20)#define IXGBE_FLAG_NEED_LINK_UPDATE (u32)(1 << 22) -#define IXGBE_FLAG_IN_WATCHDOG_TASK (u32)(1 << 23) -#define IXGBE_FLAG_IN_SFP_LINK_TASK (u32)(1 << 24) -#define IXGBE_FLAG_IN_SFP_MOD_TASK (u32)(1 << 25) -#define IXGBE_FLAG_FDIR_HASH_CAPABLE (u32)(1 << 26) -#define IXGBE_FLAG_FDIR_PERFECT_CAPABLE (u32)(1 << 27) -#define IXGBE_FLAG_FCOE_CAPABLE (u32)(1 << 28) -#define IXGBE_FLAG_FCOE_ENABLED (u32)(1 << 29) -#define IXGBE_FLAG_SRIOV_CAPABLE (u32)(1 << 30) -#define IXGBE_FLAG_SRIOV_ENABLED (u32)(1 << 31) +#define IXGBE_FLAG_IN_SFP_LINK_TASK (u32)(1 << 23) +#define IXGBE_FLAG_IN_SFP_MOD_TASK (u32)(1 << 24) +#define IXGBE_FLAG_FDIR_HASH_CAPABLE (u32)(1 << 25) +#define IXGBE_FLAG_FDIR_PERFECT_CAPABLE (u32)(1 << 26) +#define IXGBE_FLAG_FCOE_CAPABLE (u32)(1 << 27) +#define IXGBE_FLAG_FCOE_ENABLED (u32)(1 << 28) +#define IXGBE_FLAG_SRIOV_CAPABLE (u32)(1 << 29) +#define IXGBE_FLAG_SRIOV_ENABLED (u32)(1 << 30)u32 flags2;#define IXGBE_FLAG2_RSC_CAPABLE (u32)(1)
drivers/net/ixgbe/ixgbe_main.c +9 -4
@@ -5043,6 +5043,8 @@ static void ixgbe_fdir_reinit_task(struct work_struct *work)netif_tx_start_all_queues (adapter->netdev); } +static DEFINE_MUTEX (ixgbe_watchdog_lock); +/** * ixgbe_watchdog_task - worker thread to bring link up * @work: pointer to work_struct containing our data
@@ -5054,13 +5056,16 @@ static void ixgbe_watchdog_task(struct work_struct *work) watchdog_task);struct net_device *netdev = adapter->netdev;struct ixgbe_hw *hw = &adapter->hw; -u32 link_speed = adapter->link_speed; -bool link_up = adapter->link_up; +u32 link_speed; +bool link_up;int i;struct ixgbe_ring *tx_ring;int some_tx_pending =0 ; - adapter->flags |=IXGBE_FLAG_IN_WATCHDOG_TASK ; +mutex_lock (&ixgbe_watchdog_lock); + + link_up = adapter->link_up; + link_speed = adapter->link_speed;if (adapter->flags &IXGBE_FLAG_NEED_LINK_UPDATE ) { hw->mac.ops.check_link (hw, &link_speed, &link_up, false);
@@ -5149,7 +5154,7 @@ static void ixgbe_watchdog_task(struct work_struct *work) }ixgbe_update_stats (adapter); - adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK ; +mutex_unlock (&ixgbe_watchdog_lock); } static int ixgbe_tso(struct ixgbe_adapter *adapter,
From: Jeff Kirsher <[email protected]>
Date: Wed, 03 Feb 2010 16:23:32 -0800
Quoted 14 lines
From: John Fastabend <[email protected]>
Processing multiple
ixgbe_watchdog_taskcalls may cause thelink_upvariable andIXGBE_FLAG_NEED_LINK_UPDATEflag to be set incorrectly. In the worse case this is causing thenetif_carrier_offto be called inappropriately which results in an interface that can't be brought up.Although
schedule_work()will only schedule the task if it is not already on the work queue theWORK_STRUCT_PENDINGbits are cleared just before calling the work function. This allowsWORK_STRUCT_PENDINGto be cleared, the work function to start and meanwhile schedule another task.
Quoted 6 lines
This patch adds a mutex to the watchdog task. This bug is actualized by changing DCB settings or doing extended cable pull or reset tests.
Signed-off-by John Fastabend <[email protected]>Signed-off-by Jeff Kirsher <[email protected]>
Applied.