[dpdk-dev] [PATCH] net/ixgbe: fix resource leak after thread exits normally

[dpdk-dev] [PATCH] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: "Zhu, Tao" <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..f141ae4 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -4165,11 +4165,9 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
 {
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
 
 	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
 		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
 		rte_atomic32_clear(&ad->link_thread_running);
 		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
 	}
@@ -4186,6 +4184,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
-- 
1.8.3.1

[dpdk-dev] [PATCH v2] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: Zhu Tao <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

v2 changes:
	commit log.

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..f141ae4 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -4165,11 +4165,9 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
 {
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
 
 	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
 		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
 		rte_atomic32_clear(&ad->link_thread_running);
 		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
 	}
@@ -4186,6 +4184,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v2] net/ixgbe: fix resource leak after thread exits normally

[Ananyev, Konstantin]

Hi Tao,

> From: Zhu Tao <taox.zhu@intel.com>
> 
> When the thread exits normally, pthread_join() is not called, which can
> result in a resource leak. Therefore, the thread is set to separation
> mode using function pthread_detach(), so that no program call
> pthread_join() is required to recycle, and when the thread exits,
> the system automatically reclaims resources.
> 
> Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zhu Tao <taox.zhu@intel.com>
> ---
>  drivers/net/ixgbe/ixgbe_ethdev.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> v2 changes:
> 	commit log.
> 
> diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
> index 2c57976..f141ae4 100644
> --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> @@ -4165,11 +4165,9 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
>  ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
>  {
>  	struct ixgbe_adapter *ad = dev->data->dev_private;
> -	void *retval;
> 
>  	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
>  		pthread_cancel(ad->link_thread_tid);
> -		pthread_join(ad->link_thread_tid, &retval);

As you already waiting for link thread termination in
ixgbe_dev_wait_setup_link_complete(), do you still need
pthread_cancel() here? 

>  		rte_atomic32_clear(&ad->link_thread_running);
>  		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
>  	}
> @@ -4186,6 +4184,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
>  	u32 speed;
>  	bool autoneg = false;
> 
> +	pthread_detach(pthread_self());
>  	speed = hw->phy.autoneg_advertised;
>  	if (!speed)
>  		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
> --
> 1.8.3.1

Re: [dpdk-dev] [PATCH v2] net/ixgbe: fix resource leak after thread exits normally

[Ananyev, Konstantin]

Hi Tao,

 
> Because ixgbe_dev_wait_setup_link_complete() doesn't guarantee that the thread will always complete properly. So after the wait
> timeout, need to call pthread_cancel() to cancel the thread.

But pthread_cancel() also doesn't guarantee immediate thread termination.
You'll have to wait for link_thread_running again to make sure.
Another thing - in theory  if link_thread was already terminated,
then its TID can be reused by other thread and we will terminate different thread
(though chances are quite small). 

> Definition of function ixgbe_dev_wait_setup_link_complete() as below:
> 
> /* return 1: setup complete, return 0: setup not complete, and wait timeout*/ static int ixgbe_dev_wait_setup_link_complete(struct
> rte_eth_dev *dev) { #define DELAY_INTERVAL 100 /* 100ms */
> #define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
>         struct ixgbe_adapter *ad = dev->data->dev_private;
>         int timeout = MAX_TIMEOUT;
> 
>         while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
>                 msec_delay(DELAY_INTERVAL);
>                 timeout--;
>         }
> 
> 
>         return !!timeout;
> }
> 
> BR,
> Zhu, Tao
> 
> > -----Original Message-----
> > From: Ananyev, Konstantin
> > Sent: Friday, April 10, 2020 8:10 PM
> > To: Zhu, TaoX <taox.zhu@intel.com>; Lu, Wenzhuo
> > <wenzhuo.lu@intel.com>; Ye, Xiaolong <xiaolong.ye@intel.com>
> > Cc: dev@dpdk.org; martin.weiser@allegro-packets.com; stable@dpdk.org
> > Subject: RE: [PATCH v2] net/ixgbe: fix resource leak after thread exits
> > normally
> >
> > Hi Tao,
> >
> > > From: Zhu Tao <taox.zhu@intel.com>
> > >
> > > When the thread exits normally, pthread_join() is not called, which can
> > > result in a resource leak. Therefore, the thread is set to separation
> > > mode using function pthread_detach(), so that no program call
> > > pthread_join() is required to recycle, and when the thread exits,
> > > the system automatically reclaims resources.
> > >
> > > Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
> > > Cc: stable@dpdk.org
> > >
> > > Signed-off-by: Zhu Tao <taox.zhu@intel.com>
> > > ---
> > >  drivers/net/ixgbe/ixgbe_ethdev.c | 3 +--
> > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > >
> > > v2 changes:
> > > commit log.
> > >
> > > diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c
> > b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > index 2c57976..f141ae4 100644
> > > --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> > > +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > @@ -4165,11 +4165,9 @@ static int
> > ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> > >  ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
> > >  {
> > >  struct ixgbe_adapter *ad = dev->data->dev_private;
> > > -void *retval;
> > >
> > >  if (!ixgbe_dev_wait_setup_link_complete(dev)) {
> > >  pthread_cancel(ad->link_thread_tid);
> > > -pthread_join(ad->link_thread_tid, &retval);
> >
> > As you already waiting for link thread termination in
> > ixgbe_dev_wait_setup_link_complete(), do you still need
> > pthread_cancel() here?
> >
> > >  rte_atomic32_clear(&ad->link_thread_running);
> > >  PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
> > >  }
> > > @@ -4186,6 +4184,7 @@ static int
> > ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> > >  u32 speed;
> > >  bool autoneg = false;
> > >
> > > +pthread_detach(pthread_self());
> > >  speed = hw->phy.autoneg_advertised;
> > >  if (!speed)
> > >  ixgbe_get_link_capabilities(hw, &speed, &autoneg);
> > > --
> > > 1.8.3.1
> >
> 

Re: [dpdk-dev] [PATCH v2] net/ixgbe: fix resource leak after thread exits normally

[Ananyev, Konstantin]

> Hi Tao,
> 
> 
> > Because ixgbe_dev_wait_setup_link_complete() doesn't guarantee that the thread will always complete properly. So after the wait
> > timeout, need to call pthread_cancel() to cancel the thread.
> 
> But pthread_cancel() also doesn't guarantee immediate thread termination.
> You'll have to wait for link_thread_running again to make sure.
> Another thing - in theory  if link_thread was already terminated,
> then its TID can be reused by other thread and we will terminate different thread
> (though chances are quite small).
> 
> > Definition of function ixgbe_dev_wait_setup_link_complete() as below:
> >
> > /* return 1: setup complete, return 0: setup not complete, and wait timeout*/ static int ixgbe_dev_wait_setup_link_complete(struct
> > rte_eth_dev *dev) { #define DELAY_INTERVAL 100 /* 100ms */

As another thought - instead of unconditionally defining timeout
inside the function itself, can we have it a s function parameter?
Can caller can decide for how long he is ok to wait.

> > #define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
> >         struct ixgbe_adapter *ad = dev->data->dev_private;
> >         int timeout = MAX_TIMEOUT;
> >
> >         while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
> >                 msec_delay(DELAY_INTERVAL);
> >                 timeout--;
> >         }
> >
> >
> >         return !!timeout;
> > }
> >
> > BR,
> > Zhu, Tao
> >
> > > -----Original Message-----
> > > From: Ananyev, Konstantin
> > > Sent: Friday, April 10, 2020 8:10 PM
> > > To: Zhu, TaoX <taox.zhu@intel.com>; Lu, Wenzhuo
> > > <wenzhuo.lu@intel.com>; Ye, Xiaolong <xiaolong.ye@intel.com>
> > > Cc: dev@dpdk.org; martin.weiser@allegro-packets.com; stable@dpdk.org
> > > Subject: RE: [PATCH v2] net/ixgbe: fix resource leak after thread exits
> > > normally
> > >
> > > Hi Tao,
> > >
> > > > From: Zhu Tao <taox.zhu@intel.com>
> > > >
> > > > When the thread exits normally, pthread_join() is not called, which can
> > > > result in a resource leak. Therefore, the thread is set to separation
> > > > mode using function pthread_detach(), so that no program call
> > > > pthread_join() is required to recycle, and when the thread exits,
> > > > the system automatically reclaims resources.
> > > >
> > > > Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
> > > > Cc: stable@dpdk.org
> > > >
> > > > Signed-off-by: Zhu Tao <taox.zhu@intel.com>
> > > > ---
> > > >  drivers/net/ixgbe/ixgbe_ethdev.c | 3 +--
> > > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > > >
> > > > v2 changes:
> > > > commit log.
> > > >
> > > > diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c
> > > b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > > index 2c57976..f141ae4 100644
> > > > --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> > > > +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > > @@ -4165,11 +4165,9 @@ static int
> > > ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> > > >  ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
> > > >  {
> > > >  struct ixgbe_adapter *ad = dev->data->dev_private;
> > > > -void *retval;
> > > >
> > > >  if (!ixgbe_dev_wait_setup_link_complete(dev)) {
> > > >  pthread_cancel(ad->link_thread_tid);
> > > > -pthread_join(ad->link_thread_tid, &retval);
> > >
> > > As you already waiting for link thread termination in
> > > ixgbe_dev_wait_setup_link_complete(), do you still need
> > > pthread_cancel() here?
> > >
> > > >  rte_atomic32_clear(&ad->link_thread_running);
> > > >  PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
> > > >  }
> > > > @@ -4186,6 +4184,7 @@ static int
> > > ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> > > >  u32 speed;
> > > >  bool autoneg = false;
> > > >
> > > > +pthread_detach(pthread_self());
> > > >  speed = hw->phy.autoneg_advertised;
> > > >  if (!speed)
> > > >  ixgbe_get_link_capabilities(hw, &speed, &autoneg);
> > > > --
> > > > 1.8.3.1
> > >
> >

[dpdk-dev] [PATCH v3] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: Zhu Tao <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Wait for the thread to finish without setting the timeout, wait until
the thread finishes before returning. Normally, the thread will finish
in a shorter time, and give a warning message if it hasn't finished in
a longer time.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")

Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 46 +++++++++++++++-------------------------
 1 file changed, 17 insertions(+), 29 deletions(-)

v3 changes:
	1. Wait for the thread to finish without setting the timeout, and the 
	   corresponding function name has also been modified.
	2. Commit log.	

v2 changes:
        Commit log.

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..dac20a3 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
 static void ixgbe_dev_interrupt_handler(void *param);
 static void ixgbe_dev_interrupt_delayed_handler(void *param);
 static void *ixgbe_dev_setup_link_thread_handler(void *param);
-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
+static void ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev);
 
 static int ixgbe_add_rar(struct rte_eth_dev *dev,
 			struct rte_ether_addr *mac_addr,
@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	/* disable uio/vfio intr/eventfd mapping */
 	rte_intr_disable(intr_handle);
@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	/* disable interrupts */
 	ixgbe_disable_intr(hw);
@@ -4143,35 +4143,22 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	return ret_val;
 }
 
-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
-static int
+static void
 ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
 {
-#define DELAY_INTERVAL 100 /* 100ms */
-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
+#define DELAY_INTERVAL 10 /* 10ms */
+#define WARNING_TIMEOUT    900 /* 9s (900 * 10ms) in total */
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	int timeout = MAX_TIMEOUT;
+	int timeout = WARNING_TIMEOUT;
 
-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
+	while (rte_atomic32_read(&ad->link_thread_running)) {
 		msec_delay(DELAY_INTERVAL);
 		timeout--;
-	}
-
-
-	return !!timeout;
-}
 
-static void
-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
-{
-	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
-
-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
-		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
-		rte_atomic32_clear(&ad->link_thread_running);
-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
+		if (!timeout) {
+			timeout = WARNING_TIMEOUT;
+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
+		}
 	}
 }
 
@@ -4186,6 +4173,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
@@ -4282,8 +4270,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	if (link_up == 0) {
 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
+            ixgbe_dev_wait_setup_link_complete(dev);
+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
 					"ixgbe-link-handler",
 					NULL,
@@ -5323,7 +5311,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	err = hw->mac.ops.reset_hw(hw);
 	if (err) {
@@ -5421,7 +5409,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	ixgbevf_intr_disable(dev);
 
-- 
1.8.3.1

[dpdk-dev] [PATCH v4] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: Zhu Tao <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Wait for the thread to finish without setting the timeout, wait until
the thread finishes before returning. Normally, the thread will finish
in a shorter time, and give a warning message if it hasn't finished in
a longer time.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 46 +++++++++++++++-------------------------
 1 file changed, 17 insertions(+), 29 deletions(-)
v4 changes:
	Format codes.

v3 changes:
        1. Wait for the thread to finish without setting the timeout, and the
           corresponding function name has also been modified.
        2. Commit log.

v2 changes:
        Commit log.

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..9ad93c5 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
 static void ixgbe_dev_interrupt_handler(void *param);
 static void ixgbe_dev_interrupt_delayed_handler(void *param);
 static void *ixgbe_dev_setup_link_thread_handler(void *param);
-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
+static void ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev);
 
 static int ixgbe_add_rar(struct rte_eth_dev *dev,
 			struct rte_ether_addr *mac_addr,
@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	/* disable uio/vfio intr/eventfd mapping */
 	rte_intr_disable(intr_handle);
@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	/* disable interrupts */
 	ixgbe_disable_intr(hw);
@@ -4143,35 +4143,22 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	return ret_val;
 }
 
-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
-static int
+static void
 ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
 {
-#define DELAY_INTERVAL 100 /* 100ms */
-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
+#define DELAY_INTERVAL 10 /* 10ms */
+#define WARNING_TIMEOUT    900 /* 9s (900 * 10ms) in total */
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	int timeout = MAX_TIMEOUT;
+	int timeout = WARNING_TIMEOUT;
 
-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
+	while (rte_atomic32_read(&ad->link_thread_running)) {
 		msec_delay(DELAY_INTERVAL);
 		timeout--;
-	}
-
-
-	return !!timeout;
-}
 
-static void
-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
-{
-	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
-
-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
-		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
-		rte_atomic32_clear(&ad->link_thread_running);
-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
+		if (!timeout) {
+			timeout = WARNING_TIMEOUT;
+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
+		}
 	}
 }
 
@@ -4186,6 +4173,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
@@ -4282,8 +4270,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	if (link_up == 0) {
 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
+			ixgbe_dev_wait_setup_link_complete(dev);
+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
 					"ixgbe-link-handler",
 					NULL,
@@ -5323,7 +5311,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	err = hw->mac.ops.reset_hw(hw);
 	if (err) {
@@ -5421,7 +5409,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev);
 
 	ixgbevf_intr_disable(dev);
 
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v4] net/ixgbe: fix resource leak after thread exits normally

[Ye Xiaolong]

On 04/15, taox.zhu@intel.com wrote:
>From: Zhu Tao <taox.zhu@intel.com>
>
>When the thread exits normally, pthread_join() is not called, which can
>result in a resource leak. Therefore, the thread is set to separation
>mode using function pthread_detach(), so that no program call
>pthread_join() is required to recycle, and when the thread exits,
>the system automatically reclaims resources.
>
>Wait for the thread to finish without setting the timeout, wait until
>the thread finishes before returning. Normally, the thread will finish
>in a shorter time, and give a warning message if it hasn't finished in
>a longer time.
>
>Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
>Cc: stable@dpdk.org
>
>Signed-off-by: Zhu Tao <taox.zhu@intel.com>
>---
> drivers/net/ixgbe/ixgbe_ethdev.c | 46 +++++++++++++++-------------------------
> 1 file changed, 17 insertions(+), 29 deletions(-)
>v4 changes:
>	Format codes.
>
>v3 changes:
>        1. Wait for the thread to finish without setting the timeout, and the
>           corresponding function name has also been modified.
>        2. Commit log.
>
>v2 changes:
>        Commit log.
>
>diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
>index 2c57976..9ad93c5 100644
>--- a/drivers/net/ixgbe/ixgbe_ethdev.c
>+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
>@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
> static void ixgbe_dev_interrupt_handler(void *param);
> static void ixgbe_dev_interrupt_delayed_handler(void *param);
> static void *ixgbe_dev_setup_link_thread_handler(void *param);
>-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
>+static void ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev);
> 
> static int ixgbe_add_rar(struct rte_eth_dev *dev,
> 			struct rte_ether_addr *mac_addr,
>@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 	PMD_INIT_FUNC_TRACE();
> 
> 	/* Stop the link setup handler before resetting the HW. */
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev);
> 
> 	/* disable uio/vfio intr/eventfd mapping */
> 	rte_intr_disable(intr_handle);
>@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 
> 	PMD_INIT_FUNC_TRACE();
> 
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev);
> 
> 	/* disable interrupts */
> 	ixgbe_disable_intr(hw);
>@@ -4143,35 +4143,22 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	return ret_val;
> }
> 
>-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
>-static int
>+static void
> ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
> {
>-#define DELAY_INTERVAL 100 /* 100ms */
>-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
>+#define DELAY_INTERVAL 10 /* 10ms */
>+#define WARNING_TIMEOUT    900 /* 9s (900 * 10ms) in total */
> 	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	int timeout = MAX_TIMEOUT;
>+	int timeout = WARNING_TIMEOUT;
> 
>-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
>+	while (rte_atomic32_read(&ad->link_thread_running)) {
> 		msec_delay(DELAY_INTERVAL);
> 		timeout--;
>-	}
>-
>-
>-	return !!timeout;
>-}
> 
>-static void
>-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
>-{
>-	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	void *retval;
>-
>-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
>-		pthread_cancel(ad->link_thread_tid);
>-		pthread_join(ad->link_thread_tid, &retval);
>-		rte_atomic32_clear(&ad->link_thread_running);
>-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
>+		if (!timeout) {
>+			timeout = WARNING_TIMEOUT;
>+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");

In theory, is there any possibility that the while loop will never end? 
And as Konstantin suggested, I think it makes sense to provide a parameter for
caller to decide how long they want to wait.

Thanks,
Xiaolong

>+		}
> 	}
> }
> 

[dpdk-dev] [PATCH v5] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: Zhu Tao <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Wait for the thread to finish with timeout argument(0 means that it will
not return until link complete), wait until the thread finishes before
returning. Normally, the thread will finish in a shorter time, and give
a warning message if it hasn't finished in a longer time.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 57 +++++++++++++++++++---------------------
 1 file changed, 27 insertions(+), 30 deletions(-)

V5 changes:
	Modify the definition of function ixgbe_dev_wait_setup_link_complete(),
	 so that the caller can choose whether to block, etc.

v4 changes:
	Format codes.

v3 changes:
        1. Wait for the thread to finish without setting the timeout, and the
           corresponding function name has also been modified.
        2. Commit log.

v2 changes:
        Commit log.

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..92788ee 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
 static void ixgbe_dev_interrupt_handler(void *param);
 static void ixgbe_dev_interrupt_delayed_handler(void *param);
 static void *ixgbe_dev_setup_link_thread_handler(void *param);
-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
+static int ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms);
 
 static int ixgbe_add_rar(struct rte_eth_dev *dev,
 			struct rte_ether_addr *mac_addr,
@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	/* disable uio/vfio intr/eventfd mapping */
 	rte_intr_disable(intr_handle);
@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	/* disable interrupts */
 	ixgbe_disable_intr(hw);
@@ -4143,36 +4143,32 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	return ret_val;
 }
 
-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
+/*
+ * If @timeout_ms was 0, it means that it will not return until link complete.
+ * It returns 1 on complete, return 0 on timeout.
+ */
 static int
-ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
+ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms)
 {
-#define DELAY_INTERVAL 100 /* 100ms */
-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
+#define WARNING_TIMEOUT    9000 /* 9s  in total */
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	int timeout = MAX_TIMEOUT;
+	int timeout = timeout_ms ? WARNING_TIMEOUT : timeout_ms;
 
-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
-		msec_delay(DELAY_INTERVAL);
+	while (rte_atomic32_read(&ad->link_thread_running)) {
+		msec_delay(1);
 		timeout--;
-	}
-
-
-	return !!timeout;
-}
 
-static void
-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
-{
-	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
-
-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
-		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
-		rte_atomic32_clear(&ad->link_thread_running);
-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
+		if (timeout_ms) {
+			if (!timeout)
+				return 0;
+		} else if (!timeout) {
+			/* It will not return until link complete */
+			timeout = WARNING_TIMEOUT;
+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
+		}
 	}
+
+	return 1;
 }
 
 static void *
@@ -4186,6 +4182,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
@@ -4282,8 +4279,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	if (link_up == 0) {
 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
+			ixgbe_dev_wait_setup_link_complete(dev, 0);
+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
 					"ixgbe-link-handler",
 					NULL,
@@ -5323,7 +5320,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	err = hw->mac.ops.reset_hw(hw);
 	if (err) {
@@ -5421,7 +5418,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	ixgbevf_intr_disable(dev);
 
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v5] net/ixgbe: fix resource leak after thread exits normally

[Ananyev, Konstantin]

> 
> When the thread exits normally, pthread_join() is not called, which can
> result in a resource leak. Therefore, the thread is set to separation
> mode using function pthread_detach(), so that no program call
> pthread_join() is required to recycle, and when the thread exits,
> the system automatically reclaims resources.
> 
> Wait for the thread to finish with timeout argument(0 means that it will
> not return until link complete), wait until the thread finishes before
> returning. Normally, the thread will finish in a shorter time, and give
> a warning message if it hasn't finished in a longer time.
> 
> Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zhu Tao <taox.zhu@intel.com>
> ---

Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>


> --
> 1.8.3.1

Re: [dpdk-dev] [PATCH v5] net/ixgbe: fix resource leak after thread exits normally

[Ye Xiaolong]

On 04/21, taox.zhu@intel.com wrote:
>From: Zhu Tao <taox.zhu@intel.com>
>
>When the thread exits normally, pthread_join() is not called, which can
>result in a resource leak. Therefore, the thread is set to separation
>mode using function pthread_detach(), so that no program call
>pthread_join() is required to recycle, and when the thread exits,
>the system automatically reclaims resources.
>
>Wait for the thread to finish with timeout argument(0 means that it will
>not return until link complete), wait until the thread finishes before
>returning. Normally, the thread will finish in a shorter time, and give
>a warning message if it hasn't finished in a longer time.
>
>Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
>Cc: stable@dpdk.org
>
>Signed-off-by: Zhu Tao <taox.zhu@intel.com>
>---
> drivers/net/ixgbe/ixgbe_ethdev.c | 57 +++++++++++++++++++---------------------
> 1 file changed, 27 insertions(+), 30 deletions(-)
>
>V5 changes:
>	Modify the definition of function ixgbe_dev_wait_setup_link_complete(),
>	 so that the caller can choose whether to block, etc.
>
>v4 changes:
>	Format codes.
>
>v3 changes:
>        1. Wait for the thread to finish without setting the timeout, and the
>           corresponding function name has also been modified.
>        2. Commit log.
>
>v2 changes:
>        Commit log.
>
>diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
>index 2c57976..92788ee 100644
>--- a/drivers/net/ixgbe/ixgbe_ethdev.c
>+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
>@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
> static void ixgbe_dev_interrupt_handler(void *param);
> static void ixgbe_dev_interrupt_delayed_handler(void *param);
> static void *ixgbe_dev_setup_link_thread_handler(void *param);
>-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
>+static int ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms);
> 
> static int ixgbe_add_rar(struct rte_eth_dev *dev,
> 			struct rte_ether_addr *mac_addr,
>@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 	PMD_INIT_FUNC_TRACE();
> 
> 	/* Stop the link setup handler before resetting the HW. */
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	/* disable uio/vfio intr/eventfd mapping */
> 	rte_intr_disable(intr_handle);
>@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 
> 	PMD_INIT_FUNC_TRACE();
> 
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	/* disable interrupts */
> 	ixgbe_disable_intr(hw);
>@@ -4143,36 +4143,32 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	return ret_val;
> }
> 
>-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
>+/*
>+ * If @timeout_ms was 0, it means that it will not return until link complete.
>+ * It returns 1 on complete, return 0 on timeout.
>+ */
> static int
>-ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
>+ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms)
> {
>-#define DELAY_INTERVAL 100 /* 100ms */
>-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
>+#define WARNING_TIMEOUT    9000 /* 9s  in total */
> 	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	int timeout = MAX_TIMEOUT;
>+	int timeout = timeout_ms ? WARNING_TIMEOUT : timeout_ms;

Sees timeout_ms is 0, set timeout as 0? I guess what you intended to do is:

	int timeout = timeout_ms ? timeout_ms : WARNING_TIMEOUT;


Thanks,
Xiaolong
	

> 
>-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
>-		msec_delay(DELAY_INTERVAL);
>+	while (rte_atomic32_read(&ad->link_thread_running)) {
>+		msec_delay(1);
> 		timeout--;
>-	}
>-
>-
>-	return !!timeout;
>-}
> 
>-static void
>-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
>-{
>-	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	void *retval;
>-
>-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
>-		pthread_cancel(ad->link_thread_tid);
>-		pthread_join(ad->link_thread_tid, &retval);
>-		rte_atomic32_clear(&ad->link_thread_running);
>-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
>+		if (timeout_ms) {
>+			if (!timeout)
>+				return 0;
>+		} else if (!timeout) {
>+			/* It will not return until link complete */
>+			timeout = WARNING_TIMEOUT;
>+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
>+		}
> 	}
>+
>+	return 1;
> }
> 
> static void *
>@@ -4186,6 +4182,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	u32 speed;
> 	bool autoneg = false;
> 
>+	pthread_detach(pthread_self());
> 	speed = hw->phy.autoneg_advertised;
> 	if (!speed)
> 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
>@@ -4282,8 +4279,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	if (link_up == 0) {
> 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
> 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
>-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
>-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
>+			ixgbe_dev_wait_setup_link_complete(dev, 0);
>+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
> 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
> 					"ixgbe-link-handler",
> 					NULL,
>@@ -5323,7 +5320,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	PMD_INIT_FUNC_TRACE();
> 
> 	/* Stop the link setup handler before resetting the HW. */
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	err = hw->mac.ops.reset_hw(hw);
> 	if (err) {
>@@ -5421,7 +5418,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 
> 	PMD_INIT_FUNC_TRACE();
> 
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	ixgbevf_intr_disable(dev);
> 
>-- 
>1.8.3.1
>

[dpdk-dev] [PATCH v6] net/ixgbe: fix resource leak after thread exits normally

[taox.zhu]

From: Zhu Tao <taox.zhu@intel.com>

When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.

Wait for the thread to finish with timeout argument(0 means that it will
not return until link complete), wait until the thread finishes before
returning. Normally, the thread will finish in a shorter time, and give
a warning message if it hasn't finished in a longer time.

Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
Cc: stable@dpdk.org

Signed-off-by: Zhu Tao <taox.zhu@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 57 +++++++++++++++++++---------------------
 1 file changed, 27 insertions(+), 30 deletions(-)

V6 changes:
        Change code segment
                     'int timeout = timeout_ms ? WARNING_TIMEOUT : timeout_ms;'
        to
                'uint32_t timeout = timeout_ms ? timeout_ms : WARNING_TIMEOUT;'.

V5 changes:
        Modify the definition of function ixgbe_dev_wait_setup_link_complete(),
         so that the caller can choose whether to block, etc.

v4 changes:
        Format codes.

v3 changes:
        1. Wait for the thread to finish without setting the timeout, and the
           corresponding function name has also been modified.
        2. Commit log.

v2 changes:
        Commit log.

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 2c57976..a3ae4d7 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
 static void ixgbe_dev_interrupt_handler(void *param);
 static void ixgbe_dev_interrupt_delayed_handler(void *param);
 static void *ixgbe_dev_setup_link_thread_handler(void *param);
-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
+static int ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms);
 
 static int ixgbe_add_rar(struct rte_eth_dev *dev,
 			struct rte_ether_addr *mac_addr,
@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	/* disable uio/vfio intr/eventfd mapping */
 	rte_intr_disable(intr_handle);
@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	/* disable interrupts */
 	ixgbe_disable_intr(hw);
@@ -4143,36 +4143,32 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	return ret_val;
 }
 
-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
+/*
+ * If @timeout_ms was 0, it means that it will not return until link complete.
+ * It returns 1 on complete, return 0 on timeout.
+ */
 static int
-ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
+ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms)
 {
-#define DELAY_INTERVAL 100 /* 100ms */
-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
+#define WARNING_TIMEOUT    9000 /* 9s  in total */
 	struct ixgbe_adapter *ad = dev->data->dev_private;
-	int timeout = MAX_TIMEOUT;
+	uint32_t timeout = timeout_ms ? timeout_ms : WARNING_TIMEOUT;
 
-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
-		msec_delay(DELAY_INTERVAL);
+	while (rte_atomic32_read(&ad->link_thread_running)) {
+		msec_delay(1);
 		timeout--;
-	}
-
-
-	return !!timeout;
-}
 
-static void
-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
-{
-	struct ixgbe_adapter *ad = dev->data->dev_private;
-	void *retval;
-
-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
-		pthread_cancel(ad->link_thread_tid);
-		pthread_join(ad->link_thread_tid, &retval);
-		rte_atomic32_clear(&ad->link_thread_running);
-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
+		if (timeout_ms) {
+			if (!timeout)
+				return 0;
+		} else if (!timeout) {
+			/* It will not return until link complete */
+			timeout = WARNING_TIMEOUT;
+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
+		}
 	}
+
+	return 1;
 }
 
 static void *
@@ -4186,6 +4182,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	u32 speed;
 	bool autoneg = false;
 
+	pthread_detach(pthread_self());
 	speed = hw->phy.autoneg_advertised;
 	if (!speed)
 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
@@ -4282,8 +4279,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	if (link_up == 0) {
 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
+			ixgbe_dev_wait_setup_link_complete(dev, 0);
+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
 					"ixgbe-link-handler",
 					NULL,
@@ -5323,7 +5320,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 	PMD_INIT_FUNC_TRACE();
 
 	/* Stop the link setup handler before resetting the HW. */
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	err = hw->mac.ops.reset_hw(hw);
 	if (err) {
@@ -5421,7 +5418,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
 
 	PMD_INIT_FUNC_TRACE();
 
-	ixgbe_dev_cancel_link_thread(dev);
+	ixgbe_dev_wait_setup_link_complete(dev, 0);
 
 	ixgbevf_intr_disable(dev);
 
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v6] net/ixgbe: fix resource leak after thread exits normally

[Ye Xiaolong]

On 04/22, taox.zhu@intel.com wrote:
>From: Zhu Tao <taox.zhu@intel.com>
>
>When the thread exits normally, pthread_join() is not called, which can
>result in a resource leak. Therefore, the thread is set to separation
>mode using function pthread_detach(), so that no program call
>pthread_join() is required to recycle, and when the thread exits,
>the system automatically reclaims resources.
>
>Wait for the thread to finish with timeout argument(0 means that it will
>not return until link complete), wait until the thread finishes before
>returning. Normally, the thread will finish in a shorter time, and give
>a warning message if it hasn't finished in a longer time.
>
>Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events")
>Cc: stable@dpdk.org
>
>Signed-off-by: Zhu Tao <taox.zhu@intel.com>
>Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
>---
> drivers/net/ixgbe/ixgbe_ethdev.c | 57 +++++++++++++++++++---------------------
> 1 file changed, 27 insertions(+), 30 deletions(-)
>
>V6 changes:
>        Change code segment
>                     'int timeout = timeout_ms ? WARNING_TIMEOUT : timeout_ms;'
>        to
>                'uint32_t timeout = timeout_ms ? timeout_ms : WARNING_TIMEOUT;'.
>
>V5 changes:
>        Modify the definition of function ixgbe_dev_wait_setup_link_complete(),
>         so that the caller can choose whether to block, etc.
>
>v4 changes:
>        Format codes.
>
>v3 changes:
>        1. Wait for the thread to finish without setting the timeout, and the
>           corresponding function name has also been modified.
>        2. Commit log.
>
>v2 changes:
>        Commit log.
>
>diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
>index 2c57976..a3ae4d7 100644
>--- a/drivers/net/ixgbe/ixgbe_ethdev.c
>+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
>@@ -230,7 +230,7 @@ static int ixgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
> static void ixgbe_dev_interrupt_handler(void *param);
> static void ixgbe_dev_interrupt_delayed_handler(void *param);
> static void *ixgbe_dev_setup_link_thread_handler(void *param);
>-static void ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev);
>+static int ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms);
> 
> static int ixgbe_add_rar(struct rte_eth_dev *dev,
> 			struct rte_ether_addr *mac_addr,
>@@ -2601,7 +2601,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 	PMD_INIT_FUNC_TRACE();
> 
> 	/* Stop the link setup handler before resetting the HW. */
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	/* disable uio/vfio intr/eventfd mapping */
> 	rte_intr_disable(intr_handle);
>@@ -2888,7 +2888,7 @@ static int eth_ixgbevf_pci_remove(struct rte_pci_device *pci_dev)
> 
> 	PMD_INIT_FUNC_TRACE();
> 
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	/* disable interrupts */
> 	ixgbe_disable_intr(hw);
>@@ -4143,36 +4143,32 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	return ret_val;
> }
> 
>-/* return 1: setup complete, return 0: setup not complete, and wait timeout*/
>+/*
>+ * If @timeout_ms was 0, it means that it will not return until link complete.
>+ * It returns 1 on complete, return 0 on timeout.
>+ */
> static int
>-ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev)
>+ixgbe_dev_wait_setup_link_complete(struct rte_eth_dev *dev, uint32_t timeout_ms)
> {
>-#define DELAY_INTERVAL 100 /* 100ms */
>-#define MAX_TIMEOUT    90 /* 9s (90 * 100ms) in total */
>+#define WARNING_TIMEOUT    9000 /* 9s  in total */
> 	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	int timeout = MAX_TIMEOUT;
>+	uint32_t timeout = timeout_ms ? timeout_ms : WARNING_TIMEOUT;
> 
>-	while (rte_atomic32_read(&ad->link_thread_running) && timeout) {
>-		msec_delay(DELAY_INTERVAL);
>+	while (rte_atomic32_read(&ad->link_thread_running)) {
>+		msec_delay(1);
> 		timeout--;
>-	}
>-
>-
>-	return !!timeout;
>-}
> 
>-static void
>-ixgbe_dev_cancel_link_thread(struct rte_eth_dev *dev)
>-{
>-	struct ixgbe_adapter *ad = dev->data->dev_private;
>-	void *retval;
>-
>-	if (!ixgbe_dev_wait_setup_link_complete(dev)) {
>-		pthread_cancel(ad->link_thread_tid);
>-		pthread_join(ad->link_thread_tid, &retval);
>-		rte_atomic32_clear(&ad->link_thread_running);
>-		PMD_DRV_LOG(ERR, "Link thread not complete, cancel it!");
>+		if (timeout_ms) {
>+			if (!timeout)
>+				return 0;
>+		} else if (!timeout) {
>+			/* It will not return until link complete */
>+			timeout = WARNING_TIMEOUT;
>+			PMD_DRV_LOG(ERR, "IXGBE link thread not complete too long time!");
>+		}
> 	}
>+
>+	return 1;
> }
> 
> static void *
>@@ -4186,6 +4182,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	u32 speed;
> 	bool autoneg = false;
> 
>+	pthread_detach(pthread_self());
> 	speed = hw->phy.autoneg_advertised;
> 	if (!speed)
> 		ixgbe_get_link_capabilities(hw, &speed, &autoneg);
>@@ -4282,8 +4279,8 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	if (link_up == 0) {
> 		if (ixgbe_get_media_type(hw) == ixgbe_media_type_fiber) {
> 			intr->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
>-			if (ixgbe_dev_wait_setup_link_complete(dev) &&
>-			    rte_atomic32_test_and_set(&ad->link_thread_running)) {
>+			ixgbe_dev_wait_setup_link_complete(dev, 0);
>+			if (rte_atomic32_test_and_set(&ad->link_thread_running)) {
> 				if (rte_ctrl_thread_create(&ad->link_thread_tid,
> 					"ixgbe-link-handler",
> 					NULL,
>@@ -5323,7 +5320,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 	PMD_INIT_FUNC_TRACE();
> 
> 	/* Stop the link setup handler before resetting the HW. */
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	err = hw->mac.ops.reset_hw(hw);
> 	if (err) {
>@@ -5421,7 +5418,7 @@ static int ixgbevf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
> 
> 	PMD_INIT_FUNC_TRACE();
> 
>-	ixgbe_dev_cancel_link_thread(dev);
>+	ixgbe_dev_wait_setup_link_complete(dev, 0);
> 
> 	ixgbevf_intr_disable(dev);
> 
>-- 
>1.8.3.1
>

Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>

Applied to dpdk-next-net-intel, Thanks.