Skip to contentSkip to conversation

fixes for device event

PATCH Superseded wangyunjian 3 participants 18 messages Jul 3, 2020 – Oct 23, 2020

lore
Share & tools
lore
#0 lore raw

v4 cover fixes for device event

From: Yunjian Wang <[email protected]>

This series include three fixes patches for device event.

---
v4:
* add fix a wrong returned value
* remove redundant check suggested by David Marchand

v3:
* modified the format.

Yunjian Wang (3):
eal: fix memory leak when removing event_cb
eal: return error code when failure
eal: fix a wrong returned value when callback exists

 lib/librte_eal/common/eal_common_dev.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH v4 1/3 eal: fix memory leak when removing event_cb

From: Yunjian Wang <[email protected]>

The event_cb->dev_name is not freed when freeing event_cb, and this causes a memory leak.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 1 +
 1 file changed, 1 insertion(+)
lib/librte_eal/common/eal_common_dev.c +1 -0
index 9e4f09d..363a2ca 100644
@@ -526,6 +526,7 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
 		 */
 		if (event_cb->active == 0) {
 			TAILQ_REMOVE(&dev_event_cbs, event_cb, next);
+			free(event_cb->dev_name);
 			free(event_cb);
 			ret++;
 		} else {
Signature · 1 line
1.8.3.1
#2 lore raw

Re: [dpdk-dev] [PATCH v4 1/3] eal: fix memory leak when removing event_cb

Ping for review.

Thanks,
Yunjian

Quoted 40 lines

-----Original Message-----
From: wangyunjian
Sent: Friday, July 3, 2020 5:46 PM
To: [email protected]
Cc: [email protected]; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; wangyunjian <[email protected]>; [email protected]
Subject: [dpdk-dev] [PATCH v4 1/3] eal: fix memory leak when removing event_cb

From: Yunjian Wang <[email protected]>

The event_cb->dev_name is not freed when freeing event_cb, and this causes a memory leak.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 1 +
 1 file changed, 1 insertion(+)
lib/librte_eal/common/eal_common_dev.c +1 -0
index 9e4f09d..363a2ca 100644
@@ -526,6 +526,7 @@ static int cmp_dev_name(const struct rte_device *dev,
 const void *_name)
 		 */
 		if (event_cb->active == 0) {
 			TAILQ_REMOVE(&dev_event_cbs, event_cb, next);
+			free(event_cb->dev_name);
 			free(event_cb);
 			ret++;

} else {
--
1.8.3.1

Acked-by
#3 lore raw

Re: [dpdk-dev] [PATCH v4 1/3] eal: fix memory leak when removing event_cb

hi, yunjian

The patch seems no problem, but when you update your patch, please add other guys who had comment on your patch and maintainer to see if they have

other opinion, and --in-reply-to is also helpful for patch review, thanks.

Quoted 28 lines · wangyunjian

Ping for review.

Thanks,
Yunjian

Quoted 23 lines

-----Original Message-----
From: wangyunjian
Sent: Friday, July 3, 2020 5:46 PM
To: [email protected]
Cc: [email protected]; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; wangyunjian <[email protected]>; [email protected]
Subject: [dpdk-dev] [PATCH v4 1/3] eal: fix memory leak when removing event_cb

From: Yunjian Wang <[email protected]>

The event_cb->dev_name is not freed when freeing event_cb, and this causes a memory leak.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
  lib/librte_eal/common/eal_common_dev.c | 1 +
  1 file changed, 1 insertion(+)
Quoted 16 lines
Quoted 13 lines
lib/librte_eal/common/eal_common_dev.c +1 -0
index 9e4f09d..363a2ca 100644
@@ -526,6 +526,7 @@ static int cmp_dev_name(const struct rte_device *dev,
 const void *_name)
  		 */
  		if (event_cb->active == 0) {
  			TAILQ_REMOVE(&dev_event_cbs, event_cb, next);
+			free(event_cb->dev_name);
  			free(event_cb);
  			ret++;
Quoted 3 lines

} else {
--
1.8.3.1

#4 lore raw

PATCH v4 2/3 eal: return error code when failure

In reply to wangyunjian

From: Yunjian Wang <[email protected]>

Fix return value, using -EAGAIN instead of 0 when the callback is busy and using -ENOENT instead of 0 when the callback is not found.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
lib/librte_eal/common/eal_common_dev.c +7 -1
index 363a2ca..d990bfd 100644
@@ -530,9 +530,15 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
 			free(event_cb);
 			ret++;
 		} else {
-			continue;
+			ret = -EAGAIN;
+			break;
 		}
 	}
+
+	/* this callback is not be registered */
+	if (ret == 0)
+		ret = -ENOENT;
+
 	rte_spinlock_unlock(&dev_event_lock);
 	return ret;
 }
Signature · 1 line
1.8.3.1
#5 lore raw

PATCH v4 3/3 eal: fix a wrong returned value when callback exists

In reply to wangyunjian

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
lib/librte_eal/common/eal_common_dev.c +2 -2
index d990bfd..2a097aa 100644
@@ -431,7 +431,7 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
 				void *cb_arg)
 {
 	struct dev_event_callback *event_cb;
-	int ret;
+	int ret = 0;
 
 	if (!cb_fn)
 		return -EINVAL;
@@ -484,7 +484,7 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
 	}
 
 	rte_spinlock_unlock(&dev_event_lock);
-	return 0;
+	return ret;
 error:
 	free(event_cb);
 	rte_spinlock_unlock(&dev_event_lock);
Signature · 1 line
1.8.3.1
#6 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v4 3/3] eal: fix a wrong returned value when callback exists

Quoted 36 lines · wangyunjian

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
lib/librte_eal/common/eal_common_dev.c +2 -3
index d990bfd..2a097aa 100644
@@ -431,7 +431,7 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
                                void *cb_arg)
 {
        struct dev_event_callback *event_cb;
-       int ret;
+       int ret = 0;
 
        if (!cb_fn)
                return -EINVAL;
@@ -484,7 +484,7 @@ static int cmp_dev_name(const struct rte_device *dev, const void *_name)
        }
 
        rte_spinlock_unlock(&dev_event_lock);
-       return 0;
+       return ret;
 error:
        free(event_cb);
        rte_spinlock_unlock(&dev_event_lock);
--
Quoted 1 line

1.8.3.1

A simpler fix is to directly jump to the error label.
This has the advantage of having all errors go through a single cleanup code:

(quoted diff) +0 -0

b/lib/librte_eal/common/eal_common_dev.c
index 9e4f09d83e..fa47074b0b 100644

lib/librte_eal/common/eal_common_dev.c +2 -0
@@ -480,7 +480,9 @@ rte_dev_event_callback_register(const char *device_name,
                RTE_LOG(ERR, EAL,
                        "The callback is already exist, no need "
                        "to register again.\n");
+               event_cb = NULL;
                ret = -EEXIST;
+               goto error;
        }
 
        rte_spinlock_unlock(&dev_event_lock);

What do you think?

Signature · 1 line
David Marchand
#7 lore raw

Re: [dpdk-dev] [dpdk-stable] [PATCH v4 3/3] eal: fix a wrong returned value when callback exists

Quoted 71 lines

-----Original Message-----
From: David Marchand [mailto:[email protected]]
Sent: Tuesday, October 20, 2020 9:19 PM
To: wangyunjian <[email protected]>; Jeff Guo <[email protected]>
Cc: dev <[email protected]>; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; dpdk stable <[email protected]>
Subject: Re: [dpdk-stable] [dpdk-dev] [PATCH v4 3/3] eal: fix a wrong returned value when callback exists

lib/librte_eal/common/eal_common_dev.c static int cmp_dev_name(const struct rte_device

Quoted 19 lines · wangyunjian

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
lib/librte_eal/common/eal_common_dev.c +0 -0
index d990bfd..2a097aa 100644
@@ -431,7 +431,7 @@ static int cmp_dev_name(const struct rte_device

*dev, const void *_name)

lib/librte_eal/common/eal_common_dev.c static int cmp_dev_name(const struct rte_device

Quoted 8 lines
lib/librte_eal/common/eal_common_dev.c +1 -1
                                void *cb_arg)  {
        struct dev_event_callback *event_cb;
-       int ret;
+       int ret = 0;
 
        if (!cb_fn)
                return -EINVAL;
@@ -484,7 +484,7 @@ static int cmp_dev_name(const struct rte_device

*dev, const void *_name)

Quoted 10 lines
lib/librte_eal/common/eal_common_dev.c +1 -2
        }
 
        rte_spinlock_unlock(&dev_event_lock);
-       return 0;
+       return ret;
 error:
        free(event_cb);
        rte_spinlock_unlock(&dev_event_lock);
--

1.8.3.1

A simpler fix is to directly jump to the error label.
This has the advantage of having all errors go through a single cleanup code:

lib/librte_eal/common/eal_common_dev.c +2 -0
index 9e4f09d83e..fa47074b0b 100644
@@ -480,7 +480,9 @@ rte_dev_event_callback_register(const char
 *device_name,
                RTE_LOG(ERR, EAL,
                        "The callback is already exist, no need "
                        "to register again.\n");
+               event_cb = NULL;
                ret = -EEXIST;
+               goto error;
        }
 
        rte_spinlock_unlock(&dev_event_lock);
Quoted 1 line

What do you think?

Agree, I will update it in next version.

Yunjian

Quoted 4 lines

--
David Marchand

#8 lore raw
In reply to wangyunjian

Hi, Ferruh Yigit & David Marchand

Could you please give any suggestion?

Thanks,
Yunjian

Quoted 31 lines

-----Original Message-----
From: wangyunjian
Sent: Friday, July 3, 2020 5:44 PM
To: [email protected]
Cc: [email protected]; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; wangyunjian <[email protected]>
Subject: [dpdk-dev] [PATCH v4 0/3] fixes for device event

From: Yunjian Wang <[email protected]>

This series include three fixes patches for device event.

---
v4:
* add fix a wrong returned value
* remove redundant check suggested by David Marchand

v3:
* modified the format.

Yunjian Wang (3):
eal: fix memory leak when removing event_cb
eal: return error code when failure
eal: fix a wrong returned value when callback exists

 lib/librte_eal/common/eal_common_dev.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

--
1.8.3.1

#9 lore raw

v5 cover fixes for device event

In reply to wangyunjian

From: Yunjian Wang <[email protected]>

This series include three fixes patches for device event.

---
v5:
* update patch 3/3 code styles suggested by David Marchand

v4:
* add fix a wrong returned value
* remove redundant check suggested by David Marchand

v3:
* modified the format.

Yunjian Wang (3):
eal: fix memory leak when removing event_cb
eal: return error code when failure
eal: fix a wrong returned value when callback exists

 lib/librte_eal/common/eal_common_dev.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
Signature · 1 line
2.23.0
#10 lore raw

PATCH v5 1/3 eal: fix memory leak when removing event_cb

From: Yunjian Wang <[email protected]>

The event_cb->dev_name is not freed when freeing event_cb, and this causes a memory leak.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 1 +
 1 file changed, 1 insertion(+)
lib/librte_eal/common/eal_common_dev.c +1 -0
index 9e4f09d83e..363a2ca95e 100644
@@ -526,6 +526,7 @@ rte_dev_event_callback_unregister(const char *device_name,
 		 */
 		if (event_cb->active == 0) {
 			TAILQ_REMOVE(&dev_event_cbs, event_cb, next);
+			free(event_cb->dev_name);
 			free(event_cb);
 			ret++;
 		} else {
Signature · 1 line
2.23.0
#11 lore raw

PATCH v5 2/3 eal: return error code when failure

In reply to wangyunjian

From: Yunjian Wang <[email protected]>

Fix return value, using -EAGAIN instead of 0 when the callback is busy and using -ENOENT instead of 0 when the callback is not found.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
lib/librte_eal/common/eal_common_dev.c +7 -1
index 363a2ca95e..d990bfd20d 100644
@@ -530,9 +530,15 @@ rte_dev_event_callback_unregister(const char *device_name,
 			free(event_cb);
 			ret++;
 		} else {
-			continue;
+			ret = -EAGAIN;
+			break;
 		}
 	}
+
+	/* this callback is not be registered */
+	if (ret == 0)
+		ret = -ENOENT;
+
 	rte_spinlock_unlock(&dev_event_lock);
 	return ret;
 }
Signature · 1 line
2.23.0
#12 lore raw

Re: [dpdk-dev] [PATCH v5 2/3] eal: return error code when failure

Quoted 8 lines · wangyunjian

From: Yunjian Wang <[email protected]>

Fix return value, using -EAGAIN instead of 0 when the callback is busy and using -ENOENT instead of 0 when the callback is not found.

Fixes a753e53d517b ("eal: add device event monitor framework")

This API is experimental, its description is vague enough and we can change it in the current release.
But I see this as a change in behavior for existing users, not a fix and I would not backport it.

Opinions?

Quoted 31 lines
---
 lib/librte_eal/common/eal_common_dev.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
lib/librte_eal/common/eal_common_dev.c +7 -2
index 363a2ca95e..d990bfd20d 100644
@@ -530,9 +530,15 @@ rte_dev_event_callback_unregister(const char *device_name,
                        free(event_cb);
                        ret++;
                } else {
-                       continue;
+                       ret = -EAGAIN;
+                       break;
                }
        }
+
+       /* this callback is not be registered */
+       if (ret == 0)
+               ret = -ENOENT;
+
        rte_spinlock_unlock(&dev_event_lock);
        return ret;
 }
--

2.23.0

Signature · 1 line
David Marchand
#13 lore raw

Re: [dpdk-dev] [PATCH v5 2/3] eal: return error code when failure

Quoted 21 lines

-----Original Message-----
From: David Marchand [mailto:[email protected]]
Sent: Thursday, October 22, 2020 8:52 PM
To: wangyunjian <[email protected]>
Cc: dev <[email protected]>; Jeff Guo <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; dpdk stable <[email protected]>; Kevin Traynor <[email protected]>; Luca Boccassi <[email protected]>
Subject: Re: [dpdk-dev] [PATCH v5 2/3] eal: return error code when failure

Quoted 8 lines · wangyunjian

From: Yunjian Wang <[email protected]>

Fix return value, using -EAGAIN instead of 0 when the callback is busy and using -ENOENT instead of 0 when the callback is not found.

Fixes a753e53d517b ("eal: add device event monitor framework")
Quoted 6 lines

This API is experimental, its description is vague enough and we can change it in the current release.
But I see this as a change in behavior for existing users, not a fix and I would not backport it.

Opinions?

Agree, not backport it.

Yunjian

Quoted 39 lines

lib/librte_eal/common/eal_common_dev.c rte_dev_event_callback_unregister(const char

Quoted 15 lines
---
 lib/librte_eal/common/eal_common_dev.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
lib/librte_eal/common/eal_common_dev.c +0 -0
index 363a2ca95e..d990bfd20d 100644
@@ -530,9 +530,15 @@ rte_dev_event_callback_unregister(const char

*device_name,

Quoted 17 lines
lib/librte_eal/common/eal_common_dev.c +7 -2
                        free(event_cb);
                        ret++;
                } else {
-                       continue;
+                       ret = -EAGAIN;
+                       break;
                }
        }
+
+       /* this callback is not be registered */
+       if (ret == 0)
+               ret = -ENOENT;
+
        rte_spinlock_unlock(&dev_event_lock);
        return ret;
 }
--
Quoted 2 lines

2.23.0

--
David Marchand

#14 lore raw

PATCH v5 3/3 eal: fix a wrong returned value when callback exists

In reply to wangyunjian

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")
---
 lib/librte_eal/common/eal_common_dev.c | 2 ++
 1 file changed, 2 insertions(+)
lib/librte_eal/common/eal_common_dev.c +2 -0
index d990bfd20d..8a3bd3100a 100644
@@ -480,7 +480,9 @@ rte_dev_event_callback_register(const char *device_name,
 		RTE_LOG(ERR, EAL,
 			"The callback is already exist, no need "
 			"to register again.\n");
+		event_cb = NULL;
 		ret = -EEXIST;
+		goto error;
 	}
 
 	rte_spinlock_unlock(&dev_event_lock);
Signature · 1 line
2.23.0
#15 lore raw

Re: [dpdk-dev] [PATCH v5 3/3] eal: fix a wrong returned value when callback exists

Quoted 7 lines · wangyunjian

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")

Same question as for the previous patch, should we backport this?

Signature · 1 line
David Marchand
#16 lore raw

Re: [dpdk-dev] [PATCH v5 3/3] eal: fix a wrong returned value when callback exists

Quoted 14 lines

-----Original Message-----
From: David Marchand [mailto:[email protected]]
Sent: Thursday, October 22, 2020 8:55 PM
To: wangyunjian <[email protected]>
Cc: dev <[email protected]>; Jeff Guo <[email protected]>; Jerin Jacob Kollanukkaran <[email protected]>; Lilijun (Jerry) <[email protected]>; xudingke <[email protected]>; dpdk stable <[email protected]>; Kevin Traynor <[email protected]>; Luca Boccassi <[email protected]>
Subject: Re: [dpdk-dev] [PATCH v5 3/3] eal: fix a wrong returned value when callback exists

Quoted 1 line · wangyunjian
Quoted 8 lines
Quoted 6 lines

From: Yunjian Wang <[email protected]>

We should return an error value, when the callback is already exist.

Fixes a753e53d517b ("eal: add device event monitor framework")

Same question as for the previous patch, should we backport this?

Agree, not backport it.

Yunjian

Quoted 4 lines

--
David Marchand