Skip to contentSkip to conversation

lib/mempool : rte_mempool_avail_count, fixing return bigger than mempool size

PATCH In review Yasin CANER 5 participants 12 messages May 16, 2023 – May 18, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lib/mempool : rte_mempool_avail_count, fixing return bigger than mempool size

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

---
 lib/mempool/rte_mempool.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
lib/mempool/rte_mempool.c +4 -1
index cf5dea2304..17ed1eb845 100644
@@ -1009,8 +1009,11 @@ rte_mempool_avail_count(const struct rte_mempool *mp)
 
 	count = rte_mempool_ops_get_count(mp);
 
-	if (mp->cache_size == 0)
+	if (mp->cache_size == 0) {
+		if (count > mp->size)
+			return mp->size;
 		return count;
+	}
 
 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
 		count += mp->local_cache[lcore_id].len;
Signature · 1 line
2.25.1
#1 lore raw
Quoted 10 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +2 -1
index cf5dea2304a7..2406b112e7b0 100644
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;
 
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
                count += mp->local_cache[lcore_id].len;
@@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool *mp)
         * due to race condition (access to len is not locked), the
         * total can be greater than size... so fix the result
         */
+exit:
        if (count > mp->size)
                return mp->size;
        return count;
#2 lore raw
Quoted 24 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Tuesday, 16 May 2023 17.24

Quoted 10 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 12 lines
lib/mempool/rte_mempool.c +1 -0
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
                count += mp->local_cache[lcore_id].len;
@@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
         * due to race condition (access to len is not locked), the
         * total can be greater than size... so fix the result
         */
+exit:
        if (count > mp->size)
                return mp->size;
        return count;
#3 lore raw

Hello,

I don't have full knowledge of how to work rte_mempool_ops_get_count() but there is another comment about it. Maybe it relates.
/*
* due to race condition (access to len is not locked), the
* total can be greater than size... so fix the result
*/

Best regards.

Morten Brørup <[email protected]>, 16 May 2023 Sal, 19:04 tarihinde şunu yazdı:

Quoted 48 lines
Quoted 24 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Tuesday, 16 May 2023 17.24

Quoted 10 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 12 lines
lib/mempool/rte_mempool.c +1 -0
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
                count += mp->local_cache[lcore_id].len;
@@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
         * due to race condition (access to len is not locked), the
         * total can be greater than size... so fix the result
         */
+exit:
        if (count > mp->size)
                return mp->size;
        return count;
#4 lore raw

From: Yasin CANER [mailto:[email protected]]
Sent: Wednesday, 17 May 2023 10.01
To: Morten Brørup
Cc: Stephen Hemminger; [email protected]; Yasin CANER; Olivier Matz; Andrew Rybchenko
Subject: Re: [PATCH] lib/mempool : rte_mempool_avail_count, fixing return bigger than mempool size

Hello,

I don't have full knowledge of how to work rte_mempool_ops_get_count() but there is another comment about it. Maybe it relates.

/*
* due to race condition (access to len is not locked), the
* total can be greater than size... so fix the result */

MB: This comment relates to the race when accessing the per-lcore cache counters.

Best regards.

Morten Brørup <[email protected]>, 16 May 2023 Sal, 19:04 tarihinde şunu yazdı:

	> From: Stephen Hemminger [mailto:[email protected]]
	> Sent: Tuesday, 16 May 2023 17.24
	> 
	> On Tue, 16 May 2023 13:41:46 +0000
	> Yasin CANER <[email protected]> wrote:
	> 
	> > From: Yasin CANER <[email protected]>
	> >
	> > after a while working rte_mempool_avail_count function returns bigger
	> > than mempool size that cause miscalculation rte_mempool_in_use_count.
	> >
	> > it helps to avoid miscalculation rte_mempool_in_use_count.
	> >
	> > Bugzilla ID: 1229
	> >
	> > Signed-off-by: Yasin CANER <[email protected]>
	> 
	> An alternative that avoids some code duplication.
	> 
	> diff --git a/lib/mempool/rte_mempool.c b/lib/mempool/rte_mempool.c
	> index cf5dea2304a7..2406b112e7b0 100644
	> --- a/lib/mempool/rte_mempool.c
	> +++ b/lib/mempool/rte_mempool.c
	> @@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
	> *mp)
	>         count = rte_mempool_ops_get_count(mp);
	> 
	>         if (mp->cache_size == 0)
	> -               return count;
	> +               goto exit;
	
	This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.
	
	> 
	>         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
	>                 count += mp->local_cache[lcore_id].len;
	> @@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
	> *mp)
	>          * due to race condition (access to len is not locked), the
	>          * total can be greater than size... so fix the result
	>          */
	> +exit:
	>         if (count > mp->size)
	>                 return mp->size;
	>         return count;
#5 lore raw

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 17 May 2023 10.38

From: Yasin CANER [mailto:[email protected]]
Sent: Wednesday, 17 May 2023 10.01

Hello,

I don't have full knowledge of how to work rte_mempool_ops_get_count() but there is another comment about it. Maybe it relates.

/*
* due to race condition (access to len is not locked), the
* total can be greater than size... so fix the result */

MB: This comment relates to the race when accessing the per-lcore cache counters.

MB (continued): I have added more information, regarding mempool drivers, in Bugzilla: https://bugs.dpdk.org/show_bug.cgi?id=1229

Best regards.

Morten Brørup <[email protected]>, 16 May 2023 Sal, 19:04 tarihinde şunu yazdı:

	> From: Stephen Hemminger [mailto:[email protected]]
	> Sent: Tuesday, 16 May 2023 17.24
	> 
	> On Tue, 16 May 2023 13:41:46 +0000
	> Yasin CANER <[email protected]> wrote:
	> 
	> > From: Yasin CANER <[email protected]>
	> >
	> > after a while working rte_mempool_avail_count function returns bigger
	> > than mempool size that cause miscalculation rte_mempool_in_use_count.
	> >
	> > it helps to avoid miscalculation rte_mempool_in_use_count.
	> >
	> > Bugzilla ID: 1229
	> >
	> > Signed-off-by: Yasin CANER <[email protected]>
	> 
	> An alternative that avoids some code duplication.
	> 
	> diff --git a/lib/mempool/rte_mempool.c b/lib/mempool/rte_mempool.c
	> index cf5dea2304a7..2406b112e7b0 100644
	> --- a/lib/mempool/rte_mempool.c
	> +++ b/lib/mempool/rte_mempool.c
	> @@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
	> *mp)
	>         count = rte_mempool_ops_get_count(mp);
	> 
	>         if (mp->cache_size == 0)
	> -               return count;
	> +               goto exit;
	
	This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

	 

	MB (continued): The bug must be in the underlying mempool driver. I took a look at the ring and stack drivers, and they seem fine.

	
	
	> 
	>         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
	>                 count += mp->local_cache[lcore_id].len;
	> @@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
	> *mp)
	>          * due to race condition (access to len is not locked), the
	>          * total can be greater than size... so fix the result
	>          */
	> +exit:
	>         if (count > mp->size)
	>                 return mp->size;
	>         return count;
#6 lore raw

Hello,

mempool drivers is listed here.
https://doc.dpdk.org/guides/mempool/index.html

my app loads *rte_mempool_ring* and also *rte_mempool_stack* . In doc, rte_mempool_ring is default mempool driver.

"-d librte_mbuf.so -d librte_mempool.so -d librte_mempool_ring.so -d librte_mempool_stack.so -d librte_mempool_bucket.so -d librte_kni.so"

EAL: open shared lib librte_mbuf.so
EAL: open shared lib librte_mempool.so
EAL: open shared lib librte_mempool_ring.so
EAL: open shared lib librte_mempool_stack.so
EAL: lib.stack log level changed from disabled to notice
EAL: open shared lib librte_mempool_bucket.so
EAL: open shared lib librte_kni.so
EAL: open shared lib DPDK_LIBS/lib/x86_64-linux-gnu/dpdk/pmds-23.0/librte_mempool_octeontx.so
EAL: pmd.octeontx.mbox log level changed from disabled to notice
EAL: pmd.mempool.octeontx log level changed from disabled to notice

Morten Brørup <[email protected]>, 17 May 2023 Çar, 12:04 tarihinde şunu yazdı:

Quoted 95 lines

*From:* Morten Brørup [mailto:[email protected]] *Sent:* Wednesday, 17 May 2023 10.38

*From:* Yasin CANER [mailto:[email protected]] *Sent:* Wednesday, 17 May 2023 10.01

Hello,

I don't have full knowledge of how to work rte_mempool_ops_get_count() but there is another comment about it. Maybe it relates.

/*
* due to race condition (access to len is not locked), the
* total can be greater than size... so fix the result */

MB: This comment relates to the race when accessing the per-lcore cache counters.

MB (continued): I have added more information, regarding mempool drivers, in Bugzilla: https://bugs.dpdk.org/show_bug.cgi?id=1229

Best regards.

Morten Brørup <[email protected]>, 16 May 2023 Sal, 19:04 tarihinde şunu yazdı:

Quoted 24 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Tuesday, 16 May 2023 17.24

Quoted 10 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

MB (continued): The bug must be in the underlying mempool driver. I took a look at the ring and stack drivers, and they seem fine.

Quoted 12 lines
lib/mempool/rte_mempool.c +1 -0
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
                count += mp->local_cache[lcore_id].len;
@@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
         * due to race condition (access to len is not locked), the
         * total can be greater than size... so fix the result
         */
+exit:
        if (count > mp->size)
                return mp->size;
        return count;
#7 lore raw
In reply to Morten Brørup
Quoted 9 lines · Morten Brørup
Quoted 9 lines
Quoted 6 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Is this issue reproduced with an application of the reporter, or a DPDK in-tree application?

Quoted 23 lines
Quoted 12 lines
Quoted 4 lines

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 1 line

MB (continued): The bug must be in the underlying mempool driver. I took a look at the ring and stack drivers, and they seem fine.

Or it could indicate a double free (or equivalent) issue from the application (either through direct call to mempool API, or indirectly like sending/freeing an already sent/freed packet for example).

Signature · 1 line
David Marchand
#8 lore raw
Quoted 44 lines

From: David Marchand [mailto:[email protected]]
Sent: Wednesday, 17 May 2023 13.53

Quoted 9 lines · Morten Brørup
Quoted 9 lines
Quoted 6 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Is this issue reproduced with an application of the reporter, or a DPDK in-tree application?

Quoted 15 lines
Quoted 15 lines
Quoted 4 lines

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
Quoted 5 lines

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool *mp) · line 1013

Quoted 3 lines
lib/mempool/rte_mempool.c +1 -1
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if

rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 3 lines
Quoted 6 lines
Quoted 1 line

MB (continued): The bug must be in the underlying mempool driver. I took a

look at the ring and stack drivers, and they seem fine.

Or it could indicate a double free (or equivalent) issue from the application (either through direct call to mempool API, or indirectly like sending/freeing an already sent/freed packet for example).

Good point, David.

@Yasin, if you build DPDK and your application with RTE_LIBRTE_MEMPOOL_DEBUG set in config/rte_config.h, the mempool cookies should catch any double frees.

#9 lore raw

Hello,

I found a second free command in my code and removed it. David pointed to the right .

On the other hand, do you think we need to avoid miscalculations? Is it better to patch it or not?

or it needs to be aware of the second free command.

Sharing more information about env.

# ethtool -i mgmt
driver: virtio_net
version: 1.0.0
firmware-version:
expansion-rom-version:
bus-info: 0000:00:03.0
supports-statistics: yes
supports-test: no
supports-eeprom-access: no
supports-register-dump: no
supports-priv-flags: no

NAME="Ubuntu"
VERSION="20.04.4 LTS (Focal Fossa)"
ID=ubuntu
ID_LIKE=debian
PRETTY_NAME="Ubuntu 20.04.4 LTS"
VERSION_ID="20.04"

Linux spgw-dpdk 5.4.0-146-generic #163-Ubuntu SMP Fri Mar 17 18:26:02 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

Best regards.

Morten Brørup <[email protected]>, 17 May 2023 Çar, 15:23 tarihinde şunu yazdı:

Quoted 63 lines
Quoted 12 lines

From: David Marchand [mailto:[email protected]]
Sent: Wednesday, 17 May 2023 13.53

On Wed, May 17, 2023 at 11:05 AM Morten Brørup <[email protected]

Quoted 6 lines · wrote:
Quoted 6 lines
Quoted 3 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns

bigger

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

than mempool size that cause miscalculation

rte_mempool_in_use_count.

Quoted 12 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

it helps to avoid miscalculation rte_mempool_in_use_count.

Is this issue reproduced with an application of the reporter, or a DPDK in-tree application?

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines

Bugzilla ID: 1229

Quoted 7 lines
Quoted 7 lines

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct

Quoted 7 lines

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct

rte_mempool

Quoted 14 lines
Quoted 8 lines

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
 *mp)
        count = rte_mempool_ops_get_count(mp);
 
        if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if

rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 4 lines

MB (continued): The bug must be in the underlying mempool driver. I

took a

Quoted 5 lines

look at the ring and stack drivers, and they seem fine.

Or it could indicate a double free (or equivalent) issue from the application (either through direct call to mempool API, or indirectly like sending/freeing an already sent/freed packet for example).

Good point, David.

@Yasin, if you build DPDK and your application with RTE_LIBRTE_MEMPOOL_DEBUG set in config/rte_config.h, the mempool cookies should catch any double frees.

#10 lore raw

From: Yasin CANER [mailto:[email protected]]
Sent: Thursday, 18 May 2023 10.37

Hello,

I found a second free command in my code and removed it. David pointed to the right .

MB: Good to hear.

On the other hand, do you think we need to avoid miscalculations? Is it better to patch it or not?

MB: There is no bug in the library, so no need to patch it. The returned value was a consequence of using the library incorrectly (freeing twice).

or it needs to be aware of the second free command.

Sharing more information about env.

# ethtool -i mgmt
driver: virtio_net
version: 1.0.0
firmware-version:
expansion-rom-version:
bus-info: 0000:00:03.0
supports-statistics: yes
supports-test: no
supports-eeprom-access: no
supports-register-dump: no
supports-priv-flags: no

NAME="Ubuntu"
VERSION="20.04.4 LTS (Focal Fossa)"
ID=ubuntu
ID_LIKE=debian
PRETTY_NAME="Ubuntu 20.04.4 LTS"
VERSION_ID="20.04"

Linux spgw-dpdk 5.4.0-146-generic #163-Ubuntu SMP Fri Mar 17 18:26:02 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

Best regards.

Morten Brørup <[email protected]>, 17 May 2023 Çar, 15:23 tarihinde şunu yazdı:

	> From: David Marchand [mailto:[email protected]]
	> Sent: Wednesday, 17 May 2023 13.53
	> 
	> On Wed, May 17, 2023 at 11:05 AM Morten Brørup <[email protected]>
	> wrote:
	> > > On Tue, 16 May 2023 13:41:46 +0000
	> > > Yasin CANER <[email protected]> wrote:
	> > >
	> > > > From: Yasin CANER <[email protected]>
	> > > >
	> > > > after a while working rte_mempool_avail_count function returns bigger
	> > > > than mempool size that cause miscalculation rte_mempool_in_use_count.
	> > > >
	> > > > it helps to avoid miscalculation rte_mempool_in_use_count.
	> 
	> Is this issue reproduced with an application of the reporter, or a
	> DPDK in-tree application?
	> 
	> 
	> > > >
	> > > > Bugzilla ID: 1229
	> > > >
	> > > > Signed-off-by: Yasin CANER <[email protected]>
	> > >
	> > > An alternative that avoids some code duplication.
	> > >
	> > > diff --git a/lib/mempool/rte_mempool.c b/lib/mempool/rte_mempool.c
	> > > index cf5dea2304a7..2406b112e7b0 100644
	> > > --- a/lib/mempool/rte_mempool.c
	> > > +++ b/lib/mempool/rte_mempool.c
	> > > @@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
	> > > *mp)
	> > >         count = rte_mempool_ops_get_count(mp);
	> > >
	> > >         if (mp->cache_size == 0)
	> > > -               return count;
	> > > +               goto exit;
	> >
	> > This bug can only occur here (i.e. with cache_size==0) if
	> rte_mempool_ops_get_count() returns an incorrect value. The bug should be
	> fixed there instead.
	> >
	> >
	> >
	> > MB (continued): The bug must be in the underlying mempool driver. I took a
	> look at the ring and stack drivers, and they seem fine.
	> 
	> Or it could indicate a double free (or equivalent) issue from the
	> application (either through direct call to mempool API, or indirectly
	> like sending/freeing an already sent/freed packet for example).
	
	Good point, David.
	
	@Yasin, if you build DPDK and your application with RTE_LIBRTE_MEMPOOL_DEBUG set in config/rte_config.h, the mempool cookies should catch any double frees.
#11 lore raw
In reply to Morten Brørup
Quoted 34 lines · Morten Brørup
Quoted 24 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Tuesday, 16 May 2023 17.24

Quoted 10 lines · On Tue, 16 May 2023 13:41:46 +0000 Yasin CANER <[email protected]> wrote:

From: Yasin CANER <[email protected]>

after a while working rte_mempool_avail_count function returns bigger than mempool size that cause miscalculation rte_mempool_in_use_count.

it helps to avoid miscalculation rte_mempool_in_use_count.

Bugzilla ID: 1229

An alternative that avoids some code duplication.

lib/mempool/rte_mempool.c +0 -0
index cf5dea2304a7..2406b112e7b0 100644

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1014

Quoted 6 lines
lib/mempool/rte_mempool.c +1 -1
@@ -1010,7 +1010,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)
         count = rte_mempool_ops_get_count(mp);
 
         if (mp->cache_size == 0)
-               return count;
+               goto exit;

This bug can only occur here (i.e. with cache_size==0) if rte_mempool_ops_get_count() returns an incorrect value. The bug should be fixed there instead.

Quoted 1 line
Quoted 4 lines

lib/mempool/rte_mempool.c rte_mempool_avail_count(const struct rte_mempool · line 1019

Quoted 4 lines

for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
count += mp->local_cache[lcore_id].len;

lib/mempool/rte_mempool.c +0 -0
@@ -1019,6 +1019,7 @@ rte_mempool_avail_count(const struct rte_mempool
 *mp)

This loop may result in the same symptoms (i.e., count > mp->size). For example, the LF stack has a relaxed atomic load to retrieve the usage count, which may well be reordered (by the CPU and/or compiler) in relation to these non-atomic loads. The same issue may well exist on the writer side, with the order between the global and the cache counts not being ordered.

So while sanity checking the count without caching is not needed, with, you can argue it is.

Quoted 7 lines
Quoted 7 lines
lib/mempool/rte_mempool.c +1 -0
          * due to race condition (access to len is not locked), the
          * total can be greater than size... so fix the result
          */
+exit:
         if (count > mp->size)
                 return mp->size;
         return count;