[PATCH 1/2] test/distributor: flush with worker shutdown test fails

[PATCH 1/2] test/distributor: flush with worker shutdown test fails

[Harman Kalra]

From: Harman Kalra <harman.kalra@cavium.com>

On restarting worker 0 after shutdown, packets handled by
worker 0 must be incremented only when a packet is received by
it.

Signed-off-by: Harman Kalra <harman.kalra@cavium.com>
---
 test/test/test_distributor.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git test/test/test_distributor.c test/test/test_distributor.c
index 98919ec..a723254 100644
--- test/test/test_distributor.c
+++ test/test/test_distributor.c
@@ -373,7 +373,8 @@ struct worker_stats {
 				id, buf, buf, num);
 
 		while (!quit) {
-			worker_stats[id].handled_packets++, count++;
+			worker_stats[id].handled_packets += num;
+			count += num;
 			rte_pktmbuf_free(pkt);
 			num = rte_distributor_get_pkt(d, id, buf, buf, num);
 		}
-- 
1.8.3.1

[PATCH 2/2] distributor: imposing additional check on no of workers

[Harman Kalra]

From: Harman Kalra <harman.kalra@cavium.com>

No of workers should never exceed RTE_MAX_LCORE.
RTE_DIST_ALG_SINGLE also require no of workers check.

Signed-off-by: Harman Kalra <harman.kalra@cavium.com>
---
 lib/librte_distributor/rte_distributor.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git lib/librte_distributor/rte_distributor.c lib/librte_distributor/rte_distributor.c
index d505983..043b8f3 100644
--- lib/librte_distributor/rte_distributor.c
+++ lib/librte_distributor/rte_distributor.c
@@ -595,6 +595,12 @@ struct rte_distributor *
 	RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0);
 	RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0);
 
+	if (name == NULL || num_workers >=
+		(unsigned int)RTE_MIN(RTE_DISTRIB_MAX_WORKERS, RTE_MAX_LCORE)) {
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
 	if (alg_type == RTE_DIST_ALG_SINGLE) {
 		d = malloc(sizeof(struct rte_distributor));
 		if (d == NULL) {
@@ -612,11 +618,6 @@ struct rte_distributor *
 		return d;
 	}
 
-	if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) {
-		rte_errno = EINVAL;
-		return NULL;
-	}
-
 	snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
 	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
 	if (mz == NULL) {
-- 
1.8.3.1

Re: [PATCH 1/2] test/distributor: flush with worker shutdown test fails

[Yigit, Ferruh]

On 11/29/2018 3:09 PM, Harman Kalra wrote:
> From: Harman Kalra <harman.kalra@cavium.com>
> 
> On restarting worker 0 after shutdown, packets handled by
> worker 0 must be incremented only when a packet is received by
> it.
> 
> Signed-off-by: Harman Kalra <harman.kalra@cavium.com>

Hi Harman,

The patchset doesn't apply cleanly because the relative path in the diff is not
correct, can you please send a new version of it?

[PATCH v2 1/2] test/distributor: flush with worker shutdown test fails

[Harman Kalra]

On restarting worker 0 after shutdown, packets handled by
worker 0 must be incremented only when a packet is received by
it.

Signed-off-by: Harman Kalra <hkalra@marvell.com>
---
 test/test/test_distributor.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/test/test/test_distributor.c b/test/test/test_distributor.c
index 98919ec0c..a723254f7 100644
--- a/test/test/test_distributor.c
+++ b/test/test/test_distributor.c
@@ -373,7 +373,8 @@ handle_work_for_shutdown_test(void *arg)
 				id, buf, buf, num);
 
 		while (!quit) {
-			worker_stats[id].handled_packets++, count++;
+			worker_stats[id].handled_packets += num;
+			count += num;
 			rte_pktmbuf_free(pkt);
 			num = rte_distributor_get_pkt(d, id, buf, buf, num);
 		}
-- 
2.18.0

[PATCH v2 2/2] distributor: imposing additional check on no of workers

[Harman Kalra]

no of workers should never exceed RTE_MAX_LCORE

Signed-off-by: Harman Kalra <hkalra@marvell.com>
---
 lib/librte_distributor/rte_distributor.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c
index d50598377..043b8f3ae 100644
--- a/lib/librte_distributor/rte_distributor.c
+++ b/lib/librte_distributor/rte_distributor.c
@@ -595,6 +595,12 @@ rte_distributor_create_v1705(const char *name,
 	RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0);
 	RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0);
 
+	if (name == NULL || num_workers >=
+		(unsigned int)RTE_MIN(RTE_DISTRIB_MAX_WORKERS, RTE_MAX_LCORE)) {
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
 	if (alg_type == RTE_DIST_ALG_SINGLE) {
 		d = malloc(sizeof(struct rte_distributor));
 		if (d == NULL) {
@@ -612,11 +618,6 @@ rte_distributor_create_v1705(const char *name,
 		return d;
 	}
 
-	if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) {
-		rte_errno = EINVAL;
-		return NULL;
-	}
-
 	snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
 	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
 	if (mz == NULL) {
-- 
2.18.0

Re: [PATCH v2 1/2] test/distributor: flush with worker shutdown test fails

[Thomas Monjalon]

18/02/2019 15:51, Harman Kalra:
> On restarting worker 0 after shutdown, packets handled by
> worker 0 must be incremented only when a packet is received by
> it.
> 
> Signed-off-by: Harman Kalra <hkalra@marvell.com>

It looks to be a fix, so the tag Fixes: should be added
to help with backport, please.
The title may be improved by starting with "fix" verb
instead of mentioning "fails".

Re: [dpdk-dev] [PATCH v2 1/2] test/distributor: flush with worker shutdown test fails

[Thomas Monjalon]

29/03/2019 23:54, Thomas Monjalon:
> 18/02/2019 15:51, Harman Kalra:
> > On restarting worker 0 after shutdown, packets handled by
> > worker 0 must be incremented only when a packet is received by
> > it.
> > 
> > Signed-off-by: Harman Kalra <hkalra@marvell.com>
> 
> It looks to be a fix, so the tag Fixes: should be added
> to help with backport, please.
> The title may be improved by starting with "fix" verb
> instead of mentioning "fails".

No review, no reply.
Status set to "Changes Requested" in patchwork.

[dpdk-dev] [PATCH v3 1/2] test/distributor: fix flush with worker shutdown test

[Harman Kalra]

On restarting worker 0 after shutdown, packets handled by
worker 0 must be incremented only when a packet is received by
it.

Fixes: c3eabff124e6 ("distributor: add unit tests")
Cc: stable@dpdk.org

Signed-off-by: Harman Kalra <hkalra@marvell.com>
---
 app/test/test_distributor.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/app/test/test_distributor.c b/app/test/test_distributor.c
index 8084c0794..5d71bf87c 100644
--- a/app/test/test_distributor.c
+++ b/app/test/test_distributor.c
@@ -374,7 +374,8 @@ handle_work_for_shutdown_test(void *arg)
 				id, buf, buf, num);
 
 		while (!quit) {
-			worker_stats[id].handled_packets++, count++;
+			worker_stats[id].handled_packets += num;
+			count += num;
 			rte_pktmbuf_free(pkt);
 			num = rte_distributor_get_pkt(d, id, buf, buf, num);
 		}
-- 
2.18.0

[dpdk-dev] [PATCH v3 2/2] distributor: fix additional check on no of workers

[Harman Kalra]

No of workers should never exceed RTE_MAX_LCORE.
RTE_DIST_ALG_SINGLE also require no of workers check.

Fixes: 775003ad2f96 ("distributor: add new burst-capable library")
Cc: stable@dpdk.org

Signed-off-by: Harman Kalra <hkalra@marvell.com>
---
 lib/librte_distributor/rte_distributor.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c
index 208abfb1d..bfcc536fd 100644
--- a/lib/librte_distributor/rte_distributor.c
+++ b/lib/librte_distributor/rte_distributor.c
@@ -595,6 +595,12 @@ rte_distributor_create_v1705(const char *name,
 	RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0);
 	RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0);
 
+	if (name == NULL || num_workers >=
+		(unsigned int)RTE_MIN(RTE_DISTRIB_MAX_WORKERS, RTE_MAX_LCORE)) {
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
 	if (alg_type == RTE_DIST_ALG_SINGLE) {
 		d = malloc(sizeof(struct rte_distributor));
 		if (d == NULL) {
@@ -612,11 +618,6 @@ rte_distributor_create_v1705(const char *name,
 		return d;
 	}
 
-	if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) {
-		rte_errno = EINVAL;
-		return NULL;
-	}
-
 	snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
 	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
 	if (mz == NULL) {
-- 
2.18.0

Re: [dpdk-dev] [dpdk-stable] [PATCH v3 2/2] distributor: fix additional check on no of workers

[Thomas Monjalon]

Any review please?

05/07/2019 12:05, Harman Kalra:
> No of workers should never exceed RTE_MAX_LCORE.
> RTE_DIST_ALG_SINGLE also require no of workers check.
> 
> Fixes: 775003ad2f96 ("distributor: add new burst-capable library")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Harman Kalra <hkalra@marvell.com>
> ---
>  lib/librte_distributor/rte_distributor.c | 11 ++++++-----
>  1 file changed, 6 insertions(+), 5 deletions(-)
> 
> diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c
> index 208abfb1d..bfcc536fd 100644
> --- a/lib/librte_distributor/rte_distributor.c
> +++ b/lib/librte_distributor/rte_distributor.c
> @@ -595,6 +595,12 @@ rte_distributor_create_v1705(const char *name,
>  	RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0);
>  	RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0);
>  
> +	if (name == NULL || num_workers >=
> +		(unsigned int)RTE_MIN(RTE_DISTRIB_MAX_WORKERS, RTE_MAX_LCORE)) {
> +		rte_errno = EINVAL;
> +		return NULL;
> +	}
> +
>  	if (alg_type == RTE_DIST_ALG_SINGLE) {
>  		d = malloc(sizeof(struct rte_distributor));
>  		if (d == NULL) {
> @@ -612,11 +618,6 @@ rte_distributor_create_v1705(const char *name,
>  		return d;
>  	}
>  
> -	if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) {
> -		rte_errno = EINVAL;
> -		return NULL;
> -	}
> -
>  	snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
>  	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
>  	if (mz == NULL) {
> 

Re: [dpdk-dev] [PATCH v3 1/2] test/distributor: fix flush with worker shutdown test

[Hunt, David]

Hi Harman,

On 05/07/2019 11:05, Harman Kalra wrote:
> On restarting worker 0 after shutdown, packets handled by
> worker 0 must be incremented only when a packet is received by
> it.
>
> Fixes: c3eabff124e6 ("distributor: add unit tests")
> Cc: stable@dpdk.org
>
> Signed-off-by: Harman Kalra <hkalra@marvell.com>
> ---
>   app/test/test_distributor.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/app/test/test_distributor.c b/app/test/test_distributor.c
> index 8084c0794..5d71bf87c 100644
> --- a/app/test/test_distributor.c
> +++ b/app/test/test_distributor.c
> @@ -374,7 +374,8 @@ handle_work_for_shutdown_test(void *arg)
>   				id, buf, buf, num);
>   
>   		while (!quit) {
> -			worker_stats[id].handled_packets++, count++;
> +			worker_stats[id].handled_packets += num;
> +			count += num;
>   			rte_pktmbuf_free(pkt);
>   			num = rte_distributor_get_pkt(d, id, buf, buf, num);
>   		}


Change makes sense. Builds OK, Runs OK, test passes.

Acked-by: David Hunt <david.hunt@intel.com>

Re: [dpdk-dev] [PATCH v3 2/2] distributor: fix additional check on no of workers

[Hunt, David]

Hi Harman,

On 05/07/2019 11:05, Harman Kalra wrote:
> No of workers should never exceed RTE_MAX_LCORE.
> RTE_DIST_ALG_SINGLE also require no of workers check.
>
> Fixes: 775003ad2f96 ("distributor: add new burst-capable library")
> Cc: stable@dpdk.org
>
> Signed-off-by: Harman Kalra <hkalra@marvell.com>
> ---
>   lib/librte_distributor/rte_distributor.c | 11 ++++++-----
>   1 file changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c
> index 208abfb1d..bfcc536fd 100644
> --- a/lib/librte_distributor/rte_distributor.c
> +++ b/lib/librte_distributor/rte_distributor.c
> @@ -595,6 +595,12 @@ rte_distributor_create_v1705(const char *name,
>   	RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0);
>   	RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0);
>   
> +	if (name == NULL || num_workers >=
> +		(unsigned int)RTE_MIN(RTE_DISTRIB_MAX_WORKERS, RTE_MAX_LCORE)) {
> +		rte_errno = EINVAL;
> +		return NULL;
> +	}
> +
>   	if (alg_type == RTE_DIST_ALG_SINGLE) {
>   		d = malloc(sizeof(struct rte_distributor));
>   		if (d == NULL) {
> @@ -612,11 +618,6 @@ rte_distributor_create_v1705(const char *name,
>   		return d;
>   	}
>   
> -	if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) {
> -		rte_errno = EINVAL;
> -		return NULL;
> -	}
> -
>   	snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
>   	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
>   	if (mz == NULL) {


Builds OK, runs OK. Change makes sense, also good to move the check to 
the new location.

Acked-by: David Hunt <david.hunt@intel.com>

Re: [dpdk-dev] [PATCH v3 2/2] distributor: fix additional check on no of workers

[Thomas Monjalon]

17/07/2019 17:32, Hunt, David:
> Hi Harman,
> 
> On 05/07/2019 11:05, Harman Kalra wrote:
> > No of workers should never exceed RTE_MAX_LCORE.
> > RTE_DIST_ALG_SINGLE also require no of workers check.
> >
> > Fixes: 775003ad2f96 ("distributor: add new burst-capable library")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Harman Kalra <hkalra@marvell.com>
> 
> Builds OK, runs OK. Change makes sense, also good to move the check to 
> the new location.
> 
> Acked-by: David Hunt <david.hunt@intel.com>

Applied, thanks