* [dpdk-dev] [PATCH] net/af_packet: make bypass configurable
@ 2017-09-19 21:21 Chas Williams
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
` (2 more replies)
0 siblings, 3 replies; 10+ messages in thread
From: Chas Williams @ 2017-09-19 21:21 UTC (permalink / raw)
To: dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
From: "Charles (Chas) Williams" <ciwillia@brocade.com>
In certain situations, low speed interfaces, it may be desirable to
have the flow control provided by the kernel queueing disciplines.
Signed-off-by: Chas Williams <chas3@att.com>
---
drivers/net/af_packet/rte_eth_af_packet.c | 22 +++++++++++++++++-----
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index 7aa26e5..a299a4e 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -59,6 +59,7 @@
#define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
#define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
#define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
+#define ETH_AF_PACKET_BYPASS_ARG "bypass"
#define DFLT_BLOCK_SIZE (1 << 12)
#define DFLT_FRAME_SIZE (1 << 11)
@@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
ETH_AF_PACKET_BLOCKSIZE_ARG,
ETH_AF_PACKET_FRAMESIZE_ARG,
ETH_AF_PACKET_FRAMECOUNT_ARG,
+ ETH_AF_PACKET_BYPASS_ARG,
NULL
};
@@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
unsigned int blockcnt,
unsigned int framesize,
unsigned int framecnt,
+ unsigned int bypass,
struct pmd_internals **internals,
struct rte_eth_dev **eth_dev,
struct rte_kvargs *kvlist)
@@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
#if defined(PACKET_FANOUT)
int fanout_arg;
#endif
-#if defined(PACKET_QDISC_BYPASS)
- int bypass;
-#endif
for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
pair = &kvlist->pairs[k_idx];
@@ -698,7 +698,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
}
#if defined(PACKET_QDISC_BYPASS)
- bypass = 1;
rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
&bypass, sizeof(bypass));
if (rc == -1) {
@@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
unsigned int framesize = DFLT_FRAME_SIZE;
unsigned int framecount = DFLT_FRAME_COUNT;
unsigned int qpairs = 1;
+ unsigned int bypass = 1;
/* do some parameter checking */
if (*sockfd < 0)
@@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
}
continue;
}
+ if (strstr(pair->key, ETH_AF_PACKET_BYPASS_ARG) != NULL) {
+ bypass = atoi(pair->value);
+ if (bypass > 2) {
+ RTE_LOG(ERR, PMD,
+ "%s: invalid bypass value\n",
+ name);
+ return -1;
+ }
+ continue;
+ }
}
if (framesize > blocksize) {
@@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
if (rte_pmd_init_internals(dev, *sockfd, qpairs,
blocksize, blockcount,
framesize, framecount,
+ bypass,
&internals, ð_dev,
kvlist) < 0)
return -1;
@@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
"qpairs=<int> "
"blocksz=<int> "
"framesz=<int> "
- "framecnt=<int>");
+ "framecnt=<int> "
+ "bypass=<int>");
--
2.9.5
^ permalink raw reply [flat|nested] 10+ messages in thread
* [dpdk-dev] [PATCH v2] net/af_packet: make bypass configurable
2017-09-19 21:21 [dpdk-dev] [PATCH] net/af_packet: make bypass configurable Chas Williams
@ 2017-09-19 21:45 ` Chas Williams
2017-09-20 9:59 ` Luca Boccassi
2017-09-20 13:21 ` Ferruh Yigit
2017-09-21 3:25 ` [dpdk-dev] [PATCH v3] " Chas Williams
2017-09-21 9:57 ` [dpdk-dev] [PATCH v4] " Chas Williams
2 siblings, 2 replies; 10+ messages in thread
From: Chas Williams @ 2017-09-19 21:45 UTC (permalink / raw)
To: dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
From: "Charles (Chas) Williams" <ciwillia@brocade.com>
In certain situations, low speed interfaces, it may be desirable to
have the flow control provided by the kernel queueing disciplines.
Signed-off-by: Chas Williams <chas3@att.com>
---
drivers/net/af_packet/rte_eth_af_packet.c | 22 +++++++++++++++++-----
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index 7aa26e5..e3858fa 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -59,6 +59,7 @@
#define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
#define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
#define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
+#define ETH_AF_PACKET_BYPASS_ARG "bypass"
#define DFLT_BLOCK_SIZE (1 << 12)
#define DFLT_FRAME_SIZE (1 << 11)
@@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
ETH_AF_PACKET_BLOCKSIZE_ARG,
ETH_AF_PACKET_FRAMESIZE_ARG,
ETH_AF_PACKET_FRAMECOUNT_ARG,
+ ETH_AF_PACKET_BYPASS_ARG,
NULL
};
@@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
unsigned int blockcnt,
unsigned int framesize,
unsigned int framecnt,
+ unsigned int bypass,
struct pmd_internals **internals,
struct rte_eth_dev **eth_dev,
struct rte_kvargs *kvlist)
@@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
#if defined(PACKET_FANOUT)
int fanout_arg;
#endif
-#if defined(PACKET_QDISC_BYPASS)
- int bypass;
-#endif
for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
pair = &kvlist->pairs[k_idx];
@@ -698,7 +698,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
}
#if defined(PACKET_QDISC_BYPASS)
- bypass = 1;
rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
&bypass, sizeof(bypass));
if (rc == -1) {
@@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
unsigned int framesize = DFLT_FRAME_SIZE;
unsigned int framecount = DFLT_FRAME_COUNT;
unsigned int qpairs = 1;
+ unsigned int bypass = 1;
/* do some parameter checking */
if (*sockfd < 0)
@@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
}
continue;
}
+ if (strstr(pair->key, ETH_AF_PACKET_BYPASS_ARG) != NULL) {
+ bypass = atoi(pair->value);
+ if (bypass > 2) {
+ RTE_LOG(ERR, PMD,
+ "%s: invalid bypass value\n",
+ name);
+ return -1;
+ }
+ continue;
+ }
}
if (framesize > blocksize) {
@@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
if (rte_pmd_init_internals(dev, *sockfd, qpairs,
blocksize, blockcount,
framesize, framecount,
+ bypass,
&internals, ð_dev,
kvlist) < 0)
return -1;
@@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
"qpairs=<int> "
"blocksz=<int> "
"framesz=<int> "
- "framecnt=<int>");
+ "framecnt=<int> "
+ "bypass=<int>");
--
2.9.5
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v2] net/af_packet: make bypass configurable
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
@ 2017-09-20 9:59 ` Luca Boccassi
2017-09-20 13:21 ` Ferruh Yigit
1 sibling, 0 replies; 10+ messages in thread
From: Luca Boccassi @ 2017-09-20 9:59 UTC (permalink / raw)
To: Chas Williams, dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
On Tue, 2017-09-19 at 17:45 -0400, Chas Williams wrote:
> From: "Charles (Chas) Williams" <ciwillia@brocade.com>
>
> In certain situations, low speed interfaces, it may be desirable to
> have the flow control provided by the kernel queueing disciplines.
>
> Signed-off-by: Chas Williams <chas3@att.com>
> ---
> drivers/net/af_packet/rte_eth_af_packet.c | 22 +++++++++++++++++--
> ---
> 1 file changed, 17 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/af_packet/rte_eth_af_packet.c
> b/drivers/net/af_packet/rte_eth_af_packet.c
> index 7aa26e5..e3858fa 100644
> --- a/drivers/net/af_packet/rte_eth_af_packet.c
> +++ b/drivers/net/af_packet/rte_eth_af_packet.c
> @@ -59,6 +59,7 @@
> #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
> #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
> #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
> +#define ETH_AF_PACKET_BYPASS_ARG "bypass"
>
> #define DFLT_BLOCK_SIZE (1 << 12)
> #define DFLT_FRAME_SIZE (1 << 11)
> @@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
> ETH_AF_PACKET_BLOCKSIZE_ARG,
> ETH_AF_PACKET_FRAMESIZE_ARG,
> ETH_AF_PACKET_FRAMECOUNT_ARG,
> + ETH_AF_PACKET_BYPASS_ARG,
> NULL
> };
>
> @@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device
> *dev,
> unsigned int blockcnt,
> unsigned int framesize,
> unsigned int framecnt,
> + unsigned int bypass,
> struct pmd_internals **internals,
> struct rte_eth_dev **eth_dev,
> struct rte_kvargs *kvlist)
> @@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device
> *dev,
> #if defined(PACKET_FANOUT)
> int fanout_arg;
> #endif
> -#if defined(PACKET_QDISC_BYPASS)
> - int bypass;
> -#endif
>
> for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
> pair = &kvlist->pairs[k_idx];
> @@ -698,7 +698,6 @@ rte_pmd_init_internals(struct rte_vdev_device
> *dev,
> }
>
> #if defined(PACKET_QDISC_BYPASS)
> - bypass = 1;
> rc = setsockopt(qsockfd, SOL_PACKET,
> PACKET_QDISC_BYPASS,
> &bypass, sizeof(bypass));
> if (rc == -1) {
> @@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> unsigned int framesize = DFLT_FRAME_SIZE;
> unsigned int framecount = DFLT_FRAME_COUNT;
> unsigned int qpairs = 1;
> + unsigned int bypass = 1;
>
> /* do some parameter checking */
> if (*sockfd < 0)
> @@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> }
> continue;
> }
> + if (strstr(pair->key, ETH_AF_PACKET_BYPASS_ARG) !=
> NULL) {
> + bypass = atoi(pair->value);
> + if (bypass > 2) {
> + RTE_LOG(ERR, PMD,
> + "%s: invalid bypass
> value\n",
> + name);
> + return -1;
> + }
> + continue;
> + }
> }
>
> if (framesize > blocksize) {
> @@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> if (rte_pmd_init_internals(dev, *sockfd, qpairs,
> blocksize, blockcount,
> framesize, framecount,
> + bypass,
> &internals, ð_dev,
> kvlist) < 0)
> return -1;
> @@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
> "qpairs=<int> "
> "blocksz=<int> "
> "framesz=<int> "
> - "framecnt=<int>");
> + "framecnt=<int> "
> + "bypass=<int>");
Acked-by: Luca Boccassi <bluca@debian.org>
--
Kind regards,
Luca Boccassi
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v2] net/af_packet: make bypass configurable
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
2017-09-20 9:59 ` Luca Boccassi
@ 2017-09-20 13:21 ` Ferruh Yigit
2017-10-05 14:15 ` Chas Williams
1 sibling, 1 reply; 10+ messages in thread
From: Ferruh Yigit @ 2017-09-20 13:21 UTC (permalink / raw)
To: Chas Williams, dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
On 9/19/2017 10:45 PM, Chas Williams wrote:
> From: "Charles (Chas) Williams" <ciwillia@brocade.com>
>
> In certain situations, low speed interfaces, it may be desirable to
> have the flow control provided by the kernel queueing disciplines.
Out of curiosity, do you have any compression of performance numbers
with and without qdisc?
>
> Signed-off-by: Chas Williams <chas3@att.com>
> ---
> drivers/net/af_packet/rte_eth_af_packet.c | 22 +++++++++++++++++-----
> 1 file changed, 17 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
> index 7aa26e5..e3858fa 100644
> --- a/drivers/net/af_packet/rte_eth_af_packet.c
> +++ b/drivers/net/af_packet/rte_eth_af_packet.c
> @@ -59,6 +59,7 @@
> #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
> #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
> #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
> +#define ETH_AF_PACKET_BYPASS_ARG "bypass"
I thinks argument name "bypass" on its own is not clear what is
bypassed, would you mind using something like "qdisc_bypass"?
>
> #define DFLT_BLOCK_SIZE (1 << 12)
> #define DFLT_FRAME_SIZE (1 << 11)
> @@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
> ETH_AF_PACKET_BLOCKSIZE_ARG,
> ETH_AF_PACKET_FRAMESIZE_ARG,
> ETH_AF_PACKET_FRAMECOUNT_ARG,
> + ETH_AF_PACKET_BYPASS_ARG,
Same comment here, what about "ETH_AF_PACKET_QDISC_BYPASS_ARG" ?
> NULL
> };
>
> @@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
> unsigned int blockcnt,
> unsigned int framesize,
> unsigned int framecnt,
> + unsigned int bypass,
> struct pmd_internals **internals,
> struct rte_eth_dev **eth_dev,
> struct rte_kvargs *kvlist)
> @@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
> #if defined(PACKET_FANOUT)
> int fanout_arg;
> #endif
> -#if defined(PACKET_QDISC_BYPASS)
> - int bypass;
> -#endif
>
> for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
> pair = &kvlist->pairs[k_idx];
> @@ -698,7 +698,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
> }
>
> #if defined(PACKET_QDISC_BYPASS)
> - bypass = 1;
> rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
> &bypass, sizeof(bypass));
> if (rc == -1) {
> @@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> unsigned int framesize = DFLT_FRAME_SIZE;
> unsigned int framecount = DFLT_FRAME_COUNT;
> unsigned int qpairs = 1;
> + unsigned int bypass = 1;
>
> /* do some parameter checking */
> if (*sockfd < 0)
> @@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> }
> continue;
> }
> + if (strstr(pair->key, ETH_AF_PACKET_BYPASS_ARG) != NULL) {
> + bypass = atoi(pair->value);
> + if (bypass > 2) {
This accepts "2" too .
As far as I can see kernel side checks if this value is zero or not, so
perhaps this check is not required at all. But if you want to keep it I
guess intention is to limit the value to "0" and "1".
> + RTE_LOG(ERR, PMD,
> + "%s: invalid bypass value\n",
> + name);
> + return -1;
> + }
> + continue;
> + }
> }
>
> if (framesize > blocksize) {
> @@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
> if (rte_pmd_init_internals(dev, *sockfd, qpairs,
> blocksize, blockcount,
> framesize, framecount,
> + bypass,
> &internals, ð_dev,
> kvlist) < 0)
> return -1;
> @@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
> "qpairs=<int> "
> "blocksz=<int> "
> "framesz=<int> "
> - "framecnt=<int>");
> + "framecnt=<int> "
> + "bypass=<int>"
Although storage is integer, does is make sense to document it as "0|1"
to clarify the usage?
);
>
^ permalink raw reply [flat|nested] 10+ messages in thread
* [dpdk-dev] [PATCH v3] net/af_packet: make bypass configurable
2017-09-19 21:21 [dpdk-dev] [PATCH] net/af_packet: make bypass configurable Chas Williams
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
@ 2017-09-21 3:25 ` Chas Williams
2017-09-21 9:34 ` Ferruh Yigit
2017-09-21 9:57 ` [dpdk-dev] [PATCH v4] " Chas Williams
2 siblings, 1 reply; 10+ messages in thread
From: Chas Williams @ 2017-09-21 3:25 UTC (permalink / raw)
To: dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
From: "Charles (Chas) Williams" <ciwillia@brocade.com>
In certain situations, low speed interfaces, it may be desirable to
have the flow control provided by the kernel queueing disciplines.
Signed-off-by: Chas Williams <chas3@att.com>
---
drivers/net/af_packet/rte_eth_af_packet.c | 24 ++++++++++++++++++------
1 file changed, 18 insertions(+), 6 deletions(-)
diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index 7aa26e5..b95d87d 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -59,6 +59,7 @@
#define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
#define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
#define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
+#define ETH_AF_PACKET_BYPASS_ARG "qdisc_bypass"
#define DFLT_BLOCK_SIZE (1 << 12)
#define DFLT_FRAME_SIZE (1 << 11)
@@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
ETH_AF_PACKET_BLOCKSIZE_ARG,
ETH_AF_PACKET_FRAMESIZE_ARG,
ETH_AF_PACKET_FRAMECOUNT_ARG,
+ ETH_AF_PACKET_BYPASS_ARG,
NULL
};
@@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
unsigned int blockcnt,
unsigned int framesize,
unsigned int framecnt,
+ unsigned int qdisc_bypass,
struct pmd_internals **internals,
struct rte_eth_dev **eth_dev,
struct rte_kvargs *kvlist)
@@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
#if defined(PACKET_FANOUT)
int fanout_arg;
#endif
-#if defined(PACKET_QDISC_BYPASS)
- int bypass;
-#endif
for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
pair = &kvlist->pairs[k_idx];
@@ -698,9 +698,8 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
}
#if defined(PACKET_QDISC_BYPASS)
- bypass = 1;
rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
- &bypass, sizeof(bypass));
+ &qdisc_bypass, sizeof(qdisc_bypass));
if (rc == -1) {
RTE_LOG(ERR, PMD,
"%s: could not set PACKET_QDISC_BYPASS "
@@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
unsigned int framesize = DFLT_FRAME_SIZE;
unsigned int framecount = DFLT_FRAME_COUNT;
unsigned int qpairs = 1;
+ unsigned int qdisc_bypass = 1;
/* do some parameter checking */
if (*sockfd < 0)
@@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
}
continue;
}
+ if (strstr(pair->key, ETH_AF_PACKET_BYPASS_ARG) != NULL) {
+ qdisc_bypass = atoi(pair->value);
+ if (qdisc_bypass > 1) {
+ RTE_LOG(ERR, PMD,
+ "%s: invalid bypass value\n",
+ name);
+ return -1;
+ }
+ continue;
+ }
}
if (framesize > blocksize) {
@@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
if (rte_pmd_init_internals(dev, *sockfd, qpairs,
blocksize, blockcount,
framesize, framecount,
+ qdisc_bypass,
&internals, ð_dev,
kvlist) < 0)
return -1;
@@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
"qpairs=<int> "
"blocksz=<int> "
"framesz=<int> "
- "framecnt=<int>");
+ "framecnt=<int> "
+ "qdisc_bypass=<0|1>");
--
2.9.5
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v3] net/af_packet: make bypass configurable
2017-09-21 3:25 ` [dpdk-dev] [PATCH v3] " Chas Williams
@ 2017-09-21 9:34 ` Ferruh Yigit
0 siblings, 0 replies; 10+ messages in thread
From: Ferruh Yigit @ 2017-09-21 9:34 UTC (permalink / raw)
To: Chas Williams, dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
On 9/21/2017 4:25 AM, Chas Williams wrote:
> From: "Charles (Chas) Williams" <ciwillia@brocade.com>
>
> In certain situations, low speed interfaces, it may be desirable to
> have the flow control provided by the kernel queueing disciplines.
>
> Signed-off-by: Chas Williams <chas3@att.com>
> ---
> drivers/net/af_packet/rte_eth_af_packet.c | 24 ++++++++++++++++++------
> 1 file changed, 18 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
> index 7aa26e5..b95d87d 100644
> --- a/drivers/net/af_packet/rte_eth_af_packet.c
> +++ b/drivers/net/af_packet/rte_eth_af_packet.c
> @@ -59,6 +59,7 @@
> #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
> #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
> #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
> +#define ETH_AF_PACKET_BYPASS_ARG "qdisc_bypass"
>
> #define DFLT_BLOCK_SIZE (1 << 12)
> #define DFLT_FRAME_SIZE (1 << 11)
> @@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
> ETH_AF_PACKET_BLOCKSIZE_ARG,
> ETH_AF_PACKET_FRAMESIZE_ARG,
> ETH_AF_PACKET_FRAMECOUNT_ARG,
> + ETH_AF_PACKET_BYPASS_ARG,
There was a comment about the define name, just to double check if this
has been missed or intentional.
<...>
^ permalink raw reply [flat|nested] 10+ messages in thread
* [dpdk-dev] [PATCH v4] net/af_packet: make bypass configurable
2017-09-19 21:21 [dpdk-dev] [PATCH] net/af_packet: make bypass configurable Chas Williams
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
2017-09-21 3:25 ` [dpdk-dev] [PATCH v3] " Chas Williams
@ 2017-09-21 9:57 ` Chas Williams
2017-09-21 14:43 ` Ferruh Yigit
2 siblings, 1 reply; 10+ messages in thread
From: Chas Williams @ 2017-09-21 9:57 UTC (permalink / raw)
To: dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
From: "Charles (Chas) Williams" <ciwillia@brocade.com>
In certain situations, low speed interfaces, it may be desirable to
have the flow control provided by the kernel queueing disciplines.
Signed-off-by: Chas Williams <chas3@att.com>
---
drivers/net/af_packet/rte_eth_af_packet.c | 24 ++++++++++++++++++------
1 file changed, 18 insertions(+), 6 deletions(-)
diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index 7aa26e5..8089eda 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -59,6 +59,7 @@
#define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
#define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
#define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
+#define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
#define DFLT_BLOCK_SIZE (1 << 12)
#define DFLT_FRAME_SIZE (1 << 11)
@@ -115,6 +116,7 @@ static const char *valid_arguments[] = {
ETH_AF_PACKET_BLOCKSIZE_ARG,
ETH_AF_PACKET_FRAMESIZE_ARG,
ETH_AF_PACKET_FRAMECOUNT_ARG,
+ ETH_AF_PACKET_QDISC_BYPASS_ARG,
NULL
};
@@ -559,6 +561,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
unsigned int blockcnt,
unsigned int framesize,
unsigned int framecnt,
+ unsigned int qdisc_bypass,
struct pmd_internals **internals,
struct rte_eth_dev **eth_dev,
struct rte_kvargs *kvlist)
@@ -580,9 +583,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
#if defined(PACKET_FANOUT)
int fanout_arg;
#endif
-#if defined(PACKET_QDISC_BYPASS)
- int bypass;
-#endif
for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
pair = &kvlist->pairs[k_idx];
@@ -698,9 +698,8 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
}
#if defined(PACKET_QDISC_BYPASS)
- bypass = 1;
rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
- &bypass, sizeof(bypass));
+ &qdisc_bypass, sizeof(qdisc_bypass));
if (rc == -1) {
RTE_LOG(ERR, PMD,
"%s: could not set PACKET_QDISC_BYPASS "
@@ -851,6 +850,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
unsigned int framesize = DFLT_FRAME_SIZE;
unsigned int framecount = DFLT_FRAME_COUNT;
unsigned int qpairs = 1;
+ unsigned int qdisc_bypass = 1;
/* do some parameter checking */
if (*sockfd < 0)
@@ -902,6 +902,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
}
continue;
}
+ if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
+ qdisc_bypass = atoi(pair->value);
+ if (qdisc_bypass > 1) {
+ RTE_LOG(ERR, PMD,
+ "%s: invalid bypass value\n",
+ name);
+ return -1;
+ }
+ continue;
+ }
}
if (framesize > blocksize) {
@@ -927,6 +937,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
if (rte_pmd_init_internals(dev, *sockfd, qpairs,
blocksize, blockcount,
framesize, framecount,
+ qdisc_bypass,
&internals, ð_dev,
kvlist) < 0)
return -1;
@@ -1021,4 +1032,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
"qpairs=<int> "
"blocksz=<int> "
"framesz=<int> "
- "framecnt=<int>");
+ "framecnt=<int> "
+ "qdisc_bypass=<0|1>");
--
2.9.5
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v4] net/af_packet: make bypass configurable
2017-09-21 9:57 ` [dpdk-dev] [PATCH v4] " Chas Williams
@ 2017-09-21 14:43 ` Ferruh Yigit
2017-09-22 17:44 ` Ferruh Yigit
0 siblings, 1 reply; 10+ messages in thread
From: Ferruh Yigit @ 2017-09-21 14:43 UTC (permalink / raw)
To: Chas Williams, dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
On 9/21/2017 10:57 AM, Chas Williams wrote:
> From: "Charles (Chas) Williams" <ciwillia@brocade.com>
>
> In certain situations, low speed interfaces, it may be desirable to
> have the flow control provided by the kernel queueing disciplines.
>
> Signed-off-by: Chas Williams <chas3@att.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v4] net/af_packet: make bypass configurable
2017-09-21 14:43 ` Ferruh Yigit
@ 2017-09-22 17:44 ` Ferruh Yigit
0 siblings, 0 replies; 10+ messages in thread
From: Ferruh Yigit @ 2017-09-22 17:44 UTC (permalink / raw)
To: Chas Williams, dev; +Cc: linville, Charles (Chas) Williams, Chas Williams
On 9/21/2017 3:43 PM, Ferruh Yigit wrote:
> On 9/21/2017 10:57 AM, Chas Williams wrote:
>> From: "Charles (Chas) Williams" <ciwillia@brocade.com>
>>
>> In certain situations, low speed interfaces, it may be desirable to
>> have the flow control provided by the kernel queueing disciplines.
>>
>> Signed-off-by: Chas Williams <chas3@att.com>
>
> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Applied to dpdk-next-net/master, thanks.
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [dpdk-dev] [PATCH v2] net/af_packet: make bypass configurable
2017-09-20 13:21 ` Ferruh Yigit
@ 2017-10-05 14:15 ` Chas Williams
0 siblings, 0 replies; 10+ messages in thread
From: Chas Williams @ 2017-10-05 14:15 UTC (permalink / raw)
To: Ferruh Yigit
Cc: dev, John W. Linville, Charles (Chas) Williams, Chas Williams
On Wed, Sep 20, 2017 at 9:21 AM, Ferruh Yigit <ferruh.yigit@intel.com>
wrote:
> On 9/19/2017 10:45 PM, Chas Williams wrote:
> > From: "Charles (Chas) Williams" <ciwillia@brocade.com>
> >
> > In certain situations, low speed interfaces, it may be desirable to
> > have the flow control provided by the kernel queueing disciplines.
>
> Out of curiosity, do you have any compression of performance numbers
> with and without qdisc?
>
I had some time to test this. I didn't see any loss in performance but the
interfaces
I am using only go about 200-300Mbps. it likely makes a difference if you
can go
much much faster.
^ permalink raw reply [flat|nested] 10+ messages in thread
end of thread, other threads:[~2017-10-05 14:15 UTC | newest]
Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-09-19 21:21 [dpdk-dev] [PATCH] net/af_packet: make bypass configurable Chas Williams
2017-09-19 21:45 ` [dpdk-dev] [PATCH v2] " Chas Williams
2017-09-20 9:59 ` Luca Boccassi
2017-09-20 13:21 ` Ferruh Yigit
2017-10-05 14:15 ` Chas Williams
2017-09-21 3:25 ` [dpdk-dev] [PATCH v3] " Chas Williams
2017-09-21 9:34 ` Ferruh Yigit
2017-09-21 9:57 ` [dpdk-dev] [PATCH v4] " Chas Williams
2017-09-21 14:43 ` Ferruh Yigit
2017-09-22 17:44 ` Ferruh Yigit
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).