[dpdk-dev] [PATCH 0/2] Fixed/Simplified Ring PMD

[dpdk-dev] [PATCH 0/2] Fixed/Simplified Ring PMD

[Pablo de Lara]

Ring PMD library and unit test was modified last release,
causing the unit test to need ring ethdevs to be created
from the command line. This patch simplifies the test
and make it functional with devices created with EAL option vdev.

Also, ring ethdev creation from vdev has been changed, so now
only one device is created for each vdev argument (up to 3 devices
were created). 


Pablo de Lara (2):
  ring_pmd: Removed extra devices creation when passing vdev parameter
  app/test: Simplified ring pmd test

 app/test/test_pmd_ring.c           |  202 +++++++++++++-----------------------
 lib/librte_pmd_ring/rte_eth_ring.c |    3 +-
 2 files changed, 73 insertions(+), 132 deletions(-)

[dpdk-dev] [PATCH 1/2] ring_pmd: Removed extra devices creation when passing vdev parameter

[Pablo de Lara]

When passing extra arguments in EAL option --vdev, to create
ring ethdevs, API was creating three ethdevs, even if there
was just one argument, such as CREATE.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
---
 lib/librte_pmd_ring/rte_eth_ring.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/lib/librte_pmd_ring/rte_eth_ring.c b/lib/librte_pmd_ring/rte_eth_ring.c
index 73c649e..4f1b6ed 100644
--- a/lib/librte_pmd_ring/rte_eth_ring.c
+++ b/lib/librte_pmd_ring/rte_eth_ring.c
@@ -493,7 +493,6 @@ rte_pmd_ring_devinit(const char *name, const char *params)
 			eth_dev_ring_create(name, rte_socket_id(), DEV_CREATE);
 			return 0;
 		} else {
-			eth_dev_ring_create(name, rte_socket_id(), DEV_CREATE);
 			ret = rte_kvargs_count(kvlist, ETH_RING_NUMA_NODE_ACTION_ARG);
 			info = rte_zmalloc("struct node_action_list", sizeof(struct node_action_list) +
 					   (sizeof(struct node_action_pair) * ret), 0);
@@ -510,7 +509,7 @@ rte_pmd_ring_devinit(const char *name, const char *params)
 				goto out_free;
 
 			for (info->count = 0; info->count < info->total; info->count++) {
-				eth_dev_ring_pair_create(name, info->list[info->count].node,
+				eth_dev_ring_create(name, info->list[info->count].node,
 						    info->list[info->count].action);
 			}
 		}
-- 
1.7.0.7

[dpdk-dev] [PATCH 2/2] app/test: Simplified ring pmd test

[Pablo de Lara]

As this unit test does not create devices anymore,
and uses devices created by EAL option --vdev,
there were unnecesary tests that were repeated,
so they have been removed.

So now there are three tests:

1 - Test two devices that share a ring, one device
    with just one RX queue and the other with one
    TX queue.

2 - Test a device connected to itself (loopback) by
    a ring, with both RX and TX queue.

3 - Test two devices that share a ring, but both devices
    with RX and TX queue, so they can send packets to themselves
    and to the other device.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
---
 app/test/test_pmd_ring.c |  202 ++++++++++++++++-----------------------------
 1 files changed, 72 insertions(+), 130 deletions(-)

diff --git a/app/test/test_pmd_ring.c b/app/test/test_pmd_ring.c
index 0d3d95c..19ad0e9 100644
--- a/app/test/test_pmd_ring.c
+++ b/app/test/test_pmd_ring.c
@@ -39,18 +39,13 @@
 #include <rte_eth_ring.h>
 #include <rte_ethdev.h>
 
-/* two test rings, r1 is used by two ports, r2 just by one */
-static struct rte_ring *r1[2], *r2;
-
 static struct rte_mempool *mp;
-static uint8_t start_idx; /* will store the port id of the first of our new ports */
-
-#define TX_PORT (uint8_t)(start_idx + 1)
-#define RX_PORT (uint8_t)(start_idx + 2)
-#define RXTX_PORT (uint8_t)(start_idx + 3)
-#define RXTX_PORT2 (uint8_t)(start_idx + 4)
-#define RXTX_PORT4 (uint8_t)(start_idx + 6)
-#define RXTX_PORT5 (uint8_t)(start_idx + 7)
+
+#define TX_PORT 0
+#define RX_PORT 1
+#define RXTX_PORT 2
+#define RXTX_PORT2 3
+#define RXTX_PORT3 4
 #define SOCKET0 0
 
 #define RING_SIZE 256
@@ -86,7 +81,7 @@ test_ethdev_configure(void)
 		return -1;
 	}
 	if (rte_eth_dev_configure(RXTX_PORT, 1, 1, &null_conf) < 0) {
-		printf("Configure failed for RX port\n");
+		printf("Configure failed for RXTX port\n");
 		return -1;
 	}
 
@@ -250,197 +245,162 @@ test_stats_reset(void)
 }
 
 static int
-test_pmd_ring_init(void)
+test_pmd_ring_pair_create_attach(void)
 {
-	struct rte_eth_stats stats;
+	struct rte_eth_stats stats, stats2;
 	struct rte_mbuf buf, *pbuf = &buf;
 	struct rte_eth_conf null_conf;
 
-	printf("Testing ring pmd init\n");
-
-	if (RXTX_PORT2 >= RTE_MAX_ETHPORTS) {
+	if ((RXTX_PORT2 >= RTE_MAX_ETHPORTS) || (RXTX_PORT3 >= RTE_MAX_ETHPORTS)) {
 		printf(" TX/RX port exceed max eth ports\n");
 		return -1;
 	}
-	if (rte_eth_dev_configure(RXTX_PORT2, 1, 1, &null_conf) < 0) {
+	if ((rte_eth_dev_configure(RXTX_PORT2, 1, 1, &null_conf) < 0)
+		|| (rte_eth_dev_configure(RXTX_PORT3, 1, 1, &null_conf) < 0)) {
 		printf("Configure failed for RXTX port\n");
 		return -1;
 	}
 
-	if (rte_eth_tx_queue_setup(RXTX_PORT2, 0, RING_SIZE, SOCKET0, NULL) < 0) {
+	if ((rte_eth_tx_queue_setup(RXTX_PORT2, 0, RING_SIZE, SOCKET0, NULL) < 0)
+		|| (rte_eth_tx_queue_setup(RXTX_PORT3, 0, RING_SIZE, SOCKET0, NULL) < 0)) {
 		printf("TX queue setup failed\n");
 		return -1;
 	}
 
-	if (rte_eth_rx_queue_setup(RXTX_PORT2, 0, RING_SIZE, SOCKET0,
-			NULL, mp) < 0) {
+	if ((rte_eth_rx_queue_setup(RXTX_PORT2, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)
+		|| (rte_eth_rx_queue_setup(RXTX_PORT3, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)) {
 		printf("RX queue setup failed\n");
 		return -1;
 	}
 
-	if (rte_eth_dev_start(RXTX_PORT2) < 0) {
-		printf("Error starting RX port\n");
+	if ((rte_eth_dev_start(RXTX_PORT2) < 0)
+		|| (rte_eth_dev_start(RXTX_PORT3) < 0)) {
+		printf("Error starting RXTX port\n");
 		return -1;
 	}
 
-	/* send and receive 1 packet and check for stats update */
+	/*
+	 * send and receive 1 packet (RXTX_PORT2 -> RXTX_PORT3)
+	 * and check for stats update
+	 */
 	if (rte_eth_tx_burst(RXTX_PORT2, 0, &pbuf, 1) != 1) {
 		printf("Error sending packet to RXTX port\n");
 		return -1;
 	}
 
-	if (rte_eth_rx_burst(RXTX_PORT2, 0, &pbuf, 1) != 1) {
+	if (rte_eth_rx_burst(RXTX_PORT3, 0, &pbuf, 1) != 1) {
 		printf("Error receiving packet from RXTX port\n");
 		return -1;
 	}
 
 	rte_eth_stats_get(RXTX_PORT2, &stats);
-	if (stats.ipackets != 1 || stats.opackets != 1 ||
+	rte_eth_stats_get(RXTX_PORT3, &stats2);
+	if (stats.ipackets != 0 || stats.opackets != 1 ||
 			stats.ibytes != 0 || stats.obytes != 0 ||
 			stats.ierrors != 0 || stats.oerrors != 0) {
 		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	rte_eth_dev_stop(RXTX_PORT2);
-
-	return 0;
-}
-
-static int
-test_pmd_ring_pair_create(void)
-{
-	struct rte_eth_stats stats, stats2;
-	struct rte_mbuf buf, *pbuf = &buf;
-	struct rte_eth_conf null_conf;
-
-	if ((RXTX_PORT4 >= RTE_MAX_ETHPORTS) || (RXTX_PORT5 >= RTE_MAX_ETHPORTS)) {
-		printf(" TX/RX port exceed max eth ports\n");
-		return -1;
-	}
-	if ((rte_eth_dev_configure(RXTX_PORT4, 1, 1, &null_conf) < 0)
-		|| (rte_eth_dev_configure(RXTX_PORT5, 1, 1, &null_conf) < 0)) {
-		printf("Configure failed for RXTX port\n");
-		return -1;
-	}
-
-	if ((rte_eth_tx_queue_setup(RXTX_PORT4, 0, RING_SIZE, SOCKET0, NULL) < 0)
-		|| (rte_eth_tx_queue_setup(RXTX_PORT5, 0, RING_SIZE, SOCKET0, NULL) < 0)) {
-		printf("TX queue setup failed\n");
-		return -1;
-	}
-
-	if ((rte_eth_rx_queue_setup(RXTX_PORT4, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)
-		|| (rte_eth_rx_queue_setup(RXTX_PORT5, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)) {
-		printf("RX queue setup failed\n");
-		return -1;
-	}
-
-	if ((rte_eth_dev_start(RXTX_PORT4) < 0)
-		|| (rte_eth_dev_start(RXTX_PORT5) < 0)) {
-		printf("Error starting RXTX port\n");
+	if (stats2.ipackets != 1 || stats2.opackets != 0 ||
+			stats2.ibytes != 0 || stats2.obytes != 0 ||
+			stats2.ierrors != 0 || stats2.oerrors != 0) {
+		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	/* send and receive 1 packet and check for stats update */
-	if (rte_eth_tx_burst(RXTX_PORT4, 0, &pbuf, 1) != 1) {
+	/*
+	 * send and receive 1 packet (RXTX_PORT3 -> RXTX_PORT2)
+	 * and check for stats update
+	 */
+	if (rte_eth_tx_burst(RXTX_PORT3, 0, &pbuf, 1) != 1) {
 		printf("Error sending packet to RXTX port\n");
 		return -1;
 	}
 
-	if (rte_eth_rx_burst(RXTX_PORT5, 0, &pbuf, 1) != 1) {
+	if (rte_eth_rx_burst(RXTX_PORT2, 0, &pbuf, 1) != 1) {
 		printf("Error receiving packet from RXTX port\n");
 		return -1;
 	}
 
-	rte_eth_stats_get(RXTX_PORT4, &stats);
-	rte_eth_stats_get(RXTX_PORT5, &stats2);
-	if (stats.ipackets != 0 || stats.opackets != 1 ||
+	rte_eth_stats_get(RXTX_PORT2, &stats);
+	rte_eth_stats_get(RXTX_PORT3, &stats2);
+	if (stats.ipackets != 1 || stats.opackets != 1 ||
 			stats.ibytes != 0 || stats.obytes != 0 ||
 			stats.ierrors != 0 || stats.oerrors != 0) {
 		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	if (stats2.ipackets != 1 || stats2.opackets != 0 ||
+	if (stats2.ipackets != 1 || stats2.opackets != 1 ||
 			stats2.ibytes != 0 || stats2.obytes != 0 ||
 			stats2.ierrors != 0 || stats2.oerrors != 0) {
 		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	rte_eth_dev_stop(RXTX_PORT4);
-	rte_eth_dev_stop(RXTX_PORT5);
-
-	return 0;
-}
-
-static int
-test_pmd_ring_pair_attach(void)
-{
-	struct rte_eth_stats stats, stats2;
-	struct rte_mbuf buf, *pbuf = &buf;
-	struct rte_eth_conf null_conf;
-
-	if ((RXTX_PORT4 >= RTE_MAX_ETHPORTS) || (RXTX_PORT5 >= RTE_MAX_ETHPORTS)) {
-		printf(" TX/RX port exceed max eth ports\n");
-		return -1;
-	}
-	if ((rte_eth_dev_configure(RXTX_PORT4, 1, 1, &null_conf) < 0)
-		|| (rte_eth_dev_configure(RXTX_PORT5, 1, 1, &null_conf) < 0)) {
-		printf("Configure failed for RXTX port\n");
+	/*
+	 * send and receive 1 packet (RXTX_PORT2 -> RXTX_PORT2)
+	 * and check for stats update
+	 */
+	if (rte_eth_tx_burst(RXTX_PORT2, 0, &pbuf, 1) != 1) {
+		printf("Error sending packet to RXTX port\n");
 		return -1;
 	}
 
-	if ((rte_eth_tx_queue_setup(RXTX_PORT4, 0, RING_SIZE, SOCKET0, NULL) < 0)
-		|| (rte_eth_tx_queue_setup(RXTX_PORT5, 0, RING_SIZE, SOCKET0, NULL) < 0)) {
-		printf("TX queue setup failed\n");
+	if (rte_eth_rx_burst(RXTX_PORT2, 0, &pbuf, 1) != 1) {
+		printf("Error receiving packet from RXTX port\n");
 		return -1;
 	}
 
-	if ((rte_eth_rx_queue_setup(RXTX_PORT4, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)
-		|| (rte_eth_rx_queue_setup(RXTX_PORT5, 0, RING_SIZE, SOCKET0, NULL, mp) < 0)) {
-		printf("RX queue setup failed\n");
+	rte_eth_stats_get(RXTX_PORT2, &stats);
+	rte_eth_stats_get(RXTX_PORT3, &stats2);
+	if (stats.ipackets != 2 || stats.opackets != 2 ||
+			stats.ibytes != 0 || stats.obytes != 0 ||
+			stats.ierrors != 0 || stats.oerrors != 0) {
+		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	if ((rte_eth_dev_start(RXTX_PORT4) < 0)
-		|| (rte_eth_dev_start(RXTX_PORT5) < 0)) {
-		printf("Error starting RXTX port\n");
+	if (stats2.ipackets != 1 || stats2.opackets != 1 ||
+			stats2.ibytes != 0 || stats2.obytes != 0 ||
+			stats2.ierrors != 0 || stats2.oerrors != 0) {
+		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	rte_eth_stats_reset(RXTX_PORT4);
-	rte_eth_stats_reset(RXTX_PORT5);
-
-	/* send and receive 1 packet and check for stats update */
-	if (rte_eth_tx_burst(RXTX_PORT4, 0, &pbuf, 1) != 1) {
+	/*
+	 * send and receive 1 packet (RXTX_PORT3 -> RXTX_PORT3)
+	 * and check for stats update
+	 */
+	if (rte_eth_tx_burst(RXTX_PORT3, 0, &pbuf, 1) != 1) {
 		printf("Error sending packet to RXTX port\n");
 		return -1;
 	}
-	if (rte_eth_rx_burst(RXTX_PORT5, 0, &pbuf, 1) != 1) {
+
+	if (rte_eth_rx_burst(RXTX_PORT3, 0, &pbuf, 1) != 1) {
 		printf("Error receiving packet from RXTX port\n");
 		return -1;
 	}
 
-	rte_eth_stats_get(RXTX_PORT4, &stats);
-	rte_eth_stats_get(RXTX_PORT5, &stats2);
-	if (stats.ipackets != 0 || stats.opackets != 1 ||
+	rte_eth_stats_get(RXTX_PORT2, &stats);
+	rte_eth_stats_get(RXTX_PORT3, &stats2);
+	if (stats.ipackets != 2 || stats.opackets != 2 ||
 			stats.ibytes != 0 || stats.obytes != 0 ||
 			stats.ierrors != 0 || stats.oerrors != 0) {
 		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	if (stats2.ipackets != 1 || stats2.opackets != 0 ||
+	if (stats2.ipackets != 2 || stats2.opackets != 2 ||
 			stats2.ibytes != 0 || stats2.obytes != 0 ||
 			stats2.ierrors != 0 || stats2.oerrors != 0) {
 		printf("Error: RXTX port stats are not as expected\n");
 		return -1;
 	}
 
-	rte_eth_dev_stop(RXTX_PORT4);
-	rte_eth_dev_stop(RXTX_PORT5);
+	rte_eth_dev_stop(RXTX_PORT2);
+	rte_eth_dev_stop(RXTX_PORT3);
 
 	return 0;
 }
@@ -448,17 +408,6 @@ test_pmd_ring_pair_attach(void)
 int
 test_pmd_ring(void)
 {
-	r1[0] = rte_ring_create("R1", RING_SIZE, 0, 0);
-	r1[1] = rte_ring_create("R2", RING_SIZE, 0, 0);
-	if (r1[0] == NULL && (r1[0] = rte_ring_lookup("R1")) == NULL)
-		return -1;
-	if (r1[1] == NULL && (r1[1] = rte_ring_lookup("R2")) == NULL)
-		return -1;
-
-	r2 = rte_ring_create("R3", RING_SIZE, 0, RING_F_SP_ENQ|RING_F_SC_DEQ);
-	if (r2 == NULL && (r2 = rte_ring_lookup("R3")) == NULL)
-		return -1;
-
 	mp = rte_mempool_create("mbuf_pool", NB_MBUF,
 			MBUF_SIZE, 32,
 			sizeof(struct rte_pktmbuf_pool_private),
@@ -468,8 +417,6 @@ test_pmd_ring(void)
 	if (mp == NULL)
 		return -1;
 
-	start_idx = rte_eth_dev_count();
-
 	if ((TX_PORT >= RTE_MAX_ETHPORTS) || (RX_PORT >= RTE_MAX_ETHPORTS)\
 		|| (RXTX_PORT >= RTE_MAX_ETHPORTS)) {
 		printf(" TX/RX port exceed max eth ports\n");
@@ -492,14 +439,9 @@ test_pmd_ring(void)
 	rte_eth_dev_stop(TX_PORT);
 	rte_eth_dev_stop(RXTX_PORT);
 
-	if (test_pmd_ring_init() < 0)
+	if (test_pmd_ring_pair_create_attach() < 0)
 		return -1;
 
-	if (test_pmd_ring_pair_create() < 0)
-		return -1;
-
-	if (test_pmd_ring_pair_attach() < 0)
-		return -1;
 	return 0;
 }
 
-- 
1.7.0.7

Re: [dpdk-dev] [PATCH 0/2] Fixed/Simplified Ring PMD

[Neil Horman]

On Wed, Jul 09, 2014 at 03:35:34PM +0100, Pablo de Lara wrote:
> Ring PMD library and unit test was modified last release,
> causing the unit test to need ring ethdevs to be created
> from the command line. This patch simplifies the test
> and make it functional with devices created with EAL option vdev.
> 
> Also, ring ethdev creation from vdev has been changed, so now
> only one device is created for each vdev argument (up to 3 devices
> were created). 
> 
> 
> Pablo de Lara (2):
>   ring_pmd: Removed extra devices creation when passing vdev parameter
>   app/test: Simplified ring pmd test
> 
>  app/test/test_pmd_ring.c           |  202 +++++++++++++-----------------------
>  lib/librte_pmd_ring/rte_eth_ring.c |    3 +-
>  2 files changed, 73 insertions(+), 132 deletions(-)
> 
> 


Series
Acked-by: Neil Horman <nhorman@tuxdriver.com>

Nice work Pablo.
Neil

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Hobywan Kenoby]

Hi Sujith,
It was exciting to see open source code coming from Cisco ensuring a DPDK application can run on any platform and with any card....I haven't seen your patch yet. What happened?
HK
 

 

> Date : Wed, 28 May 2014 08:06> From : dev
[mailto:dev-bounces@dpdk.org] 

> To :
dev@dpdk.org

> Subject :
[dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines

> for submission

> 

> 

> Hi all,

> 

> We have been working on development of poll-mode
driver for Cisco VIC

> Ethernet NIC and integration of it with DPDK.  We would like to submit this

> poll-mode driver (ENIC PMD) to the DPDK community so
that it could be part

> of the DPDK tree.

> 

> Could someone please provide the guidelines and
steps to do this?  As of

> now, ENIC PMD is being tested with DPDK
1.6.0r2.  Is it alright to submit a

> patch for DPDK 1.6.0r2?

> 

> One aspect of ENIC PMD is that it works with
VFIO-PCI and not UIO.  Hope

> this is acceptable. 
The following thread in dpdk-dev influenced this

> decision.

> http://dpdk.org/ml/archives/dev/2013-July/000373.html

> 

> ENIC PMD uses one interrupt per interface and it is
used by the NIC for

> signalling the driver in case of any error.  Since this does not come in

> the fast path, it should be acceptable, isn¹t it?

> 

> Please give your suggestions and comments.

> 

> Thanks,

> -Sujith
 		 	   		  

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Sujith Sankar (ssujith)]

Hi Hobywan,
Thanks for the email !

We’ve been working on performance benchmarking.  Also, we felt that it would be better to push the driver after Anatoly’s patch (vfio-pci) got in so that we could make the necessary modification before submission.
Now that 1.7.0 is out, we are hopeful of submitting the patch soon.

Thanks,
-Sujith

From: Hobywan Kenoby <hobywank@hotmail.com<mailto:hobywank@hotmail.com>>
Date: Thursday, 10 July 2014 1:37 am
To: "Sujith Sankar (ssujith)" <ssujith@cisco.com<mailto:ssujith@cisco.com>>, "dev@dpdk.org<mailto:dev@dpdk.org>" <dev@dpdk.org<mailto:dev@dpdk.org>>
Subject: RE: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

Hi Sujith,

It was exciting to see open source code coming from Cisco ensuring a DPDK application can run on any platform and with any card....
I haven't seen your patch yet. What happened?

HK





> Date : Wed, 28 May 2014 08:06

> From : dev [mailto:dev-bounces@dpdk.org]

> To : dev@dpdk.org<mailto:dev@dpdk.org>

> Subject : [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines

> for submission

>

>

> Hi all,

>

> We have been working on development of poll-mode driver for Cisco VIC

> Ethernet NIC and integration of it with DPDK.  We would like to submit this

> poll-mode driver (ENIC PMD) to the DPDK community so that it could be part

> of the DPDK tree.

>

> Could someone please provide the guidelines and steps to do this?  As of

> now, ENIC PMD is being tested with DPDK 1.6.0r2.  Is it alright to submit a

> patch for DPDK 1.6.0r2?

>

> One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope

> this is acceptable.  The following thread in dpdk-dev influenced this

> decision.

> http://dpdk.org/ml/archives/dev/2013-July/000373.html

>

> ENIC PMD uses one interrupt per interface and it is used by the NIC for

> signalling the driver in case of any error.  Since this does not come in

> the fast path, it should be acceptable, isn¹t it?

>

> Please give your suggestions and comments.

>

> Thanks,

> -Sujith

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Hobywan Kenoby]

Hi Sujith,

It makes sens, using VFIO makes a far cleaner implementation. 
You worked on the performance, could you share some measurements ? Did you introduce vectorization functions as Intel did a while ago? There are allways tradeoffs between pps and latency, do you include documentation to configure the card for one or the other?
HK
From: ssujith@cisco.com
To: hobywank@hotmail.com; dev@dpdk.org
Subject: Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission
Date: Fri, 11 Jul 2014 07:46:08 +0000






Hi Hobywan,
Thanks for the email !



We’ve been working on performance benchmarking.  Also, we felt that it would be better to push the driver after Anatoly’s patch (vfio-pci) got in so that we could make the necessary modification before submission.  
Now that 1.7.0 is out, we are hopeful of submitting the patch soon.



Thanks,
-Sujith





From: Hobywan Kenoby <hobywank@hotmail.com>

Date: Thursday, 10 July 2014 1:37 am

To: "Sujith Sankar (ssujith)" <ssujith@cisco.com>, "dev@dpdk.org" <dev@dpdk.org>

Subject: RE: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission







Hi Sujith,



It was exciting to see open source code coming from Cisco ensuring a DPDK application can run on any platform and with any card....
I haven't seen your patch yet. What happened?



HK

 
 
> Date : Wed, 28 May 2014 08:06
> From : dev [mailto:dev-bounces@dpdk.org] 
> To :
dev@dpdk.org
> Subject : [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines
> for submission
> 
> 
> Hi all,
> 
> We have been working on development of poll-mode driver for Cisco VIC
> Ethernet NIC and integration of it with DPDK.  We would like to submit this
> poll-mode driver (ENIC PMD) to the DPDK community so that it could be part
> of the DPDK tree.
> 
> Could someone please provide the guidelines and steps to do this?  As of
> now, ENIC PMD is being tested with DPDK 1.6.0r2.  Is it alright to submit a
> patch for DPDK 1.6.0r2?
> 
> One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope
> this is acceptable.  The following thread in dpdk-dev influenced this
> decision.
> 
http://dpdk.org/ml/archives/dev/2013-July/000373.html
> 
> ENIC PMD uses one interrupt per interface and it is used by the NIC for
> signalling the driver in case of any error.  Since this does not come in
> the fast path, it should be acceptable, isn¹t it?
> 
> Please give your suggestions and comments.
> 
> Thanks,
> -Sujith







 		 	   		  

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Sujith Sankar (ssujith)]

Hi Hobywan,

We’re still working on benchmarking, and would share the numbers once we are done with it.
Could you please elaborate on vectorisation functions?

Thanks,
-Sujith

From: Hobywan Kenoby <hobywank@hotmail.com<mailto:hobywank@hotmail.com>>
Date: Friday, 11 July 2014 6:46 pm
To: "Sujith Sankar (ssujith)" <ssujith@cisco.com<mailto:ssujith@cisco.com>>, "dev@dpdk.org<mailto:dev@dpdk.org>" <dev@dpdk.org<mailto:dev@dpdk.org>>
Subject: RE: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

Hi Sujith,

It makes sens, using VFIO makes a far cleaner implementation.

You worked on the performance, could you share some measurements ? Did you introduce vectorization functions as Intel did a while ago? There are allways tradeoffs between pps and latency, do you include documentation to configure the card for one or the other?

HK
________________________________
From: ssujith@cisco.com<mailto:ssujith@cisco.com>
To: hobywank@hotmail.com<mailto:hobywank@hotmail.com>; dev@dpdk.org<mailto:dev@dpdk.org>
Subject: Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission
Date: Fri, 11 Jul 2014 07:46:08 +0000

Hi Hobywan,
Thanks for the email !

We’ve been working on performance benchmarking.  Also, we felt that it would be better to push the driver after Anatoly’s patch (vfio-pci) got in so that we could make the necessary modification before submission.
Now that 1.7.0 is out, we are hopeful of submitting the patch soon.

Thanks,
-Sujith

From: Hobywan Kenoby <hobywank@hotmail.com<mailto:hobywank@hotmail.com>>
Date: Thursday, 10 July 2014 1:37 am
To: "Sujith Sankar (ssujith)" <ssujith@cisco.com<mailto:ssujith@cisco.com>>, "dev@dpdk.org<mailto:dev@dpdk.org>" <dev@dpdk.org<mailto:dev@dpdk.org>>
Subject: RE: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

Hi Sujith,

It was exciting to see open source code coming from Cisco ensuring a DPDK application can run on any platform and with any card....
I haven't seen your patch yet. What happened?

HK





> Date : Wed, 28 May 2014 08:06

> From : dev [mailto:dev-bounces@dpdk.org]

> To : dev@dpdk.org<mailto:dev@dpdk.org>

> Subject : [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines

> for submission

>

>

> Hi all,

>

> We have been working on development of poll-mode driver for Cisco VIC

> Ethernet NIC and integration of it with DPDK.  We would like to submit this

> poll-mode driver (ENIC PMD) to the DPDK community so that it could be part

> of the DPDK tree.

>

> Could someone please provide the guidelines and steps to do this?  As of

> now, ENIC PMD is being tested with DPDK 1.6.0r2.  Is it alright to submit a

> patch for DPDK 1.6.0r2?

>

> One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope

> this is acceptable.  The following thread in dpdk-dev influenced this

> decision.

> http://dpdk.org/ml/archives/dev/2013-July/000373.html

>

> ENIC PMD uses one interrupt per interface and it is used by the NIC for

> signalling the driver in case of any error.  Since this does not come in

> the fast path, it should be acceptable, isn¹t it?

>

> Please give your suggestions and comments.

>

> Thanks,

> -Sujith

Re: [dpdk-dev] [PATCH 0/2] Fixed/Simplified Ring PMD

[Thomas Monjalon]

> > Ring PMD library and unit test was modified last release,
> > causing the unit test to need ring ethdevs to be created
> > from the command line. This patch simplifies the test
> > and make it functional with devices created with EAL option vdev.
> > 
> > Also, ring ethdev creation from vdev has been changed, so now
> > only one device is created for each vdev argument (up to 3 devices
> > were created). 
> > 
> > 
> > Pablo de Lara (2):
> >   ring_pmd: Removed extra devices creation when passing vdev parameter
> >   app/test: Simplified ring pmd test
> 
> Series
> Acked-by: Neil Horman <nhorman@tuxdriver.com>

Applied for version 1.7.1.

Thanks
-- 
Thomas

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Sujith Sankar (ssujith)]

Thank you Anatoly !
I¹ll take a look at it and get back in case of questions.

-Sujith

On 28/05/14 2:22 pm, "Burakov, Anatoly" <anatoly.burakov@intel.com> wrote:

>Hi Sujith,
>
>> One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope
>> this is acceptable.  The following thread in dpdk-dev influenced this
>>decision.
>> http://dpdk.org/ml/archives/dev/2013-July/000373.html
>
>You may find this [1] patchset interesting. Basically, it adds native
>support for VFIO to do the binding and unbinding in DPDK. Whether it fits
>your purposes depends on your code, but hopefully you'll be able to
>simplify your code.
>
>[1] http://dpdk.org/ml/archives/dev/2014-May/002568.html
>
>Best regards,
>Anatoly Burakov
>DPDK SW Engineer
>
>
>

Re: [dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Burakov, Anatoly]

Hi Sujith,

> One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope
> this is acceptable.  The following thread in dpdk-dev influenced this decision.
> http://dpdk.org/ml/archives/dev/2013-July/000373.html

You may find this [1] patchset interesting. Basically, it adds native support for VFIO to do the binding and unbinding in DPDK. Whether it fits your purposes depends on your code, but hopefully you'll be able to simplify your code.

[1] http://dpdk.org/ml/archives/dev/2014-May/002568.html

Best regards,
Anatoly Burakov
DPDK SW Engineer

[dpdk-dev] PMD for Cisco VIC Ethernet NIC - Request for guidelines for submission

[Sujith Sankar (ssujith)]

Hi all,

We have been working on development of poll-mode driver for Cisco VIC
Ethernet NIC and integration of it with DPDK.  We would like to submit
this poll-mode driver (ENIC PMD) to the DPDK community so that it could be
part of the DPDK tree.

Could someone please provide the guidelines and steps to do this?  As of
now, ENIC PMD is being tested with DPDK 1.6.0r2.  Is it alright to submit
a patch for DPDK 1.6.0r2?

One aspect of ENIC PMD is that it works with VFIO-PCI and not UIO.  Hope
this is acceptable.  The following thread in dpdk-dev influenced this
decision.
http://dpdk.org/ml/archives/dev/2013-July/000373.html

ENIC PMD uses one interrupt per interface and it is used by the NIC for
signalling the driver in case of any error.  Since this does not come in
the fast path, it should be acceptable, isn¹t it?

Please give your suggestions and comments.

Thanks,
-Sujith