-----Original Message-----
From: Andrew Rybchenko <redacted>
Sent: Tuesday, January 19, 2021 4:03 PM
To: Xueming(Steven) Li <redacted>
Cc: dev@dpdk.org; Slava Ovsiienko <redacted>; Asaf Penso
[off-list ref]; Bruce Richardson [off-list ref]; NBU-
Contact-Thomas Monjalon [off-list ref]; Ferruh Yigit
[off-list ref]
Subject: Re: [PATCH v5 6/9] ethdev: support multi-host in representor

On 1/19/21 10:14 AM, Xueming Li wrote:

quoted

The NIC can have multiple PCIe links and can be attached to the
multiple hosts, for example the same single NIC can be shared for
multiple server units in the rack. On each PCIe link NIC can provide
multiple PFs and VFs/SFs based on these ones. To provide the
unambiguous identification of the PCIe function the controller index
is added. The full representor identifier consists of three indices -
controller index, PF index, and VF or SF index (if any).

This patch introduces controller index to ethdev representor syntax,
examples:

[[c#]pf#]vf#: VF port representor/s, example: pf0vf1
[[c#]pf#]sf#: SF port representor/s, example: c1pf1sf[0-3]

c# is controller(host) ID/range in case of multi-host, optional.

For user application (e.g. OVS), PMD is responsible to interpret and
locate representor device based on controller ID, PF ID and VF/SF ID
in representor syntax.

Signed-off-by: Xueming Li <redacted>
Acked-by: Viacheslav Ovsiienko <redacted>
Acked-by: Thomas Monjalon <redacted>
---
 config/rte_config.h                   |  1 +
 lib/librte_ethdev/ethdev_private.c    | 12 ++++++++++--
 lib/librte_ethdev/rte_ethdev_driver.h |  4 ++++
 3 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/config/rte_config.h b/config/rte_config.h index

a0b5160ff2..23d02d51ef 100644

--- a/config/rte_config.h
+++ b/config/rte_config.h

@@ -58,6 +58,7 @@
 #define RTE_MAX_QUEUES_PER_PORT 1024
 #define RTE_ETHDEV_QUEUE_STAT_CNTRS 16 /* max 256 */  #define

RTE_ETHDEV_RXTX_CALLBACKS 1
+#define RTE_MAX_MULTI_HOST_CTRLS 4

 /* cryptodev defines */
 #define RTE_CRYPTO_MAX_DEVS 64

diff --git a/lib/librte_ethdev/ethdev_private.c

b/lib/librte_ethdev/ethdev_private.c
index b9fdbd0f72..9a0945c22e 100644

--- a/lib/librte_ethdev/ethdev_private.c
+++ b/lib/librte_ethdev/ethdev_private.c

@@ -120,8 +120,8 @@ rte_eth_devargs_process_list(char *str, uint16_t

*list, uint16_t *len_list,

quoted

  *
  * Representor format:
  *   #: range or single number of VF representor - legacy
- *   [pf#]vf#: VF port representor/s
- *   [pf#]sf#: SF port representor/s
+ *   [[c#]pf#]vf#: VF port representor/s
+ *   [[c#]pf#]sf#: SF port representor/s
  *
  * Examples of #:
  *  2               - single

@@ -133,6 +133,14 @@ rte_eth_devargs_parse_representor_ports(char

*str, void *data)  {
 	struct rte_eth_devargs *eth_da = data;

+	if (str[0] == 'c') {
+		str += 1;
+		str = rte_eth_devargs_process_list(str, eth_da-
mh_controllers,
+				&eth_da->nb_mh_controllers,
+				RTE_DIM(eth_da->mh_controllers));
+		if (str == NULL)
+			goto err;

The parser must enforce pf to follow. I.e. should not allow c1vf3. At least
above syntax description in comments says so.

Good catch, will add, thanks.

quoted

+	}
 	if (str[0] == 'p' && str[1] == 'f') {
 		eth_da->type = RTE_ETH_REPRESENTOR_PF;
 		str += 2;

diff --git a/lib/librte_ethdev/rte_ethdev_driver.h

b/lib/librte_ethdev/rte_ethdev_driver.h
index cd33184f17..b01f118965 100644

--- a/lib/librte_ethdev/rte_ethdev_driver.h
+++ b/lib/librte_ethdev/rte_ethdev_driver.h

@@ -1203,6 +1203,10 @@ enum rte_eth_representor_type {

 /** Generic Ethernet device arguments  */  struct rte_eth_devargs {
+	uint16_t mh_controllers[RTE_MAX_MULTI_HOST_CTRLS];
+	/** controller/s number in case of multi-host */
+	uint16_t nb_mh_controllers;
+	/** number of controllers in multi-host controllers field */
 	uint16_t ports[RTE_MAX_ETHPORTS];
 	/** port/s number to enable on a multi-port single function */
 	uint16_t nb_ports;