interface: support configuring RSS steering queues

This patch adds the RSS steering queues set interface, and it's
implementation in DPDK device:

/* Interface to set rss queues of the interface */
typedef clib_error_t *(vnet_interface_rss_queues_set_t)
  (struct vnet_main_t * vnm, struct vnet_hw_interface_t * hi,
   clib_bitmap_t *bitmap);

This patch also introduces a command line to set the RSS queues:
  set interface rss queues <interface> <list <queue-list>>
To display the rss queues, use "show hardware-interfaces"

Below is the example to configure rss queues for interface Gig0:
vpp# set interface rss queues Gig0 list 0,2,4-7
vpp# show hardware-interfaces brief
              Name                Idx   Link  Hardware
VirtualFunctionEthernet18/1/0      1    down  VirtualFunctionEthernet18/1/0
  Link speed: unknown
  RSS queues: 0 2 4 5 6 7
local0                             0    down  local0
  Link speed: unknown
vpp#

Users can also configure the rss queues on a dpdk interface in
startup.conf:
dpdk {
    dev 0000:18:01.0 {
        rss-queues 0,2,5-7
    }
}

Type: feature

Signed-off-by: Chenmin Sun <chenmin.sun@intel.com>
Change-Id: I1835595a1c54016a84eabee9fd62ce137935385d

3 files changed, 154 insertions, 34 deletions

diff --git a/src/plugins/dpdk/device/device.c b/src/plugins/dpdk/device/device.c
index 09a1c2f2fdf..e58dd6f4228 100644
--- a/src/plugins/dpdk/device/device.c
+++ b/src/plugins/dpdk/device/device.c
@@ -584,6 +584,109 @@ done:
   return err;
 }
 
+static clib_error_t *
+dpdk_interface_set_rss_queues (struct vnet_main_t *vnm,
+			       struct vnet_hw_interface_t *hi,
+			       clib_bitmap_t * bitmap)
+{
+  dpdk_main_t *xm = &dpdk_main;
+  u32 hw_if_index = hi->hw_if_index;
+  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
+  dpdk_device_t *xd = vec_elt_at_index (xm->devices, hw->dev_instance);
+  clib_error_t *err = 0;
+  struct rte_eth_rss_reta_entry64 *reta_conf = NULL;
+  struct rte_eth_dev_info dev_info;
+  u16 *reta = NULL;
+  u16 *valid_queue = NULL;
+  u16 valid_queue_count = 0;
+  uint32_t i, j;
+  uint32_t ret;
+
+  rte_eth_dev_info_get (xd->port_id, &dev_info);
+
+  /* parameter check */
+  if (clib_bitmap_count_set_bits (bitmap) == 0)
+    {
+      err = clib_error_return (0, "must assign at least one valid rss queue");
+      goto done;
+    }
+
+  if (clib_bitmap_count_set_bits (bitmap) > dev_info.nb_rx_queues)
+    {
+      err = clib_error_return (0, "too many rss queues");
+      goto done;
+    }
+
+  /* new RETA */
+  reta = clib_mem_alloc (dev_info.reta_size * sizeof (*reta));
+  if (reta == NULL)
+    {
+      err = clib_error_return (0, "clib_mem_alloc failed");
+      goto done;
+    }
+
+  clib_memset (reta, 0, dev_info.reta_size * sizeof (*reta));
+
+  valid_queue_count = 0;
+  /* *INDENT-OFF* */
+  clib_bitmap_foreach (i, bitmap, ({
+    if (i >= dev_info.nb_rx_queues)
+      {
+        err = clib_error_return (0, "illegal queue number");
+        goto done;
+      }
+    reta[valid_queue_count++] = i;
+  }));
+  /* *INDENT-ON* */
+
+  valid_queue = reta;
+  for (i = valid_queue_count, j = 0; i < dev_info.reta_size; i++, j++)
+    {
+      j = j % valid_queue_count;
+      reta[i] = valid_queue[j];
+    }
+
+  /* update reta table */
+  reta_conf =
+    (struct rte_eth_rss_reta_entry64 *) clib_mem_alloc (dev_info.reta_size /
+							RTE_RETA_GROUP_SIZE *
+							sizeof (*reta_conf));
+  if (reta_conf == NULL)
+    {
+      err = clib_error_return (0, "clib_mem_alloc failed");
+      goto done;
+    }
+
+  clib_memset (reta_conf, 0,
+	       dev_info.reta_size / RTE_RETA_GROUP_SIZE *
+	       sizeof (*reta_conf));
+
+  for (i = 0; i < dev_info.reta_size; i++)
+    {
+      uint32_t reta_id = i / RTE_RETA_GROUP_SIZE;
+      uint32_t reta_pos = i % RTE_RETA_GROUP_SIZE;
+
+      reta_conf[reta_id].mask = UINT64_MAX;
+      reta_conf[reta_id].reta[reta_pos] = reta[i];
+    }
+
+  ret =
+    rte_eth_dev_rss_reta_update (xd->port_id, reta_conf, dev_info.reta_size);
+  if (ret)
+    {
+      err = clib_error_return (0, "rte_eth_dev_rss_reta_update err %d", ret);
+      goto done;
+    }
+
+done:
+  if (reta)
+    clib_mem_free (reta);
+  if (reta_conf)
+    clib_mem_free (reta_conf);
+
+  return err;
+}
+
 /* *INDENT-OFF* */
 VNET_DEVICE_CLASS (dpdk_device_class) = {
   .name = "dpdk",
@@ -600,6 +703,7 @@ VNET_DEVICE_CLASS (dpdk_device_class) = {
   .mac_addr_add_del_function = dpdk_add_del_mac_address,
   .format_flow = format_dpdk_flow,
   .flow_ops_function = dpdk_flow_ops_fn,
+  .set_rss_queues_function = dpdk_interface_set_rss_queues,
 };
 /* *INDENT-ON* */
 
diff --git a/src/plugins/dpdk/device/dpdk.h b/src/plugins/dpdk/device/dpdk.h
index 963bc811ee1..40f9dce0c05 100644
--- a/src/plugins/dpdk/device/dpdk.h
+++ b/src/plugins/dpdk/device/dpdk.h
@@ -260,6 +260,7 @@ typedef struct
     clib_bitmap_t * workers;
   u8 tso;
   u8 *devargs;
+  clib_bitmap_t *rss_queues;
 
 #define DPDK_DEVICE_TSO_DEFAULT 0
 #define DPDK_DEVICE_TSO_OFF 1
@@ -373,7 +374,6 @@ typedef struct
 void dpdk_device_setup (dpdk_device_t * xd);
 void dpdk_device_start (dpdk_device_t * xd);
 void dpdk_device_stop (dpdk_device_t * xd);
-
 int dpdk_port_state_callback (dpdk_portid_t port_id,
 			      enum rte_eth_event_type type,
 			      void *param, void *ret_param);
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index 9d4aeed6924..81803ab1cb4 100644
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -616,7 +616,6 @@ dpdk_lib_init (dpdk_main_t * dm)
       /* assign interface to input thread */
       int q;
 
-
       error = ethernet_register_interface
 	(dm->vnet_main, dpdk_device_class.index, xd->device_index,
 	 /* ethernet address */ addr,
@@ -759,10 +758,20 @@ dpdk_lib_init (dpdk_main_t * dm)
 
       dpdk_device_setup (xd);
 
+      /* rss queues should be configured after dpdk_device_setup() */
+      if (devconf->rss_queues != NULL)
+        {
+          if (vnet_hw_interface_set_rss_queues
+              (vnet_get_main (), hi, devconf->rss_queues))
+          {
+            clib_warning ("%s: Failed to set rss queues", hi->name);
+          }
+        }
+
       if (vec_len (xd->errors))
-	dpdk_log_err ("setup failed for device %U. Errors:\n  %U",
-		      format_dpdk_device_name, i,
-		      format_dpdk_device_errors, xd);
+        dpdk_log_err ("setup failed for device %U. Errors:\n  %U",
+                      format_dpdk_device_name, i,
+                      format_dpdk_device_errors, xd);
 
       /*
        * A note on Cisco VIC (PMD_ENIC) and VLAN:
@@ -789,38 +798,39 @@ dpdk_lib_init (dpdk_main_t * dm)
        * otherwise in the startup config.
        */
 
-      vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id);
-      if (devconf->vlan_strip_offload == DPDK_DEVICE_VLAN_STRIP_ON)
-	{
-	  vlan_off |= ETH_VLAN_STRIP_OFFLOAD;
-	  if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
-	    dpdk_log_info ("VLAN strip enabled for interface\n");
-	  else
-	    dpdk_log_warn ("VLAN strip cannot be supported by interface\n");
-	  xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
-	}
-      else
-	{
-	  if (vlan_off & ETH_VLAN_STRIP_OFFLOAD)
-	    {
-	      vlan_off &= ~ETH_VLAN_STRIP_OFFLOAD;
-	      if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
-		dpdk_log_warn ("set VLAN offload failed\n");
-	    }
-	  xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
-	}
+        vlan_off = rte_eth_dev_get_vlan_offload (xd->port_id);
+        if (devconf->vlan_strip_offload == DPDK_DEVICE_VLAN_STRIP_ON)
+          {
+            vlan_off |= ETH_VLAN_STRIP_OFFLOAD;
+            if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
+              dpdk_log_info ("VLAN strip enabled for interface\n");
+            else
+              dpdk_log_warn ("VLAN strip cannot be supported by interface\n");
+            xd->port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
+          }
+        else
+          {
+            if (vlan_off & ETH_VLAN_STRIP_OFFLOAD)
+              {
+        	vlan_off &= ~ETH_VLAN_STRIP_OFFLOAD;
+        	if (rte_eth_dev_set_vlan_offload (xd->port_id, vlan_off) >= 0)
+        	  dpdk_log_warn ("set VLAN offload failed\n");
+              }
+            xd->port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
+          }
 
-      if (hi)
-	hi->max_packet_bytes = xd->port_conf.rxmode.max_rx_pkt_len
-	  - sizeof (ethernet_header_t);
-      else
-	dpdk_log_warn ("hi NULL");
+        if (hi)
+          hi->max_packet_bytes = xd->port_conf.rxmode.max_rx_pkt_len
+            - sizeof (ethernet_header_t);
+        else
+          dpdk_log_warn ("hi NULL");
 
-      if (dm->conf->no_multi_seg)
-	mtu = mtu > ETHER_MAX_LEN ? ETHER_MAX_LEN : mtu;
+        if (dm->conf->no_multi_seg)
+          mtu = mtu > ETHER_MAX_LEN ? ETHER_MAX_LEN : mtu;
+
+        rte_eth_dev_set_mtu (xd->port_id, mtu);
+}
 
-      rte_eth_dev_set_mtu (xd->port_id, mtu);
-    }
   /* *INDENT-ON* */
 
   return 0;
@@ -1094,6 +1104,9 @@ dpdk_device_config (dpdk_config_main_t * conf, vlib_pci_addr_t pci_addr,
 	}
       else if (unformat (input, "devargs %s", &devconf->devargs))
 	;
+      else if (unformat (input, "rss-queues %U",
+			 unformat_bitmap_list, &devconf->rss_queues))
+	;
       else
 	{
 	  error = clib_error_return (0, "unknown input `%U'",
@@ -1416,6 +1429,9 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 	/* copy tso config from default device */
 	_(devargs)
 
+	/* copy rss_queues config from default device */
+	_(rss_queues)
+
     /* add DPDK EAL whitelist/blacklist entry */
     if (num_whitelisted > 0 && devconf->is_blacklisted == 0)
     {