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

10 files changed, 320 insertions, 36 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)
     {
diff --git a/src/vnet/devices/devices.c b/src/vnet/devices/devices.c
index e78c5cbe45b..cfce2ac2856 100644
--- a/src/vnet/devices/devices.c
+++ b/src/vnet/devices/devices.c
@@ -341,8 +341,6 @@ vnet_hw_interface_get_rx_mode (vnet_main_t * vnm, u32 hw_if_index,
   return VNET_API_ERROR_INVALID_INTERFACE;
 }
 
-
-
 static clib_error_t *
 vnet_device_init (vlib_main_t * vm)
 {
diff --git a/src/vnet/interface.c b/src/vnet/interface.c
index 6d5b3561f19..18c7696e9a4 100644
--- a/src/vnet/interface.c
+++ b/src/vnet/interface.c
@@ -1692,6 +1692,42 @@ default_update_adjacency (vnet_main_t * vnm, u32 sw_if_index, u32 ai)
     }
 }
 
+clib_error_t *
+vnet_hw_interface_set_rss_queues (vnet_main_t * vnm,
+				  vnet_hw_interface_t * hi,
+				  clib_bitmap_t * bitmap)
+{
+  clib_error_t *error = 0;
+  vnet_device_class_t *dev_class =
+    vnet_get_device_class (vnm, hi->dev_class_index);
+
+  if (dev_class->set_rss_queues_function)
+    {
+      if (clib_bitmap_count_set_bits (bitmap) == 0)
+	{
+	  error = clib_error_return (0,
+				     "must assign at least one valid rss queue");
+	  goto done;
+	}
+
+      error = dev_class->set_rss_queues_function (vnm, hi, bitmap);
+    }
+  else
+    {
+      error = clib_error_return (0,
+				 "setting rss queues is not supported on this interface");
+    }
+
+  if (!error)
+    {
+      clib_bitmap_free (hi->rss_queues);
+      hi->rss_queues = clib_bitmap_dup (bitmap);
+    }
+
+done:
+  return error;
+}
+
 int collect_detailed_interface_stats_flag = 0;
 
 void
diff --git a/src/vnet/interface.h b/src/vnet/interface.h
index 07da2617d71..895ce2e9279 100644
--- a/src/vnet/interface.h
+++ b/src/vnet/interface.h
@@ -88,6 +88,11 @@ typedef clib_error_t *(vnet_interface_set_l2_mode_function_t)
   (struct vnet_main_t * vnm, struct vnet_hw_interface_t * hi,
    i32 l2_if_adjust);
 
+/* 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);
+
 typedef enum
 {
   VNET_FLOW_DEV_OP_ADD_FLOW,
@@ -273,6 +278,10 @@ typedef struct _vnet_device_class
 
   /* Function to add/delete additional MAC addresses */
   vnet_interface_add_del_mac_address_function_t *mac_addr_add_del_function;
+
+  /* Interface to set rss queues of the interface */
+  vnet_interface_rss_queues_set_t *set_rss_queues_function;
+
 } vnet_device_class_t;
 
 #ifndef CLIB_MARCH_VARIANT
@@ -601,6 +610,9 @@ typedef struct vnet_hw_interface_t
   /* numa node that hardware device connects to */
   u8 numa_node;
 
+  /* rss queues bitmap */
+  clib_bitmap_t *rss_queues;
+
   /* trace */
   i32 n_trace;
 
diff --git a/src/vnet/interface_cli.c b/src/vnet/interface_cli.c
index bee813f5874..28f24ae3977 100644
--- a/src/vnet/interface_cli.c
+++ b/src/vnet/interface_cli.c
@@ -1886,6 +1886,89 @@ VLIB_CLI_COMMAND (cmd_set_if_rx_placement,static) = {
 };
 /* *INDENT-ON* */
 
+clib_error_t *
+set_interface_rss_queues (vlib_main_t * vm, u32 hw_if_index,
+			  clib_bitmap_t * bitmap)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
+
+  return vnet_hw_interface_set_rss_queues (vnm, hi, bitmap);
+}
+
+static clib_error_t *
+set_interface_rss_queues_fn (vlib_main_t * vm,
+			     unformat_input_t * input,
+			     vlib_cli_command_t * cmd)
+{
+  clib_error_t *error = 0;
+  unformat_input_t _line_input, *line_input = &_line_input;
+  vnet_main_t *vnm = vnet_get_main ();
+  u32 hw_if_index = (u32) ~ 0;
+  clib_bitmap_t *bitmap = NULL;
+
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat
+	  (line_input, "%U", unformat_vnet_hw_interface, vnm, &hw_if_index))
+	;
+      else
+	if (unformat (line_input, "list %U", unformat_bitmap_list, &bitmap))
+	;
+      else
+	{
+	  error = clib_error_return (0, "parse error: '%U'",
+				     format_unformat_error, line_input);
+	  unformat_free (line_input);
+	  goto done;
+	}
+    }
+
+  unformat_free (line_input);
+
+  if (hw_if_index == (u32) ~ 0)
+    {
+      error = clib_error_return (0, "please specify valid interface name");
+      goto done;
+    }
+
+  if (bitmap == NULL)
+    {
+      error = clib_error_return (0, "please specify the valid rss queues");
+      goto done;
+    }
+
+  error = set_interface_rss_queues (vm, hw_if_index, bitmap);
+
+done:
+  if (bitmap)
+    clib_bitmap_free (bitmap);
+
+  return (error);
+}
+
+/*?
+ * This command is used to set the rss queues of a given interface
+ * Not all the interfaces support this operation.
+ * To display the current rss queues, use the command
+ * '<em>show hardware-interfaces</em>'.
+ *
+ * @cliexpar
+ * Example of how to set the rss queues to 0,2-5,7 of an interface:
+ * @cliexstart{set interface rss queues VirtualFunctionEthernet18/1/0 list 0,2-5,7}
+ * @cliexend
+?*/
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (cmd_set_interface_rss_queues,static) = {
+    .path = "set interface rss queues",
+    .short_help = "set interface rss queues <interface> <list <queue-list>>",
+    .function = set_interface_rss_queues_fn,
+};
+/* *INDENT-ON* */
+
 static u8 *
 format_vnet_pcap (u8 * s, va_list * args)
 {
diff --git a/src/vnet/interface_format.c b/src/vnet/interface_format.c
index 8278102c644..2b691a66071 100644
--- a/src/vnet/interface_format.c
+++ b/src/vnet/interface_format.c
@@ -130,6 +130,26 @@ format_vnet_hw_interface_link_speed (u8 * s, va_list * args)
   return format (s, "%u Kbps", link_speed);
 }
 
+u8 *
+format_vnet_hw_interface_rss_queues (u8 * s, va_list * args)
+{
+  clib_bitmap_t *bitmap = va_arg (*args, clib_bitmap_t *);
+  int i;
+
+  if (bitmap == NULL)
+    return s;
+
+  if (bitmap)
+    {
+    /* *INDENT-OFF* */
+    clib_bitmap_foreach (i, bitmap, ({
+      s = format (s, "%u ", i);
+    }));
+    /* *INDENT-ON* */
+    }
+
+  return s;
+}
 
 u8 *
 format_vnet_hw_interface (u8 * s, va_list * args)
@@ -172,6 +192,12 @@ format_vnet_hw_interface (u8 * s, va_list * args)
   s = format (s, "\n%ULink speed: %U", format_white_space, indent + 2,
 	      format_vnet_hw_interface_link_speed, hi->link_speed);
 
+  if (hi->rss_queues)
+    {
+      s = format (s, "\n%URSS queues: %U", format_white_space, indent + 2,
+		  format_vnet_hw_interface_rss_queues, hi->rss_queues);
+    }
+
   if (verbose)
     {
       if (hw_class->format_device)
diff --git a/src/vnet/interface_funcs.h b/src/vnet/interface_funcs.h
index 7d9c098b118..8f589d1d6dd 100644
--- a/src/vnet/interface_funcs.h
+++ b/src/vnet/interface_funcs.h
@@ -431,6 +431,11 @@ int vnet_sw_interface_stats_collect_enable_disable (u32 sw_if_index,
 void vnet_sw_interface_ip_directed_broadcast (vnet_main_t * vnm,
 					      u32 sw_if_index, u8 enable);
 
+/* set interface rss queues */
+clib_error_t *vnet_hw_interface_set_rss_queues (vnet_main_t * vnm,
+						vnet_hw_interface_t * hi,
+						clib_bitmap_t * bitmap);
+
 /* Formats sw/hw interface. */
 format_function_t format_vnet_hw_interface;
 format_function_t format_vnet_hw_interface_rx_mode;
diff --git a/src/vpp/conf/startup.conf b/src/vpp/conf/startup.conf
index 86b73f7a77b..56610e2e315 100644
--- a/src/vpp/conf/startup.conf
+++ b/src/vpp/conf/startup.conf
@@ -110,6 +110,10 @@ cpu {
                 ## device specific init args
                 ## Default is NULL
 		# devargs safe-mode-support=1,pipeline-mode-support=1
+
+		## rss-queues
+		## set valid rss steering queues
+		# rss-queues 0,2,5-7
 	# }
 
 	## Whitelist specific interface by specifying PCI address