dpdk: allow configure individual VMBUS devices

now startup.conf supports confuguration for VMBUS
devices as for PCI devices for whitelisting/blacklisting

dpdk { dev fa5a6e7a-cf3a-4b98-9569-addb479b84bc }
 with sub-configuration as for PCI devices
dpdk { blacklist fa5a6e7a-cf3a-4b98-9569-addb479b84bc }
 where fa5a6e7a-cf3a-4b98-9569-addb479b84bc - example of UUID

struct vlib_vmbus_addr_t changed to union with UUID described
 fields

Added device_config_index_by_vmbus_addr
      blacklist_by_vmbus_addr
to enumerate available device configs

hash_key is as_u32[0] field(last 4 bytes of UUID)
Lost of precision against full UUID, but 2^32 is enough
to handle all the devices available

Added is_blacklisted check while creating vnet devices in
order to supress creation of dev if it's blacklisted

Type: feature

Signed-off-by: Vladimir Ratnikov <vratnikov@netgate.com>
Change-Id: Id82611e54fed082190e488c7e5fbe14ecbe5b2ab

3 files changed, 220 insertions, 48 deletions

diff --git a/src/plugins/dpdk/device/common.c b/src/plugins/dpdk/device/common.c
index 0c43bfe02bb..2521abde97a 100644
--- a/src/plugins/dpdk/device/common.c
+++ b/src/plugins/dpdk/device/common.c
@@ -260,6 +260,19 @@ dpdk_get_pci_device (const struct rte_eth_dev_info *info)
     return NULL;
 }
 
+/* If this device is VMBUS return pointer to info, otherwise NULL */
+struct rte_vmbus_device *
+dpdk_get_vmbus_device (const struct rte_eth_dev_info *info)
+{
+  const struct rte_bus *bus;
+
+  bus = rte_bus_find_by_device (info->device);
+  if (bus && !strcmp (bus->name, "vmbus"))
+    return container_of (info->device, struct rte_vmbus_device, device);
+  else
+    return NULL;
+}
+
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/src/plugins/dpdk/device/dpdk.h b/src/plugins/dpdk/device/dpdk.h
index c81f23d769c..504bac5b6ee 100644
--- a/src/plugins/dpdk/device/dpdk.h
+++ b/src/plugins/dpdk/device/dpdk.h
@@ -32,6 +32,7 @@
 #include <rte_per_lcore.h>
 #include <rte_interrupts.h>
 #include <rte_pci.h>
+#include <rte_bus_vmbus.h>
 #include <rte_ether.h>
 #include <rte_ethdev.h>
 #include <rte_ring.h>
@@ -52,6 +53,7 @@
 #endif
 
 #include <vlib/pci/pci.h>
+#include <vlib/vmbus/vmbus.h>
 #include <vnet/flow/flow.h>
 
 extern vnet_device_class_t dpdk_device_class;
@@ -250,9 +252,21 @@ typedef struct
   _ (num_tx_desc) \
   _ (rss_fn)
 
+typedef enum
+{
+  VNET_DEV_ADDR_PCI,
+  VNET_DEV_ADDR_VMBUS,
+  VNET_DEV_ADDR_ANY,
+} dpdk_device_addr_type_t;
+
 typedef struct
 {
-  vlib_pci_addr_t pci_addr;
+  union
+  {
+    vlib_pci_addr_t pci_addr;
+    vlib_vmbus_addr_t vmbus_addr;
+  };
+  dpdk_device_addr_type_t dev_addr_type;
   u8 *name;
   u8 is_blacklisted;
   u8 vlan_strip_offload;
@@ -302,9 +316,12 @@ typedef struct
   dpdk_device_config_t default_devconf;
   dpdk_device_config_t *dev_confs;
   uword *device_config_index_by_pci_addr;
+  uword *device_config_index_by_vmbus_addr;
 
   /* devices blacklist by pci vendor_id, device_id */
   u32 *blacklist_by_pci_vendor_and_device;
+  /* devices blacklist by VMBUS address */
+  u32 *blacklist_by_vmbus_addr;
 
 } dpdk_config_main_t;
 
@@ -459,6 +476,8 @@ clib_error_t *unformat_rss_fn (unformat_input_t * input, uword * rss_fn);
 
 struct rte_pci_device *dpdk_get_pci_device (const struct rte_eth_dev_info
 					    *info);
+struct rte_vmbus_device *
+dpdk_get_vmbus_device (const struct rte_eth_dev_info *info);
 void dpdk_cli_reference (void);
 
 #if CLI_DEBUG
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index 19898b186fe..f3d228c5820 100644
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -249,9 +249,11 @@ dpdk_lib_init (dpdk_main_t * dm)
       int vlan_off;
       struct rte_eth_dev_info dev_info;
       struct rte_pci_device *pci_dev;
+      struct rte_vmbus_device *vmbus_dev;
       dpdk_portid_t next_port_id;
       dpdk_device_config_t *devconf = 0;
       vlib_pci_addr_t pci_addr;
+      vlib_vmbus_addr_t vmbus_addr;
       uword *p = 0;
 
       if (!rte_eth_dev_is_valid_port(i))
@@ -278,20 +280,42 @@ dpdk_lib_init (dpdk_main_t * dm)
 			pci_addr.as_u32);
 	}
 
+      vmbus_dev = dpdk_get_vmbus_device (&dev_info);
+
+      if (vmbus_dev)
+	{
+	  unformat_input_t input_vmbus;
+
+	  unformat_init_vector (&input_vmbus, (u8 *) dev_info.device->name);
+	  if (unformat (&input_vmbus, "%U", unformat_vlib_vmbus_addr,
+			&vmbus_addr))
+	    {
+	      p = hash_get (dm->conf->device_config_index_by_vmbus_addr,
+			    vmbus_addr.as_u32[0]);
+	    }
+	}
+
+      if (p)
+	{
+	  devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]);
+	  /* If device is blacklisted, we should skip it */
+	  if (devconf->is_blacklisted)
+	    {
+	      continue;
+	    }
+	}
+      else
+	devconf = &dm->conf->default_devconf;
 
       /* Create vnet interface */
       vec_add2_aligned (dm->devices, xd, 1, CLIB_CACHE_LINE_BYTES);
       xd->nb_rx_desc = DPDK_NB_RX_DESC_DEFAULT;
       xd->nb_tx_desc = DPDK_NB_TX_DESC_DEFAULT;
       xd->cpu_socket = (i8) rte_eth_dev_socket_id (i);
-
       if (p)
 	{
-	  devconf = pool_elt_at_index (dm->conf->dev_confs, p[0]);
 	  xd->name = devconf->name;
 	}
-      else
-	devconf = &dm->conf->default_devconf;
 
       /* Handle representor devices that share the same PCI ID */
       if (dev_info.switch_info.domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID)
@@ -876,9 +900,9 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf)
 
 	if (!p)
           {
-          skipped:
-            continue;
-          }
+	  skipped_pci:
+	    continue;
+	  }
 
 	devconf = pool_elt_at_index (conf->dev_confs, p[0]);
       }
@@ -902,12 +926,13 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf)
                 hash_set (conf->device_config_index_by_pci_addr, addr->as_u32,
                           devconf - conf->dev_confs);
                 devconf->pci_addr.as_u32 = addr->as_u32;
-                devconf->is_blacklisted = 1;
-                goto skipped;
-              }
-            else /* explicitly whitelisted, ignore the device blacklist  */
-              break;
-          }
+		devconf->dev_addr_type = VNET_DEV_ADDR_PCI;
+		devconf->is_blacklisted = 1;
+		goto skipped_pci;
+	      }
+	    else /* explicitly whitelisted, ignore the device blacklist  */
+	      break;
+	  }
       }
 
     /* virtio */
@@ -1002,6 +1027,7 @@ dpdk_bind_devices_to_uio (dpdk_config_main_t * conf)
 		      devconf - conf->dev_confs);
 	    devconf->pci_addr.as_u32 = addr->as_u32;
 	  }
+	devconf->dev_addr_type = VNET_DEV_ADDR_PCI;
 	devconf->is_blacklisted = 1;
 	clib_error_report (error);
       }
@@ -1016,16 +1042,72 @@ dpdk_bind_vmbus_devices_to_uio (dpdk_config_main_t * conf)
 {
   clib_error_t *error;
   vlib_vmbus_addr_t *addrs, *addr = 0;
+  int num_whitelisted = vec_len (conf->dev_confs);
+  int i;
 
   addrs = vlib_vmbus_get_all_dev_addrs ();
 
   /* *INDENT-OFF* */
   vec_foreach (addr, addrs)
     {
-      error = vlib_vmbus_bind_to_uio (addr);
+      dpdk_device_config_t *devconf = 0;
+      if (num_whitelisted)
+	{
+	  uword *p = hash_get (conf->device_config_index_by_vmbus_addr,
+			       addr->as_u32[0]);
+	  if (!p)
+	    {
+	      /* No devices blacklisted, but have whitelisted. blacklist all
+	       * non-whitelisted */
+	      pool_get (conf->dev_confs, devconf);
+	      hash_set (conf->device_config_index_by_vmbus_addr,
+			addr->as_u32[0], devconf - conf->dev_confs);
+	      devconf->vmbus_addr = *addr;
+	      devconf->dev_addr_type = VNET_DEV_ADDR_VMBUS;
+	      devconf->is_blacklisted = 1;
+	    skipped_vmbus:
+	      continue;
+	    }
+
+	  devconf = pool_elt_at_index (conf->dev_confs, p[0]);
+	}
+
+      /* Enforce Device blacklist by vmbus_addr */
+      for (i = 0; i < vec_len (conf->blacklist_by_vmbus_addr); i++)
+	{
+	  u32 vmbus_as_u32 = conf->blacklist_by_vmbus_addr[i];
+	  if (vmbus_as_u32 == addr->as_u32[0])
+	    {
+	      if (devconf == 0)
+		{
+		  /* Device not whitelisted */
+		  pool_get (conf->dev_confs, devconf);
+		  hash_set (conf->device_config_index_by_vmbus_addr,
+			    addr->as_u32[0], devconf - conf->dev_confs);
+		  devconf->vmbus_addr = *addr;
+		  devconf->dev_addr_type = VNET_DEV_ADDR_VMBUS;
+		  devconf->is_blacklisted = 1;
+		  goto skipped_vmbus;
+		}
+	      else
+		{
+		  break;
+		}
+	    }
+	}
 
+      error = vlib_vmbus_bind_to_uio (addr);
       if (error)
 	{
+	  if (devconf == 0)
+	    {
+	      pool_get (conf->dev_confs, devconf);
+	      hash_set (conf->device_config_index_by_vmbus_addr,
+			addr->as_u32[0], devconf - conf->dev_confs);
+	      devconf->vmbus_addr = *addr;
+	    }
+	  devconf->dev_addr_type = VNET_DEV_ADDR_VMBUS;
+	  devconf->is_blacklisted = 1;
 	  clib_error_report (error);
 	}
     }
@@ -1033,36 +1115,66 @@ dpdk_bind_vmbus_devices_to_uio (dpdk_config_main_t * conf)
 }
 
 static clib_error_t *
-dpdk_device_config (dpdk_config_main_t * conf, vlib_pci_addr_t pci_addr,
-		    unformat_input_t * input, u8 is_default)
+dpdk_device_config (dpdk_config_main_t *conf, void *addr,
+		    dpdk_device_addr_type_t addr_type, unformat_input_t *input,
+		    u8 is_default)
 {
   clib_error_t *error = 0;
   uword *p;
-  dpdk_device_config_t *devconf;
+  dpdk_device_config_t *devconf = 0;
   unformat_input_t sub_input;
 
   if (is_default)
     {
       devconf = &conf->default_devconf;
     }
-  else
+  else if (addr_type == VNET_DEV_ADDR_PCI)
     {
-      p = hash_get (conf->device_config_index_by_pci_addr, pci_addr.as_u32);
+      p = hash_get (conf->device_config_index_by_pci_addr,
+		    ((vlib_pci_addr_t *) (addr))->as_u32);
 
       if (!p)
 	{
 	  pool_get (conf->dev_confs, devconf);
-	  hash_set (conf->device_config_index_by_pci_addr, pci_addr.as_u32,
+	  hash_set (conf->device_config_index_by_pci_addr,
+		    ((vlib_pci_addr_t *) (addr))->as_u32,
 		    devconf - conf->dev_confs);
 	}
       else
 	return clib_error_return (0,
 				  "duplicate configuration for PCI address %U",
-				  format_vlib_pci_addr, &pci_addr);
+				  format_vlib_pci_addr, addr);
+    }
+  else if (addr_type == VNET_DEV_ADDR_VMBUS)
+    {
+      p = hash_get (conf->device_config_index_by_vmbus_addr,
+		    ((vlib_vmbus_addr_t *) (addr))->as_u32[0]);
+
+      if (!p)
+	{
+	  pool_get (conf->dev_confs, devconf);
+	  hash_set (conf->device_config_index_by_vmbus_addr,
+		    ((vlib_vmbus_addr_t *) (addr))->as_u32[0],
+		    devconf - conf->dev_confs);
+	}
+      else
+	return clib_error_return (
+	  0, "duplicate configuration for VMBUS address %U",
+	  format_vlib_vmbus_addr, addr);
     }
 
-  devconf->pci_addr.as_u32 = pci_addr.as_u32;
-  devconf->tso = DPDK_DEVICE_TSO_DEFAULT;
+  if (addr_type == VNET_DEV_ADDR_PCI)
+    {
+      devconf->pci_addr.as_u32 = ((vlib_pci_addr_t *) (addr))->as_u32;
+      devconf->tso = DPDK_DEVICE_TSO_DEFAULT;
+      devconf->dev_addr_type = VNET_DEV_ADDR_PCI;
+    }
+  else if (addr_type == VNET_DEV_ADDR_VMBUS)
+    {
+      devconf->vmbus_addr = *((vlib_vmbus_addr_t *) (addr));
+      devconf->tso = DPDK_DEVICE_TSO_DEFAULT;
+      devconf->dev_addr_type = VNET_DEV_ADDR_VMBUS;
+    }
 
   if (!input)
     return 0;
@@ -1124,11 +1236,10 @@ dpdk_device_config (dpdk_config_main_t * conf, vlib_pci_addr_t pci_addr,
   else if (devconf->workers &&
 	   clib_bitmap_count_set_bits (devconf->workers) !=
 	   devconf->num_rx_queues)
-    error =
-      clib_error_return (0,
-			 "%U: number of worker threads must be "
-			 "equal to number of rx queues", format_vlib_pci_addr,
-			 &pci_addr);
+    error = clib_error_return (0,
+			       "%U: number of worker threads must be "
+			       "equal to number of rx queues",
+			       format_vlib_pci_addr, addr);
 
   return error;
 }
@@ -1171,7 +1282,8 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
   dpdk_config_main_t *conf = &dpdk_config_main;
   vlib_thread_main_t *tm = vlib_get_thread_main ();
   dpdk_device_config_t *devconf;
-  vlib_pci_addr_t pci_addr;
+  vlib_pci_addr_t pci_addr = { 0 };
+  vlib_vmbus_addr_t vmbus_addr = { 0 };
   unformat_input_t sub_input;
   uword default_hugepage_sz, x;
   u8 *s, *tmp = 0;
@@ -1189,6 +1301,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
     format (0, "%s/hugepages%c", vlib_unix_get_runtime_dir (), 0);
 
   conf->device_config_index_by_pci_addr = hash_create (0, sizeof (uword));
+  conf->device_config_index_by_vmbus_addr = hash_create (0, sizeof (uword));
 
   while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
     {
@@ -1217,8 +1330,7 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 			 &sub_input))
 	{
 	  error =
-	    dpdk_device_config (conf, (vlib_pci_addr_t) (u32) ~ 1, &sub_input,
-				1);
+	    dpdk_device_config (conf, 0, VNET_DEV_ADDR_ANY, &sub_input, 1);
 
 	  if (error)
 	    return error;
@@ -1228,7 +1340,8 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 	    (input, "dev %U %U", unformat_vlib_pci_addr, &pci_addr,
 	     unformat_vlib_cli_sub_input, &sub_input))
 	{
-	  error = dpdk_device_config (conf, pci_addr, &sub_input, 0);
+	  error = dpdk_device_config (conf, &pci_addr, VNET_DEV_ADDR_PCI,
+				      &sub_input, 0);
 
 	  if (error)
 	    return error;
@@ -1237,7 +1350,30 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 	}
       else if (unformat (input, "dev %U", unformat_vlib_pci_addr, &pci_addr))
 	{
-	  error = dpdk_device_config (conf, pci_addr, 0, 0);
+	  error =
+	    dpdk_device_config (conf, &pci_addr, VNET_DEV_ADDR_PCI, 0, 0);
+
+	  if (error)
+	    return error;
+
+	  num_whitelisted++;
+	}
+      else if (unformat (input, "dev %U %U", unformat_vlib_vmbus_addr,
+			 &vmbus_addr, unformat_vlib_cli_sub_input, &sub_input))
+	{
+	  error = dpdk_device_config (conf, &vmbus_addr, VNET_DEV_ADDR_VMBUS,
+				      &sub_input, 0);
+
+	  if (error)
+	    return error;
+
+	  num_whitelisted++;
+	}
+      else if (unformat (input, "dev %U", unformat_vlib_vmbus_addr,
+			 &vmbus_addr))
+	{
+	  error =
+	    dpdk_device_config (conf, &vmbus_addr, VNET_DEV_ADDR_VMBUS, 0, 0);
 
 	  if (error)
 	    return error;
@@ -1259,6 +1395,11 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 	  tmp = format (0, "--no-pci%c", 0);
 	  vec_add1 (conf->eal_init_args, tmp);
 	}
+      else if (unformat (input, "blacklist %U", unformat_vlib_vmbus_addr,
+			 &vmbus_addr))
+	{
+	  vec_add1 (conf->blacklist_by_vmbus_addr, vmbus_addr.as_u32[0]);
+	}
       else
 	if (unformat
 	    (input, "blacklist %x:%x:%x.%x", &domain, &bus, &device, &func))
@@ -1434,23 +1575,21 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
     /* default per-device config items */
     foreach_dpdk_device_config_item
 
-    /* copy vlan_strip config from default device */
-	if (devconf->vlan_strip_offload == 0 &&
-		conf->default_devconf.vlan_strip_offload > 0)
-		devconf->vlan_strip_offload =
-			conf->default_devconf.vlan_strip_offload;
+      /* copy vlan_strip config from default device */
+      _ (vlan_strip_offload)
 
-	/* copy tso config from default device */
-	_(tso)
+      /* copy tso config from default device */
+      _ (tso)
 
-	/* copy tso config from default device */
-	_(devargs)
+      /* copy tso config from default device */
+      _ (devargs)
 
-	/* copy rss_queues config from default device */
-	_(rss_queues)
+      /* copy rss_queues config from default device */
+      _ (rss_queues)
 
-    /* add DPDK EAL whitelist/blacklist entry */
-    if (num_whitelisted > 0 && devconf->is_blacklisted == 0)
+      /* add DPDK EAL whitelist/blacklist entry */
+      if (num_whitelisted > 0 && devconf->is_blacklisted == 0 &&
+	  devconf->dev_addr_type == VNET_DEV_ADDR_PCI)
     {
 	  tmp = format (0, "-a%c", 0);
 	  vec_add1 (conf->eal_init_args, tmp);
@@ -1464,7 +1603,8 @@ dpdk_config (vlib_main_t * vm, unformat_input_t * input)
 	  }
 	  vec_add1 (conf->eal_init_args, tmp);
     }
-    else if (num_whitelisted == 0 && devconf->is_blacklisted != 0)
+    else if (num_whitelisted == 0 && devconf->is_blacklisted != 0 &&
+	     devconf->dev_addr_type == VNET_DEV_ADDR_PCI)
     {
 	  tmp = format (0, "-b%c", 0);
 	  vec_add1 (conf->eal_init_args, tmp);