bonding: add support for numa-only in lacp mode
If numa-only is set, Only slaves on local numa node
transmit pkts if have at least one, otherwise the bond
interface works as usual.
CLI change:
create bond mode lacp [load-balance { l2 | l23 | l34 } {numa-only}]
[hw-addr <mac-address>] [id <if-id>]
The new member "u8 numa_only;" is also added to bond_create_if_args_t.
Type: feature
Change-Id: Icdccedafb0738d8c9d4a5acce909ce562428c071
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
diff --git a/src/vnet/bonding/bond.api b/src/vnet/bonding/bond.api
index e779453..e699267 100644
--- a/src/vnet/bonding/bond.api
+++ b/src/vnet/bonding/bond.api
@@ -19,7 +19,7 @@
the bonding device driver
*/
-option version = "1.0.0";
+option version = "1.0.1";
/** \brief Initialize a new bond interface with the given paramters
@param client_index - opaque cookie to identify the sender
@@ -29,6 +29,7 @@
@param mac_address - mac addr to assign to the interface if use_custom_mac is set
@param mode - mode, required (1=round-robin, 2=active-backup, 3=xor, 4=broadcastcast, 5=lacp)
@param lb - load balance, optional (0=l2, 1=l34, 2=l23) valid for xor and lacp modes. Otherwise ignored
+ @param numa_only - if numa_only is set, pkts will be transmitted by LAG members on local numa node only if have at least one, otherwise it works as usual.
*/
define bond_create
{
@@ -39,6 +40,7 @@
u8 mac_address[6];
u8 mode;
u8 lb;
+ u8 numa_only;
};
/** \brief Reply for bond create reply
@@ -118,6 +120,7 @@
@param interface_name - name of interface
@param mode - bonding mode
@param lb - load balance algo
+ @param numa_only - enable local numa TX for lacp mode
@param active_slaves - active slaves count
@param slaves - config slave count
*/
@@ -129,6 +132,7 @@
u8 interface_name[64];
u8 mode;
u8 lb;
+ u8 numa_only;
u32 active_slaves;
u32 slaves;
};
diff --git a/src/vnet/bonding/bond_api.c b/src/vnet/bonding/bond_api.c
index 07c2cbc..8e18423 100644
--- a/src/vnet/bonding/bond_api.c
+++ b/src/vnet/bonding/bond_api.c
@@ -83,6 +83,7 @@
ap->mode = mp->mode;
ap->lb = mp->lb;
+ ap->numa_only = mp->numa_only;
bond_create_if (vm, ap);
int rv = ap->rv;
@@ -149,6 +150,7 @@
strlen ((const char *) bond_if->interface_name)));
mp->mode = bond_if->mode;
mp->lb = bond_if->lb;
+ mp->numa_only = bond_if->numa_only;
mp->active_slaves = htonl (bond_if->active_slaves);
mp->slaves = htonl (bond_if->slaves);
diff --git a/src/vnet/bonding/cli.c b/src/vnet/bonding/cli.c
index cb344c6..371e3c1 100644
--- a/src/vnet/bonding/cli.c
+++ b/src/vnet/bonding/cli.c
@@ -47,6 +47,17 @@
}
vec_del1 (bif->active_slaves, i);
hash_unset (bif->active_slave_by_sw_if_index, sif->sw_if_index);
+ if (sif->lacp_enabled && bif->numa_only)
+ {
+ /* For lacp mode, if we check it is a slave on local numa node,
+ bif->n_numa_slaves should be decreased by 1 becasue the first
+ bif->n_numa_slaves are all slaves on local numa node */
+ if (i < bif->n_numa_slaves)
+ {
+ bif->n_numa_slaves--;
+ ASSERT (bif->n_numa_slaves >= 0);
+ }
+ }
break;
}
}
@@ -104,7 +115,18 @@
{
hash_set (bif->active_slave_by_sw_if_index, sif->sw_if_index,
sif->sw_if_index);
- vec_add1 (bif->active_slaves, sif->sw_if_index);
+
+ if ((sif->lacp_enabled && bif->numa_only)
+ && (vm->numa_node == hw->numa_node))
+ {
+ vec_insert_elts (bif->active_slaves, &sif->sw_if_index, 1,
+ bif->n_numa_slaves);
+ bif->n_numa_slaves++;
+ }
+ else
+ {
+ vec_add1 (bif->active_slaves, sif->sw_if_index);
+ }
/* First slave becomes active? */
if ((vec_len (bif->active_slaves) == 1) &&
@@ -168,6 +190,7 @@
strlen ((const char *) hi->name)));
bondif->mode = bif->mode;
bondif->lb = bif->lb;
+ bondif->numa_only = bif->numa_only;
bondif->active_slaves = vec_len (bif->active_slaves);
bondif->slaves = vec_len (bif->slaves);
);
@@ -388,6 +411,7 @@
sw = vnet_get_hw_sw_interface (vnm, bif->hw_if_index);
bif->sw_if_index = sw->sw_if_index;
bif->group = bif->sw_if_index;
+ bif->numa_only = args->numa_only;
if (vlib_get_thread_main ()->n_vlib_mains > 1)
clib_spinlock_init (&bif->lockp);
@@ -428,6 +452,14 @@
args.hw_addr_set = 1;
else if (unformat (line_input, "id %u", &args.id))
;
+ else if (unformat (line_input, "numa-only"))
+ {
+ if (args.mode == BOND_MODE_LACP)
+ args.numa_only = 1;
+ else
+ return clib_error_return (0,
+ "Only lacp mode supports numa-only so far!");
+ }
else
return clib_error_return (0, "unknown input `%U'",
format_unformat_error, input);
@@ -446,7 +478,7 @@
VLIB_CLI_COMMAND (bond_create_command, static) = {
.path = "create bond",
.short_help = "create bond mode {round-robin | active-backup | broadcast | "
- "{lacp | xor} [load-balance { l2 | l23 | l34 }]} [hw-addr <mac-address>] "
+ "{lacp | xor} [load-balance { l2 | l23 | l34 } {numa-only}]} [hw-addr <mac-address>] "
"[id <if-id>]",
.function = bond_create_command_fn,
};
diff --git a/src/vnet/bonding/device.c b/src/vnet/bonding/device.c
index b76396a..c9e8b4a 100644
--- a/src/vnet/bonding/device.c
+++ b/src/vnet/bonding/device.c
@@ -699,6 +699,11 @@
goto done;
}
+ /* if have at least one slave on local numa node, only slaves on local numa
+ node will transmit pkts when bif->local_numa_only is enabled */
+ if (bif->n_numa_slaves >= 1)
+ n_slaves = bif->n_numa_slaves;
+
if (bif->lb == BOND_LB_L2)
bond_tx_inline (vm, bif, bufs, hashes, n_left, n_slaves, BOND_LB_L2);
else if (bif->lb == BOND_LB_L34)
diff --git a/src/vnet/bonding/node.h b/src/vnet/bonding/node.h
index 1782150..94b43dc 100644
--- a/src/vnet/bonding/node.h
+++ b/src/vnet/bonding/node.h
@@ -82,6 +82,7 @@
u8 hw_addr[6];
u8 mode;
u8 lb;
+ u8 numa_only;
/* return */
u32 sw_if_index;
int rv;
@@ -117,6 +118,7 @@
u8 interface_name[64];
u8 mode;
u8 lb;
+ u8 numa_only;
u32 active_slaves;
u32 slaves;
} bond_interface_details_t;
@@ -187,6 +189,14 @@
lacp_port_info_t actor;
u8 individual_aggregator;
+ /* If the flag numa_only is set, it means that only slaves
+ on local numa node works for lacp mode if have at least one,
+ otherwise it works as usual. */
+ u8 numa_only;
+
+ /* How many slaves on local numa node are there in lacp mode? */
+ word n_numa_slaves;
+
u32 group;
uword *port_number_bitmap;
u8 use_custom_mac;