summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMohsin Kazmi <sykazmi@cisco.com>2022-03-25 14:27:45 +0000
committerDamjan Marion <dmarion@me.com>2022-04-04 14:22:36 +0000
commit5a7aa51f00d562814204aca7831777651a00869f (patch)
tree33539381ecf5de1ff1f5f82e85dc8e415589f147
parent3414977152ae6362277158dc732e6b9958a6e618 (diff)
devices: add multi-queue support for af-packet
Type: feature Change-Id: I0f4e6517fcfa07ffb0aba89b159ac1337937a508 Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
-rw-r--r--src/vnet/devices/af_packet/af_packet.c532
-rw-r--r--src/vnet/devices/af_packet/af_packet.h71
-rw-r--r--src/vnet/devices/af_packet/af_packet_api.c9
-rw-r--r--src/vnet/devices/af_packet/cli.c12
-rw-r--r--src/vnet/devices/af_packet/device.c149
-rw-r--r--src/vnet/devices/af_packet/node.c68
6 files changed, 571 insertions, 270 deletions
diff --git a/src/vnet/devices/af_packet/af_packet.c b/src/vnet/devices/af_packet/af_packet.c
index 5526d72e7a5..ac6fc05a627 100644
--- a/src/vnet/devices/af_packet/af_packet.c
+++ b/src/vnet/devices/af_packet/af_packet.c
@@ -33,6 +33,7 @@
#include <vnet/devices/netlink.h>
#include <vnet/ethernet/ethernet.h>
#include <vnet/interface/rx_queue_funcs.h>
+#include <vnet/interface/tx_queue_funcs.h>
#include <vnet/devices/af_packet/af_packet.h>
@@ -47,9 +48,9 @@ VNET_HW_INTERFACE_CLASS (af_packet_ip_device_hw_interface_class, static) = {
#define AF_PACKET_DEFAULT_TX_FRAME_SIZE (2048 * 33) // GSO packet of 64KB
#define AF_PACKET_TX_BLOCK_NR 1
-#define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK 256
+#define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK 32
#define AF_PACKET_DEFAULT_RX_FRAME_SIZE 2048
-#define AF_PACKET_RX_BLOCK_NR 20
+#define AF_PACKET_RX_BLOCK_NR 160
/*defined in net/if.h but clashes with dpdk headers */
unsigned int if_nametoindex (const char *ifname);
@@ -98,13 +99,10 @@ af_packet_read_mtu (af_packet_if_t *apif)
static clib_error_t *
af_packet_fd_read_ready (clib_file_t * uf)
{
- af_packet_main_t *apm = &af_packet_main;
vnet_main_t *vnm = vnet_get_main ();
- u32 idx = uf->private_data;
- af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, idx);
/* Schedule the rx node */
- vnet_hw_if_rx_queue_set_int_pending (vnm, apif->queue_index);
+ vnet_hw_if_rx_queue_set_int_pending (vnm, uf->private_data);
return 0;
}
@@ -127,10 +125,74 @@ is_bridge (const u8 * host_if_name)
return -1;
}
+static void
+af_packet_set_rx_queues (vlib_main_t *vm, af_packet_if_t *apif)
+{
+ vnet_main_t *vnm = vnet_get_main ();
+ af_packet_queue_t *rx_queue;
+
+ vnet_hw_if_set_input_node (vnm, apif->hw_if_index,
+ af_packet_input_node.index);
+
+ vec_foreach (rx_queue, apif->rx_queues)
+ {
+ rx_queue->queue_index = vnet_hw_if_register_rx_queue (
+ vnm, apif->hw_if_index, rx_queue->queue_id, VNET_HW_IF_RXQ_THREAD_ANY);
+
+ {
+ clib_file_t template = { 0 };
+ template.read_function = af_packet_fd_read_ready;
+ template.file_descriptor = rx_queue->fd;
+ template.private_data = rx_queue->queue_index;
+ template.flags = UNIX_FILE_EVENT_EDGE_TRIGGERED;
+ template.description =
+ format (0, "%U queue %u", format_af_packet_device_name,
+ apif->dev_instance, rx_queue->queue_id);
+ rx_queue->clib_file_index = clib_file_add (&file_main, &template);
+ }
+ vnet_hw_if_set_rx_queue_file_index (vnm, rx_queue->queue_index,
+ rx_queue->clib_file_index);
+ vnet_hw_if_set_rx_queue_mode (vnm, rx_queue->queue_index,
+ VNET_HW_IF_RX_MODE_INTERRUPT);
+ rx_queue->mode = VNET_HW_IF_RX_MODE_INTERRUPT;
+ }
+ vnet_hw_if_update_runtime_data (vnm, apif->hw_if_index);
+}
+
+static void
+af_packet_set_tx_queues (vlib_main_t *vm, af_packet_if_t *apif)
+{
+ vnet_main_t *vnm = vnet_get_main ();
+ af_packet_main_t *apm = &af_packet_main;
+ af_packet_queue_t *tx_queue;
+
+ vec_foreach (tx_queue, apif->tx_queues)
+ {
+ tx_queue->queue_index = vnet_hw_if_register_tx_queue (
+ vnm, apif->hw_if_index, tx_queue->queue_id);
+ }
+
+ if (apif->num_txqs == 0)
+ {
+ vlib_log_err (apm->log_class, "Interface %U has 0 txq",
+ format_vnet_hw_if_index_name, vnm, apif->hw_if_index);
+ return;
+ }
+
+ for (u32 j = 0; j < vlib_get_n_threads (); j++)
+ {
+ u32 qi = apif->tx_queues[j % apif->num_txqs].queue_index;
+ vnet_hw_if_tx_queue_assign_thread (vnm, qi, j);
+ }
+
+ vnet_hw_if_update_runtime_data (vnm, apif->hw_if_index);
+}
+
static int
create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
- tpacket_req3_t *tx_req, int *fd, u8 **ring,
- u32 *hdrlen_ptr, u8 *is_cksum_gso_enabled)
+ tpacket_req3_t *tx_req, int *fd, af_packet_ring_t *ring,
+ u32 *hdrlen_ptr, u8 *is_cksum_gso_enabled,
+ u32 fanout_id, u8 is_fanout)
{
af_packet_main_t *apm = &af_packet_main;
struct sockaddr_ll sll;
@@ -139,14 +201,19 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
int ver = TPACKET_V3;
u32 hdrlen = 0;
u32 len = sizeof (hdrlen);
- u32 ring_sz = rx_req->tp_block_size * rx_req->tp_block_nr +
- tx_req->tp_block_size * tx_req->tp_block_nr;
+ u32 ring_sz = 0;
+
+ if (rx_req)
+ ring_sz += rx_req->tp_block_size * rx_req->tp_block_nr;
+
+ if (tx_req)
+ ring_sz += tx_req->tp_block_size * tx_req->tp_block_nr;
if ((*fd = socket (AF_PACKET, SOCK_RAW, htons (ETH_P_ALL))) < 0)
{
- vlib_log_debug (apm->log_class,
- "Failed to create AF_PACKET socket: %s (errno %d)",
- strerror (errno), errno);
+ vlib_log_err (apm->log_class,
+ "Failed to create AF_PACKET socket: %s (errno %d)",
+ strerror (errno), errno);
ret = VNET_API_ERROR_SYSCALL_ERROR_1;
goto error;
}
@@ -158,26 +225,25 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
sll.sll_ifindex = host_if_index;
if (bind (*fd, (struct sockaddr *) &sll, sizeof (sll)) < 0)
{
- vlib_log_debug (apm->log_class,
- "Failed to bind rx packet socket: %s (errno %d)",
- strerror (errno), errno);
+ vlib_log_err (apm->log_class,
+ "Failed to bind rx packet socket: %s (errno %d)",
+ strerror (errno), errno);
ret = VNET_API_ERROR_SYSCALL_ERROR_1;
goto error;
}
if (setsockopt (*fd, SOL_PACKET, PACKET_VERSION, &ver, sizeof (ver)) < 0)
{
- vlib_log_debug (
- apm->log_class,
- "Failed to set rx packet interface version: %s (errno %d)",
- strerror (errno), errno);
+ vlib_log_err (apm->log_class,
+ "Failed to set rx packet interface version: %s (errno %d)",
+ strerror (errno), errno);
ret = VNET_API_ERROR_SYSCALL_ERROR_1;
goto error;
}
if (getsockopt (*fd, SOL_PACKET, PACKET_HDRLEN, &hdrlen, &len) < 0)
{
- vlib_log_debug (
+ vlib_log_err (
apm->log_class,
"Failed to get packet hdr len error handling option: %s (errno %d)",
strerror (errno), errno);
@@ -190,7 +256,7 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
int opt = 1;
if (setsockopt (*fd, SOL_PACKET, PACKET_LOSS, &opt, sizeof (opt)) < 0)
{
- vlib_log_debug (
+ vlib_log_err (
apm->log_class,
"Failed to set packet tx ring error handling option: %s (errno %d)",
strerror (errno), errno);
@@ -221,34 +287,52 @@ create_packet_v3_sock (int host_if_index, tpacket_req3_t *rx_req,
}
#endif
- if (setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz) < 0)
+ if (is_fanout)
{
- vlib_log_debug (apm->log_class,
- "Failed to set packet rx ring options: %s (errno %d)",
- strerror (errno), errno);
- ret = VNET_API_ERROR_SYSCALL_ERROR_1;
- goto error;
+ int fanout = ((fanout_id & 0xffff) | ((PACKET_FANOUT_HASH) << 16));
+ if (setsockopt (*fd, SOL_PACKET, PACKET_FANOUT, &fanout,
+ sizeof (fanout)) < 0)
+ {
+ vlib_log_err (apm->log_class,
+ "Failed to set fanout options: %s (errno %d)",
+ strerror (errno), errno);
+ ret = VNET_API_ERROR_SYSCALL_ERROR_1;
+ goto error;
+ }
}
- if (setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz) < 0)
- {
- vlib_log_debug (apm->log_class,
+ if (rx_req)
+ if (setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz) < 0)
+ {
+ vlib_log_err (apm->log_class,
+ "Failed to set packet rx ring options: %s (errno %d)",
+ strerror (errno), errno);
+ ret = VNET_API_ERROR_SYSCALL_ERROR_1;
+ goto error;
+ }
+
+ if (tx_req)
+ if (setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz) < 0)
+ {
+ vlib_log_err (apm->log_class,
"Failed to set packet tx ring options: %s (errno %d)",
strerror (errno), errno);
- ret = VNET_API_ERROR_SYSCALL_ERROR_1;
- goto error;
- }
+ ret = VNET_API_ERROR_SYSCALL_ERROR_1;
+ goto error;
+ }
- *ring = mmap (NULL, ring_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
- *fd, 0);
- if (*ring == MAP_FAILED)
+ ring->ring_start_addr = mmap (NULL, ring_sz, PROT_READ | PROT_WRITE,
+ MAP_SHARED | MAP_LOCKED, *fd, 0);
+ if (ring->ring_start_addr == MAP_FAILED)
{
- vlib_log_debug (apm->log_class, "mmap failure: %s (errno %d)",
- strerror (errno), errno);
+ vlib_log_err (apm->log_class, "mmap failure: %s (errno %d)",
+ strerror (errno), errno);
ret = VNET_API_ERROR_SYSCALL_ERROR_1;
goto error;
}
+ ring->ring_size = ring_sz;
+
return 0;
error:
if (*fd >= 0)
@@ -260,30 +344,192 @@ error:
}
int
-af_packet_create_if (af_packet_create_if_arg_t *arg)
+af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif,
+ af_packet_create_if_arg_t *arg,
+ af_packet_queue_t *rx_queue, af_packet_queue_t *tx_queue,
+ u8 queue_id, u8 is_fanout)
{
af_packet_main_t *apm = &af_packet_main;
- vlib_main_t *vm = vlib_get_main ();
- int ret, fd = -1, fd2 = -1;
tpacket_req3_t *rx_req = 0;
tpacket_req3_t *tx_req = 0;
+ int ret, fd = -1;
+ af_packet_ring_t ring = { 0 };
+ u8 *ring_addr = 0;
+ u32 rx_frames_per_block, tx_frames_per_block;
+ u32 rx_frame_size, tx_frame_size;
+ u32 hdrlen = 0;
+ u32 i = 0;
+ u8 is_cksum_gso_enabled = 0;
+
+ if (rx_queue)
+ {
+ rx_frames_per_block = arg->rx_frames_per_block ?
+ arg->rx_frames_per_block :
+ AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK;
+
+ rx_frame_size = arg->rx_frame_size ? arg->rx_frame_size :
+ AF_PACKET_DEFAULT_RX_FRAME_SIZE;
+ vec_validate (rx_queue->rx_req, 0);
+ rx_queue->rx_req->tp_block_size = rx_frame_size * rx_frames_per_block;
+ rx_queue->rx_req->tp_frame_size = rx_frame_size;
+ rx_queue->rx_req->tp_block_nr = AF_PACKET_RX_BLOCK_NR;
+ rx_queue->rx_req->tp_frame_nr =
+ AF_PACKET_RX_BLOCK_NR * rx_frames_per_block;
+ rx_queue->rx_req->tp_retire_blk_tov = 1; // 1 ms block timout
+ rx_queue->rx_req->tp_feature_req_word = 0;
+ rx_queue->rx_req->tp_sizeof_priv = 0;
+ rx_req = rx_queue->rx_req;
+ }
+
+ if (tx_queue)
+ {
+ tx_frames_per_block = arg->tx_frames_per_block ?
+ arg->tx_frames_per_block :
+ AF_PACKET_DEFAULT_TX_FRAMES_PER_BLOCK;
+ tx_frame_size = arg->tx_frame_size ? arg->tx_frame_size :
+ AF_PACKET_DEFAULT_TX_FRAME_SIZE;
+
+ vec_validate (tx_queue->tx_req, 0);
+ tx_queue->tx_req->tp_block_size = tx_frame_size * tx_frames_per_block;
+ tx_queue->tx_req->tp_frame_size = tx_frame_size;
+ tx_queue->tx_req->tp_block_nr = AF_PACKET_TX_BLOCK_NR;
+ tx_queue->tx_req->tp_frame_nr =
+ AF_PACKET_TX_BLOCK_NR * tx_frames_per_block;
+ tx_queue->tx_req->tp_retire_blk_tov = 0;
+ tx_queue->tx_req->tp_sizeof_priv = 0;
+ tx_queue->tx_req->tp_feature_req_word = 0;
+ tx_req = tx_queue->tx_req;
+ }
+
+ ret = create_packet_v3_sock (apif->host_if_index, rx_req, tx_req, &fd, &ring,
+ &hdrlen, &is_cksum_gso_enabled,
+ apif->dev_instance, is_fanout);
+
+ if (ret != 0)
+ goto error;
+
+ vec_add1 (apif->rings, ring);
+ ring_addr = ring.ring_start_addr;
+
+ if (rx_queue)
+ {
+ rx_queue->fd = fd;
+ vec_validate (rx_queue->rx_ring, rx_queue->rx_req->tp_block_nr - 1);
+ vec_foreach_index (i, rx_queue->rx_ring)
+ {
+ rx_queue->rx_ring[i] =
+ ring_addr + i * rx_queue->rx_req->tp_block_size;
+ }
+
+ rx_queue->next_rx_block = 0;
+ rx_queue->queue_id = queue_id;
+ rx_queue->is_rx_pending = 0;
+ ring_addr = ring_addr + rx_queue->rx_req->tp_block_size *
+ rx_queue->rx_req->tp_block_nr;
+ }
+
+ if (tx_queue)
+ {
+ tx_queue->fd = fd;
+ vec_validate (tx_queue->tx_ring, tx_queue->tx_req->tp_block_nr - 1);
+ vec_foreach_index (i, tx_queue->tx_ring)
+ {
+ tx_queue->tx_ring[i] =
+ ring_addr + i * tx_queue->tx_req->tp_block_size;
+ }
+
+ tx_queue->next_tx_frame = 0;
+ tx_queue->queue_id = queue_id;
+ clib_spinlock_init (&tx_queue->lockp);
+ }
+
+ if (queue_id == 0)
+ {
+ apif->hdrlen = hdrlen;
+ apif->is_cksum_gso_enabled = is_cksum_gso_enabled;
+ }
+
+ return 0;
+error:
+ vlib_log_err (apm->log_class, "Failed to set queue %u error", queue_id);
+ vec_free (rx_queue->rx_req);
+ vec_free (tx_queue->tx_req);
+ return ret;
+}
+
+int
+af_packet_device_init (vlib_main_t *vm, af_packet_if_t *apif,
+ af_packet_create_if_arg_t *args)
+{
+ af_packet_main_t *apm = &af_packet_main;
+ af_packet_queue_t *rx_queue = 0;
+ af_packet_queue_t *tx_queue = 0;
+ u16 nq = clib_min (args->num_rxqs, args->num_txqs);
+ u16 i = 0;
+ int ret = 0;
+ u8 is_fanout = (args->num_rxqs > 1) ? 1 : 0;
+
+ vec_validate (apif->rx_queues, args->num_rxqs - 1);
+ vec_validate (apif->tx_queues, args->num_txqs - 1);
+
+ for (; i < nq; i++)
+ {
+ rx_queue = vec_elt_at_index (apif->rx_queues, i);
+ tx_queue = vec_elt_at_index (apif->tx_queues, i);
+ ret = af_packet_queue_init (vm, apif, args, rx_queue, tx_queue, i,
+ is_fanout);
+ if (ret != 0)
+ goto error;
+ }
+
+ if (args->num_rxqs > args->num_txqs)
+ {
+ for (; i < args->num_rxqs; i++)
+ {
+ rx_queue = vec_elt_at_index (apif->rx_queues, i);
+ ret =
+ af_packet_queue_init (vm, apif, args, rx_queue, 0, i, is_fanout);
+ if (ret != 0)
+ goto error;
+ }
+ }
+ else if (args->num_txqs > args->num_rxqs)
+ {
+ for (; i < args->num_txqs; i++)
+ {
+ tx_queue = vec_elt_at_index (apif->tx_queues, i);
+ ret = af_packet_queue_init (vm, apif, args, 0, tx_queue, i, 0);
+ if (ret != 0)
+ goto error;
+ }
+ }
+
+ apif->num_rxqs = args->num_rxqs;
+ apif->num_txqs = args->num_txqs;
+
+ return 0;
+error:
+ vlib_log_err (apm->log_class, "Failed to init device error");
+ return ret;
+}
+
+int
+af_packet_create_if (af_packet_create_if_arg_t *arg)
+{
+ af_packet_main_t *apm = &af_packet_main;
+ vlib_main_t *vm = vlib_get_main ();
+ int fd2 = -1;
struct ifreq ifr;
- u8 *ring = 0;
af_packet_if_t *apif = 0;
u8 hw_addr[6];
vnet_sw_interface_t *sw;
- vlib_thread_main_t *tm = vlib_get_thread_main ();
vnet_main_t *vnm = vnet_get_main ();
vnet_hw_if_caps_t caps = VNET_HW_IF_CAP_INT_MODE;
uword *p;
uword if_index;
u8 *host_if_name_dup = 0;
int host_if_index = -1;
- u32 rx_frames_per_block, tx_frames_per_block;
- u32 rx_frame_size, tx_frame_size;
- u32 hdrlen = 0;
- u32 i = 0;
- u8 is_cksum_gso_enabled = 0;
+ int ret = 0;
p = mhash_get (&apm->if_index_by_host_if_name, arg->host_if_name);
if (p)
@@ -295,35 +541,6 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
host_if_name_dup = vec_dup (arg->host_if_name);
- rx_frames_per_block = arg->rx_frames_per_block ?
- arg->rx_frames_per_block :
- AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK;
- tx_frames_per_block = arg->tx_frames_per_block ?
- arg->tx_frames_per_block :
- AF_PACKET_DEFAULT_TX_FRAMES_PER_BLOCK;
- rx_frame_size =
- arg->rx_frame_size ? arg->rx_frame_size : AF_PACKET_DEFAULT_RX_FRAME_SIZE;
- tx_frame_size =
- arg->tx_frame_size ? arg->tx_frame_size : AF_PACKET_DEFAULT_TX_FRAME_SIZE;
-
- vec_validate (rx_req, 0);
- rx_req->tp_block_size = rx_frame_size * rx_frames_per_block;
- rx_req->tp_frame_size = rx_frame_size;
- rx_req->tp_block_nr = AF_PACKET_RX_BLOCK_NR;
- rx_req->tp_frame_nr = AF_PACKET_RX_BLOCK_NR * rx_frames_per_block;
- rx_req->tp_retire_blk_tov = 0;
- rx_req->tp_feature_req_word = 0;
- rx_req->tp_sizeof_priv = 0;
-
- vec_validate (tx_req, 0);
- tx_req->tp_block_size = tx_frame_size * tx_frames_per_block;
- tx_req->tp_frame_size = tx_frame_size;
- tx_req->tp_block_nr = AF_PACKET_TX_BLOCK_NR;
- tx_req->tp_frame_nr = AF_PACKET_TX_BLOCK_NR * tx_frames_per_block;
- tx_req->tp_retire_blk_tov = 0;
- tx_req->tp_sizeof_priv = 0;
- tx_req->tp_feature_req_word = 0;
-
/*
* make sure host side of interface is 'UP' before binding AF_PACKET
* socket on it.
@@ -378,14 +595,7 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
fd2 = -1;
}
- ret = create_packet_v3_sock (host_if_index, rx_req, tx_req, &fd, &ring,
- &hdrlen, &is_cksum_gso_enabled);
-
- if (ret != 0)
- goto error;
-
ret = is_bridge (arg->host_if_name);
-
if (ret == 0) /* is a bridge, ignore state */
host_if_index = -1;
@@ -393,40 +603,20 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
pool_get (apm->interfaces, apif);
if_index = apif - apm->interfaces;
+ apif->dev_instance = if_index;
apif->host_if_index = host_if_index;
- apif->fd = fd;
-
- vec_validate (apif->rx_ring, rx_req->tp_block_nr - 1);
- vec_foreach_index (i, apif->rx_ring)
- {
- apif->rx_ring[i] = ring + i * rx_req->tp_block_size;
- }
-
- ring = ring + rx_req->tp_block_size * rx_req->tp_block_nr;
-
- vec_validate (apif->tx_ring, tx_req->tp_block_nr - 1);
- vec_foreach_index (i, apif->tx_ring)
- {
- apif->tx_ring[i] = ring + i * tx_req->tp_block_size;
- }
-
- apif->rx_req = rx_req;
- apif->tx_req = tx_req;
apif->host_if_name = host_if_name_dup;
apif->per_interface_next_index = ~0;
- apif->next_tx_frame = 0;
- apif->next_rx_block = 0;
apif->mode = arg->mode;
- apif->hdrlen = hdrlen;
- apif->is_cksum_gso_enabled = is_cksum_gso_enabled;
- apif->ss.is_save = 0;
+
+ ret = af_packet_device_init (vm, apif, arg);
+ if (ret != 0)
+ goto error;
ret = af_packet_read_mtu (apif);
if (ret != 0)
goto error;
- if (tm->n_vlib_mains > 1)
- clib_spinlock_init (&apif->lockp);
if (apif->mode != AF_PACKET_IF_MODE_IP)
{
@@ -447,7 +637,7 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
}
eir.dev_class_index = af_packet_device_class.index;
- eir.dev_instance = if_index;
+ eir.dev_instance = apif->dev_instance;
eir.address = hw_addr;
eir.cb.set_max_frame_size = af_packet_eth_set_max_frame_size;
apif->hw_if_index = vnet_eth_register_interface (vnm, &eir);
@@ -455,15 +645,15 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
else
{
apif->hw_if_index = vnet_register_interface (
- vnm, af_packet_device_class.index, if_index,
- af_packet_ip_device_hw_interface_class.index, if_index);
+ vnm, af_packet_device_class.index, apif->dev_instance,
+ af_packet_ip_device_hw_interface_class.index, apif->dev_instance);
}
+
sw = vnet_get_hw_sw_interface (vnm, apif->hw_if_index);
apif->sw_if_index = sw->sw_if_index;
- vnet_hw_if_set_input_node (vnm, apif->hw_if_index,
- af_packet_input_node.index);
- apif->queue_index = vnet_hw_if_register_rx_queue (vnm, apif->hw_if_index, 0,
- VNET_HW_IF_RXQ_THREAD_ANY);
+
+ af_packet_set_rx_queues (vm, apif);
+ af_packet_set_tx_queues (vm, apif);
if (apif->is_cksum_gso_enabled)
caps |= VNET_HW_IF_CAP_TCP_GSO | VNET_HW_IF_CAP_TX_IP4_CKSUM |
@@ -473,22 +663,6 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
vnet_hw_interface_set_flags (vnm, apif->hw_if_index,
VNET_HW_INTERFACE_FLAG_LINK_UP);
- vnet_hw_if_set_rx_queue_mode (vnm, apif->queue_index,
- VNET_HW_IF_RX_MODE_INTERRUPT);
- vnet_hw_if_update_runtime_data (vnm, apif->hw_if_index);
- {
- clib_file_t template = { 0 };
- template.read_function = af_packet_fd_read_ready;
- template.file_descriptor = fd;
- template.private_data = if_index;
- template.flags = UNIX_FILE_EVENT_EDGE_TRIGGERED;
- template.description =
- format (0, "%U", format_af_packet_device_name, if_index);
- apif->clib_file_index = clib_file_add (&file_main, &template);
- }
- vnet_hw_if_set_rx_queue_file_index (vnm, apif->queue_index,
- apif->clib_file_index);
-
mhash_set_mem (&apm->if_index_by_host_if_name, host_if_name_dup, &if_index,
0);
arg->sw_if_index = apif->sw_if_index;
@@ -502,20 +676,65 @@ error:
fd2 = -1;
}
vec_free (host_if_name_dup);
- vec_free (rx_req);
- vec_free (tx_req);
+ memset (apif, 0, sizeof (*apif));
+ pool_put (apm->interfaces, apif);
return ret;
}
+static int
+af_packet_rx_queue_free (af_packet_if_t *apif, af_packet_queue_t *rx_queue)
+{
+ clib_file_del_by_index (&file_main, rx_queue->clib_file_index);
+ close (rx_queue->fd);
+ rx_queue->fd = -1;
+ rx_queue->rx_ring = NULL;
+ vec_free (rx_queue->rx_req);
+ rx_queue->rx_req = NULL;
+ return 0;
+}
+
+static int
+af_packet_tx_queue_free (af_packet_if_t *apif, af_packet_queue_t *tx_queue)
+{
+ close (tx_queue->fd);
+ tx_queue->fd = -1;
+ clib_spinlock_free (&tx_queue->lockp);
+ tx_queue->tx_ring = NULL;
+ vec_free (tx_queue->tx_req);
+ tx_queue->tx_req = NULL;
+ return 0;
+}
+
+static int
+af_packet_ring_free (af_packet_if_t *apif, af_packet_ring_t *ring)
+{
+ af_packet_main_t *apm = &af_packet_main;
+
+ if (ring)
+ {
+ // FIXME: unmap the memory
+ if (munmap (ring->ring_start_addr, ring->ring_size))
+ vlib_log_warn (apm->log_class,
+ "Host interface %s could not free ring %p of size %u",
+ apif->host_if_name, ring->ring_start_addr,
+ ring->ring_size);
+ else
+ ring->ring_start_addr = 0;
+ }
+
+ return 0;
+}
+
int
af_packet_delete_if (u8 *host_if_name)
{
vnet_main_t *vnm = vnet_get_main ();
af_packet_main_t *apm = &af_packet_main;
af_packet_if_t *apif;
+ af_packet_queue_t *rx_queue;
+ af_packet_queue_t *tx_queue;
+ af_packet_ring_t *ring;
uword *p;
- uword if_index;
- u32 ring_sz;
p = mhash_get (&apm->if_index_by_host_if_name, host_if_name);
if (p == NULL)
@@ -525,46 +744,37 @@ af_packet_delete_if (u8 *host_if_name)
return VNET_API_ERROR_SYSCALL_ERROR_1;
}
apif = pool_elt_at_index (apm->interfaces, p[0]);
- if_index = apif - apm->interfaces;
/* bring down the interface */
vnet_hw_interface_set_flags (vnm, apif->hw_if_index, 0);
/* clean up */
- if (apif->clib_file_index != ~0)
- {
- clib_file_del (&file_main, file_main.file_pool + apif->clib_file_index);
- apif->clib_file_index = ~0;
- }
- else
- close (apif->fd);
-
- ring_sz = apif->rx_req->tp_block_size * apif->rx_req->tp_block_nr +
- apif->tx_req->tp_block_size * apif->tx_req->tp_block_nr;
- if (munmap (apif->rx_ring, ring_sz))
- vlib_log_warn (apm->log_class,
- "Host interface %s could not free rx/tx ring",
- host_if_name);
- apif->rx_ring = NULL;
- apif->tx_ring = NULL;
- apif->fd = -1;
-
- vec_free (apif->rx_req);
- apif->rx_req = NULL;
- vec_free (apif->tx_req);
- apif->tx_req = NULL;
+ vec_foreach (rx_queue, apif->rx_queues)
+ af_packet_rx_queue_free (apif, rx_queue);
+ vec_foreach (tx_queue, apif->tx_queues)
+ af_packet_tx_queue_free (apif, tx_queue);
+ vec_foreach (ring, apif->rings)
+ af_packet_ring_free (apif, ring);
+
+ vec_free (apif->rx_queues);
+ apif->rx_queues = NULL;
+ vec_free (apif->tx_queues);
+ apif->tx_queues = NULL;
+ vec_free (apif->rings);
+ apif->rings = NULL;
vec_free (apif->host_if_name);
apif->host_if_name = NULL;
apif->host_if_index = -1;
- mhash_unset (&apm->if_index_by_host_if_name, host_if_name, &if_index);
+ mhash_unset (&apm->if_index_by_host_if_name, host_if_name, p);
if (apif->mode != AF_PACKET_IF_MODE_IP)
ethernet_delete_interface (vnm, apif->hw_if_index);
else
vnet_delete_hw_interface (vnm, apif->hw_if_index);
+ memset (apif, 0, sizeof (*apif));
pool_put (apm->interfaces, apif);
return 0;
diff --git a/src/vnet/devices/af_packet/af_packet.h b/src/vnet/devices/af_packet/af_packet.h
index 217d2f6b009..4cae700909a 100644
--- a/src/vnet/devices/af_packet/af_packet.h
+++ b/src/vnet/devices/af_packet/af_packet.h
@@ -40,37 +40,68 @@ typedef struct
typedef struct
{
- u32 rx_frame_offset;
- u32 num_pkts;
- u8 is_save;
-} save_state_t;
+ u8 *ring_start_addr;
+ u32 ring_size;
+} af_packet_ring_t;
typedef struct
{
CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
clib_spinlock_t lockp;
- u8 *host_if_name;
- int host_if_index;
int fd;
- tpacket_req3_t *rx_req;
- tpacket_req3_t *tx_req;
- u8 **rx_ring;
- u8 **tx_ring;
- u8 is_cksum_gso_enabled;
- u32 hdrlen;
- u32 hw_if_index;
- u32 sw_if_index;
+ union
+ {
+ tpacket_req3_t *rx_req;
+ tpacket_req3_t *tx_req;
+ };
+
+ union
+ {
+ u8 **rx_ring;
+ u8 **tx_ring;
+ };
+
+ union
+ {
+ u32 next_rx_block;
+ u32 next_tx_frame;
+ };
+
+ u16 queue_id;
+ u32 queue_index;
+
u32 clib_file_index;
- u32 next_rx_block;
- u32 next_tx_frame;
+ u32 rx_frame_offset;
+ u16 num_rx_pkts;
+ u8 is_rx_pending;
+ vnet_hw_if_rx_mode mode;
+} af_packet_queue_t;
+typedef struct
+{
+ CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+ u32 hw_if_index;
+ u32 sw_if_index;
u32 per_interface_next_index;
+ af_packet_if_mode_t mode;
u8 is_admin_up;
- u32 queue_index;
+ u8 is_cksum_gso_enabled;
+
+ af_packet_queue_t *rx_queues;
+ af_packet_queue_t *tx_queues;
+
+ u8 num_rxqs;
+ u8 num_txqs;
+
+ u8 *host_if_name;
+ int host_if_index;
+ u32 hdrlen;
+
u32 host_mtu;
- af_packet_if_mode_t mode;
- save_state_t ss;
+ u32 dev_instance;
+
+ af_packet_ring_t *rings;
} af_packet_if_t;
typedef struct
@@ -96,6 +127,8 @@ typedef struct
u32 tx_frame_size;
u32 rx_frames_per_block;
u32 tx_frames_per_block;
+ u8 num_rxqs;
+ u8 num_txqs;
af_packet_if_mode_t mode;
/* return */
diff --git a/src/vnet/devices/af_packet/af_packet_api.c b/src/vnet/devices/af_packet/af_packet_api.c
index 3d57977eb49..693380d8ae5 100644
--- a/src/vnet/devices/af_packet/af_packet_api.c
+++ b/src/vnet/devices/af_packet/af_packet_api.c
@@ -68,6 +68,9 @@ vl_api_af_packet_create_v2_t_handler (vl_api_af_packet_create_v2_t *mp)
arg->host_if_name = format (0, "%s", mp->host_if_name);
vec_add1 (arg->host_if_name, 0);
+ // Default number of rx/tx queue(s)
+ arg->num_rxqs = 1;
+ arg->num_txqs = 1;
arg->rx_frame_size = clib_net_to_host_u32 (mp->rx_frame_size);
arg->tx_frame_size = clib_net_to_host_u32 (mp->tx_frame_size);
arg->rx_frames_per_block = clib_net_to_host_u32 (mp->rx_frames_per_block);
@@ -76,14 +79,10 @@ vl_api_af_packet_create_v2_t_handler (vl_api_af_packet_create_v2_t *mp)
arg->mode = AF_PACKET_IF_MODE_ETHERNET;
if (mp->num_rx_queues > 1)
- {
- rv = VNET_API_ERROR_INVALID_VALUE;
- goto out;
- }
+ arg->num_rxqs = clib_net_to_host_u16 (mp->num_rx_queues);
rv = af_packet_create_if (arg);
-out:
vec_free (arg->host_if_name);
REPLY_MACRO2 (VL_API_AF_PACKET_CREATE_V2_REPLY, ({
rmp->sw_if_index = clib_host_to_net_u32 (arg->sw_if_index);
diff --git a/src/vnet/devices/af_packet/cli.c b/src/vnet/devices/af_packet/cli.c
index 3dd3c8ee848..c90d5251f9f 100644
--- a/src/vnet/devices/af_packet/cli.c
+++ b/src/vnet/devices/af_packet/cli.c
@@ -54,6 +54,10 @@ af_packet_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
// Default mode
arg->mode = AF_PACKET_IF_MODE_ETHERNET;
+ // Default number of rx/tx queue(s)
+ arg->num_rxqs = 1;
+ arg->num_txqs = 1;
+
/* Get a line of input. */
if (!unformat_user (input, unformat_line_input, line_input))
return 0;
@@ -72,6 +76,10 @@ af_packet_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
else if (unformat (line_input, "tx-per-block %u",
&arg->tx_frames_per_block))
;
+ else if (unformat (line_input, "num-rx-queues %u", &arg->num_rxqs))
+ ;
+ else if (unformat (line_input, "num-tx-queues %u", &arg->num_txqs))
+ ;
else if (unformat (line_input, "mode ip"))
arg->mode = AF_PACKET_IF_MODE_IP;
else if (unformat (line_input, "hw-addr %U", unformat_ethernet_address,
@@ -145,8 +153,8 @@ done:
?*/
VLIB_CLI_COMMAND (af_packet_create_command, static) = {
.path = "create host-interface",
- .short_help =
- "create host-interface name <ifname> [hw-addr <mac-addr>] [mode ip]",
+ .short_help = "create host-interface name <ifname> [num-rx-queues <n>] "
+ "[num-tx-queues <n>] [hw-addr <mac-addr>] [mode ip]",
.function = af_packet_create_command_fn,
};
diff --git a/src/vnet/devices/af_packet/device.c b/src/vnet/devices/af_packet/device.c
index f7f006ebd7f..013d9f71733 100644
--- a/src/vnet/devices/af_packet/device.c
+++ b/src/vnet/devices/af_packet/device.c
@@ -60,6 +60,7 @@ typedef struct
{
u32 buffer_index;
u32 hw_if_index;
+ u16 queue_id;
tpacket3_hdr_t tph;
vnet_virtio_net_hdr_t vnet_hdr;
vlib_buffer_t buffer;
@@ -87,51 +88,74 @@ format_af_packet_device (u8 * s, va_list * args)
af_packet_main_t *apm = &af_packet_main;
af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, dev_instance);
- clib_spinlock_lock_if_init (&apif->lockp);
- u32 tx_block_sz = apif->tx_req->tp_block_size;
- u32 tx_frame_sz = apif->tx_req->tp_frame_size;
- u32 tx_frame_nr = apif->tx_req->tp_frame_nr;
- u32 tx_block_nr = apif->tx_req->tp_block_nr;
- u32 rx_block_size = apif->rx_req->tp_block_size;
- u32 rx_frame_size = apif->rx_req->tp_frame_size;
- u32 rx_frame_nr = apif->rx_req->tp_frame_nr;
- u32 rx_block_nr = apif->rx_req->tp_block_nr;
- int block = 0;
- u8 *tx_block_start = apif->tx_ring[block];
- u32 tx_frame = apif->next_tx_frame;
- tpacket3_hdr_t *tph;
-
- s = format (s, "Linux PACKET socket interface\n");
- s = format (s, "%UTX block size:%d nr:%d TX frame size:%d nr:%d\n",
- format_white_space, indent, tx_block_sz, tx_block_nr,
- tx_frame_sz, tx_frame_nr);
- s = format (s, "%URX block size:%d nr:%d RX frame size:%d nr:%d\n",
- format_white_space, indent, rx_block_size, rx_block_nr,
- rx_frame_size, rx_frame_nr);
- s = format (s, "%Unext frame:%d\n", format_white_space, indent,
- apif->next_tx_frame);
-
- int n_send_req = 0, n_avail = 0, n_sending = 0, n_tot = 0, n_wrong = 0;
- do
+ af_packet_queue_t *rx_queue = 0;
+ af_packet_queue_t *tx_queue = 0;
+
+ s = format (s, "Linux PACKET socket interface");
+
+ vec_foreach (rx_queue, apif->rx_queues)
{
- tph = (tpacket3_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz);
- tx_frame = (tx_frame + 1) % tx_frame_nr;
- if (tph->tp_status == 0)
- n_avail++;
- else if (tph->tp_status & TP_STATUS_SEND_REQUEST)
- n_send_req++;
- else if (tph->tp_status & TP_STATUS_SENDING)
- n_sending++;
- else
- n_wrong++;
- n_tot++;
+ u32 rx_block_size = rx_queue->rx_req->tp_block_size;
+ u32 rx_frame_size = rx_queue->rx_req->tp_frame_size;
+ u32 rx_frame_nr = rx_queue->rx_req->tp_frame_nr;
+ u32 rx_block_nr = rx_queue->rx_req->tp_block_nr;
+
+ s = format (s, "\n%URX Queue %u:", format_white_space, indent,
+ rx_queue->queue_id);
+ s = format (s, "\n%Ublock size:%d nr:%d frame size:%d nr:%d",
+ format_white_space, indent + 2, rx_block_size, rx_block_nr,
+ rx_frame_size, rx_frame_nr);
+ s = format (s, " next block:%d", rx_queue->next_rx_block);
+ if (rx_queue->is_rx_pending)
+ {
+ s = format (
+ s, "\n%UPending Request: num-rx-pkts:%d next-frame-offset:%d",
+ format_white_space, indent + 2, rx_queue->num_rx_pkts,
+ rx_queue->rx_frame_offset);
+ }
}
- while (tx_frame != apif->next_tx_frame);
- s = format (s, "%Uavailable:%d request:%d sending:%d wrong:%d total:%d\n",
- format_white_space, indent, n_avail, n_send_req, n_sending,
- n_wrong, n_tot);
- clib_spinlock_unlock_if_init (&apif->lockp);
+ vec_foreach (tx_queue, apif->tx_queues)
+ {
+ clib_spinlock_lock (&tx_queue->lockp);
+ u32 tx_block_sz = tx_queue->tx_req->tp_block_size;
+ u32 tx_frame_sz = tx_queue->tx_req->tp_frame_size;
+ u32 tx_frame_nr = tx_queue->tx_req->tp_frame_nr;
+ u32 tx_block_nr = tx_queue->tx_req->tp_block_nr;
+ int block = 0;
+ int n_send_req = 0, n_avail = 0, n_sending = 0, n_tot = 0, n_wrong = 0;
+ u8 *tx_block_start = tx_queue->tx_ring[block];
+ u32 tx_frame = tx_queue->next_tx_frame;
+ tpacket3_hdr_t *tph;
+
+ s = format (s, "\n%UTX Queue %u:", format_white_space, indent,
+ tx_queue->queue_id);
+ s = format (s, "\n%Ublock size:%d nr:%d frame size:%d nr:%d",
+ format_white_space, indent + 2, tx_block_sz, tx_block_nr,
+ tx_frame_sz, tx_frame_nr);
+ s = format (s, " next frame:%d", tx_queue->next_tx_frame);
+
+ do
+ {
+ tph = (tpacket3_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz);
+ tx_frame = (tx_frame + 1) % tx_frame_nr;
+ if (tph->tp_status == 0)
+ n_avail++;
+ else if (tph->tp_status & TP_STATUS_SEND_REQUEST)
+ n_send_req++;
+ else if (tph->tp_status & TP_STATUS_SENDING)
+ n_sending++;
+ else
+ n_wrong++;
+ n_tot++;
+ }
+ while (tx_frame != tx_queue->next_tx_frame);
+ s =
+ format (s, "\n%Uavailable:%d request:%d sending:%d wrong:%d total:%d",
+ format_white_space, indent + 2, n_avail, n_send_req, n_sending,
+ n_wrong, n_tot);
+ clib_spinlock_unlock (&tx_queue->lockp);
+ }
return s;
}
@@ -143,7 +167,8 @@ format_af_packet_tx_trace (u8 *s, va_list *va)
af_packet_tx_trace_t *t = va_arg (*va, af_packet_tx_trace_t *);
u32 indent = format_get_indent (s);
- s = format (s, "af_packet: hw_if_index %u", t->hw_if_index);
+ s = format (s, "af_packet: hw_if_index %u tx-queue %u", t->hw_if_index,
+ t->queue_id);
s =
format (s,
@@ -183,11 +208,13 @@ format_af_packet_tx_trace (u8 *s, va_list *va)
static void
af_packet_tx_trace (vlib_main_t *vm, vlib_node_runtime_t *node,
vlib_buffer_t *b0, u32 bi, tpacket3_hdr_t *tph,
- vnet_virtio_net_hdr_t *vnet_hdr, u32 hw_if_index)
+ vnet_virtio_net_hdr_t *vnet_hdr, u32 hw_if_index,
+ u16 queue_id)
{
af_packet_tx_trace_t *t;
t = vlib_add_trace (vm, node, b0, sizeof (t[0]));
t->hw_if_index = hw_if_index;
+ t->queue_id = queue_id;
t->buffer_index = bi;
clib_memcpy_fast (&t->tph, tph, sizeof (*tph));
@@ -289,22 +316,29 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm,
vlib_frame_t * frame)
{
af_packet_main_t *apm = &af_packet_main;
+ vnet_hw_if_tx_frame_t *tf = vlib_frame_scalar_args (frame);
u32 *buffers = vlib_frame_vector_args (frame);
u32 n_left = frame->n_vectors;
u32 n_sent = 0;
vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
af_packet_if_t *apif =
pool_elt_at_index (apm->interfaces, rd->dev_instance);
- clib_spinlock_lock_if_init (&apif->lockp);
- u32 block = 0;
- u32 frame_size = apif->tx_req->tp_frame_size;
- u32 frame_num = apif->tx_req->tp_frame_nr;
- u8 *block_start = apif->tx_ring[block];
- u32 tx_frame = apif->next_tx_frame;
- tpacket3_hdr_t *tph;
+ u16 queue_id = tf->queue_id;
+ af_packet_queue_t *tx_queue = vec_elt_at_index (apif->tx_queues, queue_id);
+ u32 block = 0, frame_size = 0, frame_num = 0, tx_frame = 0;
+ u8 *block_start = 0;
+ tpacket3_hdr_t *tph = 0;
u32 frame_not_ready = 0;
u8 is_cksum_gso_enabled = (apif->is_cksum_gso_enabled == 1) ? 1 : 0;
+ if (tf->shared_queue)
+ clib_spinlock_lock (&tx_queue->lockp);
+
+ frame_size = tx_queue->tx_req->tp_frame_size;
+ frame_num = tx_queue->tx_req->tp_frame_nr;
+ block_start = tx_queue->tx_ring[block];
+ tx_frame = tx_queue->next_tx_frame;
+
while (n_left)
{
u32 len;
@@ -366,12 +400,12 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm,
{
if (PREDICT_TRUE (is_cksum_gso_enabled))
af_packet_tx_trace (vm, node, b0_first, bi_first, tph, vnet_hdr,
- apif->hw_if_index);
+ apif->hw_if_index, queue_id);
else
{
vnet_virtio_net_hdr_t vnet_hdr2 = {};
af_packet_tx_trace (vm, node, b0_first, bi_first, tph,
- &vnet_hdr2, apif->hw_if_index);
+ &vnet_hdr2, apif->hw_if_index, queue_id);
}
}
tx_frame = (tx_frame + 1) % frame_num;
@@ -386,10 +420,10 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm,
if (PREDICT_TRUE (n_sent))
{
- apif->next_tx_frame = tx_frame;
+ tx_queue->next_tx_frame = tx_frame;
- if (PREDICT_FALSE (sendto (apif->fd, NULL, 0, MSG_DONTWAIT, NULL, 0) ==
- -1))
+ if (PREDICT_FALSE (
+ sendto (tx_queue->fd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1))
{
/* Uh-oh, drop & move on, but count whether it was fatal or not.
* Note that we have no reliable way to properly determine the
@@ -403,7 +437,8 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm,
}
}
- clib_spinlock_unlock_if_init (&apif->lockp);
+ if (tf->shared_queue)
+ clib_spinlock_unlock (&tx_queue->lockp);
if (PREDICT_FALSE (frame_not_ready))
vlib_error_count (vm, node->node_index,
diff --git a/src/vnet/devices/af_packet/node.c b/src/vnet/devices/af_packet/node.c
index 06012fd263a..323508bd304 100644
--- a/src/vnet/devices/af_packet/node.c
+++ b/src/vnet/devices/af_packet/node.c
@@ -30,9 +30,10 @@
#include <vnet/devices/af_packet/af_packet.h>
#include <vnet/devices/virtio/virtio_std.h>
-#define foreach_af_packet_input_error \
- _(PARTIAL_PKT, "partial packet")
-
+#define foreach_af_packet_input_error \
+ _ (PARTIAL_PKT, "partial packet") \
+ _ (TIMEDOUT_BLK, "timed out block") \
+ _ (TOTAL_RECV_BLK, "total received block")
typedef enum
{
#define _(f,s) AF_PACKET_INPUT_ERROR_##f,
@@ -51,6 +52,7 @@ typedef struct
{
u32 next_index;
u32 hw_if_index;
+ u16 queue_id;
int block;
u32 pkt_num;
void *block_start;
@@ -67,8 +69,8 @@ format_af_packet_input_trace (u8 * s, va_list * args)
af_packet_input_trace_t *t = va_arg (*args, af_packet_input_trace_t *);
u32 indent = format_get_indent (s);
- s = format (s, "af_packet: hw_if_index %d next-index %d",
- t->hw_if_index, t->next_index);
+ s = format (s, "af_packet: hw_if_index %d rx-queue %u next-index %d",
+ t->hw_if_index, t->queue_id, t->next_index);
s = format (
s, "\n%Ublock %u:\n%Uaddress %p version %u seq_num %lu pkt_num %u",
@@ -222,22 +224,25 @@ fill_cksum_offload (vlib_buffer_t *b, u8 *l4_hdr_sz, u8 is_ip)
always_inline uword
af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
vlib_frame_t *frame, af_packet_if_t *apif,
- u8 is_cksum_gso_enabled)
+ u16 queue_id, u8 is_cksum_gso_enabled)
{
af_packet_main_t *apm = &af_packet_main;
+ af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, queue_id);
tpacket3_hdr_t *tph;
u32 next_index;
u32 n_free_bufs;
u32 n_rx_packets = 0;
u32 n_rx_bytes = 0;
+ u32 timedout_blk = 0;
+ u32 total = 0;
u32 *to_next = 0;
- u32 block = apif->next_rx_block;
- u32 block_nr = apif->rx_req->tp_block_nr;
+ u32 block = rx_queue->next_rx_block;
+ u32 block_nr = rx_queue->rx_req->tp_block_nr;
u8 *block_start = 0;
uword n_trace = vlib_get_trace_count (vm, node);
u32 thread_index = vm->thread_index;
u32 n_buffer_bytes = vlib_buffer_get_default_data_size (vm);
- u32 min_bufs = apif->rx_req->tp_frame_size / n_buffer_bytes;
+ u32 min_bufs = rx_queue->rx_req->tp_frame_size / n_buffer_bytes;
u32 num_pkts = 0;
u32 rx_frame_offset = 0;
block_desc_t *bd = 0;
@@ -256,18 +261,23 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
vnet_feature_start_device_input_x1 (apif->sw_if_index, &next_index, &bt);
}
- if ((((block_desc_t *) (block_start = apif->rx_ring[block]))
+ if ((((block_desc_t *) (block_start = rx_queue->rx_ring[block]))
->hdr.bh1.block_status &
TP_STATUS_USER) != 0)
{
u32 n_required = 0;
bd = (block_desc_t *) block_start;
- if (PREDICT_FALSE (apif->ss.is_save))
+ total++;
+
+ if (TP_STATUS_BLK_TMO & bd->hdr.bh1.block_status)
+ timedout_blk++;
+
+ if (PREDICT_FALSE (rx_queue->is_rx_pending))
{
- num_pkts = apif->ss.num_pkts;
- rx_frame_offset = apif->ss.rx_frame_offset;
- apif->ss.is_save = 0;
+ num_pkts = rx_queue->num_rx_pkts;
+ rx_frame_offset = rx_queue->rx_frame_offset;
+ rx_queue->is_rx_pending = 0;
}
else
{
@@ -318,9 +328,9 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
if (PREDICT_FALSE (((data_len / n_buffer_bytes) + 1) >
vec_len (apm->rx_buffers[thread_index])))
{
- apif->ss.rx_frame_offset = rx_frame_offset;
- apif->ss.num_pkts = num_pkts;
- apif->ss.is_save = 1;
+ rx_queue->rx_frame_offset = rx_frame_offset;
+ rx_queue->num_rx_pkts = num_pkts;
+ rx_queue->is_rx_pending = 1;
vlib_put_next_frame (vm, node, next_index, n_left_to_next);
goto done;
}
@@ -450,6 +460,7 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
tr = vlib_add_trace (vm, node, first_b0, sizeof (*tr));
tr->next_index = next0;
tr->hw_if_index = apif->hw_if_index;
+ tr->queue_id = queue_id;
tr->block = block;
tr->block_start = bd;
tr->pkt_num = bd->hdr.bh1.num_pkts - num_pkts;
@@ -483,23 +494,28 @@ af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node,
}
else
{
- apif->ss.rx_frame_offset = rx_frame_offset;
- apif->ss.num_pkts = num_pkts;
- apif->ss.is_save = 1;
+ rx_queue->rx_frame_offset = rx_frame_offset;
+ rx_queue->num_rx_pkts = num_pkts;
+ rx_queue->is_rx_pending = 1;
}
}
- apif->next_rx_block = block;
+ rx_queue->next_rx_block = block;
done:
- if ((((block_desc_t *) (block_start = apif->rx_ring[block]))
+ if ((((block_desc_t *) (block_start = rx_queue->rx_ring[block]))
->hdr.bh1.block_status &
TP_STATUS_USER) != 0)
vlib_node_set_state (vm, node->node_index, VLIB_NODE_STATE_POLLING);
else
vlib_node_set_state (vm, node->node_index, VLIB_NODE_STATE_INTERRUPT);
+ vlib_error_count (vm, node->node_index, AF_PACKET_INPUT_ERROR_TOTAL_RECV_BLK,
+ total);
+ vlib_error_count (vm, node->node_index, AF_PACKET_INPUT_ERROR_TIMEDOUT_BLK,
+ timedout_blk);
+
vlib_increment_combined_counter
(vnet_get_main ()->interface_main.combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_RX,
@@ -524,11 +540,11 @@ VLIB_NODE_FN (af_packet_input_node) (vlib_main_t * vm,
if (apif->is_admin_up)
{
if (apif->is_cksum_gso_enabled)
- n_rx_packets +=
- af_packet_device_input_fn (vm, node, frame, apif, 1);
+ n_rx_packets += af_packet_device_input_fn (vm, node, frame, apif,
+ pv[i].queue_id, 1);
else
- n_rx_packets +=
- af_packet_device_input_fn (vm, node, frame, apif, 0);
+ n_rx_packets += af_packet_device_input_fn (vm, node, frame, apif,
+ pv[i].queue_id, 0);
}
}
return n_rx_packets;