aboutsummaryrefslogtreecommitdiffstats
path: root/src/plugins/dpdk/hqos
diff options
context:
space:
mode:
authorDamjan Marion <damarion@cisco.com>2017-02-28 19:22:22 +0100
committerDave Barach <openvpp@barachs.net>2017-03-01 22:06:31 +0000
commitc3a814be9dc769be942ff8029c7b6eccd4b3af05 (patch)
treedd152b3e92281ab52f806ed2ad237c9fb0607e8b /src/plugins/dpdk/hqos
parentf71ef1dddfc7cf34944c510b85bec074f431bb12 (diff)
dpdk: be a plugin
Change-Id: I238258cdeb77035adc5e88903d824593d0a1da90 Signed-off-by: Damjan Marion <damarion@cisco.com>
Diffstat (limited to 'src/plugins/dpdk/hqos')
-rw-r--r--src/plugins/dpdk/hqos/hqos.c775
-rw-r--r--src/plugins/dpdk/hqos/qos_doc.md411
2 files changed, 1186 insertions, 0 deletions
diff --git a/src/plugins/dpdk/hqos/hqos.c b/src/plugins/dpdk/hqos/hqos.c
new file mode 100644
index 00000000000..a288fca7e5e
--- /dev/null
+++ b/src/plugins/dpdk/hqos/hqos.c
@@ -0,0 +1,775 @@
+/*
+ * Copyright(c) 2016 Intel Corporation. All rights reserved.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <sys/mount.h>
+#include <string.h>
+#include <fcntl.h>
+
+#include <vppinfra/vec.h>
+#include <vppinfra/error.h>
+#include <vppinfra/format.h>
+#include <vppinfra/bitmap.h>
+
+#include <vnet/vnet.h>
+#include <vnet/ethernet/ethernet.h>
+#include <dpdk/device/dpdk.h>
+
+#include <vlib/unix/physmem.h>
+#include <vlib/pci/pci.h>
+#include <vlibmemory/api.h>
+#include <vlibmemory/vl_memory_msg_enum.h> /* enumerate all vlib messages */
+
+#define vl_typedefs /* define message structures */
+#include <vlibmemory/vl_memory_api_h.h>
+#undef vl_typedefs
+
+/* instantiate all the print functions we know about */
+#define vl_print(handle, ...) vlib_cli_output (handle, __VA_ARGS__)
+#define vl_printfun
+#include <vlibmemory/vl_memory_api_h.h>
+#undef vl_printfun
+
+#include <dpdk/device/dpdk_priv.h>
+
+dpdk_main_t dpdk_main;
+
+/***
+ *
+ * HQoS default configuration values
+ *
+ ***/
+
+static dpdk_device_config_hqos_t hqos_params_default = {
+ .hqos_thread_valid = 0,
+
+ .swq_size = 4096,
+ .burst_enq = 256,
+ .burst_deq = 220,
+
+ /*
+ * Packet field to identify the subport.
+ *
+ * Default value: Since only one subport is defined by default (see below:
+ * n_subports_per_port = 1), the subport ID is hardcoded to 0.
+ */
+ .pktfield0_slabpos = 0,
+ .pktfield0_slabmask = 0,
+
+ /*
+ * Packet field to identify the pipe.
+ *
+ * Default value: Assuming Ethernet/IPv4/UDP packets, UDP payload bits 12 .. 23
+ */
+ .pktfield1_slabpos = 40,
+ .pktfield1_slabmask = 0x0000000FFF000000LLU,
+
+ /* Packet field used as index into TC translation table to identify the traffic
+ * class and queue.
+ *
+ * Default value: Assuming Ethernet/IPv4 packets, IPv4 DSCP field
+ */
+ .pktfield2_slabpos = 8,
+ .pktfield2_slabmask = 0x00000000000000FCLLU,
+ .tc_table = {
+ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+ },
+
+ /* port */
+ .port = {
+ .name = NULL, /* Set at init */
+ .socket = 0, /* Set at init */
+ .rate = 1250000000, /* Assuming 10GbE port */
+ .mtu = 14 + 1500, /* Assuming Ethernet/IPv4 pkt (Ethernet FCS not included) */
+ .frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
+ .n_subports_per_port = 1,
+ .n_pipes_per_subport = 4096,
+ .qsize = {64, 64, 64, 64},
+ .pipe_profiles = NULL, /* Set at config */
+ .n_pipe_profiles = 1,
+
+#ifdef RTE_SCHED_RED
+ .red_params = {
+ /* Traffic Class 0 Colors Green / Yellow / Red */
+ [0][0] = {.min_th = 48,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [0][1] = {.min_th = 40,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [0][2] = {.min_th = 32,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+
+ /* Traffic Class 1 - Colors Green / Yellow / Red */
+ [1][0] = {.min_th = 48,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [1][1] = {.min_th = 40,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [1][2] = {.min_th = 32,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+
+ /* Traffic Class 2 - Colors Green / Yellow / Red */
+ [2][0] = {.min_th = 48,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [2][1] = {.min_th = 40,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [2][2] = {.min_th = 32,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+
+ /* Traffic Class 3 - Colors Green / Yellow / Red */
+ [3][0] = {.min_th = 48,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [3][1] = {.min_th = 40,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9},
+ [3][2] = {.min_th = 32,.max_th = 64,.maxp_inv =
+ 10,.wq_log2 = 9}
+ },
+#endif /* RTE_SCHED_RED */
+ },
+};
+
+static struct rte_sched_subport_params hqos_subport_params_default = {
+ .tb_rate = 1250000000, /* 10GbE line rate (measured in bytes/second) */
+ .tb_size = 1000000,
+ .tc_rate = {1250000000, 1250000000, 1250000000, 1250000000},
+ .tc_period = 10,
+};
+
+static struct rte_sched_pipe_params hqos_pipe_params_default = {
+ .tb_rate = 305175, /* 10GbE line rate divided by 4K pipes */
+ .tb_size = 1000000,
+ .tc_rate = {305175, 305175, 305175, 305175},
+ .tc_period = 40,
+#ifdef RTE_SCHED_SUBPORT_TC_OV
+ .tc_ov_weight = 1,
+#endif
+ .wrr_weights = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1},
+};
+
+/***
+ *
+ * HQoS configuration
+ *
+ ***/
+
+int
+dpdk_hqos_validate_mask (u64 mask, u32 n)
+{
+ int count = __builtin_popcountll (mask);
+ int pos_lead = sizeof (u64) * 8 - __builtin_clzll (mask);
+ int pos_trail = __builtin_ctzll (mask);
+ int count_expected = __builtin_popcount (n - 1);
+
+ /* Handle the exceptions */
+ if (n == 0)
+ return -1; /* Error */
+
+ if ((mask == 0) && (n == 1))
+ return 0; /* OK */
+
+ if (((mask == 0) && (n != 1)) || ((mask != 0) && (n == 1)))
+ return -2; /* Error */
+
+ /* Check that mask is contiguous */
+ if ((pos_lead - pos_trail) != count)
+ return -3; /* Error */
+
+ /* Check that mask contains the expected number of bits set */
+ if (count != count_expected)
+ return -4; /* Error */
+
+ return 0; /* OK */
+}
+
+void
+dpdk_device_config_hqos_pipe_profile_default (dpdk_device_config_hqos_t *
+ hqos, u32 pipe_profile_id)
+{
+ memcpy (&hqos->pipe[pipe_profile_id], &hqos_pipe_params_default,
+ sizeof (hqos_pipe_params_default));
+}
+
+void
+dpdk_device_config_hqos_default (dpdk_device_config_hqos_t * hqos)
+{
+ struct rte_sched_subport_params *subport_params;
+ struct rte_sched_pipe_params *pipe_params;
+ u32 *pipe_map;
+ u32 i;
+
+ memcpy (hqos, &hqos_params_default, sizeof (hqos_params_default));
+
+ /* pipe */
+ vec_add2 (hqos->pipe, pipe_params, hqos->port.n_pipe_profiles);
+
+ for (i = 0; i < vec_len (hqos->pipe); i++)
+ memcpy (&pipe_params[i],
+ &hqos_pipe_params_default, sizeof (hqos_pipe_params_default));
+
+ hqos->port.pipe_profiles = hqos->pipe;
+
+ /* subport */
+ vec_add2 (hqos->subport, subport_params, hqos->port.n_subports_per_port);
+
+ for (i = 0; i < vec_len (hqos->subport); i++)
+ memcpy (&subport_params[i],
+ &hqos_subport_params_default,
+ sizeof (hqos_subport_params_default));
+
+ /* pipe profile */
+ vec_add2 (hqos->pipe_map,
+ pipe_map,
+ hqos->port.n_subports_per_port * hqos->port.n_pipes_per_subport);
+
+ for (i = 0; i < vec_len (hqos->pipe_map); i++)
+ pipe_map[i] = 0;
+}
+
+/***
+ *
+ * HQoS init
+ *
+ ***/
+
+clib_error_t *
+dpdk_port_setup_hqos (dpdk_device_t * xd, dpdk_device_config_hqos_t * hqos)
+{
+ vlib_thread_main_t *tm = vlib_get_thread_main ();
+ char name[32];
+ u32 subport_id, i;
+ int rv;
+
+ /* Detect the set of worker threads */
+ int worker_thread_first = 0;
+ int worker_thread_count = 0;
+
+ uword *p = hash_get_mem (tm->thread_registrations_by_name, "workers");
+ vlib_thread_registration_t *tr =
+ p ? (vlib_thread_registration_t *) p[0] : 0;
+
+ if (tr && tr->count > 0)
+ {
+ worker_thread_first = tr->first_index;
+ worker_thread_count = tr->count;
+ }
+
+ /* Allocate the per-thread device data array */
+ vec_validate_aligned (xd->hqos_wt, tm->n_vlib_mains - 1,
+ CLIB_CACHE_LINE_BYTES);
+ memset (xd->hqos_wt, 0, tm->n_vlib_mains * sizeof (xd->hqos_wt[0]));
+
+ vec_validate_aligned (xd->hqos_ht, 0, CLIB_CACHE_LINE_BYTES);
+ memset (xd->hqos_ht, 0, sizeof (xd->hqos_ht[0]));
+
+ /* Allocate space for one SWQ per worker thread in the I/O TX thread data structure */
+ vec_validate (xd->hqos_ht->swq, worker_thread_count);
+
+ /* SWQ */
+ for (i = 0; i < worker_thread_count + 1; i++)
+ {
+ u32 swq_flags = RING_F_SP_ENQ | RING_F_SC_DEQ;
+
+ snprintf (name, sizeof (name), "SWQ-worker%u-to-device%u", i,
+ xd->device_index);
+ xd->hqos_ht->swq[i] =
+ rte_ring_create (name, hqos->swq_size, xd->cpu_socket, swq_flags);
+ if (xd->hqos_ht->swq[i] == NULL)
+ return clib_error_return (0,
+ "SWQ-worker%u-to-device%u: rte_ring_create err",
+ i, xd->device_index);
+ }
+
+ /*
+ * HQoS
+ */
+
+ /* HQoS port */
+ snprintf (name, sizeof (name), "HQoS%u", xd->device_index);
+ hqos->port.name = strdup (name);
+ if (hqos->port.name == NULL)
+ return clib_error_return (0, "HQoS%u: strdup err", xd->device_index);
+
+ hqos->port.socket = rte_eth_dev_socket_id (xd->device_index);
+ if (hqos->port.socket == SOCKET_ID_ANY)
+ hqos->port.socket = 0;
+
+ xd->hqos_ht->hqos = rte_sched_port_config (&hqos->port);
+ if (xd->hqos_ht->hqos == NULL)
+ return clib_error_return (0, "HQoS%u: rte_sched_port_config err",
+ xd->device_index);
+
+ /* HQoS subport */
+ for (subport_id = 0; subport_id < hqos->port.n_subports_per_port;
+ subport_id++)
+ {
+ u32 pipe_id;
+
+ rv =
+ rte_sched_subport_config (xd->hqos_ht->hqos, subport_id,
+ &hqos->subport[subport_id]);
+ if (rv)
+ return clib_error_return (0,
+ "HQoS%u subport %u: rte_sched_subport_config err (%d)",
+ xd->device_index, subport_id, rv);
+
+ /* HQoS pipe */
+ for (pipe_id = 0; pipe_id < hqos->port.n_pipes_per_subport; pipe_id++)
+ {
+ u32 pos = subport_id * hqos->port.n_pipes_per_subport + pipe_id;
+ u32 profile_id = hqos->pipe_map[pos];
+
+ rv =
+ rte_sched_pipe_config (xd->hqos_ht->hqos, subport_id, pipe_id,
+ profile_id);
+ if (rv)
+ return clib_error_return (0,
+ "HQoS%u subport %u pipe %u: rte_sched_pipe_config err (%d)",
+ xd->device_index, subport_id, pipe_id,
+ rv);
+ }
+ }
+
+ /* Set up per-thread device data for the I/O TX thread */
+ xd->hqos_ht->hqos_burst_enq = hqos->burst_enq;
+ xd->hqos_ht->hqos_burst_deq = hqos->burst_deq;
+ vec_validate (xd->hqos_ht->pkts_enq, 2 * hqos->burst_enq - 1);
+ vec_validate (xd->hqos_ht->pkts_deq, hqos->burst_deq - 1);
+ xd->hqos_ht->pkts_enq_len = 0;
+ xd->hqos_ht->swq_pos = 0;
+ xd->hqos_ht->flush_count = 0;
+
+ /* Set up per-thread device data for each worker thread */
+ for (i = 0; i < worker_thread_count + 1; i++)
+ {
+ u32 tid;
+ if (i)
+ tid = worker_thread_first + (i - 1);
+ else
+ tid = i;
+
+ xd->hqos_wt[tid].swq = xd->hqos_ht->swq[i];
+ xd->hqos_wt[tid].hqos_field0_slabpos = hqos->pktfield0_slabpos;
+ xd->hqos_wt[tid].hqos_field0_slabmask = hqos->pktfield0_slabmask;
+ xd->hqos_wt[tid].hqos_field0_slabshr =
+ __builtin_ctzll (hqos->pktfield0_slabmask);
+ xd->hqos_wt[tid].hqos_field1_slabpos = hqos->pktfield1_slabpos;
+ xd->hqos_wt[tid].hqos_field1_slabmask = hqos->pktfield1_slabmask;
+ xd->hqos_wt[tid].hqos_field1_slabshr =
+ __builtin_ctzll (hqos->pktfield1_slabmask);
+ xd->hqos_wt[tid].hqos_field2_slabpos = hqos->pktfield2_slabpos;
+ xd->hqos_wt[tid].hqos_field2_slabmask = hqos->pktfield2_slabmask;
+ xd->hqos_wt[tid].hqos_field2_slabshr =
+ __builtin_ctzll (hqos->pktfield2_slabmask);
+ memcpy (xd->hqos_wt[tid].hqos_tc_table, hqos->tc_table,
+ sizeof (hqos->tc_table));
+ }
+
+ return 0;
+}
+
+/***
+ *
+ * HQoS run-time
+ *
+ ***/
+/*
+ * dpdk_hqos_thread - Contains the main loop of an HQoS thread.
+ *
+ * w
+ * Information for the current thread
+ */
+static_always_inline void
+dpdk_hqos_thread_internal_hqos_dbg_bypass (vlib_main_t * vm)
+{
+ dpdk_main_t *dm = &dpdk_main;
+ u32 cpu_index = vm->cpu_index;
+ u32 dev_pos;
+
+ dev_pos = 0;
+ while (1)
+ {
+ vlib_worker_thread_barrier_check ();
+
+ u32 n_devs = vec_len (dm->devices_by_hqos_cpu[cpu_index]);
+ if (dev_pos >= n_devs)
+ dev_pos = 0;
+
+ dpdk_device_and_queue_t *dq =
+ vec_elt_at_index (dm->devices_by_hqos_cpu[cpu_index], dev_pos);
+ dpdk_device_t *xd = vec_elt_at_index (dm->devices, dq->device);
+
+ dpdk_device_hqos_per_hqos_thread_t *hqos = xd->hqos_ht;
+ u32 device_index = xd->device_index;
+ u16 queue_id = dq->queue_id;
+
+ struct rte_mbuf **pkts_enq = hqos->pkts_enq;
+ u32 pkts_enq_len = hqos->pkts_enq_len;
+ u32 swq_pos = hqos->swq_pos;
+ u32 n_swq = vec_len (hqos->swq), i;
+ u32 flush_count = hqos->flush_count;
+
+ for (i = 0; i < n_swq; i++)
+ {
+ /* Get current SWQ for this device */
+ struct rte_ring *swq = hqos->swq[swq_pos];
+
+ /* Read SWQ burst to packet buffer of this device */
+ pkts_enq_len += rte_ring_sc_dequeue_burst (swq,
+ (void **)
+ &pkts_enq[pkts_enq_len],
+ hqos->hqos_burst_enq);
+
+ /* Get next SWQ for this device */
+ swq_pos++;
+ if (swq_pos >= n_swq)
+ swq_pos = 0;
+ hqos->swq_pos = swq_pos;
+
+ /* HWQ TX enqueue when burst available */
+ if (pkts_enq_len >= hqos->hqos_burst_enq)
+ {
+ u32 n_pkts = rte_eth_tx_burst (device_index,
+ (uint16_t) queue_id,
+ pkts_enq,
+ (uint16_t) pkts_enq_len);
+
+ for (; n_pkts < pkts_enq_len; n_pkts++)
+ rte_pktmbuf_free (pkts_enq[n_pkts]);
+
+ pkts_enq_len = 0;
+ flush_count = 0;
+ break;
+ }
+ }
+ if (pkts_enq_len)
+ {
+ flush_count++;
+ if (PREDICT_FALSE (flush_count == HQOS_FLUSH_COUNT_THRESHOLD))
+ {
+ rte_sched_port_enqueue (hqos->hqos, pkts_enq, pkts_enq_len);
+
+ pkts_enq_len = 0;
+ flush_count = 0;
+ }
+ }
+ hqos->pkts_enq_len = pkts_enq_len;
+ hqos->flush_count = flush_count;
+
+ /* Advance to next device */
+ dev_pos++;
+ }
+}
+
+static_always_inline void
+dpdk_hqos_thread_internal (vlib_main_t * vm)
+{
+ dpdk_main_t *dm = &dpdk_main;
+ u32 cpu_index = vm->cpu_index;
+ u32 dev_pos;
+
+ dev_pos = 0;
+ while (1)
+ {
+ vlib_worker_thread_barrier_check ();
+
+ u32 n_devs = vec_len (dm->devices_by_hqos_cpu[cpu_index]);
+ if (PREDICT_FALSE (n_devs == 0))
+ {
+ dev_pos = 0;
+ continue;
+ }
+ if (dev_pos >= n_devs)
+ dev_pos = 0;
+
+ dpdk_device_and_queue_t *dq =
+ vec_elt_at_index (dm->devices_by_hqos_cpu[cpu_index], dev_pos);
+ dpdk_device_t *xd = vec_elt_at_index (dm->devices, dq->device);
+
+ dpdk_device_hqos_per_hqos_thread_t *hqos = xd->hqos_ht;
+ u32 device_index = xd->device_index;
+ u16 queue_id = dq->queue_id;
+
+ struct rte_mbuf **pkts_enq = hqos->pkts_enq;
+ struct rte_mbuf **pkts_deq = hqos->pkts_deq;
+ u32 pkts_enq_len = hqos->pkts_enq_len;
+ u32 swq_pos = hqos->swq_pos;
+ u32 n_swq = vec_len (hqos->swq), i;
+ u32 flush_count = hqos->flush_count;
+
+ /*
+ * SWQ dequeue and HQoS enqueue for current device
+ */
+ for (i = 0; i < n_swq; i++)
+ {
+ /* Get current SWQ for this device */
+ struct rte_ring *swq = hqos->swq[swq_pos];
+
+ /* Read SWQ burst to packet buffer of this device */
+ pkts_enq_len += rte_ring_sc_dequeue_burst (swq,
+ (void **)
+ &pkts_enq[pkts_enq_len],
+ hqos->hqos_burst_enq);
+
+ /* Get next SWQ for this device */
+ swq_pos++;
+ if (swq_pos >= n_swq)
+ swq_pos = 0;
+ hqos->swq_pos = swq_pos;
+
+ /* HQoS enqueue when burst available */
+ if (pkts_enq_len >= hqos->hqos_burst_enq)
+ {
+ rte_sched_port_enqueue (hqos->hqos, pkts_enq, pkts_enq_len);
+
+ pkts_enq_len = 0;
+ flush_count = 0;
+ break;
+ }
+ }
+ if (pkts_enq_len)
+ {
+ flush_count++;
+ if (PREDICT_FALSE (flush_count == HQOS_FLUSH_COUNT_THRESHOLD))
+ {
+ rte_sched_port_enqueue (hqos->hqos, pkts_enq, pkts_enq_len);
+
+ pkts_enq_len = 0;
+ flush_count = 0;
+ }
+ }
+ hqos->pkts_enq_len = pkts_enq_len;
+ hqos->flush_count = flush_count;
+
+ /*
+ * HQoS dequeue and HWQ TX enqueue for current device
+ */
+ {
+ u32 pkts_deq_len, n_pkts;
+
+ pkts_deq_len = rte_sched_port_dequeue (hqos->hqos,
+ pkts_deq,
+ hqos->hqos_burst_deq);
+
+ for (n_pkts = 0; n_pkts < pkts_deq_len;)
+ n_pkts += rte_eth_tx_burst (device_index,
+ (uint16_t) queue_id,
+ &pkts_deq[n_pkts],
+ (uint16_t) (pkts_deq_len - n_pkts));
+ }
+
+ /* Advance to next device */
+ dev_pos++;
+ }
+}
+
+void
+dpdk_hqos_thread (vlib_worker_thread_t * w)
+{
+ vlib_main_t *vm;
+ vlib_thread_main_t *tm = vlib_get_thread_main ();
+ dpdk_main_t *dm = &dpdk_main;
+
+ vm = vlib_get_main ();
+
+ ASSERT (vm->cpu_index == os_get_cpu_number ());
+
+ clib_time_init (&vm->clib_time);
+ clib_mem_set_heap (w->thread_mheap);
+
+ /* Wait until the dpdk init sequence is complete */
+ while (tm->worker_thread_release == 0)
+ vlib_worker_thread_barrier_check ();
+
+ if (vec_len (dm->devices_by_hqos_cpu[vm->cpu_index]) == 0)
+ return
+ clib_error
+ ("current I/O TX thread does not have any devices assigned to it");
+
+ if (DPDK_HQOS_DBG_BYPASS)
+ dpdk_hqos_thread_internal_hqos_dbg_bypass (vm);
+ else
+ dpdk_hqos_thread_internal (vm);
+}
+
+void
+dpdk_hqos_thread_fn (void *arg)
+{
+ vlib_worker_thread_t *w = (vlib_worker_thread_t *) arg;
+ vlib_worker_thread_init (w);
+ dpdk_hqos_thread (w);
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_THREAD (hqos_thread_reg, static) =
+{
+ .name = "hqos-threads",
+ .short_name = "hqos-threads",
+ .function = dpdk_hqos_thread_fn,
+};
+/* *INDENT-ON* */
+
+/*
+ * HQoS run-time code to be called by the worker threads
+ */
+#define BITFIELD(byte_array, slab_pos, slab_mask, slab_shr) \
+({ \
+ u64 slab = *((u64 *) &byte_array[slab_pos]); \
+ u64 val = (rte_be_to_cpu_64(slab) & slab_mask) >> slab_shr; \
+ val; \
+})
+
+#define RTE_SCHED_PORT_HIERARCHY(subport, pipe, traffic_class, queue, color) \
+ ((((u64) (queue)) & 0x3) | \
+ ((((u64) (traffic_class)) & 0x3) << 2) | \
+ ((((u64) (color)) & 0x3) << 4) | \
+ ((((u64) (subport)) & 0xFFFF) << 16) | \
+ ((((u64) (pipe)) & 0xFFFFFFFF) << 32))
+
+void
+dpdk_hqos_metadata_set (dpdk_device_hqos_per_worker_thread_t * hqos,
+ struct rte_mbuf **pkts, u32 n_pkts)
+{
+ u32 i;
+
+ for (i = 0; i < (n_pkts & (~0x3)); i += 4)
+ {
+ struct rte_mbuf *pkt0 = pkts[i];
+ struct rte_mbuf *pkt1 = pkts[i + 1];
+ struct rte_mbuf *pkt2 = pkts[i + 2];
+ struct rte_mbuf *pkt3 = pkts[i + 3];
+
+ u8 *pkt0_data = rte_pktmbuf_mtod (pkt0, u8 *);
+ u8 *pkt1_data = rte_pktmbuf_mtod (pkt1, u8 *);
+ u8 *pkt2_data = rte_pktmbuf_mtod (pkt2, u8 *);
+ u8 *pkt3_data = rte_pktmbuf_mtod (pkt3, u8 *);
+
+ u64 pkt0_subport = BITFIELD (pkt0_data, hqos->hqos_field0_slabpos,
+ hqos->hqos_field0_slabmask,
+ hqos->hqos_field0_slabshr);
+ u64 pkt0_pipe = BITFIELD (pkt0_data, hqos->hqos_field1_slabpos,
+ hqos->hqos_field1_slabmask,
+ hqos->hqos_field1_slabshr);
+ u64 pkt0_dscp = BITFIELD (pkt0_data, hqos->hqos_field2_slabpos,
+ hqos->hqos_field2_slabmask,
+ hqos->hqos_field2_slabshr);
+ u32 pkt0_tc = hqos->hqos_tc_table[pkt0_dscp & 0x3F] >> 2;
+ u32 pkt0_tc_q = hqos->hqos_tc_table[pkt0_dscp & 0x3F] & 0x3;
+
+ u64 pkt1_subport = BITFIELD (pkt1_data, hqos->hqos_field0_slabpos,
+ hqos->hqos_field0_slabmask,
+ hqos->hqos_field0_slabshr);
+ u64 pkt1_pipe = BITFIELD (pkt1_data, hqos->hqos_field1_slabpos,
+ hqos->hqos_field1_slabmask,
+ hqos->hqos_field1_slabshr);
+ u64 pkt1_dscp = BITFIELD (pkt1_data, hqos->hqos_field2_slabpos,
+ hqos->hqos_field2_slabmask,
+ hqos->hqos_field2_slabshr);
+ u32 pkt1_tc = hqos->hqos_tc_table[pkt1_dscp & 0x3F] >> 2;
+ u32 pkt1_tc_q = hqos->hqos_tc_table[pkt1_dscp & 0x3F] & 0x3;
+
+ u64 pkt2_subport = BITFIELD (pkt2_data, hqos->hqos_field0_slabpos,
+ hqos->hqos_field0_slabmask,
+ hqos->hqos_field0_slabshr);
+ u64 pkt2_pipe = BITFIELD (pkt2_data, hqos->hqos_field1_slabpos,
+ hqos->hqos_field1_slabmask,
+ hqos->hqos_field1_slabshr);
+ u64 pkt2_dscp = BITFIELD (pkt2_data, hqos->hqos_field2_slabpos,
+ hqos->hqos_field2_slabmask,
+ hqos->hqos_field2_slabshr);
+ u32 pkt2_tc = hqos->hqos_tc_table[pkt2_dscp & 0x3F] >> 2;
+ u32 pkt2_tc_q = hqos->hqos_tc_table[pkt2_dscp & 0x3F] & 0x3;
+
+ u64 pkt3_subport = BITFIELD (pkt3_data, hqos->hqos_field0_slabpos,
+ hqos->hqos_field0_slabmask,
+ hqos->hqos_field0_slabshr);
+ u64 pkt3_pipe = BITFIELD (pkt3_data, hqos->hqos_field1_slabpos,
+ hqos->hqos_field1_slabmask,
+ hqos->hqos_field1_slabshr);
+ u64 pkt3_dscp = BITFIELD (pkt3_data, hqos->hqos_field2_slabpos,
+ hqos->hqos_field2_slabmask,
+ hqos->hqos_field2_slabshr);
+ u32 pkt3_tc = hqos->hqos_tc_table[pkt3_dscp & 0x3F] >> 2;
+ u32 pkt3_tc_q = hqos->hqos_tc_table[pkt3_dscp & 0x3F] & 0x3;
+
+ u64 pkt0_sched = RTE_SCHED_PORT_HIERARCHY (pkt0_subport,
+ pkt0_pipe,
+ pkt0_tc,
+ pkt0_tc_q,
+ 0);
+ u64 pkt1_sched = RTE_SCHED_PORT_HIERARCHY (pkt1_subport,
+ pkt1_pipe,
+ pkt1_tc,
+ pkt1_tc_q,
+ 0);
+ u64 pkt2_sched = RTE_SCHED_PORT_HIERARCHY (pkt2_subport,
+ pkt2_pipe,
+ pkt2_tc,
+ pkt2_tc_q,
+ 0);
+ u64 pkt3_sched = RTE_SCHED_PORT_HIERARCHY (pkt3_subport,
+ pkt3_pipe,
+ pkt3_tc,
+ pkt3_tc_q,
+ 0);
+
+ pkt0->hash.sched.lo = pkt0_sched & 0xFFFFFFFF;
+ pkt0->hash.sched.hi = pkt0_sched >> 32;
+ pkt1->hash.sched.lo = pkt1_sched & 0xFFFFFFFF;
+ pkt1->hash.sched.hi = pkt1_sched >> 32;
+ pkt2->hash.sched.lo = pkt2_sched & 0xFFFFFFFF;
+ pkt2->hash.sched.hi = pkt2_sched >> 32;
+ pkt3->hash.sched.lo = pkt3_sched & 0xFFFFFFFF;
+ pkt3->hash.sched.hi = pkt3_sched >> 32;
+ }
+
+ for (; i < n_pkts; i++)
+ {
+ struct rte_mbuf *pkt = pkts[i];
+
+ u8 *pkt_data = rte_pktmbuf_mtod (pkt, u8 *);
+
+ u64 pkt_subport = BITFIELD (pkt_data, hqos->hqos_field0_slabpos,
+ hqos->hqos_field0_slabmask,
+ hqos->hqos_field0_slabshr);
+ u64 pkt_pipe = BITFIELD (pkt_data, hqos->hqos_field1_slabpos,
+ hqos->hqos_field1_slabmask,
+ hqos->hqos_field1_slabshr);
+ u64 pkt_dscp = BITFIELD (pkt_data, hqos->hqos_field2_slabpos,
+ hqos->hqos_field2_slabmask,
+ hqos->hqos_field2_slabshr);
+ u32 pkt_tc = hqos->hqos_tc_table[pkt_dscp & 0x3F] >> 2;
+ u32 pkt_tc_q = hqos->hqos_tc_table[pkt_dscp & 0x3F] & 0x3;
+
+ u64 pkt_sched = RTE_SCHED_PORT_HIERARCHY (pkt_subport,
+ pkt_pipe,
+ pkt_tc,
+ pkt_tc_q,
+ 0);
+
+ pkt->hash.sched.lo = pkt_sched & 0xFFFFFFFF;
+ pkt->hash.sched.hi = pkt_sched >> 32;
+ }
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/dpdk/hqos/qos_doc.md b/src/plugins/dpdk/hqos/qos_doc.md
new file mode 100644
index 00000000000..7c064246e60
--- /dev/null
+++ b/src/plugins/dpdk/hqos/qos_doc.md
@@ -0,0 +1,411 @@
+# QoS Hierarchical Scheduler {#qos_doc}
+
+The Quality-of-Service (QoS) scheduler performs egress-traffic management by
+prioritizing the transmission of the packets of different type services and
+subcribers based on the Service Level Agreements (SLAs). The QoS scheduler can
+be enabled on one or more NIC output interfaces depending upon the
+requirement.
+
+
+## Overview
+
+The QoS schdeuler supports a number of scheduling and shaping levels which
+construct hierarchical-tree. The first level in the hierarchy is port (i.e.
+the physical interface) that constitutes the root node of the tree. The
+subsequent level is subport which represents the group of the
+users/subscribers. The individual user/subscriber is represented by the pipe
+at the next level. Each user can have different traffic type based on the
+criteria of specific loss rate, jitter, and latency. These traffic types are
+represented at the traffic-class level in the form of different traffic-
+classes. The last level contains number of queues which are grouped together
+to host the packets of the specific class type traffic.
+
+The QoS scheduler implementation requires flow classification, enqueue and
+dequeue operations. The flow classification is mandatory stage for HQoS where
+incoming packets are classified by mapping the packet fields information to
+5-tuple (HQoS subport, pipe, traffic class, queue within traffic class, and
+color) and storing that information in mbuf sched field. The enqueue operation
+uses this information to determine the queue for storing the packet, and at
+this stage, if the specific queue is full, QoS drops the packet. The dequeue
+operation consists of scheduling the packet based on its length and available
+credits, and handing over the scheduled packet to the output interface.
+
+For more information on QoS Scheduler, please refer DPDK Programmer's Guide-
+http://dpdk.org/doc/guides/prog_guide/qos_framework.html
+
+
+### QoS Schdeuler Parameters
+
+Following illustrates the default HQoS configuration for each 10GbE output
+port:
+
+Single subport (subport 0):
+ - Subport rate set to 100% of port rate
+ - Each of the 4 traffic classes has rate set to 100% of port rate
+
+4K pipes per subport 0 (pipes 0 .. 4095) with identical configuration:
+ - Pipe rate set to 1/4K of port rate
+ - Each of the 4 traffic classes has rate set to 100% of pipe rate
+ - Within each traffic class, the byte-level WRR weights for the 4 queues are set to 1:1:1:1
+
+
+#### Port configuration
+
+```
+port {
+ rate 1250000000 /* Assuming 10GbE port */
+ frame_overhead 24 /* Overhead fields per Ethernet frame:
+ * 7B (Preamble) +
+ * 1B (Start of Frame Delimiter (SFD)) +
+ * 4B (Frame Check Sequence (FCS)) +
+ * 12B (Inter Frame Gap (IFG))
+ */
+ mtu 1522 /* Assuming Ethernet/IPv4 pkt (FCS not included) */
+ n_subports_per_port 1 /* Number of subports per output interface */
+ n_pipes_per_subport 4096 /* Number of pipes (users/subscribers) */
+ queue_sizes 64 64 64 64 /* Packet queue size for each traffic class.
+ * All queues within the same pipe traffic class
+ * have the same size. Queues from different
+ * pipes serving the same traffic class have
+ * the same size. */
+}
+```
+
+
+#### Subport configuration
+
+```
+subport 0 {
+ tb_rate 1250000000 /* Subport level token bucket rate (bytes per second) */
+ tb_size 1000000 /* Subport level token bucket size (bytes) */
+ tc0_rate 1250000000 /* Subport level token bucket rate for traffic class 0 (bytes per second) */
+ tc1_rate 1250000000 /* Subport level token bucket rate for traffic class 1 (bytes per second) */
+ tc2_rate 1250000000 /* Subport level token bucket rate for traffic class 2 (bytes per second) */
+ tc3_rate 1250000000 /* Subport level token bucket rate for traffic class 3 (bytes per second) */
+ tc_period 10 /* Time interval for refilling the token bucket associated with traffic class (Milliseconds) */
+ pipe 0 4095 profile 0 /* pipes (users/subscribers) configured with pipe profile 0 */
+}
+```
+
+
+#### Pipe configuration
+
+```
+pipe_profile 0 {
+ tb_rate 305175 /* Pipe level token bucket rate (bytes per second) */
+ tb_size 1000000 /* Pipe level token bucket size (bytes) */
+ tc0_rate 305175 /* Pipe level token bucket rate for traffic class 0 (bytes per second) */
+ tc1_rate 305175 /* Pipe level token bucket rate for traffic class 1 (bytes per second) */
+ tc2_rate 305175 /* Pipe level token bucket rate for traffic class 2 (bytes per second) */
+ tc3_rate 305175 /* Pipe level token bucket rate for traffic class 3 (bytes per second) */
+ tc_period 40 /* Time interval for refilling the token bucket associated with traffic class at pipe level (Milliseconds) */
+ tc3_oversubscription_weight 1 /* Weight traffic class 3 oversubscription */
+ tc0_wrr_weights 1 1 1 1 /* Pipe queues WRR weights for traffic class 0 */
+ tc1_wrr_weights 1 1 1 1 /* Pipe queues WRR weights for traffic class 1 */
+ tc2_wrr_weights 1 1 1 1 /* Pipe queues WRR weights for traffic class 2 */
+ tc3_wrr_weights 1 1 1 1 /* Pipe queues WRR weights for traffic class 3 */
+}
+```
+
+
+#### Random Early Detection (RED) parameters per traffic class and color (Green / Yellow / Red)
+
+```
+red {
+ tc0_wred_min 48 40 32 /* Minimum threshold for traffic class 0 queue (min_th) in number of packets */
+ tc0_wred_max 64 64 64 /* Maximum threshold for traffic class 0 queue (max_th) in number of packets */
+ tc0_wred_inv_prob 10 10 10 /* Inverse of packet marking probability for traffic class 0 queue (maxp = 1 / maxp_inv) */
+ tc0_wred_weight 9 9 9 /* Traffic Class 0 queue weight */
+ tc1_wred_min 48 40 32 /* Minimum threshold for traffic class 1 queue (min_th) in number of packets */
+ tc1_wred_max 64 64 64 /* Maximum threshold for traffic class 1 queue (max_th) in number of packets */
+ tc1_wred_inv_prob 10 10 10 /* Inverse of packet marking probability for traffic class 1 queue (maxp = 1 / maxp_inv) */
+ tc1_wred_weight 9 9 9 /* Traffic Class 1 queue weight */
+ tc2_wred_min 48 40 32 /* Minimum threshold for traffic class 2 queue (min_th) in number of packets */
+ tc2_wred_max 64 64 64 /* Maximum threshold for traffic class 2 queue (max_th) in number of packets */
+ tc2_wred_inv_prob 10 10 10 /* Inverse of packet marking probability for traffic class 2 queue (maxp = 1 / maxp_inv) */
+ tc2_wred_weight 9 9 9 /* Traffic Class 2 queue weight */
+ tc3_wred_min 48 40 32 /* Minimum threshold for traffic class 3 queue (min_th) in number of packets */
+ tc3_wred_max 64 64 64 /* Maximum threshold for traffic class 3 queue (max_th) in number of packets */
+ tc3_wred_inv_prob 10 10 10 /* Inverse of packet marking probability for traffic class 3 queue (maxp = 1 / maxp_inv) */
+ tc3_wred_weight 9 9 9 /* Traffic Class 3 queue weight */
+}
+```
+
+
+### DPDK QoS Scheduler Integration in VPP
+
+The Hierarchical Quaity-of-Service (HQoS) scheduler object could be seen as
+part of the logical NIC output interface. To enable HQoS on specific output
+interface, vpp startup.conf file has to be configured accordingly. The output
+interface that requires HQoS, should have "hqos" parameter specified in dpdk
+section. Another optional parameter "hqos-thread" has been defined which can
+be used to associate the output interface with specific hqos thread. In cpu
+section of the config file, "corelist-hqos-threads" is introduced to assign
+logical cpu cores to run the HQoS threads. A HQoS thread can run multiple HQoS
+objects each associated with different output interfaces. All worker threads
+instead of writing packets to NIC TX queue directly, write the packets to a
+software queues. The hqos_threads read the software queues, and enqueue the
+packets to HQoS objects, as well as dequeue packets from HQOS objects and
+write them to NIC output interfaces. The worker threads need to be able to
+send the packets to any output interface, therefore, each HQoS object
+associated with NIC output interface should have software queues equal to
+worker threads count.
+
+Following illustrates the sample startup configuration file with 4x worker
+threads feeding 2x hqos threads that handle each QoS scheduler for 1x output
+interface.
+
+```
+dpdk {
+ socket-mem 16384,16384
+
+ dev 0000:02:00.0 {
+ num-rx-queues 2
+ hqos
+ }
+ dev 0000:06:00.0 {
+ num-rx-queues 2
+ hqos
+ }
+
+ num-mbufs 1000000
+}
+
+cpu {
+ main-core 0
+ corelist-workers 1, 2, 3, 4
+ corelist-hqos-threads 5, 6
+}
+```
+
+
+### QoS scheduler CLI Commands
+
+Each QoS scheduler instance is initialised with default parameters required to
+configure hqos port, subport, pipe and queues. Some of the parameters can be
+re-configured in run-time through CLI commands.
+
+
+#### Configuration
+
+Following commands can be used to configure QoS scheduler parameters.
+
+The command below can be used to set the subport level parameters such as
+token bucket rate (bytes per seconds), token bucket size (bytes), traffic
+class rates (bytes per seconds) and token update period (Milliseconds).
+
+```
+set dpdk interface hqos subport <interface> subport <subport_id> [rate <n>]
+ [bktsize <n>] [tc0 <n>] [tc1 <n>] [tc2 <n>] [tc3 <n>] [period <n>]
+```
+
+For setting the pipe profile, following command can be used.
+
+```
+set dpdk interface hqos pipe <interface> subport <subport_id> pipe <pipe_id>
+ profile <profile_id>
+```
+
+To assign QoS scheduler instance to the specific thread, following command can
+be used.
+
+```
+set dpdk interface hqos placement <interface> thread <n>
+```
+
+The command below is used to set the packet fields required for classifiying
+the incoming packet. As a result of classification process, packet field
+information will be mapped to 5 tuples (subport, pipe, traffic class, pipe,
+color) and stored in packet mbuf.
+
+```
+set dpdk interface hqos pktfield <interface> id subport|pipe|tc offset <n>
+ mask <hex-mask>
+```
+
+The DSCP table entries used for idenfiying the traffic class and queue can be set using the command below;
+
+```
+set dpdk interface hqos tctbl <interface> entry <map_val> tc <tc_id> queue <queue_id>
+```
+
+
+#### Show Command
+
+The QoS Scheduler configuration can displayed using the command below.
+
+```
+ vpp# show dpdk interface hqos TenGigabitEthernet2/0/0
+ Thread:
+ Input SWQ size = 4096 packets
+ Enqueue burst size = 256 packets
+ Dequeue burst size = 220 packets
+ Packet field 0: slab position = 0, slab bitmask = 0x0000000000000000 (subport)
+ Packet field 1: slab position = 40, slab bitmask = 0x0000000fff000000 (pipe)
+ Packet field 2: slab position = 8, slab bitmask = 0x00000000000000fc (tc)
+ Packet field 2 tc translation table: ([Mapped Value Range]: tc/queue tc/queue ...)
+ [ 0 .. 15]: 0/0 0/1 0/2 0/3 1/0 1/1 1/2 1/3 2/0 2/1 2/2 2/3 3/0 3/1 3/2 3/3
+ [16 .. 31]: 0/0 0/1 0/2 0/3 1/0 1/1 1/2 1/3 2/0 2/1 2/2 2/3 3/0 3/1 3/2 3/3
+ [32 .. 47]: 0/0 0/1 0/2 0/3 1/0 1/1 1/2 1/3 2/0 2/1 2/2 2/3 3/0 3/1 3/2 3/3
+ [48 .. 63]: 0/0 0/1 0/2 0/3 1/0 1/1 1/2 1/3 2/0 2/1 2/2 2/3 3/0 3/1 3/2 3/3
+ Port:
+ Rate = 1250000000 bytes/second
+ MTU = 1514 bytes
+ Frame overhead = 24 bytes
+ Number of subports = 1
+ Number of pipes per subport = 4096
+ Packet queue size: TC0 = 64, TC1 = 64, TC2 = 64, TC3 = 64 packets
+ Number of pipe profiles = 1
+ Subport 0:
+ Rate = 120000000 bytes/second
+ Token bucket size = 1000000 bytes
+ Traffic class rate: TC0 = 120000000, TC1 = 120000000, TC2 = 120000000, TC3 = 120000000 bytes/second
+ TC period = 10 milliseconds
+ Pipe profile 0:
+ Rate = 305175 bytes/second
+ Token bucket size = 1000000 bytes
+ Traffic class rate: TC0 = 305175, TC1 = 305175, TC2 = 305175, TC3 = 305175 bytes/second
+ TC period = 40 milliseconds
+ TC0 WRR weights: Q0 = 1, Q1 = 1, Q2 = 1, Q3 = 1
+ TC1 WRR weights: Q0 = 1, Q1 = 1, Q2 = 1, Q3 = 1
+ TC2 WRR weights: Q0 = 1, Q1 = 1, Q2 = 1, Q3 = 1
+ TC3 WRR weights: Q0 = 1, Q1 = 1, Q2 = 1, Q3 = 1
+```
+
+The QoS Scheduler placement over the logical cpu cores can be displayed using
+below command.
+
+```
+ vpp# show dpdk interface hqos placement
+ Thread 5 (vpp_hqos-threads_0 at lcore 5):
+ TenGigabitEthernet2/0/0 queue 0
+ Thread 6 (vpp_hqos-threads_1 at lcore 6):
+ TenGigabitEthernet4/0/1 queue 0
+```
+
+
+### QoS Scheduler Binary APIs
+
+This section explans the available binary APIs for configuring QoS scheduler
+parameters in run-time.
+
+The following API can be used to set the pipe profile of a pipe that belongs
+to a given subport:
+
+```
+sw_interface_set_dpdk_hqos_pipe rx <intfc> | sw_if_index <id>
+ subport <subport-id> pipe <pipe-id> profile <profile-id>
+```
+
+The data structures used for set the pipe profile parameter are as follows;
+
+```
+ /** \\brief DPDK interface HQoS pipe profile set request
+ @param client_index - opaque cookie to identify the sender
+ @param context - sender context, to match reply w/ request
+ @param sw_if_index - the interface
+ @param subport - subport ID
+ @param pipe - pipe ID within its subport
+ @param profile - pipe profile ID
+ */
+ define sw_interface_set_dpdk_hqos_pipe {
+ u32 client_index;
+ u32 context;
+ u32 sw_if_index;
+ u32 subport;
+ u32 pipe;
+ u32 profile;
+ };
+
+ /** \\brief DPDK interface HQoS pipe profile set reply
+ @param context - sender context, to match reply w/ request
+ @param retval - request return code
+ */
+ define sw_interface_set_dpdk_hqos_pipe_reply {
+ u32 context;
+ i32 retval;
+ };
+```
+
+The following API can be used to set the subport level parameters, for
+example- token bucket rate (bytes per seconds), token bucket size (bytes),
+traffic class rate (bytes per seconds) and tokens update period.
+
+```
+sw_interface_set_dpdk_hqos_subport rx <intfc> | sw_if_index <id>
+ subport <subport-id> [rate <n>] [bktsize <n>]
+ [tc0 <n>] [tc1 <n>] [tc2 <n>] [tc3 <n>] [period <n>]
+```
+
+The data structures used for set the subport level parameter are as follows;
+
+```
+ /** \\brief DPDK interface HQoS subport parameters set request
+ @param client_index - opaque cookie to identify the sender
+ @param context - sender context, to match reply w/ request
+ @param sw_if_index - the interface
+ @param subport - subport ID
+ @param tb_rate - subport token bucket rate (measured in bytes/second)
+ @param tb_size - subport token bucket size (measured in credits)
+ @param tc_rate - subport traffic class 0 .. 3 rates (measured in bytes/second)
+ @param tc_period - enforcement period for rates (measured in milliseconds)
+ */
+ define sw_interface_set_dpdk_hqos_subport {
+ u32 client_index;
+ u32 context;
+ u32 sw_if_index;
+ u32 subport;
+ u32 tb_rate;
+ u32 tb_size;
+ u32 tc_rate[4];
+ u32 tc_period;
+ };
+
+ /** \\brief DPDK interface HQoS subport parameters set reply
+ @param context - sender context, to match reply w/ request
+ @param retval - request return code
+ */
+ define sw_interface_set_dpdk_hqos_subport_reply {
+ u32 context;
+ i32 retval;
+ };
+```
+
+The following API can be used set the DSCP table entry. The DSCP table have
+64 entries to map the packet DSCP field onto traffic class and hqos input
+queue.
+
+```
+sw_interface_set_dpdk_hqos_tctbl rx <intfc> | sw_if_index <id>
+ entry <n> tc <n> queue <n>
+```
+
+The data structures used for setting DSCP table entries are given below.
+
+```
+ /** \\brief DPDK interface HQoS tctbl entry set request
+ @param client_index - opaque cookie to identify the sender
+ @param context - sender context, to match reply w/ request
+ @param sw_if_index - the interface
+ @param entry - entry index ID
+ @param tc - traffic class (0 .. 3)
+ @param queue - traffic class queue (0 .. 3)
+ */
+ define sw_interface_set_dpdk_hqos_tctbl {
+ u32 client_index;
+ u32 context;
+ u32 sw_if_index;
+ u32 entry;
+ u32 tc;
+ u32 queue;
+ };
+
+ /** \\brief DPDK interface HQoS tctbl entry set reply
+ @param context - sender context, to match reply w/ request
+ @param retval - request return code
+ */
+ define sw_interface_set_dpdk_hqos_tctbl_reply {
+ u32 context;
+ i32 retval;
+ };
+```