4 * Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #ifndef __INCLUDE_RTE_SCHED_H__
36 #define __INCLUDE_RTE_SCHED_H__
44 * RTE Hierarchical Scheduler
46 * The hierarchical scheduler prioritizes the transmission of packets from different
47 * users and traffic classes according to the Service Level Agreements (SLAs) defined
48 * for the current network node.
50 * The scheduler supports thousands of packet queues grouped under a 5-level hierarchy:
52 * - Typical usage: output Ethernet port;
53 * - Multiple ports are scheduled in round robin order with equal priority;
55 * - Typical usage: group of users;
56 * - Traffic shaping using the token bucket algorithm (one bucket per subport);
57 * - Upper limit enforced per traffic class at subport level;
58 * - Lower priority traffic classes able to reuse subport bandwidth currently
59 * unused by higher priority traffic classes of the same subport;
60 * - When any subport traffic class is oversubscribed (configuration time
61 * event), the usage of subport member pipes with high demand for that
62 * traffic class pipes is truncated to a dynamically adjusted value with no
63 * impact to low demand pipes;
65 * - Typical usage: individual user/subscriber;
66 * - Traffic shaping using the token bucket algorithm (one bucket per pipe);
68 * - Traffic classes of the same pipe handled in strict priority order;
69 * - Upper limit enforced per traffic class at the pipe level;
70 * - Lower priority traffic classes able to reuse pipe bandwidth currently
71 * unused by higher priority traffic classes of the same pipe;
73 * - Typical usage: queue hosting packets from one or multiple connections
74 * of same traffic class belonging to the same user;
75 * - Weighted Round Robin (WRR) is used to service the queues within same
80 #include <sys/types.h>
82 #include <rte_meter.h>
84 /** Random Early Detection (RED) */
89 /** Number of traffic classes per pipe (as well as subport). Cannot be changed. */
90 #define RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE 4
92 /** Number of queues per pipe traffic class. Cannot be changed. */
93 #define RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS 4
95 /** Number of queues per pipe. */
96 #define RTE_SCHED_QUEUES_PER_PIPE \
97 (RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE * \
98 RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS)
100 /** Maximum number of pipe profiles that can be defined per port. Compile-time configurable.*/
101 #ifndef RTE_SCHED_PIPE_PROFILES_PER_PORT
102 #define RTE_SCHED_PIPE_PROFILES_PER_PORT 256
105 /** Ethernet framing overhead. Overhead fields per Ethernet frame:
106 1. Preamble: 7 bytes;
107 2. Start of Frame Delimiter (SFD): 1 byte;
108 3. Frame Check Sequence (FCS): 4 bytes;
109 4. Inter Frame Gap (IFG): 12 bytes.
110 The FCS is considered overhead only if not included in the packet length (field pkt.pkt_len
111 of struct rte_mbuf). */
112 #ifndef RTE_SCHED_FRAME_OVERHEAD_DEFAULT
113 #define RTE_SCHED_FRAME_OVERHEAD_DEFAULT 24
116 /** Subport configuration parameters. The period and credits_per_period parameters are measured
117 in bytes, with one byte meaning the time duration associated with the transmission of one byte
118 on the physical medium of the output port, with pipe or pipe traffic class rate (measured as
119 percentage of output port rate) determined as credits_per_period divided by period. One credit
120 represents one byte. */
121 struct rte_sched_subport_params {
122 /* Subport token bucket */
123 uint32_t tb_rate; /**< Subport token bucket rate (measured in bytes per second) */
124 uint32_t tb_size; /**< Subport token bucket size (measured in credits) */
126 /* Subport traffic classes */
127 uint32_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Subport traffic class rates (measured in bytes per second) */
128 uint32_t tc_period; /**< Enforcement period for traffic class rates (measured in milliseconds) */
131 /** Subport statistics */
132 struct rte_sched_subport_stats {
134 uint32_t n_pkts_tc[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Number of packets successfully written to current
135 subport for each traffic class */
136 uint32_t n_pkts_tc_dropped[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Number of packets dropped by the current
137 subport for each traffic class due to subport queues being full or congested*/
140 uint32_t n_bytes_tc[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Number of bytes successfully written to current
141 subport for each traffic class*/
142 uint32_t n_bytes_tc_dropped[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Number of bytes dropped by the current
143 subport for each traffic class due to subport queues being full or congested */
146 /** Pipe configuration parameters. The period and credits_per_period parameters are measured
147 in bytes, with one byte meaning the time duration associated with the transmission of one byte
148 on the physical medium of the output port, with pipe or pipe traffic class rate (measured as
149 percentage of output port rate) determined as credits_per_period divided by period. One credit
150 represents one byte. */
151 struct rte_sched_pipe_params {
152 /* Pipe token bucket */
153 uint32_t tb_rate; /**< Pipe token bucket rate (measured in bytes per second) */
154 uint32_t tb_size; /**< Pipe token bucket size (measured in credits) */
156 /* Pipe traffic classes */
157 uint32_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Pipe traffic class rates (measured in bytes per second) */
158 uint32_t tc_period; /**< Enforcement period for pipe traffic class rates (measured in milliseconds) */
159 #ifdef RTE_SCHED_SUBPORT_TC_OV
160 uint8_t tc_ov_weight; /**< Weight for the current pipe in the event of subport traffic class 3 oversubscription */
164 uint8_t wrr_weights[RTE_SCHED_QUEUES_PER_PIPE]; /**< WRR weights for the queues of the current pipe */
167 /** Queue statistics */
168 struct rte_sched_queue_stats {
170 uint32_t n_pkts; /**< Number of packets successfully written to current queue */
171 uint32_t n_pkts_dropped; /**< Number of packets dropped due to current queue being full or congested */
174 uint32_t n_bytes; /**< Number of bytes successfully written to current queue */
175 uint32_t n_bytes_dropped; /**< Number of bytes dropped due to current queue being full or congested */
178 /** Port configuration parameters. */
179 struct rte_sched_port_params {
180 const char *name; /**< Literal string to be associated to the current port scheduler instance */
181 int socket; /**< CPU socket ID where the memory for port scheduler should be allocated */
182 uint32_t rate; /**< Output port rate (measured in bytes per second) */
183 uint32_t mtu; /**< Maximum Ethernet frame size (measured in bytes). Should not include the framing overhead. */
184 uint32_t frame_overhead; /**< Framing overhead per packet (measured in bytes) */
185 uint32_t n_subports_per_port; /**< Number of subports for the current port scheduler instance*/
186 uint32_t n_pipes_per_subport; /**< Number of pipes for each port scheduler subport */
187 uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; /**< Packet queue size for each traffic class. All queues
188 within the same pipe traffic class have the same size. Queues from
189 different pipes serving the same traffic class have the same size. */
190 struct rte_sched_pipe_params *pipe_profiles; /**< Pipe profile table defined for current port scheduler instance.
191 Every pipe of the current port scheduler is configured using one of the
192 profiles from this table. */
193 uint32_t n_pipe_profiles; /**< Number of profiles in the pipe profile table */
195 struct rte_red_params red_params[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][e_RTE_METER_COLORS]; /**< RED parameters */
199 /** Path through the scheduler hierarchy used by the scheduler enqueue operation to
200 identify the destination queue for the current packet. Stored in the field pkt.hash.sched
201 of struct rte_mbuf of each packet, typically written by the classification stage and read by
203 struct rte_sched_port_hierarchy {
204 uint32_t queue:2; /**< Queue ID (0 .. 3) */
205 uint32_t traffic_class:2; /**< Traffic class ID (0 .. 3)*/
206 uint32_t pipe:20; /**< Pipe ID */
207 uint32_t subport:6; /**< Subport ID */
208 uint32_t color:2; /**< Color */
217 * Hierarchical scheduler port configuration
220 * Port scheduler configuration parameter structure
222 * Handle to port scheduler instance upon success or NULL otherwise.
224 struct rte_sched_port *
225 rte_sched_port_config(struct rte_sched_port_params *params);
228 * Hierarchical scheduler port free
231 * Handle to port scheduler instance
234 rte_sched_port_free(struct rte_sched_port *port);
237 * Hierarchical scheduler subport configuration
240 * Handle to port scheduler instance
244 * Subport configuration parameters
246 * 0 upon success, error code otherwise
249 rte_sched_subport_config(struct rte_sched_port *port,
251 struct rte_sched_subport_params *params);
254 * Hierarchical scheduler pipe configuration
257 * Handle to port scheduler instance
261 * Pipe ID within subport
262 * @param pipe_profile
263 * ID of port-level pre-configured pipe profile
265 * 0 upon success, error code otherwise
268 rte_sched_pipe_config(struct rte_sched_port *port,
271 int32_t pipe_profile);
274 * Hierarchical scheduler memory footprint size per port
277 * Port scheduler configuration parameter structure
279 * Memory footprint size in bytes upon success, 0 otherwise
282 rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params);
290 * Hierarchical scheduler subport statistics read
293 * Handle to port scheduler instance
297 * Pointer to pre-allocated subport statistics structure where the statistics
298 * counters should be stored
300 * Pointer to pre-allocated 4-entry array where the oversubscription status for
301 * each of the 4 subport traffic classes should be stored.
303 * 0 upon success, error code otherwise
306 rte_sched_subport_read_stats(struct rte_sched_port *port,
308 struct rte_sched_subport_stats *stats,
312 * Hierarchical scheduler queue statistics read
315 * Handle to port scheduler instance
317 * Queue ID within port scheduler
319 * Pointer to pre-allocated subport statistics structure where the statistics
320 * counters should be stored
322 * Pointer to pre-allocated variable where the current queue length should be stored.
324 * 0 upon success, error code otherwise
327 rte_sched_queue_read_stats(struct rte_sched_port *port,
329 struct rte_sched_queue_stats *stats,
338 * Scheduler hierarchy path write to packet descriptor. Typically called by the
339 * packet classification stage.
342 * Packet descriptor handle
346 * Pipe ID within subport
347 * @param traffic_class
348 * Traffic class ID within pipe (0 .. 3)
350 * Queue ID within pipe traffic class (0 .. 3)
353 rte_sched_port_pkt_write(struct rte_mbuf *pkt,
354 uint32_t subport, uint32_t pipe, uint32_t traffic_class, uint32_t queue, enum rte_meter_color color)
356 struct rte_sched_port_hierarchy *sched = (struct rte_sched_port_hierarchy *) &pkt->pkt.hash.sched;
358 sched->color = (uint32_t) color;
359 sched->subport = subport;
361 sched->traffic_class = traffic_class;
362 sched->queue = queue;
366 * Scheduler hierarchy path read from packet descriptor (struct rte_mbuf). Typically
367 * called as part of the hierarchical scheduler enqueue operation. The subport,
368 * pipe, traffic class and queue parameters need to be pre-allocated by the caller.
371 * Packet descriptor handle
375 * Pipe ID within subport
376 * @param traffic_class
377 * Traffic class ID within pipe (0 .. 3)
379 * Queue ID within pipe traffic class (0 .. 3)
383 rte_sched_port_pkt_read_tree_path(struct rte_mbuf *pkt, uint32_t *subport, uint32_t *pipe, uint32_t *traffic_class, uint32_t *queue)
385 struct rte_sched_port_hierarchy *sched = (struct rte_sched_port_hierarchy *) &pkt->pkt.hash.sched;
387 *subport = sched->subport;
389 *traffic_class = sched->traffic_class;
390 *queue = sched->queue;
393 static inline enum rte_meter_color
394 rte_sched_port_pkt_read_color(struct rte_mbuf *pkt)
396 struct rte_sched_port_hierarchy *sched = (struct rte_sched_port_hierarchy *) &pkt->pkt.hash.sched;
398 return (enum rte_meter_color) sched->color;
402 * Hierarchical scheduler port enqueue. Writes up to n_pkts to port scheduler and
403 * returns the number of packets actually written. For each packet, the port scheduler
404 * queue to write the packet to is identified by reading the hierarchy path from the
405 * packet descriptor; if the queue is full or congested and the packet is not written
406 * to the queue, then the packet is automatically dropped without any action required
410 * Handle to port scheduler instance
412 * Array storing the packet descriptor handles
414 * Number of packets to enqueue from the pkts array into the port scheduler
416 * Number of packets successfully enqueued
419 rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts);
422 * Hierarchical scheduler port dequeue. Reads up to n_pkts from the port scheduler
423 * and stores them in the pkts array and returns the number of packets actually read.
424 * The pkts array needs to be pre-allocated by the caller with at least n_pkts entries.
427 * Handle to port scheduler instance
429 * Pre-allocated packet descriptor array where the packets dequeued from the port
430 * scheduler should be stored
432 * Number of packets to dequeue from the port scheduler
434 * Number of packets successfully dequeued and placed in the pkts array
437 rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts);
443 #endif /* __INCLUDE_RTE_SCHED_H__ */