]>
Commit | Line | Data |
---|---|---|
72865317 | 1 | /* |
ff073a71 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. |
72865317 BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
db73f716 | 18 | #include "dpif-netdev.h" |
72865317 | 19 | |
72865317 BP |
20 | #include <ctype.h> |
21 | #include <errno.h> | |
22 | #include <fcntl.h> | |
23 | #include <inttypes.h> | |
72865317 | 24 | #include <netinet/in.h> |
9d82ec47 | 25 | #include <sys/socket.h> |
7f3adc00 | 26 | #include <net/if.h> |
cdee00fd | 27 | #include <stdint.h> |
72865317 BP |
28 | #include <stdlib.h> |
29 | #include <string.h> | |
30 | #include <sys/ioctl.h> | |
31 | #include <sys/stat.h> | |
72865317 BP |
32 | #include <unistd.h> |
33 | ||
59e6d833 | 34 | #include "cmap.h" |
72865317 | 35 | #include "csum.h" |
e14deea0 | 36 | #include "dp-packet.h" |
614c4892 | 37 | #include "dpif.h" |
72865317 | 38 | #include "dpif-provider.h" |
614c4892 | 39 | #include "dummy.h" |
36956a7d | 40 | #include "dynamic-string.h" |
afae68b1 | 41 | #include "fat-rwlock.h" |
72865317 | 42 | #include "flow.h" |
9f361d6b | 43 | #include "cmap.h" |
6c3eee82 | 44 | #include "latch.h" |
72865317 | 45 | #include "list.h" |
0de8783a | 46 | #include "match.h" |
8c301900 | 47 | #include "meta-flow.h" |
72865317 | 48 | #include "netdev.h" |
8617afff | 49 | #include "netdev-dpdk.h" |
de281153 | 50 | #include "netdev-vport.h" |
cdee00fd | 51 | #include "netlink.h" |
f094af7b | 52 | #include "odp-execute.h" |
72865317 BP |
53 | #include "odp-util.h" |
54 | #include "ofp-print.h" | |
55 | #include "ofpbuf.h" | |
5a034064 | 56 | #include "ovs-numa.h" |
61e7deb1 | 57 | #include "ovs-rcu.h" |
72865317 BP |
58 | #include "packets.h" |
59 | #include "poll-loop.h" | |
0de8783a | 60 | #include "pvector.h" |
26c6b6cd | 61 | #include "random.h" |
d33ed218 | 62 | #include "seq.h" |
462278db | 63 | #include "shash.h" |
0cbfe35d | 64 | #include "sset.h" |
72865317 | 65 | #include "timeval.h" |
a36de779 | 66 | #include "tnl-arp-cache.h" |
74cc3969 | 67 | #include "unixctl.h" |
72865317 | 68 | #include "util.h" |
e6211adc | 69 | #include "openvswitch/vlog.h" |
5136ce49 | 70 | |
d98e6007 | 71 | VLOG_DEFINE_THIS_MODULE(dpif_netdev); |
72865317 | 72 | |
8bb113da | 73 | #define FLOW_DUMP_MAX_BATCH 50 |
adcf00ba AZ |
74 | /* Use per thread recirc_depth to prevent recirculation loop. */ |
75 | #define MAX_RECIRC_DEPTH 5 | |
76 | DEFINE_STATIC_PER_THREAD_DATA(uint32_t, recirc_depth, 0) | |
e4cfed38 | 77 | |
72865317 | 78 | /* Configuration parameters. */ |
72865317 BP |
79 | enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ |
80 | ||
8a4e3a85 BP |
81 | /* Protects against changes to 'dp_netdevs'. */ |
82 | static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER; | |
83 | ||
84 | /* Contains all 'struct dp_netdev's. */ | |
85 | static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex) | |
86 | = SHASH_INITIALIZER(&dp_netdevs); | |
87 | ||
623540e4 | 88 | static struct vlog_rate_limit upcall_rl = VLOG_RATE_LIMIT_INIT(600, 600); |
6b31e073 | 89 | |
79df317f | 90 | /* Stores a miniflow with inline values */ |
9bbf1c3d | 91 | |
9bbf1c3d | 92 | struct netdev_flow_key { |
caeb4906 JR |
93 | uint32_t hash; /* Hash function differs for different users. */ |
94 | uint32_t len; /* Length of the following miniflow (incl. map). */ | |
0de8783a | 95 | struct miniflow mf; |
d70e8c28 | 96 | uint64_t buf[FLOW_MAX_PACKET_U64S - MINI_N_INLINE]; |
9bbf1c3d DDP |
97 | }; |
98 | ||
99 | /* Exact match cache for frequently used flows | |
100 | * | |
101 | * The cache uses a 32-bit hash of the packet (which can be the RSS hash) to | |
102 | * search its entries for a miniflow that matches exactly the miniflow of the | |
0de8783a | 103 | * packet. It stores the 'dpcls_rule' (rule) that matches the miniflow. |
9bbf1c3d DDP |
104 | * |
105 | * A cache entry holds a reference to its 'dp_netdev_flow'. | |
106 | * | |
107 | * A miniflow with a given hash can be in one of EM_FLOW_HASH_SEGS different | |
108 | * entries. The 32-bit hash is split into EM_FLOW_HASH_SEGS values (each of | |
109 | * them is EM_FLOW_HASH_SHIFT bits wide and the remainder is thrown away). Each | |
110 | * value is the index of a cache entry where the miniflow could be. | |
111 | * | |
112 | * | |
113 | * Thread-safety | |
114 | * ============= | |
115 | * | |
116 | * Each pmd_thread has its own private exact match cache. | |
117 | * If dp_netdev_input is not called from a pmd thread, a mutex is used. | |
118 | */ | |
119 | ||
fc82e877 | 120 | #define EM_FLOW_HASH_SHIFT 13 |
9bbf1c3d DDP |
121 | #define EM_FLOW_HASH_ENTRIES (1u << EM_FLOW_HASH_SHIFT) |
122 | #define EM_FLOW_HASH_MASK (EM_FLOW_HASH_ENTRIES - 1) | |
123 | #define EM_FLOW_HASH_SEGS 2 | |
124 | ||
125 | struct emc_entry { | |
9bbf1c3d | 126 | struct dp_netdev_flow *flow; |
0de8783a | 127 | struct netdev_flow_key key; /* key.hash used for emc hash value. */ |
9bbf1c3d DDP |
128 | }; |
129 | ||
130 | struct emc_cache { | |
131 | struct emc_entry entries[EM_FLOW_HASH_ENTRIES]; | |
67ad54cb | 132 | int sweep_idx; /* For emc_cache_slow_sweep(). */ |
9bbf1c3d DDP |
133 | }; |
134 | ||
135 | /* Iterate in the exact match cache through every entry that might contain a | |
136 | * miniflow with hash 'HASH'. */ | |
137 | #define EMC_FOR_EACH_POS_WITH_HASH(EMC, CURRENT_ENTRY, HASH) \ | |
138 | for (uint32_t i__ = 0, srch_hash__ = (HASH); \ | |
139 | (CURRENT_ENTRY) = &(EMC)->entries[srch_hash__ & EM_FLOW_HASH_MASK], \ | |
140 | i__ < EM_FLOW_HASH_SEGS; \ | |
141 | i__++, srch_hash__ >>= EM_FLOW_HASH_SHIFT) | |
0de8783a JR |
142 | \f |
143 | /* Simple non-wildcarding single-priority classifier. */ | |
144 | ||
145 | struct dpcls { | |
146 | struct cmap subtables_map; | |
147 | struct pvector subtables; | |
148 | }; | |
9bbf1c3d | 149 | |
0de8783a JR |
150 | /* A rule to be inserted to the classifier. */ |
151 | struct dpcls_rule { | |
152 | struct cmap_node cmap_node; /* Within struct dpcls_subtable 'rules'. */ | |
153 | struct netdev_flow_key *mask; /* Subtable's mask. */ | |
154 | struct netdev_flow_key flow; /* Matching key. */ | |
155 | /* 'flow' must be the last field, additional space is allocated here. */ | |
156 | }; | |
157 | ||
158 | static void dpcls_init(struct dpcls *); | |
159 | static void dpcls_destroy(struct dpcls *); | |
160 | static void dpcls_insert(struct dpcls *, struct dpcls_rule *, | |
161 | const struct netdev_flow_key *mask); | |
162 | static void dpcls_remove(struct dpcls *, struct dpcls_rule *); | |
163 | static bool dpcls_lookup(const struct dpcls *cls, | |
164 | const struct netdev_flow_key keys[], | |
165 | struct dpcls_rule **rules, size_t cnt); | |
166 | \f | |
8a4e3a85 BP |
167 | /* Datapath based on the network device interface from netdev.h. |
168 | * | |
169 | * | |
170 | * Thread-safety | |
171 | * ============= | |
172 | * | |
173 | * Some members, marked 'const', are immutable. Accessing other members | |
174 | * requires synchronization, as noted in more detail below. | |
175 | * | |
176 | * Acquisition order is, from outermost to innermost: | |
177 | * | |
178 | * dp_netdev_mutex (global) | |
59e6d833 | 179 | * port_mutex |
8a4e3a85 | 180 | */ |
72865317 | 181 | struct dp_netdev { |
8a4e3a85 BP |
182 | const struct dpif_class *const class; |
183 | const char *const name; | |
6b31e073 | 184 | struct dpif *dpif; |
6a8267c5 BP |
185 | struct ovs_refcount ref_cnt; |
186 | atomic_flag destroyed; | |
72865317 | 187 | |
8a4e3a85 BP |
188 | /* Ports. |
189 | * | |
59e6d833 BP |
190 | * Protected by RCU. Take the mutex to add or remove ports. */ |
191 | struct ovs_mutex port_mutex; | |
192 | struct cmap ports; | |
d33ed218 | 193 | struct seq *port_seq; /* Incremented whenever a port changes. */ |
6c3eee82 | 194 | |
6b31e073 RW |
195 | /* Protects access to ofproto-dpif-upcall interface during revalidator |
196 | * thread synchronization. */ | |
197 | struct fat_rwlock upcall_rwlock; | |
623540e4 EJ |
198 | upcall_callback *upcall_cb; /* Callback function for executing upcalls. */ |
199 | void *upcall_aux; | |
6b31e073 | 200 | |
65f13b50 AW |
201 | /* Stores all 'struct dp_netdev_pmd_thread's. */ |
202 | struct cmap poll_threads; | |
203 | ||
204 | /* Protects the access of the 'struct dp_netdev_pmd_thread' | |
205 | * instance for non-pmd thread. */ | |
206 | struct ovs_mutex non_pmd_mutex; | |
207 | ||
208 | /* Each pmd thread will store its pointer to | |
209 | * 'struct dp_netdev_pmd_thread' in 'per_pmd_key'. */ | |
210 | ovsthread_key_t per_pmd_key; | |
f2eee189 AW |
211 | |
212 | /* Number of rx queues for each dpdk interface and the cpu mask | |
213 | * for pin of pmd threads. */ | |
214 | size_t n_dpdk_rxqs; | |
215 | char *pmd_cmask; | |
a36de779 | 216 | uint64_t last_tnl_conf_seq; |
72865317 BP |
217 | }; |
218 | ||
8a4e3a85 | 219 | static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp, |
59e6d833 | 220 | odp_port_t); |
ff073a71 | 221 | |
51852a57 | 222 | enum dp_stat_type { |
abcf3ef4 DDP |
223 | DP_STAT_EXACT_HIT, /* Packets that had an exact match (emc). */ |
224 | DP_STAT_MASKED_HIT, /* Packets that matched in the flow table. */ | |
51852a57 BP |
225 | DP_STAT_MISS, /* Packets that did not match. */ |
226 | DP_STAT_LOST, /* Packets not passed up to the client. */ | |
227 | DP_N_STATS | |
228 | }; | |
229 | ||
55e3ca97 DDP |
230 | enum pmd_cycles_counter_type { |
231 | PMD_CYCLES_POLLING, /* Cycles spent polling NICs. */ | |
232 | PMD_CYCLES_PROCESSING, /* Cycles spent processing packets */ | |
233 | PMD_N_CYCLES | |
234 | }; | |
235 | ||
72865317 BP |
236 | /* A port in a netdev-based datapath. */ |
237 | struct dp_netdev_port { | |
efa2bcbb | 238 | struct pkt_metadata md; |
72865317 | 239 | struct netdev *netdev; |
efa2bcbb | 240 | struct cmap_node node; /* Node in dp_netdev's 'ports'. */ |
4b609110 | 241 | struct netdev_saved_flags *sf; |
55c955bd | 242 | struct netdev_rxq **rxq; |
b284085e | 243 | struct ovs_refcount ref_cnt; |
0cbfe35d | 244 | char *type; /* Port type as requested by user. */ |
72865317 BP |
245 | }; |
246 | ||
1c1e46ed AW |
247 | /* Contained by struct dp_netdev_flow's 'stats' member. */ |
248 | struct dp_netdev_flow_stats { | |
eb94da30 DDP |
249 | atomic_llong used; /* Last used time, in monotonic msecs. */ |
250 | atomic_ullong packet_count; /* Number of packets matched. */ | |
251 | atomic_ullong byte_count; /* Number of bytes matched. */ | |
252 | atomic_uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ | |
1c1e46ed AW |
253 | }; |
254 | ||
255 | /* A flow in 'dp_netdev_pmd_thread's 'flow_table'. | |
8a4e3a85 BP |
256 | * |
257 | * | |
258 | * Thread-safety | |
259 | * ============= | |
260 | * | |
261 | * Except near the beginning or ending of its lifespan, rule 'rule' belongs to | |
1c1e46ed | 262 | * its pmd thread's classifier. The text below calls this classifier 'cls'. |
8a4e3a85 BP |
263 | * |
264 | * Motivation | |
265 | * ---------- | |
266 | * | |
267 | * The thread safety rules described here for "struct dp_netdev_flow" are | |
268 | * motivated by two goals: | |
269 | * | |
270 | * - Prevent threads that read members of "struct dp_netdev_flow" from | |
271 | * reading bad data due to changes by some thread concurrently modifying | |
272 | * those members. | |
273 | * | |
274 | * - Prevent two threads making changes to members of a given "struct | |
275 | * dp_netdev_flow" from interfering with each other. | |
276 | * | |
277 | * | |
278 | * Rules | |
279 | * ----- | |
280 | * | |
ed79f89a DDP |
281 | * A flow 'flow' may be accessed without a risk of being freed during an RCU |
282 | * grace period. Code that needs to hold onto a flow for a while | |
283 | * should try incrementing 'flow->ref_cnt' with dp_netdev_flow_ref(). | |
8a4e3a85 BP |
284 | * |
285 | * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the | |
ed79f89a DDP |
286 | * flow from being deleted from 'cls' and it doesn't protect members of 'flow' |
287 | * from modification. | |
8a4e3a85 BP |
288 | * |
289 | * Some members, marked 'const', are immutable. Accessing other members | |
290 | * requires synchronization, as noted in more detail below. | |
291 | */ | |
72865317 | 292 | struct dp_netdev_flow { |
11e5cf1f | 293 | const struct flow flow; /* Unmasked flow that created this entry. */ |
8a4e3a85 | 294 | /* Hash table index by unmasked flow. */ |
1c1e46ed AW |
295 | const struct cmap_node node; /* In owning dp_netdev_pmd_thread's */ |
296 | /* 'flow_table'. */ | |
70e5ed6f | 297 | const ovs_u128 ufid; /* Unique flow identifier. */ |
bd5131ba | 298 | const unsigned pmd_id; /* The 'core_id' of pmd thread owning this */ |
1c1e46ed | 299 | /* flow. */ |
72865317 | 300 | |
ed79f89a DDP |
301 | /* Number of references. |
302 | * The classifier owns one reference. | |
303 | * Any thread trying to keep a rule from being freed should hold its own | |
304 | * reference. */ | |
305 | struct ovs_refcount ref_cnt; | |
306 | ||
11e5cf1f DDP |
307 | bool dead; |
308 | ||
1c1e46ed AW |
309 | /* Statistics. */ |
310 | struct dp_netdev_flow_stats stats; | |
8a4e3a85 | 311 | |
45c626a3 | 312 | /* Actions. */ |
61e7deb1 | 313 | OVSRCU_TYPE(struct dp_netdev_actions *) actions; |
0de8783a | 314 | |
11e5cf1f DDP |
315 | /* While processing a group of input packets, the datapath uses the next |
316 | * member to store a pointer to the output batch for the flow. It is | |
317 | * reset after the batch has been sent out (See dp_netdev_queue_batches(), | |
318 | * packet_batch_init() and packet_batch_execute()). */ | |
319 | struct packet_batch *batch; | |
320 | ||
0de8783a JR |
321 | /* Packet classification. */ |
322 | struct dpcls_rule cr; /* In owning dp_netdev's 'cls'. */ | |
323 | /* 'cr' must be the last member. */ | |
72865317 BP |
324 | }; |
325 | ||
ed79f89a | 326 | static void dp_netdev_flow_unref(struct dp_netdev_flow *); |
9bbf1c3d | 327 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *); |
70e5ed6f JS |
328 | static int dpif_netdev_flow_from_nlattrs(const struct nlattr *, uint32_t, |
329 | struct flow *); | |
8a4e3a85 | 330 | |
a84cb64a BP |
331 | /* A set of datapath actions within a "struct dp_netdev_flow". |
332 | * | |
333 | * | |
334 | * Thread-safety | |
335 | * ============= | |
336 | * | |
45c626a3 | 337 | * A struct dp_netdev_actions 'actions' is protected with RCU. */ |
a84cb64a | 338 | struct dp_netdev_actions { |
a84cb64a BP |
339 | /* These members are immutable: they do not change during the struct's |
340 | * lifetime. */ | |
a84cb64a | 341 | unsigned int size; /* Size of 'actions', in bytes. */ |
9ff55ae2 | 342 | struct nlattr actions[]; /* Sequence of OVS_ACTION_ATTR_* attributes. */ |
a84cb64a BP |
343 | }; |
344 | ||
345 | struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *, | |
346 | size_t); | |
61e7deb1 BP |
347 | struct dp_netdev_actions *dp_netdev_flow_get_actions( |
348 | const struct dp_netdev_flow *); | |
349 | static void dp_netdev_actions_free(struct dp_netdev_actions *); | |
a84cb64a | 350 | |
1c1e46ed AW |
351 | /* Contained by struct dp_netdev_pmd_thread's 'stats' member. */ |
352 | struct dp_netdev_pmd_stats { | |
353 | /* Indexed by DP_STAT_*. */ | |
eb94da30 | 354 | atomic_ullong n[DP_N_STATS]; |
1c1e46ed AW |
355 | }; |
356 | ||
55e3ca97 DDP |
357 | /* Contained by struct dp_netdev_pmd_thread's 'cycle' member. */ |
358 | struct dp_netdev_pmd_cycles { | |
359 | /* Indexed by PMD_CYCLES_*. */ | |
360 | atomic_ullong n[PMD_N_CYCLES]; | |
361 | }; | |
362 | ||
e4cfed38 PS |
363 | /* PMD: Poll modes drivers. PMD accesses devices via polling to eliminate |
364 | * the performance overhead of interrupt processing. Therefore netdev can | |
365 | * not implement rx-wait for these devices. dpif-netdev needs to poll | |
366 | * these device to check for recv buffer. pmd-thread does polling for | |
1c1e46ed | 367 | * devices assigned to itself. |
e4cfed38 PS |
368 | * |
369 | * DPDK used PMD for accessing NIC. | |
370 | * | |
65f13b50 AW |
371 | * Note, instance with cpu core id NON_PMD_CORE_ID will be reserved for |
372 | * I/O of all non-pmd threads. There will be no actual thread created | |
373 | * for the instance. | |
1c1e46ed AW |
374 | * |
375 | * Each struct has its own flow table and classifier. Packets received | |
376 | * from managed ports are looked up in the corresponding pmd thread's | |
377 | * flow table, and are executed with the found actions. | |
378 | * */ | |
65f13b50 | 379 | struct dp_netdev_pmd_thread { |
6c3eee82 | 380 | struct dp_netdev *dp; |
1c1e46ed | 381 | struct ovs_refcount ref_cnt; /* Every reference must be refcount'ed. */ |
65f13b50 | 382 | struct cmap_node node; /* In 'dp->poll_threads'. */ |
accf8626 AW |
383 | |
384 | pthread_cond_t cond; /* For synchronizing pmd thread reload. */ | |
385 | struct ovs_mutex cond_mutex; /* Mutex for condition variable. */ | |
386 | ||
65f13b50 AW |
387 | /* Per thread exact-match cache. Note, the instance for cpu core |
388 | * NON_PMD_CORE_ID can be accessed by multiple threads, and thusly | |
389 | * need to be protected (e.g. by 'dp_netdev_mutex'). All other | |
390 | * instances will only be accessed by its own pmd thread. */ | |
9bbf1c3d | 391 | struct emc_cache flow_cache; |
1c1e46ed AW |
392 | |
393 | /* Classifier and Flow-Table. | |
394 | * | |
395 | * Writers of 'flow_table' must take the 'flow_mutex'. Corresponding | |
396 | * changes to 'cls' must be made while still holding the 'flow_mutex'. | |
397 | */ | |
398 | struct ovs_mutex flow_mutex; | |
399 | struct dpcls cls; | |
400 | struct cmap flow_table OVS_GUARDED; /* Flow table. */ | |
401 | ||
402 | /* Statistics. */ | |
403 | struct dp_netdev_pmd_stats stats; | |
404 | ||
55e3ca97 DDP |
405 | /* Cycles counters */ |
406 | struct dp_netdev_pmd_cycles cycles; | |
407 | ||
408 | /* Used to count cicles. See 'cycles_counter_end()' */ | |
409 | unsigned long long last_cycles; | |
410 | ||
65f13b50 AW |
411 | struct latch exit_latch; /* For terminating the pmd thread. */ |
412 | atomic_uint change_seq; /* For reloading pmd ports. */ | |
6c3eee82 | 413 | pthread_t thread; |
65f13b50 AW |
414 | int index; /* Idx of this pmd thread among pmd*/ |
415 | /* threads on same numa node. */ | |
bd5131ba | 416 | unsigned core_id; /* CPU core id of this pmd thread. */ |
65f13b50 | 417 | int numa_id; /* numa node id of this pmd thread. */ |
3bcc10c0 DDP |
418 | int tx_qid; /* Queue id used by this pmd thread to |
419 | * send packets on all netdevs */ | |
6553d06b DDP |
420 | |
421 | /* Only a pmd thread can write on its own 'cycles' and 'stats'. | |
422 | * The main thread keeps 'stats_zero' and 'cycles_zero' as base | |
423 | * values and subtracts them from 'stats' and 'cycles' before | |
424 | * reporting to the user */ | |
425 | unsigned long long stats_zero[DP_N_STATS]; | |
426 | uint64_t cycles_zero[PMD_N_CYCLES]; | |
6c3eee82 BP |
427 | }; |
428 | ||
84067a4c JR |
429 | #define PMD_INITIAL_SEQ 1 |
430 | ||
72865317 BP |
431 | /* Interface to netdev-based datapath. */ |
432 | struct dpif_netdev { | |
433 | struct dpif dpif; | |
434 | struct dp_netdev *dp; | |
d33ed218 | 435 | uint64_t last_port_seq; |
72865317 BP |
436 | }; |
437 | ||
8a4e3a85 | 438 | static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no, |
59e6d833 | 439 | struct dp_netdev_port **portp); |
8a4e3a85 | 440 | static int get_port_by_name(struct dp_netdev *dp, const char *devname, |
59e6d833 | 441 | struct dp_netdev_port **portp); |
8a4e3a85 BP |
442 | static void dp_netdev_free(struct dp_netdev *) |
443 | OVS_REQUIRES(dp_netdev_mutex); | |
8a4e3a85 BP |
444 | static int do_add_port(struct dp_netdev *dp, const char *devname, |
445 | const char *type, odp_port_t port_no) | |
59e6d833 | 446 | OVS_REQUIRES(dp->port_mutex); |
c40b890f | 447 | static void do_del_port(struct dp_netdev *dp, struct dp_netdev_port *) |
59e6d833 | 448 | OVS_REQUIRES(dp->port_mutex); |
614c4892 BP |
449 | static int dpif_netdev_open(const struct dpif_class *, const char *name, |
450 | bool create, struct dpif **); | |
65f13b50 | 451 | static void dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 452 | struct dp_packet **, int c, |
41ccaa24 | 453 | bool may_steal, |
4edb9ae9 | 454 | const struct nlattr *actions, |
e4cfed38 | 455 | size_t actions_len); |
65f13b50 | 456 | static void dp_netdev_input(struct dp_netdev_pmd_thread *, |
e14deea0 | 457 | struct dp_packet **, int cnt); |
41ccaa24 | 458 | |
6b31e073 | 459 | static void dp_netdev_disable_upcall(struct dp_netdev *); |
accf8626 | 460 | void dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd); |
65f13b50 AW |
461 | static void dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, |
462 | struct dp_netdev *dp, int index, | |
bd5131ba | 463 | unsigned core_id, int numa_id); |
1c1e46ed | 464 | static void dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd); |
f2eee189 | 465 | static void dp_netdev_set_nonpmd(struct dp_netdev *dp); |
b19befae | 466 | static struct dp_netdev_pmd_thread *dp_netdev_get_pmd(struct dp_netdev *dp, |
bd5131ba | 467 | unsigned core_id); |
1c1e46ed AW |
468 | static struct dp_netdev_pmd_thread * |
469 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos); | |
65f13b50 AW |
470 | static void dp_netdev_destroy_all_pmds(struct dp_netdev *dp); |
471 | static void dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
472 | static void dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
f2eee189 | 473 | static void dp_netdev_reset_pmd_threads(struct dp_netdev *dp); |
1c1e46ed AW |
474 | static bool dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd); |
475 | static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd); | |
476 | static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd); | |
72865317 | 477 | |
67ad54cb | 478 | static inline bool emc_entry_alive(struct emc_entry *ce); |
9bbf1c3d DDP |
479 | static void emc_clear_entry(struct emc_entry *ce); |
480 | ||
481 | static void | |
482 | emc_cache_init(struct emc_cache *flow_cache) | |
483 | { | |
484 | int i; | |
485 | ||
caeb4906 JR |
486 | BUILD_ASSERT(offsetof(struct miniflow, inline_values) == sizeof(uint64_t)); |
487 | ||
67ad54cb | 488 | flow_cache->sweep_idx = 0; |
9bbf1c3d DDP |
489 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { |
490 | flow_cache->entries[i].flow = NULL; | |
0de8783a | 491 | flow_cache->entries[i].key.hash = 0; |
caeb4906 JR |
492 | flow_cache->entries[i].key.len |
493 | = offsetof(struct miniflow, inline_values); | |
0de8783a JR |
494 | miniflow_initialize(&flow_cache->entries[i].key.mf, |
495 | flow_cache->entries[i].key.buf); | |
9bbf1c3d DDP |
496 | } |
497 | } | |
498 | ||
499 | static void | |
500 | emc_cache_uninit(struct emc_cache *flow_cache) | |
501 | { | |
502 | int i; | |
503 | ||
504 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { | |
505 | emc_clear_entry(&flow_cache->entries[i]); | |
506 | } | |
507 | } | |
508 | ||
67ad54cb AW |
509 | /* Check and clear dead flow references slowly (one entry at each |
510 | * invocation). */ | |
511 | static void | |
512 | emc_cache_slow_sweep(struct emc_cache *flow_cache) | |
513 | { | |
514 | struct emc_entry *entry = &flow_cache->entries[flow_cache->sweep_idx]; | |
515 | ||
516 | if (!emc_entry_alive(entry)) { | |
517 | emc_clear_entry(entry); | |
518 | } | |
519 | flow_cache->sweep_idx = (flow_cache->sweep_idx + 1) & EM_FLOW_HASH_MASK; | |
520 | } | |
521 | ||
c4ea7529 BP |
522 | /* Returns true if 'dpif' is a netdev or dummy dpif, false otherwise. */ |
523 | bool | |
524 | dpif_is_netdev(const struct dpif *dpif) | |
525 | { | |
526 | return dpif->dpif_class->open == dpif_netdev_open; | |
527 | } | |
528 | ||
72865317 BP |
529 | static struct dpif_netdev * |
530 | dpif_netdev_cast(const struct dpif *dpif) | |
531 | { | |
c4ea7529 | 532 | ovs_assert(dpif_is_netdev(dpif)); |
72865317 BP |
533 | return CONTAINER_OF(dpif, struct dpif_netdev, dpif); |
534 | } | |
535 | ||
536 | static struct dp_netdev * | |
537 | get_dp_netdev(const struct dpif *dpif) | |
538 | { | |
539 | return dpif_netdev_cast(dpif)->dp; | |
540 | } | |
6553d06b DDP |
541 | \f |
542 | enum pmd_info_type { | |
543 | PMD_INFO_SHOW_STATS, /* show how cpu cycles are spent */ | |
544 | PMD_INFO_CLEAR_STATS /* set the cycles count to 0 */ | |
545 | }; | |
546 | ||
547 | static void | |
548 | pmd_info_show_stats(struct ds *reply, | |
549 | struct dp_netdev_pmd_thread *pmd, | |
550 | unsigned long long stats[DP_N_STATS], | |
551 | uint64_t cycles[PMD_N_CYCLES]) | |
552 | { | |
553 | unsigned long long total_packets = 0; | |
554 | uint64_t total_cycles = 0; | |
555 | int i; | |
556 | ||
557 | /* These loops subtracts reference values ('*_zero') from the counters. | |
558 | * Since loads and stores are relaxed, it might be possible for a '*_zero' | |
559 | * value to be more recent than the current value we're reading from the | |
560 | * counter. This is not a big problem, since these numbers are not | |
561 | * supposed to be too accurate, but we should at least make sure that | |
562 | * the result is not negative. */ | |
563 | for (i = 0; i < DP_N_STATS; i++) { | |
564 | if (stats[i] > pmd->stats_zero[i]) { | |
565 | stats[i] -= pmd->stats_zero[i]; | |
566 | } else { | |
567 | stats[i] = 0; | |
568 | } | |
569 | ||
570 | if (i != DP_STAT_LOST) { | |
571 | /* Lost packets are already included in DP_STAT_MISS */ | |
572 | total_packets += stats[i]; | |
573 | } | |
574 | } | |
575 | ||
576 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
577 | if (cycles[i] > pmd->cycles_zero[i]) { | |
578 | cycles[i] -= pmd->cycles_zero[i]; | |
579 | } else { | |
580 | cycles[i] = 0; | |
581 | } | |
582 | ||
583 | total_cycles += cycles[i]; | |
584 | } | |
585 | ||
586 | ds_put_cstr(reply, (pmd->core_id == NON_PMD_CORE_ID) | |
587 | ? "main thread" : "pmd thread"); | |
588 | ||
589 | if (pmd->numa_id != OVS_NUMA_UNSPEC) { | |
590 | ds_put_format(reply, " numa_id %d", pmd->numa_id); | |
591 | } | |
d5c199ea | 592 | if (pmd->core_id != OVS_CORE_UNSPEC && pmd->core_id != NON_PMD_CORE_ID) { |
bd5131ba | 593 | ds_put_format(reply, " core_id %u", pmd->core_id); |
6553d06b DDP |
594 | } |
595 | ds_put_cstr(reply, ":\n"); | |
596 | ||
597 | ds_put_format(reply, | |
598 | "\temc hits:%llu\n\tmegaflow hits:%llu\n" | |
599 | "\tmiss:%llu\n\tlost:%llu\n", | |
600 | stats[DP_STAT_EXACT_HIT], stats[DP_STAT_MASKED_HIT], | |
601 | stats[DP_STAT_MISS], stats[DP_STAT_LOST]); | |
602 | ||
603 | if (total_cycles == 0) { | |
604 | return; | |
605 | } | |
606 | ||
607 | ds_put_format(reply, | |
608 | "\tpolling cycles:%"PRIu64" (%.02f%%)\n" | |
609 | "\tprocessing cycles:%"PRIu64" (%.02f%%)\n", | |
610 | cycles[PMD_CYCLES_POLLING], | |
611 | cycles[PMD_CYCLES_POLLING] / (double)total_cycles * 100, | |
612 | cycles[PMD_CYCLES_PROCESSING], | |
613 | cycles[PMD_CYCLES_PROCESSING] / (double)total_cycles * 100); | |
614 | ||
615 | if (total_packets == 0) { | |
616 | return; | |
617 | } | |
618 | ||
619 | ds_put_format(reply, | |
620 | "\tavg cycles per packet: %.02f (%"PRIu64"/%llu)\n", | |
621 | total_cycles / (double)total_packets, | |
622 | total_cycles, total_packets); | |
623 | ||
624 | ds_put_format(reply, | |
625 | "\tavg processing cycles per packet: " | |
626 | "%.02f (%"PRIu64"/%llu)\n", | |
627 | cycles[PMD_CYCLES_PROCESSING] / (double)total_packets, | |
628 | cycles[PMD_CYCLES_PROCESSING], total_packets); | |
629 | } | |
630 | ||
631 | static void | |
632 | pmd_info_clear_stats(struct ds *reply OVS_UNUSED, | |
633 | struct dp_netdev_pmd_thread *pmd, | |
634 | unsigned long long stats[DP_N_STATS], | |
635 | uint64_t cycles[PMD_N_CYCLES]) | |
636 | { | |
637 | int i; | |
638 | ||
639 | /* We cannot write 'stats' and 'cycles' (because they're written by other | |
640 | * threads) and we shouldn't change 'stats' (because they're used to count | |
641 | * datapath stats, which must not be cleared here). Instead, we save the | |
642 | * current values and subtract them from the values to be displayed in the | |
643 | * future */ | |
644 | for (i = 0; i < DP_N_STATS; i++) { | |
645 | pmd->stats_zero[i] = stats[i]; | |
646 | } | |
647 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
648 | pmd->cycles_zero[i] = cycles[i]; | |
649 | } | |
650 | } | |
651 | ||
652 | static void | |
653 | dpif_netdev_pmd_info(struct unixctl_conn *conn, int argc, const char *argv[], | |
654 | void *aux) | |
655 | { | |
656 | struct ds reply = DS_EMPTY_INITIALIZER; | |
657 | struct dp_netdev_pmd_thread *pmd; | |
658 | struct dp_netdev *dp = NULL; | |
659 | enum pmd_info_type type = *(enum pmd_info_type *) aux; | |
660 | ||
661 | ovs_mutex_lock(&dp_netdev_mutex); | |
662 | ||
663 | if (argc == 2) { | |
664 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
665 | } else if (shash_count(&dp_netdevs) == 1) { | |
666 | /* There's only one datapath */ | |
667 | dp = shash_first(&dp_netdevs)->data; | |
668 | } | |
669 | ||
670 | if (!dp) { | |
671 | ovs_mutex_unlock(&dp_netdev_mutex); | |
672 | unixctl_command_reply_error(conn, | |
673 | "please specify an existing datapath"); | |
674 | return; | |
675 | } | |
676 | ||
677 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
678 | unsigned long long stats[DP_N_STATS]; | |
679 | uint64_t cycles[PMD_N_CYCLES]; | |
680 | int i; | |
681 | ||
682 | /* Read current stats and cycle counters */ | |
683 | for (i = 0; i < ARRAY_SIZE(stats); i++) { | |
684 | atomic_read_relaxed(&pmd->stats.n[i], &stats[i]); | |
685 | } | |
686 | for (i = 0; i < ARRAY_SIZE(cycles); i++) { | |
687 | atomic_read_relaxed(&pmd->cycles.n[i], &cycles[i]); | |
688 | } | |
689 | ||
690 | if (type == PMD_INFO_CLEAR_STATS) { | |
691 | pmd_info_clear_stats(&reply, pmd, stats, cycles); | |
692 | } else if (type == PMD_INFO_SHOW_STATS) { | |
693 | pmd_info_show_stats(&reply, pmd, stats, cycles); | |
694 | } | |
695 | } | |
696 | ||
697 | ovs_mutex_unlock(&dp_netdev_mutex); | |
698 | ||
699 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
700 | ds_destroy(&reply); | |
701 | } | |
702 | \f | |
703 | static int | |
704 | dpif_netdev_init(void) | |
705 | { | |
706 | static enum pmd_info_type show_aux = PMD_INFO_SHOW_STATS, | |
707 | clear_aux = PMD_INFO_CLEAR_STATS; | |
708 | ||
709 | unixctl_command_register("dpif-netdev/pmd-stats-show", "[dp]", | |
710 | 0, 1, dpif_netdev_pmd_info, | |
711 | (void *)&show_aux); | |
712 | unixctl_command_register("dpif-netdev/pmd-stats-clear", "[dp]", | |
713 | 0, 1, dpif_netdev_pmd_info, | |
714 | (void *)&clear_aux); | |
715 | return 0; | |
716 | } | |
72865317 | 717 | |
2197d7ab | 718 | static int |
2240af25 DDP |
719 | dpif_netdev_enumerate(struct sset *all_dps, |
720 | const struct dpif_class *dpif_class) | |
2197d7ab GL |
721 | { |
722 | struct shash_node *node; | |
723 | ||
97be1538 | 724 | ovs_mutex_lock(&dp_netdev_mutex); |
2197d7ab | 725 | SHASH_FOR_EACH(node, &dp_netdevs) { |
2240af25 DDP |
726 | struct dp_netdev *dp = node->data; |
727 | if (dpif_class != dp->class) { | |
728 | /* 'dp_netdevs' contains both "netdev" and "dummy" dpifs. | |
729 | * If the class doesn't match, skip this dpif. */ | |
730 | continue; | |
731 | } | |
2197d7ab GL |
732 | sset_add(all_dps, node->name); |
733 | } | |
97be1538 | 734 | ovs_mutex_unlock(&dp_netdev_mutex); |
5279f8fd | 735 | |
2197d7ab GL |
736 | return 0; |
737 | } | |
738 | ||
add90f6f EJ |
739 | static bool |
740 | dpif_netdev_class_is_dummy(const struct dpif_class *class) | |
741 | { | |
742 | return class != &dpif_netdev_class; | |
743 | } | |
744 | ||
0aeaabc8 JP |
745 | static const char * |
746 | dpif_netdev_port_open_type(const struct dpif_class *class, const char *type) | |
747 | { | |
748 | return strcmp(type, "internal") ? type | |
add90f6f | 749 | : dpif_netdev_class_is_dummy(class) ? "dummy" |
0aeaabc8 JP |
750 | : "tap"; |
751 | } | |
752 | ||
72865317 BP |
753 | static struct dpif * |
754 | create_dpif_netdev(struct dp_netdev *dp) | |
755 | { | |
462278db | 756 | uint16_t netflow_id = hash_string(dp->name, 0); |
72865317 | 757 | struct dpif_netdev *dpif; |
72865317 | 758 | |
6a8267c5 | 759 | ovs_refcount_ref(&dp->ref_cnt); |
72865317 | 760 | |
72865317 | 761 | dpif = xmalloc(sizeof *dpif); |
614c4892 | 762 | dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id); |
72865317 | 763 | dpif->dp = dp; |
d33ed218 | 764 | dpif->last_port_seq = seq_read(dp->port_seq); |
72865317 BP |
765 | |
766 | return &dpif->dpif; | |
767 | } | |
768 | ||
4e022ec0 AW |
769 | /* Choose an unused, non-zero port number and return it on success. |
770 | * Return ODPP_NONE on failure. */ | |
771 | static odp_port_t | |
e44768b7 | 772 | choose_port(struct dp_netdev *dp, const char *name) |
59e6d833 | 773 | OVS_REQUIRES(dp->port_mutex) |
e44768b7 | 774 | { |
4e022ec0 | 775 | uint32_t port_no; |
e44768b7 JP |
776 | |
777 | if (dp->class != &dpif_netdev_class) { | |
778 | const char *p; | |
779 | int start_no = 0; | |
780 | ||
781 | /* If the port name begins with "br", start the number search at | |
782 | * 100 to make writing tests easier. */ | |
783 | if (!strncmp(name, "br", 2)) { | |
784 | start_no = 100; | |
785 | } | |
786 | ||
787 | /* If the port name contains a number, try to assign that port number. | |
788 | * This can make writing unit tests easier because port numbers are | |
789 | * predictable. */ | |
790 | for (p = name; *p != '\0'; p++) { | |
791 | if (isdigit((unsigned char) *p)) { | |
792 | port_no = start_no + strtol(p, NULL, 10); | |
ff073a71 BP |
793 | if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE) |
794 | && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 795 | return u32_to_odp(port_no); |
e44768b7 JP |
796 | } |
797 | break; | |
798 | } | |
799 | } | |
800 | } | |
801 | ||
ff073a71 BP |
802 | for (port_no = 1; port_no <= UINT16_MAX; port_no++) { |
803 | if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 804 | return u32_to_odp(port_no); |
e44768b7 JP |
805 | } |
806 | } | |
807 | ||
4e022ec0 | 808 | return ODPP_NONE; |
e44768b7 JP |
809 | } |
810 | ||
72865317 | 811 | static int |
614c4892 BP |
812 | create_dp_netdev(const char *name, const struct dpif_class *class, |
813 | struct dp_netdev **dpp) | |
8a4e3a85 | 814 | OVS_REQUIRES(dp_netdev_mutex) |
72865317 BP |
815 | { |
816 | struct dp_netdev *dp; | |
817 | int error; | |
72865317 | 818 | |
462278db | 819 | dp = xzalloc(sizeof *dp); |
8a4e3a85 BP |
820 | shash_add(&dp_netdevs, name, dp); |
821 | ||
822 | *CONST_CAST(const struct dpif_class **, &dp->class) = class; | |
823 | *CONST_CAST(const char **, &dp->name) = xstrdup(name); | |
6a8267c5 | 824 | ovs_refcount_init(&dp->ref_cnt); |
1a65ba85 | 825 | atomic_flag_clear(&dp->destroyed); |
8a4e3a85 | 826 | |
59e6d833 BP |
827 | ovs_mutex_init(&dp->port_mutex); |
828 | cmap_init(&dp->ports); | |
d33ed218 | 829 | dp->port_seq = seq_create(); |
6b31e073 RW |
830 | fat_rwlock_init(&dp->upcall_rwlock); |
831 | ||
832 | /* Disable upcalls by default. */ | |
833 | dp_netdev_disable_upcall(dp); | |
623540e4 | 834 | dp->upcall_aux = NULL; |
6b31e073 | 835 | dp->upcall_cb = NULL; |
e44768b7 | 836 | |
65f13b50 AW |
837 | cmap_init(&dp->poll_threads); |
838 | ovs_mutex_init_recursive(&dp->non_pmd_mutex); | |
839 | ovsthread_key_create(&dp->per_pmd_key, NULL); | |
840 | ||
f2eee189 AW |
841 | dp_netdev_set_nonpmd(dp); |
842 | dp->n_dpdk_rxqs = NR_QUEUE; | |
65f13b50 | 843 | |
59e6d833 | 844 | ovs_mutex_lock(&dp->port_mutex); |
4e022ec0 | 845 | error = do_add_port(dp, name, "internal", ODPP_LOCAL); |
59e6d833 | 846 | ovs_mutex_unlock(&dp->port_mutex); |
72865317 BP |
847 | if (error) { |
848 | dp_netdev_free(dp); | |
462278db | 849 | return error; |
72865317 BP |
850 | } |
851 | ||
a36de779 | 852 | dp->last_tnl_conf_seq = seq_read(tnl_conf_seq); |
462278db | 853 | *dpp = dp; |
72865317 BP |
854 | return 0; |
855 | } | |
856 | ||
857 | static int | |
614c4892 | 858 | dpif_netdev_open(const struct dpif_class *class, const char *name, |
4a387741 | 859 | bool create, struct dpif **dpifp) |
72865317 | 860 | { |
462278db | 861 | struct dp_netdev *dp; |
5279f8fd | 862 | int error; |
462278db | 863 | |
97be1538 | 864 | ovs_mutex_lock(&dp_netdev_mutex); |
462278db BP |
865 | dp = shash_find_data(&dp_netdevs, name); |
866 | if (!dp) { | |
5279f8fd | 867 | error = create ? create_dp_netdev(name, class, &dp) : ENODEV; |
72865317 | 868 | } else { |
5279f8fd BP |
869 | error = (dp->class != class ? EINVAL |
870 | : create ? EEXIST | |
871 | : 0); | |
872 | } | |
873 | if (!error) { | |
874 | *dpifp = create_dpif_netdev(dp); | |
6b31e073 | 875 | dp->dpif = *dpifp; |
72865317 | 876 | } |
97be1538 | 877 | ovs_mutex_unlock(&dp_netdev_mutex); |
462278db | 878 | |
5279f8fd | 879 | return error; |
72865317 BP |
880 | } |
881 | ||
88ace79b DDP |
882 | static void |
883 | dp_netdev_destroy_upcall_lock(struct dp_netdev *dp) | |
884 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
885 | { | |
886 | /* Check that upcalls are disabled, i.e. that the rwlock is taken */ | |
887 | ovs_assert(fat_rwlock_tryrdlock(&dp->upcall_rwlock)); | |
888 | ||
889 | /* Before freeing a lock we should release it */ | |
890 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
891 | fat_rwlock_destroy(&dp->upcall_rwlock); | |
892 | } | |
893 | ||
8a4e3a85 BP |
894 | /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp' |
895 | * through the 'dp_netdevs' shash while freeing 'dp'. */ | |
1ba530f4 BP |
896 | static void |
897 | dp_netdev_free(struct dp_netdev *dp) | |
8a4e3a85 | 898 | OVS_REQUIRES(dp_netdev_mutex) |
1ba530f4 | 899 | { |
59e6d833 | 900 | struct dp_netdev_port *port; |
4ad28026 | 901 | |
8a4e3a85 BP |
902 | shash_find_and_delete(&dp_netdevs, dp->name); |
903 | ||
65f13b50 | 904 | dp_netdev_destroy_all_pmds(dp); |
8bd89cdc | 905 | cmap_destroy(&dp->poll_threads); |
65f13b50 AW |
906 | ovs_mutex_destroy(&dp->non_pmd_mutex); |
907 | ovsthread_key_delete(dp->per_pmd_key); | |
6c3eee82 | 908 | |
59e6d833 | 909 | ovs_mutex_lock(&dp->port_mutex); |
a532e683 | 910 | CMAP_FOR_EACH (port, node, &dp->ports) { |
c40b890f | 911 | do_del_port(dp, port); |
1ba530f4 | 912 | } |
59e6d833 | 913 | ovs_mutex_unlock(&dp->port_mutex); |
51852a57 | 914 | |
d33ed218 | 915 | seq_destroy(dp->port_seq); |
59e6d833 | 916 | cmap_destroy(&dp->ports); |
88ace79b DDP |
917 | |
918 | /* Upcalls must be disabled at this point */ | |
919 | dp_netdev_destroy_upcall_lock(dp); | |
9bbf1c3d | 920 | |
f2eee189 | 921 | free(dp->pmd_cmask); |
8a4e3a85 | 922 | free(CONST_CAST(char *, dp->name)); |
72865317 BP |
923 | free(dp); |
924 | } | |
925 | ||
8a4e3a85 BP |
926 | static void |
927 | dp_netdev_unref(struct dp_netdev *dp) | |
928 | { | |
929 | if (dp) { | |
930 | /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't | |
931 | * get a new reference to 'dp' through the 'dp_netdevs' shash. */ | |
932 | ovs_mutex_lock(&dp_netdev_mutex); | |
24f83812 | 933 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
8a4e3a85 BP |
934 | dp_netdev_free(dp); |
935 | } | |
936 | ovs_mutex_unlock(&dp_netdev_mutex); | |
937 | } | |
938 | } | |
939 | ||
72865317 BP |
940 | static void |
941 | dpif_netdev_close(struct dpif *dpif) | |
942 | { | |
943 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 944 | |
8a4e3a85 | 945 | dp_netdev_unref(dp); |
72865317 BP |
946 | free(dpif); |
947 | } | |
948 | ||
949 | static int | |
7dab847a | 950 | dpif_netdev_destroy(struct dpif *dpif) |
72865317 BP |
951 | { |
952 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 953 | |
6a8267c5 | 954 | if (!atomic_flag_test_and_set(&dp->destroyed)) { |
24f83812 | 955 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
6a8267c5 BP |
956 | /* Can't happen: 'dpif' still owns a reference to 'dp'. */ |
957 | OVS_NOT_REACHED(); | |
958 | } | |
959 | } | |
5279f8fd | 960 | |
72865317 BP |
961 | return 0; |
962 | } | |
963 | ||
eb94da30 DDP |
964 | /* Add 'n' to the atomic variable 'var' non-atomically and using relaxed |
965 | * load/store semantics. While the increment is not atomic, the load and | |
966 | * store operations are, making it impossible to read inconsistent values. | |
967 | * | |
968 | * This is used to update thread local stats counters. */ | |
969 | static void | |
970 | non_atomic_ullong_add(atomic_ullong *var, unsigned long long n) | |
971 | { | |
972 | unsigned long long tmp; | |
973 | ||
974 | atomic_read_relaxed(var, &tmp); | |
975 | tmp += n; | |
976 | atomic_store_relaxed(var, tmp); | |
977 | } | |
978 | ||
72865317 | 979 | static int |
a8d9304d | 980 | dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) |
72865317 BP |
981 | { |
982 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed | 983 | struct dp_netdev_pmd_thread *pmd; |
8a4e3a85 | 984 | |
1c1e46ed AW |
985 | stats->n_flows = stats->n_hit = stats->n_missed = stats->n_lost = 0; |
986 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
eb94da30 | 987 | unsigned long long n; |
1c1e46ed | 988 | stats->n_flows += cmap_count(&pmd->flow_table); |
eb94da30 | 989 | |
abcf3ef4 DDP |
990 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MASKED_HIT], &n); |
991 | stats->n_hit += n; | |
992 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_EXACT_HIT], &n); | |
eb94da30 DDP |
993 | stats->n_hit += n; |
994 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MISS], &n); | |
995 | stats->n_missed += n; | |
996 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_LOST], &n); | |
997 | stats->n_lost += n; | |
51852a57 | 998 | } |
1ce3fa06 | 999 | stats->n_masks = UINT32_MAX; |
847108dc | 1000 | stats->n_mask_hit = UINT64_MAX; |
5279f8fd | 1001 | |
72865317 BP |
1002 | return 0; |
1003 | } | |
1004 | ||
e4cfed38 | 1005 | static void |
65f13b50 | 1006 | dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd) |
e4cfed38 | 1007 | { |
65f13b50 AW |
1008 | int old_seq; |
1009 | ||
accf8626 AW |
1010 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1011 | return; | |
1012 | } | |
1013 | ||
1014 | ovs_mutex_lock(&pmd->cond_mutex); | |
65f13b50 | 1015 | atomic_add_relaxed(&pmd->change_seq, 1, &old_seq); |
accf8626 AW |
1016 | ovs_mutex_cond_wait(&pmd->cond, &pmd->cond_mutex); |
1017 | ovs_mutex_unlock(&pmd->cond_mutex); | |
65f13b50 | 1018 | } |
e4cfed38 | 1019 | |
65f13b50 AW |
1020 | /* Causes all pmd threads to reload its tx/rx devices. |
1021 | * Must be called after adding/removing ports. */ | |
1022 | static void | |
1023 | dp_netdev_reload_pmds(struct dp_netdev *dp) | |
1024 | { | |
1025 | struct dp_netdev_pmd_thread *pmd; | |
e4cfed38 | 1026 | |
65f13b50 AW |
1027 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { |
1028 | dp_netdev_reload_pmd__(pmd); | |
84067a4c | 1029 | } |
e4cfed38 PS |
1030 | } |
1031 | ||
59e6d833 BP |
1032 | static uint32_t |
1033 | hash_port_no(odp_port_t port_no) | |
1034 | { | |
1035 | return hash_int(odp_to_u32(port_no), 0); | |
1036 | } | |
1037 | ||
72865317 | 1038 | static int |
c3827f61 | 1039 | do_add_port(struct dp_netdev *dp, const char *devname, const char *type, |
4e022ec0 | 1040 | odp_port_t port_no) |
59e6d833 | 1041 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1042 | { |
4b609110 | 1043 | struct netdev_saved_flags *sf; |
72865317 BP |
1044 | struct dp_netdev_port *port; |
1045 | struct netdev *netdev; | |
2499a8ce | 1046 | enum netdev_flags flags; |
0cbfe35d | 1047 | const char *open_type; |
72865317 | 1048 | int error; |
55c955bd | 1049 | int i; |
72865317 | 1050 | |
17050610 BP |
1051 | /* Reject devices already in 'dp'. */ |
1052 | if (!get_port_by_name(dp, devname, &port)) { | |
1053 | return EEXIST; | |
1054 | } | |
72865317 BP |
1055 | |
1056 | /* Open and validate network device. */ | |
0aeaabc8 | 1057 | open_type = dpif_netdev_port_open_type(dp->class, type); |
0cbfe35d | 1058 | error = netdev_open(devname, open_type, &netdev); |
72865317 BP |
1059 | if (error) { |
1060 | return error; | |
1061 | } | |
72865317 BP |
1062 | /* XXX reject non-Ethernet devices */ |
1063 | ||
2499a8ce AC |
1064 | netdev_get_flags(netdev, &flags); |
1065 | if (flags & NETDEV_LOOPBACK) { | |
1066 | VLOG_ERR("%s: cannot add a loopback device", devname); | |
1067 | netdev_close(netdev); | |
1068 | return EINVAL; | |
1069 | } | |
1070 | ||
5a034064 AW |
1071 | if (netdev_is_pmd(netdev)) { |
1072 | int n_cores = ovs_numa_get_n_cores(); | |
1073 | ||
1074 | if (n_cores == OVS_CORE_UNSPEC) { | |
1075 | VLOG_ERR("%s, cannot get cpu core info", devname); | |
1076 | return ENOENT; | |
1077 | } | |
1078 | /* There can only be ovs_numa_get_n_cores() pmd threads, | |
3bcc10c0 DDP |
1079 | * so creates a txq for each, and one extra for the non |
1080 | * pmd threads. */ | |
1081 | error = netdev_set_multiq(netdev, n_cores + 1, dp->n_dpdk_rxqs); | |
7251515e | 1082 | if (error && (error != EOPNOTSUPP)) { |
5a034064 AW |
1083 | VLOG_ERR("%s, cannot set multiq", devname); |
1084 | return errno; | |
1085 | } | |
1086 | } | |
e4cfed38 | 1087 | port = xzalloc(sizeof *port); |
efa2bcbb | 1088 | port->md = PKT_METADATA_INITIALIZER(port_no); |
e4cfed38 | 1089 | port->netdev = netdev; |
55c955bd | 1090 | port->rxq = xmalloc(sizeof *port->rxq * netdev_n_rxq(netdev)); |
e4cfed38 | 1091 | port->type = xstrdup(type); |
55c955bd PS |
1092 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1093 | error = netdev_rxq_open(netdev, &port->rxq[i], i); | |
1094 | if (error | |
1095 | && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) { | |
1096 | VLOG_ERR("%s: cannot receive packets on this network device (%s)", | |
1097 | devname, ovs_strerror(errno)); | |
1098 | netdev_close(netdev); | |
16bea12c TG |
1099 | free(port->type); |
1100 | free(port->rxq); | |
1101 | free(port); | |
55c955bd PS |
1102 | return error; |
1103 | } | |
7b6b0ef4 BP |
1104 | } |
1105 | ||
4b609110 | 1106 | error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); |
72865317 | 1107 | if (error) { |
55c955bd PS |
1108 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1109 | netdev_rxq_close(port->rxq[i]); | |
1110 | } | |
72865317 | 1111 | netdev_close(netdev); |
16bea12c | 1112 | free(port->type); |
f7791740 | 1113 | free(port->rxq); |
e4cfed38 | 1114 | free(port); |
72865317 BP |
1115 | return error; |
1116 | } | |
4b609110 | 1117 | port->sf = sf; |
e4cfed38 | 1118 | |
f7d63652 AW |
1119 | ovs_refcount_init(&port->ref_cnt); |
1120 | cmap_insert(&dp->ports, &port->node, hash_port_no(port_no)); | |
1121 | ||
e4cfed38 | 1122 | if (netdev_is_pmd(netdev)) { |
65f13b50 AW |
1123 | dp_netdev_set_pmds_on_numa(dp, netdev_get_numa_id(netdev)); |
1124 | dp_netdev_reload_pmds(dp); | |
e4cfed38 | 1125 | } |
d33ed218 | 1126 | seq_change(dp->port_seq); |
72865317 BP |
1127 | |
1128 | return 0; | |
1129 | } | |
1130 | ||
247527db BP |
1131 | static int |
1132 | dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, | |
4e022ec0 | 1133 | odp_port_t *port_nop) |
247527db BP |
1134 | { |
1135 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3aa30359 BP |
1136 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1137 | const char *dpif_port; | |
4e022ec0 | 1138 | odp_port_t port_no; |
5279f8fd | 1139 | int error; |
247527db | 1140 | |
59e6d833 | 1141 | ovs_mutex_lock(&dp->port_mutex); |
3aa30359 | 1142 | dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
4e022ec0 | 1143 | if (*port_nop != ODPP_NONE) { |
ff073a71 BP |
1144 | port_no = *port_nop; |
1145 | error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0; | |
232dfa4a | 1146 | } else { |
3aa30359 | 1147 | port_no = choose_port(dp, dpif_port); |
5279f8fd | 1148 | error = port_no == ODPP_NONE ? EFBIG : 0; |
232dfa4a | 1149 | } |
5279f8fd | 1150 | if (!error) { |
247527db | 1151 | *port_nop = port_no; |
5279f8fd | 1152 | error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); |
247527db | 1153 | } |
59e6d833 | 1154 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1155 | |
1156 | return error; | |
72865317 BP |
1157 | } |
1158 | ||
1159 | static int | |
4e022ec0 | 1160 | dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no) |
72865317 BP |
1161 | { |
1162 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd BP |
1163 | int error; |
1164 | ||
59e6d833 | 1165 | ovs_mutex_lock(&dp->port_mutex); |
c40b890f BP |
1166 | if (port_no == ODPP_LOCAL) { |
1167 | error = EINVAL; | |
1168 | } else { | |
1169 | struct dp_netdev_port *port; | |
1170 | ||
1171 | error = get_port_by_number(dp, port_no, &port); | |
1172 | if (!error) { | |
1173 | do_del_port(dp, port); | |
1174 | } | |
1175 | } | |
59e6d833 | 1176 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1177 | |
1178 | return error; | |
72865317 BP |
1179 | } |
1180 | ||
1181 | static bool | |
4e022ec0 | 1182 | is_valid_port_number(odp_port_t port_no) |
72865317 | 1183 | { |
ff073a71 BP |
1184 | return port_no != ODPP_NONE; |
1185 | } | |
1186 | ||
1187 | static struct dp_netdev_port * | |
1188 | dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no) | |
1189 | { | |
1190 | struct dp_netdev_port *port; | |
1191 | ||
59e6d833 | 1192 | CMAP_FOR_EACH_WITH_HASH (port, node, hash_port_no(port_no), &dp->ports) { |
efa2bcbb | 1193 | if (port->md.in_port.odp_port == port_no) { |
ff073a71 BP |
1194 | return port; |
1195 | } | |
1196 | } | |
1197 | return NULL; | |
72865317 BP |
1198 | } |
1199 | ||
1200 | static int | |
1201 | get_port_by_number(struct dp_netdev *dp, | |
4e022ec0 | 1202 | odp_port_t port_no, struct dp_netdev_port **portp) |
72865317 BP |
1203 | { |
1204 | if (!is_valid_port_number(port_no)) { | |
1205 | *portp = NULL; | |
1206 | return EINVAL; | |
1207 | } else { | |
ff073a71 | 1208 | *portp = dp_netdev_lookup_port(dp, port_no); |
72865317 BP |
1209 | return *portp ? 0 : ENOENT; |
1210 | } | |
1211 | } | |
1212 | ||
b284085e PS |
1213 | static void |
1214 | port_ref(struct dp_netdev_port *port) | |
1215 | { | |
1216 | if (port) { | |
1217 | ovs_refcount_ref(&port->ref_cnt); | |
1218 | } | |
1219 | } | |
1220 | ||
a1fdee13 AW |
1221 | static bool |
1222 | port_try_ref(struct dp_netdev_port *port) | |
1223 | { | |
1224 | if (port) { | |
1225 | return ovs_refcount_try_ref_rcu(&port->ref_cnt); | |
1226 | } | |
1227 | ||
1228 | return false; | |
1229 | } | |
1230 | ||
59e6d833 BP |
1231 | static void |
1232 | port_unref(struct dp_netdev_port *port) | |
1233 | { | |
24f83812 | 1234 | if (port && ovs_refcount_unref_relaxed(&port->ref_cnt) == 1) { |
accf8626 AW |
1235 | int n_rxq = netdev_n_rxq(port->netdev); |
1236 | int i; | |
1237 | ||
1238 | netdev_close(port->netdev); | |
1239 | netdev_restore_flags(port->sf); | |
1240 | ||
1241 | for (i = 0; i < n_rxq; i++) { | |
1242 | netdev_rxq_close(port->rxq[i]); | |
1243 | } | |
1244 | free(port->rxq); | |
1245 | free(port->type); | |
1246 | free(port); | |
b284085e PS |
1247 | } |
1248 | } | |
1249 | ||
72865317 BP |
1250 | static int |
1251 | get_port_by_name(struct dp_netdev *dp, | |
1252 | const char *devname, struct dp_netdev_port **portp) | |
59e6d833 | 1253 | OVS_REQUIRES(dp->port_mutex) |
72865317 BP |
1254 | { |
1255 | struct dp_netdev_port *port; | |
1256 | ||
a532e683 | 1257 | CMAP_FOR_EACH (port, node, &dp->ports) { |
3efb6063 | 1258 | if (!strcmp(netdev_get_name(port->netdev), devname)) { |
72865317 BP |
1259 | *portp = port; |
1260 | return 0; | |
1261 | } | |
1262 | } | |
1263 | return ENOENT; | |
1264 | } | |
1265 | ||
65f13b50 AW |
1266 | static int |
1267 | get_n_pmd_threads_on_numa(struct dp_netdev *dp, int numa_id) | |
1268 | { | |
1269 | struct dp_netdev_pmd_thread *pmd; | |
1270 | int n_pmds = 0; | |
1271 | ||
1272 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1273 | if (pmd->numa_id == numa_id) { | |
1274 | n_pmds++; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | return n_pmds; | |
1279 | } | |
1280 | ||
1281 | /* Returns 'true' if there is a port with pmd netdev and the netdev | |
1282 | * is on numa node 'numa_id'. */ | |
1283 | static bool | |
1284 | has_pmd_port_for_numa(struct dp_netdev *dp, int numa_id) | |
1285 | { | |
1286 | struct dp_netdev_port *port; | |
1287 | ||
1288 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
1289 | if (netdev_is_pmd(port->netdev) | |
1290 | && netdev_get_numa_id(port->netdev) == numa_id) { | |
1291 | return true; | |
1292 | } | |
1293 | } | |
1294 | ||
1295 | return false; | |
1296 | } | |
1297 | ||
1298 | ||
c40b890f BP |
1299 | static void |
1300 | do_del_port(struct dp_netdev *dp, struct dp_netdev_port *port) | |
59e6d833 | 1301 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1302 | { |
efa2bcbb DDP |
1303 | cmap_remove(&dp->ports, &port->node, |
1304 | hash_odp_port(port->md.in_port.odp_port)); | |
d33ed218 | 1305 | seq_change(dp->port_seq); |
e4cfed38 | 1306 | if (netdev_is_pmd(port->netdev)) { |
65f13b50 AW |
1307 | int numa_id = netdev_get_numa_id(port->netdev); |
1308 | ||
1309 | /* If there is no netdev on the numa node, deletes the pmd threads | |
1310 | * for that numa. Else, just reloads the queues. */ | |
1311 | if (!has_pmd_port_for_numa(dp, numa_id)) { | |
1312 | dp_netdev_del_pmds_on_numa(dp, numa_id); | |
1313 | } | |
1314 | dp_netdev_reload_pmds(dp); | |
e4cfed38 | 1315 | } |
72865317 | 1316 | |
b284085e | 1317 | port_unref(port); |
72865317 BP |
1318 | } |
1319 | ||
1320 | static void | |
4c738a8d BP |
1321 | answer_port_query(const struct dp_netdev_port *port, |
1322 | struct dpif_port *dpif_port) | |
72865317 | 1323 | { |
3efb6063 | 1324 | dpif_port->name = xstrdup(netdev_get_name(port->netdev)); |
0cbfe35d | 1325 | dpif_port->type = xstrdup(port->type); |
efa2bcbb | 1326 | dpif_port->port_no = port->md.in_port.odp_port; |
72865317 BP |
1327 | } |
1328 | ||
1329 | static int | |
4e022ec0 | 1330 | dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, |
4c738a8d | 1331 | struct dpif_port *dpif_port) |
72865317 BP |
1332 | { |
1333 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1334 | struct dp_netdev_port *port; | |
1335 | int error; | |
1336 | ||
1337 | error = get_port_by_number(dp, port_no, &port); | |
4afba28d | 1338 | if (!error && dpif_port) { |
4c738a8d | 1339 | answer_port_query(port, dpif_port); |
72865317 | 1340 | } |
5279f8fd | 1341 | |
72865317 BP |
1342 | return error; |
1343 | } | |
1344 | ||
1345 | static int | |
1346 | dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, | |
4c738a8d | 1347 | struct dpif_port *dpif_port) |
72865317 BP |
1348 | { |
1349 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1350 | struct dp_netdev_port *port; | |
1351 | int error; | |
1352 | ||
59e6d833 | 1353 | ovs_mutex_lock(&dp->port_mutex); |
72865317 | 1354 | error = get_port_by_name(dp, devname, &port); |
4afba28d | 1355 | if (!error && dpif_port) { |
4c738a8d | 1356 | answer_port_query(port, dpif_port); |
72865317 | 1357 | } |
59e6d833 | 1358 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1359 | |
72865317 BP |
1360 | return error; |
1361 | } | |
1362 | ||
61e7deb1 BP |
1363 | static void |
1364 | dp_netdev_flow_free(struct dp_netdev_flow *flow) | |
1365 | { | |
61e7deb1 | 1366 | dp_netdev_actions_free(dp_netdev_flow_get_actions(flow)); |
61e7deb1 BP |
1367 | free(flow); |
1368 | } | |
1369 | ||
ed79f89a DDP |
1370 | static void dp_netdev_flow_unref(struct dp_netdev_flow *flow) |
1371 | { | |
1372 | if (ovs_refcount_unref_relaxed(&flow->ref_cnt) == 1) { | |
1373 | ovsrcu_postpone(dp_netdev_flow_free, flow); | |
1374 | } | |
1375 | } | |
1376 | ||
70e5ed6f JS |
1377 | static uint32_t |
1378 | dp_netdev_flow_hash(const ovs_u128 *ufid) | |
1379 | { | |
1380 | return ufid->u32[0]; | |
1381 | } | |
1382 | ||
72865317 | 1383 | static void |
1c1e46ed AW |
1384 | dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, |
1385 | struct dp_netdev_flow *flow) | |
1386 | OVS_REQUIRES(pmd->flow_mutex) | |
72865317 | 1387 | { |
9f361d6b | 1388 | struct cmap_node *node = CONST_CAST(struct cmap_node *, &flow->node); |
2c0ea78f | 1389 | |
1c1e46ed AW |
1390 | dpcls_remove(&pmd->cls, &flow->cr); |
1391 | cmap_remove(&pmd->flow_table, node, dp_netdev_flow_hash(&flow->ufid)); | |
9bbf1c3d | 1392 | flow->dead = true; |
ed79f89a DDP |
1393 | |
1394 | dp_netdev_flow_unref(flow); | |
72865317 BP |
1395 | } |
1396 | ||
1397 | static void | |
1c1e46ed | 1398 | dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd) |
72865317 | 1399 | { |
78c8df12 | 1400 | struct dp_netdev_flow *netdev_flow; |
72865317 | 1401 | |
1c1e46ed AW |
1402 | ovs_mutex_lock(&pmd->flow_mutex); |
1403 | CMAP_FOR_EACH (netdev_flow, node, &pmd->flow_table) { | |
1404 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); | |
72865317 | 1405 | } |
1c1e46ed | 1406 | ovs_mutex_unlock(&pmd->flow_mutex); |
72865317 BP |
1407 | } |
1408 | ||
1409 | static int | |
1410 | dpif_netdev_flow_flush(struct dpif *dpif) | |
1411 | { | |
1412 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed AW |
1413 | struct dp_netdev_pmd_thread *pmd; |
1414 | ||
1415 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1416 | dp_netdev_pmd_flow_flush(pmd); | |
1417 | } | |
5279f8fd | 1418 | |
72865317 BP |
1419 | return 0; |
1420 | } | |
1421 | ||
b0ec0f27 | 1422 | struct dp_netdev_port_state { |
59e6d833 | 1423 | struct cmap_position position; |
4c738a8d | 1424 | char *name; |
b0ec0f27 BP |
1425 | }; |
1426 | ||
1427 | static int | |
1428 | dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) | |
1429 | { | |
1430 | *statep = xzalloc(sizeof(struct dp_netdev_port_state)); | |
1431 | return 0; | |
1432 | } | |
1433 | ||
72865317 | 1434 | static int |
b0ec0f27 | 1435 | dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, |
4c738a8d | 1436 | struct dpif_port *dpif_port) |
72865317 | 1437 | { |
b0ec0f27 | 1438 | struct dp_netdev_port_state *state = state_; |
72865317 | 1439 | struct dp_netdev *dp = get_dp_netdev(dpif); |
59e6d833 | 1440 | struct cmap_node *node; |
ff073a71 | 1441 | int retval; |
72865317 | 1442 | |
59e6d833 | 1443 | node = cmap_next_position(&dp->ports, &state->position); |
ff073a71 BP |
1444 | if (node) { |
1445 | struct dp_netdev_port *port; | |
5279f8fd | 1446 | |
ff073a71 BP |
1447 | port = CONTAINER_OF(node, struct dp_netdev_port, node); |
1448 | ||
1449 | free(state->name); | |
1450 | state->name = xstrdup(netdev_get_name(port->netdev)); | |
1451 | dpif_port->name = state->name; | |
1452 | dpif_port->type = port->type; | |
efa2bcbb | 1453 | dpif_port->port_no = port->md.in_port.odp_port; |
ff073a71 BP |
1454 | |
1455 | retval = 0; | |
1456 | } else { | |
1457 | retval = EOF; | |
72865317 | 1458 | } |
5279f8fd | 1459 | |
ff073a71 | 1460 | return retval; |
b0ec0f27 BP |
1461 | } |
1462 | ||
1463 | static int | |
4c738a8d | 1464 | dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) |
b0ec0f27 | 1465 | { |
4c738a8d BP |
1466 | struct dp_netdev_port_state *state = state_; |
1467 | free(state->name); | |
b0ec0f27 BP |
1468 | free(state); |
1469 | return 0; | |
72865317 BP |
1470 | } |
1471 | ||
1472 | static int | |
67a4917b | 1473 | dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) |
72865317 BP |
1474 | { |
1475 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
d33ed218 | 1476 | uint64_t new_port_seq; |
5279f8fd BP |
1477 | int error; |
1478 | ||
d33ed218 BP |
1479 | new_port_seq = seq_read(dpif->dp->port_seq); |
1480 | if (dpif->last_port_seq != new_port_seq) { | |
1481 | dpif->last_port_seq = new_port_seq; | |
5279f8fd | 1482 | error = ENOBUFS; |
72865317 | 1483 | } else { |
5279f8fd | 1484 | error = EAGAIN; |
72865317 | 1485 | } |
5279f8fd BP |
1486 | |
1487 | return error; | |
72865317 BP |
1488 | } |
1489 | ||
1490 | static void | |
1491 | dpif_netdev_port_poll_wait(const struct dpif *dpif_) | |
1492 | { | |
1493 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
5279f8fd | 1494 | |
d33ed218 | 1495 | seq_wait(dpif->dp->port_seq, dpif->last_port_seq); |
8a4e3a85 BP |
1496 | } |
1497 | ||
1498 | static struct dp_netdev_flow * | |
0de8783a | 1499 | dp_netdev_flow_cast(const struct dpcls_rule *cr) |
8a4e3a85 BP |
1500 | { |
1501 | return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL; | |
72865317 BP |
1502 | } |
1503 | ||
9bbf1c3d DDP |
1504 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *flow) |
1505 | { | |
1506 | return ovs_refcount_try_ref_rcu(&flow->ref_cnt); | |
1507 | } | |
1508 | ||
79df317f DDP |
1509 | /* netdev_flow_key utilities. |
1510 | * | |
1511 | * netdev_flow_key is basically a miniflow. We use these functions | |
1512 | * (netdev_flow_key_clone, netdev_flow_key_equal, ...) instead of the miniflow | |
1513 | * functions (miniflow_clone_inline, miniflow_equal, ...), because: | |
1514 | * | |
1515 | * - Since we are dealing exclusively with miniflows created by | |
1516 | * miniflow_extract(), if the map is different the miniflow is different. | |
1517 | * Therefore we can be faster by comparing the map and the miniflow in a | |
1518 | * single memcmp(). | |
1519 | * _ netdev_flow_key's miniflow has always inline values. | |
1520 | * - These functions can be inlined by the compiler. | |
1521 | * | |
1522 | * The following assertions make sure that what we're doing with miniflow is | |
1523 | * safe | |
1524 | */ | |
1525 | BUILD_ASSERT_DECL(offsetof(struct miniflow, inline_values) | |
1526 | == sizeof(uint64_t)); | |
79df317f DDP |
1527 | |
1528 | /* Given the number of bits set in the miniflow map, returns the size of the | |
caeb4906 | 1529 | * 'netdev_flow_key.mf' */ |
79df317f DDP |
1530 | static inline uint32_t |
1531 | netdev_flow_key_size(uint32_t flow_u32s) | |
1532 | { | |
caeb4906 | 1533 | return offsetof(struct miniflow, inline_values) + |
0de8783a | 1534 | MINIFLOW_VALUES_SIZE(flow_u32s); |
79df317f DDP |
1535 | } |
1536 | ||
79df317f DDP |
1537 | static inline bool |
1538 | netdev_flow_key_equal(const struct netdev_flow_key *a, | |
0de8783a JR |
1539 | const struct netdev_flow_key *b) |
1540 | { | |
caeb4906 JR |
1541 | /* 'b->len' may be not set yet. */ |
1542 | return a->hash == b->hash && !memcmp(&a->mf, &b->mf, a->len); | |
0de8783a JR |
1543 | } |
1544 | ||
1545 | /* Used to compare 'netdev_flow_key' in the exact match cache to a miniflow. | |
1546 | * The maps are compared bitwise, so both 'key->mf' 'mf' must have been | |
1547 | * generated by miniflow_extract. */ | |
1548 | static inline bool | |
1549 | netdev_flow_key_equal_mf(const struct netdev_flow_key *key, | |
1550 | const struct miniflow *mf) | |
79df317f | 1551 | { |
caeb4906 | 1552 | return !memcmp(&key->mf, mf, key->len); |
79df317f DDP |
1553 | } |
1554 | ||
1555 | static inline void | |
1556 | netdev_flow_key_clone(struct netdev_flow_key *dst, | |
0de8783a JR |
1557 | const struct netdev_flow_key *src) |
1558 | { | |
caeb4906 JR |
1559 | memcpy(dst, src, |
1560 | offsetof(struct netdev_flow_key, mf) + src->len); | |
0de8783a JR |
1561 | } |
1562 | ||
1563 | /* Slow. */ | |
1564 | static void | |
1565 | netdev_flow_key_from_flow(struct netdev_flow_key *dst, | |
1566 | const struct flow *src) | |
1567 | { | |
cf62fa4c | 1568 | struct dp_packet packet; |
0de8783a | 1569 | uint64_t buf_stub[512 / 8]; |
0de8783a JR |
1570 | |
1571 | miniflow_initialize(&dst->mf, dst->buf); | |
1572 | ||
cf62fa4c PS |
1573 | dp_packet_use_stub(&packet, buf_stub, sizeof buf_stub); |
1574 | pkt_metadata_from_flow(&packet.md, src); | |
0de8783a | 1575 | flow_compose(&packet, src); |
cf62fa4c PS |
1576 | miniflow_extract(&packet, &dst->mf); |
1577 | dp_packet_uninit(&packet); | |
0de8783a JR |
1578 | |
1579 | dst->len = netdev_flow_key_size(count_1bits(dst->mf.map)); | |
1580 | dst->hash = 0; /* Not computed yet. */ | |
1581 | } | |
1582 | ||
1583 | /* Initialize a netdev_flow_key 'mask' from 'match'. */ | |
1584 | static inline void | |
1585 | netdev_flow_mask_init(struct netdev_flow_key *mask, | |
1586 | const struct match *match) | |
1587 | { | |
d70e8c28 JR |
1588 | const uint64_t *mask_u64 = (const uint64_t *) &match->wc.masks; |
1589 | uint64_t *dst = mask->mf.inline_values; | |
0de8783a JR |
1590 | uint64_t map, mask_map = 0; |
1591 | uint32_t hash = 0; | |
1592 | int n; | |
1593 | ||
1594 | /* Only check masks that make sense for the flow. */ | |
1595 | map = flow_wc_map(&match->flow); | |
1596 | ||
1597 | while (map) { | |
1598 | uint64_t rm1bit = rightmost_1bit(map); | |
1599 | int i = raw_ctz(map); | |
1600 | ||
d70e8c28 | 1601 | if (mask_u64[i]) { |
0de8783a | 1602 | mask_map |= rm1bit; |
d70e8c28 JR |
1603 | *dst++ = mask_u64[i]; |
1604 | hash = hash_add64(hash, mask_u64[i]); | |
0de8783a JR |
1605 | } |
1606 | map -= rm1bit; | |
1607 | } | |
1608 | ||
1609 | mask->mf.values_inline = true; | |
1610 | mask->mf.map = mask_map; | |
1611 | ||
aae7c34f | 1612 | hash = hash_add64(hash, mask_map); |
0de8783a JR |
1613 | |
1614 | n = dst - mask->mf.inline_values; | |
1615 | ||
d70e8c28 | 1616 | mask->hash = hash_finish(hash, n * 8); |
0de8783a JR |
1617 | mask->len = netdev_flow_key_size(n); |
1618 | } | |
1619 | ||
1620 | /* Initializes 'dst' as a copy of 'src' masked with 'mask'. */ | |
1621 | static inline void | |
1622 | netdev_flow_key_init_masked(struct netdev_flow_key *dst, | |
1623 | const struct flow *flow, | |
1624 | const struct netdev_flow_key *mask) | |
79df317f | 1625 | { |
d70e8c28 JR |
1626 | uint64_t *dst_u64 = dst->mf.inline_values; |
1627 | const uint64_t *mask_u64 = mask->mf.inline_values; | |
0de8783a | 1628 | uint32_t hash = 0; |
d70e8c28 | 1629 | uint64_t value; |
0de8783a JR |
1630 | |
1631 | dst->len = mask->len; | |
1632 | dst->mf.values_inline = true; | |
1633 | dst->mf.map = mask->mf.map; | |
1634 | ||
1635 | FLOW_FOR_EACH_IN_MAP(value, flow, mask->mf.map) { | |
d70e8c28 JR |
1636 | *dst_u64 = value & *mask_u64++; |
1637 | hash = hash_add64(hash, *dst_u64++); | |
0de8783a | 1638 | } |
d70e8c28 | 1639 | dst->hash = hash_finish(hash, (dst_u64 - dst->mf.inline_values) * 8); |
0de8783a JR |
1640 | } |
1641 | ||
d70e8c28 | 1642 | /* Iterate through all netdev_flow_key u64 values specified by 'MAP' */ |
0de8783a JR |
1643 | #define NETDEV_FLOW_KEY_FOR_EACH_IN_MAP(VALUE, KEY, MAP) \ |
1644 | for (struct mf_for_each_in_map_aux aux__ \ | |
1645 | = { (KEY)->mf.inline_values, (KEY)->mf.map, MAP }; \ | |
1646 | mf_get_next_in_map(&aux__, &(VALUE)); \ | |
1647 | ) | |
1648 | ||
1649 | /* Returns a hash value for the bits of 'key' where there are 1-bits in | |
1650 | * 'mask'. */ | |
1651 | static inline uint32_t | |
1652 | netdev_flow_key_hash_in_mask(const struct netdev_flow_key *key, | |
1653 | const struct netdev_flow_key *mask) | |
1654 | { | |
d70e8c28 | 1655 | const uint64_t *p = mask->mf.inline_values; |
0de8783a | 1656 | uint32_t hash = 0; |
d70e8c28 | 1657 | uint64_t key_u64; |
0de8783a | 1658 | |
d70e8c28 JR |
1659 | NETDEV_FLOW_KEY_FOR_EACH_IN_MAP(key_u64, key, mask->mf.map) { |
1660 | hash = hash_add64(hash, key_u64 & *p++); | |
0de8783a JR |
1661 | } |
1662 | ||
d70e8c28 | 1663 | return hash_finish(hash, (p - mask->mf.inline_values) * 8); |
79df317f DDP |
1664 | } |
1665 | ||
9bbf1c3d DDP |
1666 | static inline bool |
1667 | emc_entry_alive(struct emc_entry *ce) | |
1668 | { | |
1669 | return ce->flow && !ce->flow->dead; | |
1670 | } | |
1671 | ||
1672 | static void | |
1673 | emc_clear_entry(struct emc_entry *ce) | |
1674 | { | |
1675 | if (ce->flow) { | |
1676 | dp_netdev_flow_unref(ce->flow); | |
1677 | ce->flow = NULL; | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | static inline void | |
1682 | emc_change_entry(struct emc_entry *ce, struct dp_netdev_flow *flow, | |
0de8783a | 1683 | const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1684 | { |
1685 | if (ce->flow != flow) { | |
1686 | if (ce->flow) { | |
1687 | dp_netdev_flow_unref(ce->flow); | |
1688 | } | |
1689 | ||
1690 | if (dp_netdev_flow_ref(flow)) { | |
1691 | ce->flow = flow; | |
1692 | } else { | |
1693 | ce->flow = NULL; | |
1694 | } | |
1695 | } | |
0de8783a JR |
1696 | if (key) { |
1697 | netdev_flow_key_clone(&ce->key, key); | |
9bbf1c3d DDP |
1698 | } |
1699 | } | |
1700 | ||
1701 | static inline void | |
0de8783a | 1702 | emc_insert(struct emc_cache *cache, const struct netdev_flow_key *key, |
9bbf1c3d DDP |
1703 | struct dp_netdev_flow *flow) |
1704 | { | |
1705 | struct emc_entry *to_be_replaced = NULL; | |
1706 | struct emc_entry *current_entry; | |
1707 | ||
0de8783a JR |
1708 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1709 | if (netdev_flow_key_equal(¤t_entry->key, key)) { | |
9bbf1c3d | 1710 | /* We found the entry with the 'mf' miniflow */ |
0de8783a | 1711 | emc_change_entry(current_entry, flow, NULL); |
9bbf1c3d DDP |
1712 | return; |
1713 | } | |
1714 | ||
1715 | /* Replacement policy: put the flow in an empty (not alive) entry, or | |
1716 | * in the first entry where it can be */ | |
1717 | if (!to_be_replaced | |
1718 | || (emc_entry_alive(to_be_replaced) | |
1719 | && !emc_entry_alive(current_entry)) | |
0de8783a | 1720 | || current_entry->key.hash < to_be_replaced->key.hash) { |
9bbf1c3d DDP |
1721 | to_be_replaced = current_entry; |
1722 | } | |
1723 | } | |
1724 | /* We didn't find the miniflow in the cache. | |
1725 | * The 'to_be_replaced' entry is where the new flow will be stored */ | |
1726 | ||
0de8783a | 1727 | emc_change_entry(to_be_replaced, flow, key); |
9bbf1c3d DDP |
1728 | } |
1729 | ||
1730 | static inline struct dp_netdev_flow * | |
0de8783a | 1731 | emc_lookup(struct emc_cache *cache, const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1732 | { |
1733 | struct emc_entry *current_entry; | |
1734 | ||
0de8783a JR |
1735 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1736 | if (current_entry->key.hash == key->hash | |
1737 | && emc_entry_alive(current_entry) | |
1738 | && netdev_flow_key_equal_mf(¤t_entry->key, &key->mf)) { | |
9bbf1c3d | 1739 | |
0de8783a | 1740 | /* We found the entry with the 'key->mf' miniflow */ |
9bbf1c3d DDP |
1741 | return current_entry->flow; |
1742 | } | |
1743 | } | |
1744 | ||
1745 | return NULL; | |
1746 | } | |
1747 | ||
72865317 | 1748 | static struct dp_netdev_flow * |
1c1e46ed AW |
1749 | dp_netdev_pmd_lookup_flow(const struct dp_netdev_pmd_thread *pmd, |
1750 | const struct netdev_flow_key *key) | |
2c0ea78f | 1751 | { |
8a4e3a85 | 1752 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 1753 | struct dpcls_rule *rule; |
2c0ea78f | 1754 | |
1c1e46ed | 1755 | dpcls_lookup(&pmd->cls, key, &rule, 1); |
4f150744 | 1756 | netdev_flow = dp_netdev_flow_cast(rule); |
2c0ea78f | 1757 | |
8a4e3a85 | 1758 | return netdev_flow; |
2c0ea78f GS |
1759 | } |
1760 | ||
1761 | static struct dp_netdev_flow * | |
1c1e46ed AW |
1762 | dp_netdev_pmd_find_flow(const struct dp_netdev_pmd_thread *pmd, |
1763 | const ovs_u128 *ufidp, const struct nlattr *key, | |
1764 | size_t key_len) | |
72865317 | 1765 | { |
1763b4b8 | 1766 | struct dp_netdev_flow *netdev_flow; |
70e5ed6f JS |
1767 | struct flow flow; |
1768 | ovs_u128 ufid; | |
1769 | ||
1770 | /* If a UFID is not provided, determine one based on the key. */ | |
1771 | if (!ufidp && key && key_len | |
1772 | && !dpif_netdev_flow_from_nlattrs(key, key_len, &flow)) { | |
1c1e46ed | 1773 | dpif_flow_hash(pmd->dp->dpif, &flow, sizeof flow, &ufid); |
70e5ed6f JS |
1774 | ufidp = &ufid; |
1775 | } | |
72865317 | 1776 | |
70e5ed6f JS |
1777 | if (ufidp) { |
1778 | CMAP_FOR_EACH_WITH_HASH (netdev_flow, node, dp_netdev_flow_hash(ufidp), | |
1c1e46ed | 1779 | &pmd->flow_table) { |
bdd7ecf5 | 1780 | if (ovs_u128_equals(&netdev_flow->ufid, ufidp)) { |
70e5ed6f JS |
1781 | return netdev_flow; |
1782 | } | |
72865317 BP |
1783 | } |
1784 | } | |
8a4e3a85 | 1785 | |
72865317 BP |
1786 | return NULL; |
1787 | } | |
1788 | ||
1789 | static void | |
eb94da30 | 1790 | get_dpif_flow_stats(const struct dp_netdev_flow *netdev_flow_, |
1763b4b8 | 1791 | struct dpif_flow_stats *stats) |
feebdea2 | 1792 | { |
eb94da30 DDP |
1793 | struct dp_netdev_flow *netdev_flow; |
1794 | unsigned long long n; | |
1795 | long long used; | |
1796 | uint16_t flags; | |
1797 | ||
1798 | netdev_flow = CONST_CAST(struct dp_netdev_flow *, netdev_flow_); | |
1799 | ||
1800 | atomic_read_relaxed(&netdev_flow->stats.packet_count, &n); | |
1801 | stats->n_packets = n; | |
1802 | atomic_read_relaxed(&netdev_flow->stats.byte_count, &n); | |
1803 | stats->n_bytes = n; | |
1804 | atomic_read_relaxed(&netdev_flow->stats.used, &used); | |
1805 | stats->used = used; | |
1806 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
1807 | stats->tcp_flags = flags; | |
72865317 BP |
1808 | } |
1809 | ||
7af12bd7 JS |
1810 | /* Converts to the dpif_flow format, using 'key_buf' and 'mask_buf' for |
1811 | * storing the netlink-formatted key/mask. 'key_buf' may be the same as | |
1812 | * 'mask_buf'. Actions will be returned without copying, by relying on RCU to | |
1813 | * protect them. */ | |
6fe09f8c | 1814 | static void |
70e5ed6f | 1815 | dp_netdev_flow_to_dpif_flow(const struct dp_netdev_flow *netdev_flow, |
7af12bd7 | 1816 | struct ofpbuf *key_buf, struct ofpbuf *mask_buf, |
64bb477f | 1817 | struct dpif_flow *flow, bool terse) |
6fe09f8c | 1818 | { |
64bb477f JS |
1819 | if (terse) { |
1820 | memset(flow, 0, sizeof *flow); | |
1821 | } else { | |
1822 | struct flow_wildcards wc; | |
1823 | struct dp_netdev_actions *actions; | |
1824 | size_t offset; | |
5262eea1 JG |
1825 | struct odp_flow_key_parms odp_parms = { |
1826 | .flow = &netdev_flow->flow, | |
1827 | .mask = &wc.masks, | |
1828 | .recirc = true, | |
1829 | .max_mpls_depth = SIZE_MAX, | |
1830 | }; | |
64bb477f JS |
1831 | |
1832 | miniflow_expand(&netdev_flow->cr.mask->mf, &wc.masks); | |
1833 | ||
1834 | /* Key */ | |
6fd6ed71 | 1835 | offset = key_buf->size; |
64bb477f | 1836 | flow->key = ofpbuf_tail(key_buf); |
5262eea1 JG |
1837 | odp_parms.odp_in_port = netdev_flow->flow.in_port.odp_port; |
1838 | odp_flow_key_from_flow(&odp_parms, key_buf); | |
6fd6ed71 | 1839 | flow->key_len = key_buf->size - offset; |
64bb477f JS |
1840 | |
1841 | /* Mask */ | |
6fd6ed71 | 1842 | offset = mask_buf->size; |
64bb477f | 1843 | flow->mask = ofpbuf_tail(mask_buf); |
5262eea1 JG |
1844 | odp_parms.odp_in_port = wc.masks.in_port.odp_port; |
1845 | odp_flow_key_from_mask(&odp_parms, mask_buf); | |
6fd6ed71 | 1846 | flow->mask_len = mask_buf->size - offset; |
64bb477f JS |
1847 | |
1848 | /* Actions */ | |
1849 | actions = dp_netdev_flow_get_actions(netdev_flow); | |
1850 | flow->actions = actions->actions; | |
1851 | flow->actions_len = actions->size; | |
1852 | } | |
6fe09f8c | 1853 | |
70e5ed6f JS |
1854 | flow->ufid = netdev_flow->ufid; |
1855 | flow->ufid_present = true; | |
1c1e46ed | 1856 | flow->pmd_id = netdev_flow->pmd_id; |
6fe09f8c JS |
1857 | get_dpif_flow_stats(netdev_flow, &flow->stats); |
1858 | } | |
1859 | ||
36956a7d | 1860 | static int |
8c301900 JR |
1861 | dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len, |
1862 | const struct nlattr *mask_key, | |
1863 | uint32_t mask_key_len, const struct flow *flow, | |
1864 | struct flow *mask) | |
1865 | { | |
1866 | if (mask_key_len) { | |
80e44883 BP |
1867 | enum odp_key_fitness fitness; |
1868 | ||
1869 | fitness = odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow); | |
1870 | if (fitness) { | |
8c301900 JR |
1871 | /* This should not happen: it indicates that |
1872 | * odp_flow_key_from_mask() and odp_flow_key_to_mask() | |
1873 | * disagree on the acceptable form of a mask. Log the problem | |
1874 | * as an error, with enough details to enable debugging. */ | |
1875 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1876 | ||
1877 | if (!VLOG_DROP_ERR(&rl)) { | |
1878 | struct ds s; | |
1879 | ||
1880 | ds_init(&s); | |
1881 | odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s, | |
1882 | true); | |
80e44883 BP |
1883 | VLOG_ERR("internal error parsing flow mask %s (%s)", |
1884 | ds_cstr(&s), odp_key_fitness_to_string(fitness)); | |
8c301900 JR |
1885 | ds_destroy(&s); |
1886 | } | |
1887 | ||
1888 | return EINVAL; | |
1889 | } | |
8c301900 JR |
1890 | } else { |
1891 | enum mf_field_id id; | |
1892 | /* No mask key, unwildcard everything except fields whose | |
1893 | * prerequisities are not met. */ | |
1894 | memset(mask, 0x0, sizeof *mask); | |
1895 | ||
1896 | for (id = 0; id < MFF_N_IDS; ++id) { | |
1897 | /* Skip registers and metadata. */ | |
1898 | if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS) | |
1899 | && id != MFF_METADATA) { | |
1900 | const struct mf_field *mf = mf_from_id(id); | |
1901 | if (mf_are_prereqs_ok(mf, flow)) { | |
1902 | mf_mask_field(mf, mask); | |
1903 | } | |
1904 | } | |
1905 | } | |
1906 | } | |
1907 | ||
f3f750e5 BP |
1908 | /* Force unwildcard the in_port. |
1909 | * | |
1910 | * We need to do this even in the case where we unwildcard "everything" | |
1911 | * above because "everything" only includes the 16-bit OpenFlow port number | |
1912 | * mask->in_port.ofp_port, which only covers half of the 32-bit datapath | |
1913 | * port number mask->in_port.odp_port. */ | |
1914 | mask->in_port.odp_port = u32_to_odp(UINT32_MAX); | |
1915 | ||
8c301900 JR |
1916 | return 0; |
1917 | } | |
1918 | ||
1919 | static int | |
1920 | dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, | |
1921 | struct flow *flow) | |
36956a7d | 1922 | { |
586ddea5 BP |
1923 | odp_port_t in_port; |
1924 | ||
8c301900 | 1925 | if (odp_flow_key_to_flow(key, key_len, flow)) { |
36956a7d | 1926 | /* This should not happen: it indicates that odp_flow_key_from_flow() |
8c301900 JR |
1927 | * and odp_flow_key_to_flow() disagree on the acceptable form of a |
1928 | * flow. Log the problem as an error, with enough details to enable | |
1929 | * debugging. */ | |
36956a7d BP |
1930 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
1931 | ||
1932 | if (!VLOG_DROP_ERR(&rl)) { | |
1933 | struct ds s; | |
1934 | ||
1935 | ds_init(&s); | |
8c301900 | 1936 | odp_flow_format(key, key_len, NULL, 0, NULL, &s, true); |
36956a7d BP |
1937 | VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s)); |
1938 | ds_destroy(&s); | |
1939 | } | |
1940 | ||
1941 | return EINVAL; | |
1942 | } | |
1943 | ||
586ddea5 BP |
1944 | in_port = flow->in_port.odp_port; |
1945 | if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) { | |
18886b60 BP |
1946 | return EINVAL; |
1947 | } | |
1948 | ||
36956a7d BP |
1949 | return 0; |
1950 | } | |
1951 | ||
72865317 | 1952 | static int |
6fe09f8c | 1953 | dpif_netdev_flow_get(const struct dpif *dpif, const struct dpif_flow_get *get) |
72865317 BP |
1954 | { |
1955 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 1956 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 1957 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
1958 | unsigned pmd_id = get->pmd_id == PMD_ID_NULL |
1959 | ? NON_PMD_CORE_ID : get->pmd_id; | |
70e5ed6f | 1960 | int error = 0; |
8a4e3a85 | 1961 | |
1c1e46ed AW |
1962 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
1963 | if (!pmd) { | |
1964 | return EINVAL; | |
1965 | } | |
1966 | ||
1967 | netdev_flow = dp_netdev_pmd_find_flow(pmd, get->ufid, get->key, | |
1968 | get->key_len); | |
1763b4b8 | 1969 | if (netdev_flow) { |
70e5ed6f | 1970 | dp_netdev_flow_to_dpif_flow(netdev_flow, get->buffer, get->buffer, |
64bb477f | 1971 | get->flow, false); |
70e5ed6f | 1972 | } else { |
5279f8fd | 1973 | error = ENOENT; |
72865317 | 1974 | } |
1c1e46ed AW |
1975 | dp_netdev_pmd_unref(pmd); |
1976 | ||
bc4a05c6 | 1977 | |
5279f8fd | 1978 | return error; |
72865317 BP |
1979 | } |
1980 | ||
0de8783a | 1981 | static struct dp_netdev_flow * |
1c1e46ed AW |
1982 | dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd, |
1983 | struct match *match, const ovs_u128 *ufid, | |
ae2ceebd | 1984 | const struct nlattr *actions, size_t actions_len) |
1c1e46ed | 1985 | OVS_REQUIRES(pmd->flow_mutex) |
72865317 | 1986 | { |
0de8783a JR |
1987 | struct dp_netdev_flow *flow; |
1988 | struct netdev_flow_key mask; | |
ed79f89a | 1989 | |
0de8783a JR |
1990 | netdev_flow_mask_init(&mask, match); |
1991 | /* Make sure wc does not have metadata. */ | |
1992 | ovs_assert(!(mask.mf.map & (MINIFLOW_MAP(metadata) | MINIFLOW_MAP(regs)))); | |
679ba04c | 1993 | |
0de8783a | 1994 | /* Do not allocate extra space. */ |
caeb4906 | 1995 | flow = xmalloc(sizeof *flow - sizeof flow->cr.flow.mf + mask.len); |
1c1e46ed | 1996 | memset(&flow->stats, 0, sizeof flow->stats); |
0de8783a | 1997 | flow->dead = false; |
11e5cf1f | 1998 | flow->batch = NULL; |
bd5131ba | 1999 | *CONST_CAST(unsigned *, &flow->pmd_id) = pmd->core_id; |
0de8783a | 2000 | *CONST_CAST(struct flow *, &flow->flow) = match->flow; |
70e5ed6f | 2001 | *CONST_CAST(ovs_u128 *, &flow->ufid) = *ufid; |
0de8783a | 2002 | ovs_refcount_init(&flow->ref_cnt); |
0de8783a | 2003 | ovsrcu_set(&flow->actions, dp_netdev_actions_create(actions, actions_len)); |
2c0ea78f | 2004 | |
0de8783a | 2005 | netdev_flow_key_init_masked(&flow->cr.flow, &match->flow, &mask); |
1c1e46ed | 2006 | dpcls_insert(&pmd->cls, &flow->cr, &mask); |
72865317 | 2007 | |
4c75aaab EJ |
2008 | cmap_insert(&pmd->flow_table, CONST_CAST(struct cmap_node *, &flow->node), |
2009 | dp_netdev_flow_hash(&flow->ufid)); | |
2010 | ||
623540e4 | 2011 | if (OVS_UNLIKELY(VLOG_IS_DBG_ENABLED())) { |
0de8783a | 2012 | struct match match; |
623540e4 EJ |
2013 | struct ds ds = DS_EMPTY_INITIALIZER; |
2014 | ||
0de8783a JR |
2015 | match.flow = flow->flow; |
2016 | miniflow_expand(&flow->cr.mask->mf, &match.wc.masks); | |
2017 | ||
623540e4 | 2018 | ds_put_cstr(&ds, "flow_add: "); |
70e5ed6f JS |
2019 | odp_format_ufid(ufid, &ds); |
2020 | ds_put_cstr(&ds, " "); | |
0de8783a | 2021 | match_format(&match, &ds, OFP_DEFAULT_PRIORITY); |
623540e4 EJ |
2022 | ds_put_cstr(&ds, ", actions:"); |
2023 | format_odp_actions(&ds, actions, actions_len); | |
2024 | ||
2025 | VLOG_DBG_RL(&upcall_rl, "%s", ds_cstr(&ds)); | |
2026 | ||
2027 | ds_destroy(&ds); | |
2028 | } | |
2029 | ||
0de8783a | 2030 | return flow; |
72865317 BP |
2031 | } |
2032 | ||
72865317 | 2033 | static int |
89625d1e | 2034 | dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put) |
72865317 BP |
2035 | { |
2036 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2037 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 2038 | struct netdev_flow_key key; |
1c1e46ed | 2039 | struct dp_netdev_pmd_thread *pmd; |
ae2ceebd | 2040 | struct match match; |
70e5ed6f | 2041 | ovs_u128 ufid; |
bd5131ba DDP |
2042 | unsigned pmd_id = put->pmd_id == PMD_ID_NULL |
2043 | ? NON_PMD_CORE_ID : put->pmd_id; | |
36956a7d BP |
2044 | int error; |
2045 | ||
ae2ceebd | 2046 | error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &match.flow); |
8c301900 JR |
2047 | if (error) { |
2048 | return error; | |
2049 | } | |
2050 | error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len, | |
2051 | put->mask, put->mask_len, | |
ae2ceebd | 2052 | &match.flow, &match.wc.masks); |
36956a7d BP |
2053 | if (error) { |
2054 | return error; | |
2055 | } | |
0de8783a | 2056 | |
1c1e46ed AW |
2057 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2058 | if (!pmd) { | |
2059 | return EINVAL; | |
2060 | } | |
2061 | ||
0de8783a JR |
2062 | /* Must produce a netdev_flow_key for lookup. |
2063 | * This interface is no longer performance critical, since it is not used | |
2064 | * for upcall processing any more. */ | |
2065 | netdev_flow_key_from_flow(&key, &match.flow); | |
72865317 | 2066 | |
70e5ed6f JS |
2067 | if (put->ufid) { |
2068 | ufid = *put->ufid; | |
2069 | } else { | |
2070 | dpif_flow_hash(dpif, &match.flow, sizeof match.flow, &ufid); | |
2071 | } | |
2072 | ||
1c1e46ed AW |
2073 | ovs_mutex_lock(&pmd->flow_mutex); |
2074 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &key); | |
1763b4b8 | 2075 | if (!netdev_flow) { |
89625d1e | 2076 | if (put->flags & DPIF_FP_CREATE) { |
1c1e46ed | 2077 | if (cmap_count(&pmd->flow_table) < MAX_FLOWS) { |
89625d1e BP |
2078 | if (put->stats) { |
2079 | memset(put->stats, 0, sizeof *put->stats); | |
feebdea2 | 2080 | } |
1c1e46ed | 2081 | dp_netdev_flow_add(pmd, &match, &ufid, put->actions, |
70e5ed6f | 2082 | put->actions_len); |
0de8783a | 2083 | error = 0; |
72865317 | 2084 | } else { |
5279f8fd | 2085 | error = EFBIG; |
72865317 BP |
2086 | } |
2087 | } else { | |
5279f8fd | 2088 | error = ENOENT; |
72865317 BP |
2089 | } |
2090 | } else { | |
2c0ea78f | 2091 | if (put->flags & DPIF_FP_MODIFY |
ae2ceebd | 2092 | && flow_equal(&match.flow, &netdev_flow->flow)) { |
8a4e3a85 BP |
2093 | struct dp_netdev_actions *new_actions; |
2094 | struct dp_netdev_actions *old_actions; | |
2095 | ||
2096 | new_actions = dp_netdev_actions_create(put->actions, | |
2097 | put->actions_len); | |
2098 | ||
61e7deb1 BP |
2099 | old_actions = dp_netdev_flow_get_actions(netdev_flow); |
2100 | ovsrcu_set(&netdev_flow->actions, new_actions); | |
679ba04c | 2101 | |
a84cb64a BP |
2102 | if (put->stats) { |
2103 | get_dpif_flow_stats(netdev_flow, put->stats); | |
2104 | } | |
2105 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
97447f55 DDP |
2106 | /* XXX: The userspace datapath uses thread local statistics |
2107 | * (for flows), which should be updated only by the owning | |
2108 | * thread. Since we cannot write on stats memory here, | |
2109 | * we choose not to support this flag. Please note: | |
2110 | * - This feature is currently used only by dpctl commands with | |
2111 | * option --clear. | |
2112 | * - Should the need arise, this operation can be implemented | |
2113 | * by keeping a base value (to be update here) for each | |
2114 | * counter, and subtracting it before outputting the stats */ | |
2115 | error = EOPNOTSUPP; | |
72865317 | 2116 | } |
8a4e3a85 | 2117 | |
61e7deb1 | 2118 | ovsrcu_postpone(dp_netdev_actions_free, old_actions); |
2c0ea78f | 2119 | } else if (put->flags & DPIF_FP_CREATE) { |
5279f8fd | 2120 | error = EEXIST; |
2c0ea78f GS |
2121 | } else { |
2122 | /* Overlapping flow. */ | |
2123 | error = EINVAL; | |
72865317 BP |
2124 | } |
2125 | } | |
1c1e46ed AW |
2126 | ovs_mutex_unlock(&pmd->flow_mutex); |
2127 | dp_netdev_pmd_unref(pmd); | |
5279f8fd BP |
2128 | |
2129 | return error; | |
72865317 BP |
2130 | } |
2131 | ||
72865317 | 2132 | static int |
b99d3cee | 2133 | dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del) |
72865317 BP |
2134 | { |
2135 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2136 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 2137 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
2138 | unsigned pmd_id = del->pmd_id == PMD_ID_NULL |
2139 | ? NON_PMD_CORE_ID : del->pmd_id; | |
70e5ed6f | 2140 | int error = 0; |
72865317 | 2141 | |
1c1e46ed AW |
2142 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2143 | if (!pmd) { | |
2144 | return EINVAL; | |
2145 | } | |
2146 | ||
2147 | ovs_mutex_lock(&pmd->flow_mutex); | |
2148 | netdev_flow = dp_netdev_pmd_find_flow(pmd, del->ufid, del->key, | |
2149 | del->key_len); | |
1763b4b8 | 2150 | if (netdev_flow) { |
b99d3cee | 2151 | if (del->stats) { |
1763b4b8 | 2152 | get_dpif_flow_stats(netdev_flow, del->stats); |
feebdea2 | 2153 | } |
1c1e46ed | 2154 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); |
72865317 | 2155 | } else { |
5279f8fd | 2156 | error = ENOENT; |
72865317 | 2157 | } |
1c1e46ed AW |
2158 | ovs_mutex_unlock(&pmd->flow_mutex); |
2159 | dp_netdev_pmd_unref(pmd); | |
5279f8fd BP |
2160 | |
2161 | return error; | |
72865317 BP |
2162 | } |
2163 | ||
ac64794a BP |
2164 | struct dpif_netdev_flow_dump { |
2165 | struct dpif_flow_dump up; | |
1c1e46ed AW |
2166 | struct cmap_position poll_thread_pos; |
2167 | struct cmap_position flow_pos; | |
2168 | struct dp_netdev_pmd_thread *cur_pmd; | |
d2ad7ef1 JS |
2169 | int status; |
2170 | struct ovs_mutex mutex; | |
e723fd32 JS |
2171 | }; |
2172 | ||
ac64794a BP |
2173 | static struct dpif_netdev_flow_dump * |
2174 | dpif_netdev_flow_dump_cast(struct dpif_flow_dump *dump) | |
72865317 | 2175 | { |
ac64794a | 2176 | return CONTAINER_OF(dump, struct dpif_netdev_flow_dump, up); |
e723fd32 JS |
2177 | } |
2178 | ||
ac64794a | 2179 | static struct dpif_flow_dump * |
64bb477f | 2180 | dpif_netdev_flow_dump_create(const struct dpif *dpif_, bool terse) |
e723fd32 | 2181 | { |
ac64794a | 2182 | struct dpif_netdev_flow_dump *dump; |
e723fd32 | 2183 | |
1c1e46ed | 2184 | dump = xzalloc(sizeof *dump); |
ac64794a | 2185 | dpif_flow_dump_init(&dump->up, dpif_); |
64bb477f | 2186 | dump->up.terse = terse; |
ac64794a BP |
2187 | ovs_mutex_init(&dump->mutex); |
2188 | ||
2189 | return &dump->up; | |
e723fd32 JS |
2190 | } |
2191 | ||
2192 | static int | |
ac64794a | 2193 | dpif_netdev_flow_dump_destroy(struct dpif_flow_dump *dump_) |
e723fd32 | 2194 | { |
ac64794a | 2195 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); |
e723fd32 | 2196 | |
ac64794a BP |
2197 | ovs_mutex_destroy(&dump->mutex); |
2198 | free(dump); | |
704a1e09 BP |
2199 | return 0; |
2200 | } | |
2201 | ||
ac64794a BP |
2202 | struct dpif_netdev_flow_dump_thread { |
2203 | struct dpif_flow_dump_thread up; | |
2204 | struct dpif_netdev_flow_dump *dump; | |
8bb113da RW |
2205 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; |
2206 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
2207 | }; |
2208 | ||
2209 | static struct dpif_netdev_flow_dump_thread * | |
2210 | dpif_netdev_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
2211 | { | |
2212 | return CONTAINER_OF(thread, struct dpif_netdev_flow_dump_thread, up); | |
2213 | } | |
2214 | ||
2215 | static struct dpif_flow_dump_thread * | |
2216 | dpif_netdev_flow_dump_thread_create(struct dpif_flow_dump *dump_) | |
2217 | { | |
2218 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); | |
2219 | struct dpif_netdev_flow_dump_thread *thread; | |
2220 | ||
2221 | thread = xmalloc(sizeof *thread); | |
2222 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
2223 | thread->dump = dump; | |
2224 | return &thread->up; | |
2225 | } | |
2226 | ||
2227 | static void | |
2228 | dpif_netdev_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) | |
2229 | { | |
2230 | struct dpif_netdev_flow_dump_thread *thread | |
2231 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2232 | ||
2233 | free(thread); | |
2234 | } | |
2235 | ||
704a1e09 | 2236 | static int |
ac64794a | 2237 | dpif_netdev_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
8bb113da | 2238 | struct dpif_flow *flows, int max_flows) |
ac64794a BP |
2239 | { |
2240 | struct dpif_netdev_flow_dump_thread *thread | |
2241 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2242 | struct dpif_netdev_flow_dump *dump = thread->dump; | |
8bb113da | 2243 | struct dp_netdev_flow *netdev_flows[FLOW_DUMP_MAX_BATCH]; |
8bb113da RW |
2244 | int n_flows = 0; |
2245 | int i; | |
14608a15 | 2246 | |
ac64794a | 2247 | ovs_mutex_lock(&dump->mutex); |
8bb113da | 2248 | if (!dump->status) { |
1c1e46ed AW |
2249 | struct dpif_netdev *dpif = dpif_netdev_cast(thread->up.dpif); |
2250 | struct dp_netdev *dp = get_dp_netdev(&dpif->dpif); | |
2251 | struct dp_netdev_pmd_thread *pmd = dump->cur_pmd; | |
2252 | int flow_limit = MIN(max_flows, FLOW_DUMP_MAX_BATCH); | |
2253 | ||
2254 | /* First call to dump_next(), extracts the first pmd thread. | |
2255 | * If there is no pmd thread, returns immediately. */ | |
2256 | if (!pmd) { | |
2257 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2258 | if (!pmd) { | |
2259 | ovs_mutex_unlock(&dump->mutex); | |
2260 | return n_flows; | |
8bb113da | 2261 | |
8bb113da | 2262 | } |
d2ad7ef1 | 2263 | } |
1c1e46ed AW |
2264 | |
2265 | do { | |
2266 | for (n_flows = 0; n_flows < flow_limit; n_flows++) { | |
2267 | struct cmap_node *node; | |
2268 | ||
2269 | node = cmap_next_position(&pmd->flow_table, &dump->flow_pos); | |
2270 | if (!node) { | |
2271 | break; | |
2272 | } | |
2273 | netdev_flows[n_flows] = CONTAINER_OF(node, | |
2274 | struct dp_netdev_flow, | |
2275 | node); | |
2276 | } | |
2277 | /* When finishing dumping the current pmd thread, moves to | |
2278 | * the next. */ | |
2279 | if (n_flows < flow_limit) { | |
2280 | memset(&dump->flow_pos, 0, sizeof dump->flow_pos); | |
2281 | dp_netdev_pmd_unref(pmd); | |
2282 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2283 | if (!pmd) { | |
2284 | dump->status = EOF; | |
2285 | break; | |
2286 | } | |
2287 | } | |
2288 | /* Keeps the reference to next caller. */ | |
2289 | dump->cur_pmd = pmd; | |
2290 | ||
2291 | /* If the current dump is empty, do not exit the loop, since the | |
2292 | * remaining pmds could have flows to be dumped. Just dumps again | |
2293 | * on the new 'pmd'. */ | |
2294 | } while (!n_flows); | |
8a4e3a85 | 2295 | } |
ac64794a | 2296 | ovs_mutex_unlock(&dump->mutex); |
ac64794a | 2297 | |
8bb113da RW |
2298 | for (i = 0; i < n_flows; i++) { |
2299 | struct odputil_keybuf *maskbuf = &thread->maskbuf[i]; | |
2300 | struct odputil_keybuf *keybuf = &thread->keybuf[i]; | |
2301 | struct dp_netdev_flow *netdev_flow = netdev_flows[i]; | |
2302 | struct dpif_flow *f = &flows[i]; | |
7af12bd7 | 2303 | struct ofpbuf key, mask; |
8bb113da | 2304 | |
7af12bd7 JS |
2305 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); |
2306 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
64bb477f JS |
2307 | dp_netdev_flow_to_dpif_flow(netdev_flow, &key, &mask, f, |
2308 | dump->up.terse); | |
8bb113da | 2309 | } |
feebdea2 | 2310 | |
8bb113da | 2311 | return n_flows; |
72865317 BP |
2312 | } |
2313 | ||
2314 | static int | |
758c456d | 2315 | dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute) |
65f13b50 | 2316 | OVS_NO_THREAD_SAFETY_ANALYSIS |
72865317 BP |
2317 | { |
2318 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
65f13b50 | 2319 | struct dp_netdev_pmd_thread *pmd; |
cf62fa4c | 2320 | struct dp_packet *pp; |
72865317 | 2321 | |
cf62fa4c PS |
2322 | if (dp_packet_size(execute->packet) < ETH_HEADER_LEN || |
2323 | dp_packet_size(execute->packet) > UINT16_MAX) { | |
72865317 BP |
2324 | return EINVAL; |
2325 | } | |
2326 | ||
65f13b50 AW |
2327 | /* Tries finding the 'pmd'. If NULL is returned, that means |
2328 | * the current thread is a non-pmd thread and should use | |
b19befae | 2329 | * dp_netdev_get_pmd(dp, NON_PMD_CORE_ID). */ |
65f13b50 AW |
2330 | pmd = ovsthread_getspecific(dp->per_pmd_key); |
2331 | if (!pmd) { | |
b19befae | 2332 | pmd = dp_netdev_get_pmd(dp, NON_PMD_CORE_ID); |
65f13b50 AW |
2333 | } |
2334 | ||
2335 | /* If the current thread is non-pmd thread, acquires | |
2336 | * the 'non_pmd_mutex'. */ | |
2337 | if (pmd->core_id == NON_PMD_CORE_ID) { | |
2338 | ovs_mutex_lock(&dp->non_pmd_mutex); | |
433330a8 | 2339 | ovs_mutex_lock(&dp->port_mutex); |
65f13b50 | 2340 | } |
1c1e46ed | 2341 | |
cf62fa4c | 2342 | pp = execute->packet; |
41ccaa24 | 2343 | dp_netdev_execute_actions(pmd, &pp, 1, false, execute->actions, |
9bbf1c3d | 2344 | execute->actions_len); |
65f13b50 | 2345 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1c1e46ed | 2346 | dp_netdev_pmd_unref(pmd); |
433330a8 | 2347 | ovs_mutex_unlock(&dp->port_mutex); |
65f13b50 AW |
2348 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
2349 | } | |
8a4e3a85 | 2350 | |
758c456d | 2351 | return 0; |
72865317 BP |
2352 | } |
2353 | ||
1a0c894a BP |
2354 | static void |
2355 | dpif_netdev_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) | |
2356 | { | |
2357 | size_t i; | |
2358 | ||
2359 | for (i = 0; i < n_ops; i++) { | |
2360 | struct dpif_op *op = ops[i]; | |
2361 | ||
2362 | switch (op->type) { | |
2363 | case DPIF_OP_FLOW_PUT: | |
2364 | op->error = dpif_netdev_flow_put(dpif, &op->u.flow_put); | |
2365 | break; | |
2366 | ||
2367 | case DPIF_OP_FLOW_DEL: | |
2368 | op->error = dpif_netdev_flow_del(dpif, &op->u.flow_del); | |
2369 | break; | |
2370 | ||
2371 | case DPIF_OP_EXECUTE: | |
2372 | op->error = dpif_netdev_execute(dpif, &op->u.execute); | |
2373 | break; | |
6fe09f8c JS |
2374 | |
2375 | case DPIF_OP_FLOW_GET: | |
2376 | op->error = dpif_netdev_flow_get(dpif, &op->u.flow_get); | |
2377 | break; | |
1a0c894a BP |
2378 | } |
2379 | } | |
2380 | } | |
2381 | ||
f2eee189 AW |
2382 | /* Returns true if the configuration for rx queues or cpu mask |
2383 | * is changed. */ | |
2384 | static bool | |
2385 | pmd_config_changed(const struct dp_netdev *dp, size_t rxqs, const char *cmask) | |
2386 | { | |
2387 | if (dp->n_dpdk_rxqs != rxqs) { | |
2388 | return true; | |
2389 | } else { | |
2390 | if (dp->pmd_cmask != NULL && cmask != NULL) { | |
2391 | return strcmp(dp->pmd_cmask, cmask); | |
2392 | } else { | |
2393 | return (dp->pmd_cmask != NULL || cmask != NULL); | |
2394 | } | |
2395 | } | |
2396 | } | |
2397 | ||
2398 | /* Resets pmd threads if the configuration for 'rxq's or cpu mask changes. */ | |
2399 | static int | |
2400 | dpif_netdev_pmd_set(struct dpif *dpif, unsigned int n_rxqs, const char *cmask) | |
2401 | { | |
2402 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2403 | ||
2404 | if (pmd_config_changed(dp, n_rxqs, cmask)) { | |
2405 | struct dp_netdev_port *port; | |
2406 | ||
2407 | dp_netdev_destroy_all_pmds(dp); | |
2408 | ||
2409 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
2410 | if (netdev_is_pmd(port->netdev)) { | |
2411 | int i, err; | |
2412 | ||
2413 | /* Closes the existing 'rxq's. */ | |
2414 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2415 | netdev_rxq_close(port->rxq[i]); | |
2416 | port->rxq[i] = NULL; | |
2417 | } | |
2418 | ||
2419 | /* Sets the new rx queue config. */ | |
3bcc10c0 DDP |
2420 | err = netdev_set_multiq(port->netdev, |
2421 | ovs_numa_get_n_cores() + 1, | |
f2eee189 | 2422 | n_rxqs); |
7251515e | 2423 | if (err && (err != EOPNOTSUPP)) { |
f2eee189 AW |
2424 | VLOG_ERR("Failed to set dpdk interface %s rx_queue to:" |
2425 | " %u", netdev_get_name(port->netdev), | |
2426 | n_rxqs); | |
2427 | return err; | |
2428 | } | |
2429 | ||
2430 | /* If the set_multiq() above succeeds, reopens the 'rxq's. */ | |
2431 | port->rxq = xrealloc(port->rxq, sizeof *port->rxq | |
2432 | * netdev_n_rxq(port->netdev)); | |
2433 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2434 | netdev_rxq_open(port->netdev, &port->rxq[i], i); | |
2435 | } | |
2436 | } | |
2437 | } | |
2438 | dp->n_dpdk_rxqs = n_rxqs; | |
2439 | ||
2440 | /* Reconfigures the cpu mask. */ | |
2441 | ovs_numa_set_cpu_mask(cmask); | |
2442 | free(dp->pmd_cmask); | |
2443 | dp->pmd_cmask = cmask ? xstrdup(cmask) : NULL; | |
2444 | ||
2445 | /* Restores the non-pmd. */ | |
2446 | dp_netdev_set_nonpmd(dp); | |
2447 | /* Restores all pmd threads. */ | |
2448 | dp_netdev_reset_pmd_threads(dp); | |
2449 | } | |
2450 | ||
2451 | return 0; | |
2452 | } | |
2453 | ||
5bf93d67 EJ |
2454 | static int |
2455 | dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED, | |
2456 | uint32_t queue_id, uint32_t *priority) | |
2457 | { | |
2458 | *priority = queue_id; | |
2459 | return 0; | |
2460 | } | |
2461 | ||
72865317 | 2462 | \f |
9ff55ae2 DDP |
2463 | /* Creates and returns a new 'struct dp_netdev_actions', whose actions are |
2464 | * a copy of the 'ofpacts_len' bytes of 'ofpacts'. */ | |
a84cb64a BP |
2465 | struct dp_netdev_actions * |
2466 | dp_netdev_actions_create(const struct nlattr *actions, size_t size) | |
2467 | { | |
2468 | struct dp_netdev_actions *netdev_actions; | |
2469 | ||
9ff55ae2 DDP |
2470 | netdev_actions = xmalloc(sizeof *netdev_actions + size); |
2471 | memcpy(netdev_actions->actions, actions, size); | |
a84cb64a BP |
2472 | netdev_actions->size = size; |
2473 | ||
2474 | return netdev_actions; | |
2475 | } | |
2476 | ||
a84cb64a | 2477 | struct dp_netdev_actions * |
61e7deb1 | 2478 | dp_netdev_flow_get_actions(const struct dp_netdev_flow *flow) |
a84cb64a | 2479 | { |
61e7deb1 | 2480 | return ovsrcu_get(struct dp_netdev_actions *, &flow->actions); |
a84cb64a BP |
2481 | } |
2482 | ||
61e7deb1 BP |
2483 | static void |
2484 | dp_netdev_actions_free(struct dp_netdev_actions *actions) | |
a84cb64a | 2485 | { |
61e7deb1 | 2486 | free(actions); |
a84cb64a BP |
2487 | } |
2488 | \f | |
55e3ca97 DDP |
2489 | static inline unsigned long long |
2490 | cycles_counter(void) | |
2491 | { | |
2492 | #ifdef DPDK_NETDEV | |
2493 | return rte_get_tsc_cycles(); | |
2494 | #else | |
2495 | return 0; | |
2496 | #endif | |
2497 | } | |
2498 | ||
2499 | /* Fake mutex to make sure that the calls to cycles_count_* are balanced */ | |
2500 | extern struct ovs_mutex cycles_counter_fake_mutex; | |
2501 | ||
2502 | /* Start counting cycles. Must be followed by 'cycles_count_end()' */ | |
2503 | static inline void | |
2504 | cycles_count_start(struct dp_netdev_pmd_thread *pmd) | |
2505 | OVS_ACQUIRES(&cycles_counter_fake_mutex) | |
2506 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2507 | { | |
2508 | pmd->last_cycles = cycles_counter(); | |
2509 | } | |
2510 | ||
2511 | /* Stop counting cycles and add them to the counter 'type' */ | |
2512 | static inline void | |
2513 | cycles_count_end(struct dp_netdev_pmd_thread *pmd, | |
2514 | enum pmd_cycles_counter_type type) | |
2515 | OVS_RELEASES(&cycles_counter_fake_mutex) | |
2516 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2517 | { | |
2518 | unsigned long long interval = cycles_counter() - pmd->last_cycles; | |
2519 | ||
2520 | non_atomic_ullong_add(&pmd->cycles.n[type], interval); | |
2521 | } | |
e4cfed38 | 2522 | |
5794e276 | 2523 | static void |
65f13b50 | 2524 | dp_netdev_process_rxq_port(struct dp_netdev_pmd_thread *pmd, |
9bbf1c3d DDP |
2525 | struct dp_netdev_port *port, |
2526 | struct netdev_rxq *rxq) | |
e4cfed38 | 2527 | { |
cd159f1a | 2528 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 | 2529 | int error, cnt; |
e4cfed38 | 2530 | |
55e3ca97 | 2531 | cycles_count_start(pmd); |
8cbf4f47 | 2532 | error = netdev_rxq_recv(rxq, packets, &cnt); |
55e3ca97 | 2533 | cycles_count_end(pmd, PMD_CYCLES_POLLING); |
e4cfed38 | 2534 | if (!error) { |
41ccaa24 | 2535 | int i; |
3c33f0ff JR |
2536 | |
2537 | *recirc_depth_get() = 0; | |
41ccaa24 PS |
2538 | |
2539 | /* XXX: initialize md in netdev implementation. */ | |
2540 | for (i = 0; i < cnt; i++) { | |
efa2bcbb | 2541 | packets[i]->md = port->md; |
41ccaa24 | 2542 | } |
55e3ca97 | 2543 | cycles_count_start(pmd); |
41ccaa24 | 2544 | dp_netdev_input(pmd, packets, cnt); |
55e3ca97 | 2545 | cycles_count_end(pmd, PMD_CYCLES_PROCESSING); |
e4cfed38 | 2546 | } else if (error != EAGAIN && error != EOPNOTSUPP) { |
3c33f0ff | 2547 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
e4cfed38 PS |
2548 | |
2549 | VLOG_ERR_RL(&rl, "error receiving data from %s: %s", | |
3c33f0ff | 2550 | netdev_get_name(port->netdev), ovs_strerror(error)); |
e4cfed38 PS |
2551 | } |
2552 | } | |
2553 | ||
a36de779 PS |
2554 | /* Return true if needs to revalidate datapath flows. */ |
2555 | static bool | |
e4cfed38 PS |
2556 | dpif_netdev_run(struct dpif *dpif) |
2557 | { | |
2558 | struct dp_netdev_port *port; | |
2559 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
b19befae AW |
2560 | struct dp_netdev_pmd_thread *non_pmd = dp_netdev_get_pmd(dp, |
2561 | NON_PMD_CORE_ID); | |
a36de779 | 2562 | uint64_t new_tnl_seq; |
e4cfed38 | 2563 | |
65f13b50 | 2564 | ovs_mutex_lock(&dp->non_pmd_mutex); |
a532e683 | 2565 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2566 | if (!netdev_is_pmd(port->netdev)) { |
2567 | int i; | |
2568 | ||
2569 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
65f13b50 | 2570 | dp_netdev_process_rxq_port(non_pmd, port, port->rxq[i]); |
55c955bd | 2571 | } |
e4cfed38 PS |
2572 | } |
2573 | } | |
65f13b50 | 2574 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
1c1e46ed AW |
2575 | dp_netdev_pmd_unref(non_pmd); |
2576 | ||
a36de779 PS |
2577 | tnl_arp_cache_run(); |
2578 | new_tnl_seq = seq_read(tnl_conf_seq); | |
2579 | ||
2580 | if (dp->last_tnl_conf_seq != new_tnl_seq) { | |
2581 | dp->last_tnl_conf_seq = new_tnl_seq; | |
2582 | return true; | |
2583 | } | |
2584 | return false; | |
e4cfed38 PS |
2585 | } |
2586 | ||
2587 | static void | |
2588 | dpif_netdev_wait(struct dpif *dpif) | |
2589 | { | |
2590 | struct dp_netdev_port *port; | |
2591 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2592 | ||
59e6d833 | 2593 | ovs_mutex_lock(&dp_netdev_mutex); |
a532e683 | 2594 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2595 | if (!netdev_is_pmd(port->netdev)) { |
2596 | int i; | |
2597 | ||
2598 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2599 | netdev_rxq_wait(port->rxq[i]); | |
2600 | } | |
e4cfed38 PS |
2601 | } |
2602 | } | |
59e6d833 | 2603 | ovs_mutex_unlock(&dp_netdev_mutex); |
a36de779 | 2604 | seq_wait(tnl_conf_seq, dp->last_tnl_conf_seq); |
e4cfed38 PS |
2605 | } |
2606 | ||
f7791740 | 2607 | struct rxq_poll { |
e4cfed38 | 2608 | struct dp_netdev_port *port; |
55c955bd | 2609 | struct netdev_rxq *rx; |
e4cfed38 PS |
2610 | }; |
2611 | ||
2612 | static int | |
65f13b50 | 2613 | pmd_load_queues(struct dp_netdev_pmd_thread *pmd, |
f7791740 | 2614 | struct rxq_poll **ppoll_list, int poll_cnt) |
e4cfed38 | 2615 | { |
f7791740 | 2616 | struct rxq_poll *poll_list = *ppoll_list; |
e4cfed38 | 2617 | struct dp_netdev_port *port; |
65f13b50 | 2618 | int n_pmds_on_numa, index, i; |
e4cfed38 PS |
2619 | |
2620 | /* Simple scheduler for netdev rx polling. */ | |
e4cfed38 | 2621 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2622 | port_unref(poll_list[i].port); |
e4cfed38 PS |
2623 | } |
2624 | ||
2625 | poll_cnt = 0; | |
65f13b50 | 2626 | n_pmds_on_numa = get_n_pmd_threads_on_numa(pmd->dp, pmd->numa_id); |
e4cfed38 PS |
2627 | index = 0; |
2628 | ||
65f13b50 | 2629 | CMAP_FOR_EACH (port, node, &pmd->dp->ports) { |
a1fdee13 AW |
2630 | /* Calls port_try_ref() to prevent the main thread |
2631 | * from deleting the port. */ | |
2632 | if (port_try_ref(port)) { | |
65f13b50 AW |
2633 | if (netdev_is_pmd(port->netdev) |
2634 | && netdev_get_numa_id(port->netdev) == pmd->numa_id) { | |
a1fdee13 AW |
2635 | int i; |
2636 | ||
2637 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
65f13b50 | 2638 | if ((index % n_pmds_on_numa) == pmd->index) { |
a1fdee13 AW |
2639 | poll_list = xrealloc(poll_list, |
2640 | sizeof *poll_list * (poll_cnt + 1)); | |
2641 | ||
2642 | port_ref(port); | |
2643 | poll_list[poll_cnt].port = port; | |
2644 | poll_list[poll_cnt].rx = port->rxq[i]; | |
2645 | poll_cnt++; | |
2646 | } | |
2647 | index++; | |
55c955bd | 2648 | } |
e4cfed38 | 2649 | } |
a1fdee13 AW |
2650 | /* Unrefs the port_try_ref(). */ |
2651 | port_unref(port); | |
e4cfed38 PS |
2652 | } |
2653 | } | |
2654 | ||
e4cfed38 PS |
2655 | *ppoll_list = poll_list; |
2656 | return poll_cnt; | |
2657 | } | |
2658 | ||
6c3eee82 | 2659 | static void * |
e4cfed38 | 2660 | pmd_thread_main(void *f_) |
6c3eee82 | 2661 | { |
65f13b50 | 2662 | struct dp_netdev_pmd_thread *pmd = f_; |
e4cfed38 | 2663 | unsigned int lc = 0; |
f7791740 | 2664 | struct rxq_poll *poll_list; |
84067a4c | 2665 | unsigned int port_seq = PMD_INITIAL_SEQ; |
e4cfed38 PS |
2666 | int poll_cnt; |
2667 | int i; | |
6c3eee82 | 2668 | |
e4cfed38 PS |
2669 | poll_cnt = 0; |
2670 | poll_list = NULL; | |
2671 | ||
65f13b50 AW |
2672 | /* Stores the pmd thread's 'pmd' to 'per_pmd_key'. */ |
2673 | ovsthread_setspecific(pmd->dp->per_pmd_key, pmd); | |
2674 | pmd_thread_setaffinity_cpu(pmd->core_id); | |
e4cfed38 | 2675 | reload: |
65f13b50 AW |
2676 | emc_cache_init(&pmd->flow_cache); |
2677 | poll_cnt = pmd_load_queues(pmd, &poll_list, poll_cnt); | |
6c3eee82 | 2678 | |
accf8626 AW |
2679 | /* Signal here to make sure the pmd finishes |
2680 | * reloading the updated configuration. */ | |
2681 | dp_netdev_pmd_reload_done(pmd); | |
2682 | ||
e4cfed38 | 2683 | for (;;) { |
6c3eee82 BP |
2684 | int i; |
2685 | ||
e4cfed38 | 2686 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2687 | dp_netdev_process_rxq_port(pmd, poll_list[i].port, poll_list[i].rx); |
e4cfed38 PS |
2688 | } |
2689 | ||
2690 | if (lc++ > 1024) { | |
84067a4c | 2691 | unsigned int seq; |
6c3eee82 | 2692 | |
e4cfed38 | 2693 | lc = 0; |
84067a4c | 2694 | |
67ad54cb | 2695 | emc_cache_slow_sweep(&pmd->flow_cache); |
84067a4c JR |
2696 | ovsrcu_quiesce(); |
2697 | ||
65f13b50 | 2698 | atomic_read_relaxed(&pmd->change_seq, &seq); |
84067a4c JR |
2699 | if (seq != port_seq) { |
2700 | port_seq = seq; | |
6c3eee82 BP |
2701 | break; |
2702 | } | |
2703 | } | |
e4cfed38 | 2704 | } |
6c3eee82 | 2705 | |
65f13b50 | 2706 | emc_cache_uninit(&pmd->flow_cache); |
9bbf1c3d | 2707 | |
65f13b50 | 2708 | if (!latch_is_set(&pmd->exit_latch)){ |
e4cfed38 PS |
2709 | goto reload; |
2710 | } | |
6c3eee82 | 2711 | |
e4cfed38 PS |
2712 | for (i = 0; i < poll_cnt; i++) { |
2713 | port_unref(poll_list[i].port); | |
6c3eee82 | 2714 | } |
6c3eee82 | 2715 | |
accf8626 AW |
2716 | dp_netdev_pmd_reload_done(pmd); |
2717 | ||
e4cfed38 | 2718 | free(poll_list); |
6c3eee82 BP |
2719 | return NULL; |
2720 | } | |
2721 | ||
6b31e073 RW |
2722 | static void |
2723 | dp_netdev_disable_upcall(struct dp_netdev *dp) | |
2724 | OVS_ACQUIRES(dp->upcall_rwlock) | |
2725 | { | |
2726 | fat_rwlock_wrlock(&dp->upcall_rwlock); | |
2727 | } | |
2728 | ||
2729 | static void | |
2730 | dpif_netdev_disable_upcall(struct dpif *dpif) | |
2731 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2732 | { | |
2733 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2734 | dp_netdev_disable_upcall(dp); | |
2735 | } | |
2736 | ||
2737 | static void | |
2738 | dp_netdev_enable_upcall(struct dp_netdev *dp) | |
2739 | OVS_RELEASES(dp->upcall_rwlock) | |
2740 | { | |
2741 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
2742 | } | |
2743 | ||
2744 | static void | |
2745 | dpif_netdev_enable_upcall(struct dpif *dpif) | |
2746 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2747 | { | |
2748 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2749 | dp_netdev_enable_upcall(dp); | |
2750 | } | |
2751 | ||
accf8626 AW |
2752 | void |
2753 | dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd) | |
2754 | { | |
2755 | ovs_mutex_lock(&pmd->cond_mutex); | |
2756 | xpthread_cond_signal(&pmd->cond); | |
2757 | ovs_mutex_unlock(&pmd->cond_mutex); | |
2758 | } | |
2759 | ||
1c1e46ed AW |
2760 | /* Finds and refs the dp_netdev_pmd_thread on core 'core_id'. Returns |
2761 | * the pointer if succeeds, otherwise, NULL. | |
2762 | * | |
2763 | * Caller must unrefs the returned reference. */ | |
65f13b50 | 2764 | static struct dp_netdev_pmd_thread * |
bd5131ba | 2765 | dp_netdev_get_pmd(struct dp_netdev *dp, unsigned core_id) |
65f13b50 AW |
2766 | { |
2767 | struct dp_netdev_pmd_thread *pmd; | |
55847abe | 2768 | const struct cmap_node *pnode; |
65f13b50 | 2769 | |
b19befae | 2770 | pnode = cmap_find(&dp->poll_threads, hash_int(core_id, 0)); |
1c1e46ed AW |
2771 | if (!pnode) { |
2772 | return NULL; | |
2773 | } | |
65f13b50 AW |
2774 | pmd = CONTAINER_OF(pnode, struct dp_netdev_pmd_thread, node); |
2775 | ||
1c1e46ed | 2776 | return dp_netdev_pmd_try_ref(pmd) ? pmd : NULL; |
65f13b50 AW |
2777 | } |
2778 | ||
f2eee189 AW |
2779 | /* Sets the 'struct dp_netdev_pmd_thread' for non-pmd threads. */ |
2780 | static void | |
2781 | dp_netdev_set_nonpmd(struct dp_netdev *dp) | |
2782 | { | |
2783 | struct dp_netdev_pmd_thread *non_pmd; | |
2784 | ||
2785 | non_pmd = xzalloc(sizeof *non_pmd); | |
2786 | dp_netdev_configure_pmd(non_pmd, dp, 0, NON_PMD_CORE_ID, | |
2787 | OVS_NUMA_UNSPEC); | |
2788 | } | |
2789 | ||
1c1e46ed AW |
2790 | /* Caller must have valid pointer to 'pmd'. */ |
2791 | static bool | |
2792 | dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd) | |
2793 | { | |
2794 | return ovs_refcount_try_ref_rcu(&pmd->ref_cnt); | |
2795 | } | |
2796 | ||
2797 | static void | |
2798 | dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd) | |
2799 | { | |
2800 | if (pmd && ovs_refcount_unref(&pmd->ref_cnt) == 1) { | |
2801 | ovsrcu_postpone(dp_netdev_destroy_pmd, pmd); | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | /* Given cmap position 'pos', tries to ref the next node. If try_ref() | |
2806 | * fails, keeps checking for next node until reaching the end of cmap. | |
2807 | * | |
2808 | * Caller must unrefs the returned reference. */ | |
2809 | static struct dp_netdev_pmd_thread * | |
2810 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos) | |
2811 | { | |
2812 | struct dp_netdev_pmd_thread *next; | |
2813 | ||
2814 | do { | |
2815 | struct cmap_node *node; | |
2816 | ||
2817 | node = cmap_next_position(&dp->poll_threads, pos); | |
2818 | next = node ? CONTAINER_OF(node, struct dp_netdev_pmd_thread, node) | |
2819 | : NULL; | |
2820 | } while (next && !dp_netdev_pmd_try_ref(next)); | |
2821 | ||
2822 | return next; | |
2823 | } | |
2824 | ||
3bcc10c0 DDP |
2825 | static int |
2826 | core_id_to_qid(unsigned core_id) | |
2827 | { | |
2828 | if (core_id != NON_PMD_CORE_ID) { | |
2829 | return core_id; | |
2830 | } else { | |
2831 | return ovs_numa_get_n_cores(); | |
2832 | } | |
2833 | } | |
2834 | ||
65f13b50 | 2835 | /* Configures the 'pmd' based on the input argument. */ |
6c3eee82 | 2836 | static void |
65f13b50 | 2837 | dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp, |
bd5131ba | 2838 | int index, unsigned core_id, int numa_id) |
65f13b50 AW |
2839 | { |
2840 | pmd->dp = dp; | |
2841 | pmd->index = index; | |
2842 | pmd->core_id = core_id; | |
3bcc10c0 | 2843 | pmd->tx_qid = core_id_to_qid(core_id); |
65f13b50 | 2844 | pmd->numa_id = numa_id; |
1c1e46ed AW |
2845 | |
2846 | ovs_refcount_init(&pmd->ref_cnt); | |
65f13b50 AW |
2847 | latch_init(&pmd->exit_latch); |
2848 | atomic_init(&pmd->change_seq, PMD_INITIAL_SEQ); | |
accf8626 AW |
2849 | xpthread_cond_init(&pmd->cond, NULL); |
2850 | ovs_mutex_init(&pmd->cond_mutex); | |
1c1e46ed AW |
2851 | ovs_mutex_init(&pmd->flow_mutex); |
2852 | dpcls_init(&pmd->cls); | |
2853 | cmap_init(&pmd->flow_table); | |
65f13b50 AW |
2854 | /* init the 'flow_cache' since there is no |
2855 | * actual thread created for NON_PMD_CORE_ID. */ | |
2856 | if (core_id == NON_PMD_CORE_ID) { | |
2857 | emc_cache_init(&pmd->flow_cache); | |
2858 | } | |
2859 | cmap_insert(&dp->poll_threads, CONST_CAST(struct cmap_node *, &pmd->node), | |
2860 | hash_int(core_id, 0)); | |
2861 | } | |
2862 | ||
1c1e46ed AW |
2863 | static void |
2864 | dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd) | |
2865 | { | |
2866 | dp_netdev_pmd_flow_flush(pmd); | |
2867 | dpcls_destroy(&pmd->cls); | |
2868 | cmap_destroy(&pmd->flow_table); | |
2869 | ovs_mutex_destroy(&pmd->flow_mutex); | |
2870 | latch_destroy(&pmd->exit_latch); | |
2871 | xpthread_cond_destroy(&pmd->cond); | |
2872 | ovs_mutex_destroy(&pmd->cond_mutex); | |
2873 | free(pmd); | |
2874 | } | |
2875 | ||
2876 | /* Stops the pmd thread, removes it from the 'dp->poll_threads', | |
2877 | * and unrefs the struct. */ | |
65f13b50 AW |
2878 | static void |
2879 | dp_netdev_del_pmd(struct dp_netdev_pmd_thread *pmd) | |
6c3eee82 | 2880 | { |
65f13b50 | 2881 | /* Uninit the 'flow_cache' since there is |
1c1e46ed | 2882 | * no actual thread uninit it for NON_PMD_CORE_ID. */ |
65f13b50 AW |
2883 | if (pmd->core_id == NON_PMD_CORE_ID) { |
2884 | emc_cache_uninit(&pmd->flow_cache); | |
2885 | } else { | |
2886 | latch_set(&pmd->exit_latch); | |
2887 | dp_netdev_reload_pmd__(pmd); | |
2888 | ovs_numa_unpin_core(pmd->core_id); | |
2889 | xpthread_join(pmd->thread, NULL); | |
2890 | } | |
2891 | cmap_remove(&pmd->dp->poll_threads, &pmd->node, hash_int(pmd->core_id, 0)); | |
1c1e46ed | 2892 | dp_netdev_pmd_unref(pmd); |
65f13b50 | 2893 | } |
6c3eee82 | 2894 | |
65f13b50 AW |
2895 | /* Destroys all pmd threads. */ |
2896 | static void | |
2897 | dp_netdev_destroy_all_pmds(struct dp_netdev *dp) | |
2898 | { | |
2899 | struct dp_netdev_pmd_thread *pmd; | |
2900 | ||
2901 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2902 | dp_netdev_del_pmd(pmd); | |
6c3eee82 | 2903 | } |
65f13b50 | 2904 | } |
6c3eee82 | 2905 | |
65f13b50 AW |
2906 | /* Deletes all pmd threads on numa node 'numa_id'. */ |
2907 | static void | |
2908 | dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
2909 | { | |
2910 | struct dp_netdev_pmd_thread *pmd; | |
6c3eee82 | 2911 | |
65f13b50 AW |
2912 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { |
2913 | if (pmd->numa_id == numa_id) { | |
2914 | dp_netdev_del_pmd(pmd); | |
2915 | } | |
6c3eee82 | 2916 | } |
65f13b50 | 2917 | } |
6c3eee82 | 2918 | |
65f13b50 AW |
2919 | /* Checks the numa node id of 'netdev' and starts pmd threads for |
2920 | * the numa node. */ | |
2921 | static void | |
2922 | dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
2923 | { | |
2924 | int n_pmds; | |
e4cfed38 | 2925 | |
65f13b50 AW |
2926 | if (!ovs_numa_numa_id_is_valid(numa_id)) { |
2927 | VLOG_ERR("Cannot create pmd threads due to numa id (%d)" | |
2928 | "invalid", numa_id); | |
2929 | return ; | |
2930 | } | |
2931 | ||
2932 | n_pmds = get_n_pmd_threads_on_numa(dp, numa_id); | |
2933 | ||
2934 | /* If there are already pmd threads created for the numa node | |
2935 | * in which 'netdev' is on, do nothing. Else, creates the | |
2936 | * pmd threads for the numa node. */ | |
2937 | if (!n_pmds) { | |
2938 | int can_have, n_unpinned, i; | |
2939 | ||
2940 | n_unpinned = ovs_numa_get_n_unpinned_cores_on_numa(numa_id); | |
2941 | if (!n_unpinned) { | |
2942 | VLOG_ERR("Cannot create pmd threads due to out of unpinned " | |
2943 | "cores on numa node"); | |
2944 | return; | |
2945 | } | |
6c3eee82 | 2946 | |
f2eee189 AW |
2947 | /* If cpu mask is specified, uses all unpinned cores, otherwise |
2948 | * tries creating NR_PMD_THREADS pmd threads. */ | |
2949 | can_have = dp->pmd_cmask ? n_unpinned : MIN(n_unpinned, NR_PMD_THREADS); | |
65f13b50 AW |
2950 | for (i = 0; i < can_have; i++) { |
2951 | struct dp_netdev_pmd_thread *pmd = xzalloc(sizeof *pmd); | |
bd5131ba | 2952 | unsigned core_id = ovs_numa_get_unpinned_core_on_numa(numa_id); |
e4cfed38 | 2953 | |
65f13b50 AW |
2954 | dp_netdev_configure_pmd(pmd, dp, i, core_id, numa_id); |
2955 | /* Each thread will distribute all devices rx-queues among | |
2956 | * themselves. */ | |
2957 | pmd->thread = ovs_thread_create("pmd", pmd_thread_main, pmd); | |
2958 | } | |
2959 | VLOG_INFO("Created %d pmd threads on numa node %d", can_have, numa_id); | |
6c3eee82 BP |
2960 | } |
2961 | } | |
e4cfed38 | 2962 | |
6c3eee82 | 2963 | \f |
f2eee189 AW |
2964 | /* Called after pmd threads config change. Restarts pmd threads with |
2965 | * new configuration. */ | |
2966 | static void | |
2967 | dp_netdev_reset_pmd_threads(struct dp_netdev *dp) | |
2968 | { | |
2969 | struct dp_netdev_port *port; | |
2970 | ||
2971 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
2972 | if (netdev_is_pmd(port->netdev)) { | |
2973 | int numa_id = netdev_get_numa_id(port->netdev); | |
2974 | ||
2975 | dp_netdev_set_pmds_on_numa(dp, numa_id); | |
2976 | } | |
2977 | } | |
2978 | } | |
2979 | ||
b5cbbcf6 AZ |
2980 | static char * |
2981 | dpif_netdev_get_datapath_version(void) | |
2982 | { | |
2983 | return xstrdup("<built-in>"); | |
2984 | } | |
2985 | ||
72865317 | 2986 | static void |
1c1e46ed | 2987 | dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow, int cnt, int size, |
11bfdadd | 2988 | uint16_t tcp_flags, long long now) |
72865317 | 2989 | { |
eb94da30 | 2990 | uint16_t flags; |
72865317 | 2991 | |
eb94da30 DDP |
2992 | atomic_store_relaxed(&netdev_flow->stats.used, now); |
2993 | non_atomic_ullong_add(&netdev_flow->stats.packet_count, cnt); | |
2994 | non_atomic_ullong_add(&netdev_flow->stats.byte_count, size); | |
2995 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
2996 | flags |= tcp_flags; | |
2997 | atomic_store_relaxed(&netdev_flow->stats.tcp_flags, flags); | |
51852a57 BP |
2998 | } |
2999 | ||
3000 | static void | |
1c1e46ed AW |
3001 | dp_netdev_count_packet(struct dp_netdev_pmd_thread *pmd, |
3002 | enum dp_stat_type type, int cnt) | |
51852a57 | 3003 | { |
eb94da30 | 3004 | non_atomic_ullong_add(&pmd->stats.n[type], cnt); |
51852a57 BP |
3005 | } |
3006 | ||
623540e4 | 3007 | static int |
e14deea0 | 3008 | dp_netdev_upcall(struct dp_netdev_pmd_thread *pmd, struct dp_packet *packet_, |
7af12bd7 | 3009 | struct flow *flow, struct flow_wildcards *wc, ovs_u128 *ufid, |
623540e4 EJ |
3010 | enum dpif_upcall_type type, const struct nlattr *userdata, |
3011 | struct ofpbuf *actions, struct ofpbuf *put_actions) | |
3012 | { | |
1c1e46ed | 3013 | struct dp_netdev *dp = pmd->dp; |
623540e4 | 3014 | |
623540e4 EJ |
3015 | if (OVS_UNLIKELY(!dp->upcall_cb)) { |
3016 | return ENODEV; | |
3017 | } | |
3018 | ||
3019 | if (OVS_UNLIKELY(!VLOG_DROP_DBG(&upcall_rl))) { | |
3020 | struct ds ds = DS_EMPTY_INITIALIZER; | |
623540e4 | 3021 | char *packet_str; |
cf62fa4c | 3022 | struct ofpbuf key; |
5262eea1 JG |
3023 | struct odp_flow_key_parms odp_parms = { |
3024 | .flow = flow, | |
3025 | .mask = &wc->masks, | |
3026 | .odp_in_port = flow->in_port.odp_port, | |
3027 | .recirc = true, | |
3028 | }; | |
623540e4 EJ |
3029 | |
3030 | ofpbuf_init(&key, 0); | |
5262eea1 | 3031 | odp_flow_key_from_flow(&odp_parms, &key); |
cf62fa4c PS |
3032 | packet_str = ofp_packet_to_string(dp_packet_data(packet_), |
3033 | dp_packet_size(packet_)); | |
623540e4 | 3034 | |
6fd6ed71 | 3035 | odp_flow_key_format(key.data, key.size, &ds); |
623540e4 EJ |
3036 | |
3037 | VLOG_DBG("%s: %s upcall:\n%s\n%s", dp->name, | |
3038 | dpif_upcall_type_to_string(type), ds_cstr(&ds), packet_str); | |
3039 | ||
3040 | ofpbuf_uninit(&key); | |
3041 | free(packet_str); | |
6fd6ed71 | 3042 | |
623540e4 EJ |
3043 | ds_destroy(&ds); |
3044 | } | |
3045 | ||
cf62fa4c | 3046 | return dp->upcall_cb(packet_, flow, ufid, pmd->core_id, type, userdata, |
1c1e46ed | 3047 | actions, wc, put_actions, dp->upcall_aux); |
623540e4 EJ |
3048 | } |
3049 | ||
9bbf1c3d | 3050 | static inline uint32_t |
048963aa DDP |
3051 | dpif_netdev_packet_get_rss_hash(struct dp_packet *packet, |
3052 | const struct miniflow *mf) | |
9bbf1c3d | 3053 | { |
048963aa | 3054 | uint32_t hash, recirc_depth; |
9bbf1c3d | 3055 | |
2bc1bbd2 | 3056 | hash = dp_packet_get_rss_hash(packet); |
9bbf1c3d DDP |
3057 | if (OVS_UNLIKELY(!hash)) { |
3058 | hash = miniflow_hash_5tuple(mf, 0); | |
2bc1bbd2 | 3059 | dp_packet_set_rss_hash(packet, hash); |
9bbf1c3d | 3060 | } |
048963aa DDP |
3061 | |
3062 | /* The RSS hash must account for the recirculation depth to avoid | |
3063 | * collisions in the exact match cache */ | |
3064 | recirc_depth = *recirc_depth_get_unsafe(); | |
3065 | if (OVS_UNLIKELY(recirc_depth)) { | |
3066 | hash = hash_finish(hash, recirc_depth); | |
3067 | dp_packet_set_rss_hash(packet, hash); | |
3068 | } | |
9bbf1c3d DDP |
3069 | return hash; |
3070 | } | |
3071 | ||
567bbb2e | 3072 | struct packet_batch { |
8cbf4f47 DDP |
3073 | unsigned int packet_count; |
3074 | unsigned int byte_count; | |
3075 | uint16_t tcp_flags; | |
3076 | ||
3077 | struct dp_netdev_flow *flow; | |
3078 | ||
cd159f1a | 3079 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 DDP |
3080 | }; |
3081 | ||
3082 | static inline void | |
e14deea0 | 3083 | packet_batch_update(struct packet_batch *batch, struct dp_packet *packet, |
9bbf1c3d | 3084 | const struct miniflow *mf) |
8cbf4f47 DDP |
3085 | { |
3086 | batch->tcp_flags |= miniflow_get_tcp_flags(mf); | |
3087 | batch->packets[batch->packet_count++] = packet; | |
cf62fa4c | 3088 | batch->byte_count += dp_packet_size(packet); |
8cbf4f47 DDP |
3089 | } |
3090 | ||
3091 | static inline void | |
41ccaa24 | 3092 | packet_batch_init(struct packet_batch *batch, struct dp_netdev_flow *flow) |
8cbf4f47 | 3093 | { |
11e5cf1f | 3094 | flow->batch = batch; |
8cbf4f47 | 3095 | |
11e5cf1f | 3096 | batch->flow = flow; |
8cbf4f47 DDP |
3097 | batch->packet_count = 0; |
3098 | batch->byte_count = 0; | |
3099 | batch->tcp_flags = 0; | |
8cbf4f47 DDP |
3100 | } |
3101 | ||
3102 | static inline void | |
65f13b50 | 3103 | packet_batch_execute(struct packet_batch *batch, |
abcf3ef4 | 3104 | struct dp_netdev_pmd_thread *pmd, |
11bfdadd | 3105 | long long now) |
8cbf4f47 DDP |
3106 | { |
3107 | struct dp_netdev_actions *actions; | |
3108 | struct dp_netdev_flow *flow = batch->flow; | |
3109 | ||
11e5cf1f | 3110 | dp_netdev_flow_used(flow, batch->packet_count, batch->byte_count, |
11bfdadd | 3111 | batch->tcp_flags, now); |
8cbf4f47 DDP |
3112 | |
3113 | actions = dp_netdev_flow_get_actions(flow); | |
3114 | ||
65f13b50 | 3115 | dp_netdev_execute_actions(pmd, batch->packets, batch->packet_count, true, |
41ccaa24 | 3116 | actions->actions, actions->size); |
8cbf4f47 DDP |
3117 | } |
3118 | ||
8aaa125d | 3119 | static inline void |
e14deea0 | 3120 | dp_netdev_queue_batches(struct dp_packet *pkt, |
9bbf1c3d | 3121 | struct dp_netdev_flow *flow, const struct miniflow *mf, |
8aaa125d | 3122 | struct packet_batch *batches, size_t *n_batches) |
9bbf1c3d | 3123 | { |
8aaa125d | 3124 | struct packet_batch *batch = flow->batch; |
11e5cf1f DDP |
3125 | |
3126 | if (OVS_LIKELY(batch)) { | |
3127 | packet_batch_update(batch, pkt, mf); | |
8aaa125d | 3128 | return; |
9bbf1c3d DDP |
3129 | } |
3130 | ||
3131 | batch = &batches[(*n_batches)++]; | |
41ccaa24 | 3132 | packet_batch_init(batch, flow); |
9bbf1c3d | 3133 | packet_batch_update(batch, pkt, mf); |
9bbf1c3d DDP |
3134 | } |
3135 | ||
3136 | static inline void | |
e14deea0 | 3137 | dp_packet_swap(struct dp_packet **a, struct dp_packet **b) |
9bbf1c3d | 3138 | { |
e14deea0 | 3139 | struct dp_packet *tmp = *a; |
9bbf1c3d DDP |
3140 | *a = *b; |
3141 | *b = tmp; | |
3142 | } | |
3143 | ||
3144 | /* Try to process all ('cnt') the 'packets' using only the exact match cache | |
8aaa125d DDP |
3145 | * 'flow_cache'. If a flow is not found for a packet 'packets[i]', the |
3146 | * miniflow is copied into 'keys' and the packet pointer is moved at the | |
3147 | * beginning of the 'packets' array. | |
9bbf1c3d DDP |
3148 | * |
3149 | * The function returns the number of packets that needs to be processed in the | |
3150 | * 'packets' array (they have been moved to the beginning of the vector). | |
3151 | */ | |
3152 | static inline size_t | |
e14deea0 | 3153 | emc_processing(struct dp_netdev_pmd_thread *pmd, struct dp_packet **packets, |
8aaa125d DDP |
3154 | size_t cnt, struct netdev_flow_key *keys, |
3155 | struct packet_batch batches[], size_t *n_batches) | |
72865317 | 3156 | { |
65f13b50 | 3157 | struct emc_cache *flow_cache = &pmd->flow_cache; |
8aaa125d DDP |
3158 | struct netdev_flow_key key; |
3159 | size_t i, notfound_cnt = 0; | |
8cbf4f47 | 3160 | |
0de8783a | 3161 | miniflow_initialize(&key.mf, key.buf); |
84d6d5eb | 3162 | for (i = 0; i < cnt; i++) { |
9bbf1c3d | 3163 | struct dp_netdev_flow *flow; |
9bbf1c3d | 3164 | |
cf62fa4c | 3165 | if (OVS_UNLIKELY(dp_packet_size(packets[i]) < ETH_HEADER_LEN)) { |
e14deea0 | 3166 | dp_packet_delete(packets[i]); |
84d6d5eb EJ |
3167 | continue; |
3168 | } | |
8cbf4f47 | 3169 | |
72a5e2b8 DDP |
3170 | if (i != cnt - 1) { |
3171 | /* Prefetch next packet data */ | |
3172 | OVS_PREFETCH(dp_packet_data(packets[i+1])); | |
3173 | } | |
3174 | ||
cf62fa4c | 3175 | miniflow_extract(packets[i], &key.mf); |
0de8783a | 3176 | key.len = 0; /* Not computed yet. */ |
048963aa | 3177 | key.hash = dpif_netdev_packet_get_rss_hash(packets[i], &key.mf); |
9bbf1c3d | 3178 | |
0de8783a | 3179 | flow = emc_lookup(flow_cache, &key); |
8aaa125d DDP |
3180 | if (OVS_LIKELY(flow)) { |
3181 | dp_netdev_queue_batches(packets[i], flow, &key.mf, batches, | |
3182 | n_batches); | |
3183 | } else { | |
9bbf1c3d | 3184 | if (i != notfound_cnt) { |
e14deea0 | 3185 | dp_packet_swap(&packets[i], &packets[notfound_cnt]); |
9bbf1c3d DDP |
3186 | } |
3187 | ||
3188 | keys[notfound_cnt++] = key; | |
3189 | } | |
3190 | } | |
3191 | ||
8aaa125d | 3192 | dp_netdev_count_packet(pmd, DP_STAT_EXACT_HIT, cnt - notfound_cnt); |
4f150744 | 3193 | |
9bbf1c3d DDP |
3194 | return notfound_cnt; |
3195 | } | |
3196 | ||
3197 | static inline void | |
65f13b50 | 3198 | fast_path_processing(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3199 | struct dp_packet **packets, size_t cnt, |
8aaa125d DDP |
3200 | struct netdev_flow_key *keys, |
3201 | struct packet_batch batches[], size_t *n_batches) | |
9bbf1c3d | 3202 | { |
1a0d5831 | 3203 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3204 | const size_t PKT_ARRAY_SIZE = cnt; |
3205 | #else | |
1a0d5831 | 3206 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3207 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d | 3208 | #endif |
0de8783a | 3209 | struct dpcls_rule *rules[PKT_ARRAY_SIZE]; |
65f13b50 AW |
3210 | struct dp_netdev *dp = pmd->dp; |
3211 | struct emc_cache *flow_cache = &pmd->flow_cache; | |
8aaa125d | 3212 | int miss_cnt = 0, lost_cnt = 0; |
9bbf1c3d | 3213 | bool any_miss; |
8aaa125d | 3214 | size_t i; |
9bbf1c3d DDP |
3215 | |
3216 | for (i = 0; i < cnt; i++) { | |
0de8783a JR |
3217 | /* Key length is needed in all the cases, hash computed on demand. */ |
3218 | keys[i].len = netdev_flow_key_size(count_1bits(keys[i].mf.map)); | |
9bbf1c3d | 3219 | } |
1c1e46ed | 3220 | any_miss = !dpcls_lookup(&pmd->cls, keys, rules, cnt); |
623540e4 EJ |
3221 | if (OVS_UNLIKELY(any_miss) && !fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { |
3222 | uint64_t actions_stub[512 / 8], slow_stub[512 / 8]; | |
3223 | struct ofpbuf actions, put_actions; | |
7af12bd7 | 3224 | ovs_u128 ufid; |
623540e4 EJ |
3225 | |
3226 | ofpbuf_use_stub(&actions, actions_stub, sizeof actions_stub); | |
3227 | ofpbuf_use_stub(&put_actions, slow_stub, sizeof slow_stub); | |
3228 | ||
3229 | for (i = 0; i < cnt; i++) { | |
0de8783a | 3230 | struct dp_netdev_flow *netdev_flow; |
623540e4 | 3231 | struct ofpbuf *add_actions; |
0de8783a | 3232 | struct match match; |
623540e4 EJ |
3233 | int error; |
3234 | ||
0de8783a | 3235 | if (OVS_LIKELY(rules[i])) { |
623540e4 EJ |
3236 | continue; |
3237 | } | |
3238 | ||
3239 | /* It's possible that an earlier slow path execution installed | |
0de8783a | 3240 | * a rule covering this flow. In this case, it's a lot cheaper |
623540e4 | 3241 | * to catch it here than execute a miss. */ |
1c1e46ed | 3242 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); |
623540e4 | 3243 | if (netdev_flow) { |
0de8783a | 3244 | rules[i] = &netdev_flow->cr; |
623540e4 EJ |
3245 | continue; |
3246 | } | |
3247 | ||
60fc3b7b DDP |
3248 | miss_cnt++; |
3249 | ||
0de8783a | 3250 | miniflow_expand(&keys[i].mf, &match.flow); |
623540e4 EJ |
3251 | |
3252 | ofpbuf_clear(&actions); | |
3253 | ofpbuf_clear(&put_actions); | |
3254 | ||
7af12bd7 | 3255 | dpif_flow_hash(dp->dpif, &match.flow, sizeof match.flow, &ufid); |
1c1e46ed | 3256 | error = dp_netdev_upcall(pmd, packets[i], &match.flow, &match.wc, |
7af12bd7 | 3257 | &ufid, DPIF_UC_MISS, NULL, &actions, |
0de8783a | 3258 | &put_actions); |
623540e4 | 3259 | if (OVS_UNLIKELY(error && error != ENOSPC)) { |
7ad20cbd | 3260 | dp_packet_delete(packets[i]); |
60fc3b7b | 3261 | lost_cnt++; |
623540e4 EJ |
3262 | continue; |
3263 | } | |
3264 | ||
3265 | /* We can't allow the packet batching in the next loop to execute | |
3266 | * the actions. Otherwise, if there are any slow path actions, | |
3267 | * we'll send the packet up twice. */ | |
41ccaa24 | 3268 | dp_netdev_execute_actions(pmd, &packets[i], 1, true, |
6fd6ed71 | 3269 | actions.data, actions.size); |
623540e4 | 3270 | |
6fd6ed71 | 3271 | add_actions = put_actions.size ? &put_actions : &actions; |
0de8783a JR |
3272 | if (OVS_LIKELY(error != ENOSPC)) { |
3273 | /* XXX: There's a race window where a flow covering this packet | |
3274 | * could have already been installed since we last did the flow | |
3275 | * lookup before upcall. This could be solved by moving the | |
3276 | * mutex lock outside the loop, but that's an awful long time | |
3277 | * to be locking everyone out of making flow installs. If we | |
3278 | * move to a per-core classifier, it would be reasonable. */ | |
1c1e46ed AW |
3279 | ovs_mutex_lock(&pmd->flow_mutex); |
3280 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); | |
0de8783a | 3281 | if (OVS_LIKELY(!netdev_flow)) { |
1c1e46ed | 3282 | netdev_flow = dp_netdev_flow_add(pmd, &match, &ufid, |
6fd6ed71 PS |
3283 | add_actions->data, |
3284 | add_actions->size); | |
0de8783a | 3285 | } |
1c1e46ed | 3286 | ovs_mutex_unlock(&pmd->flow_mutex); |
0de8783a | 3287 | |
0de8783a | 3288 | emc_insert(flow_cache, &keys[i], netdev_flow); |
623540e4 | 3289 | } |
623540e4 EJ |
3290 | } |
3291 | ||
3292 | ofpbuf_uninit(&actions); | |
3293 | ofpbuf_uninit(&put_actions); | |
3294 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
60fc3b7b | 3295 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); |
ac8c2081 | 3296 | } else if (OVS_UNLIKELY(any_miss)) { |
ac8c2081 | 3297 | for (i = 0; i < cnt; i++) { |
0de8783a | 3298 | if (OVS_UNLIKELY(!rules[i])) { |
e14deea0 | 3299 | dp_packet_delete(packets[i]); |
8aaa125d DDP |
3300 | lost_cnt++; |
3301 | miss_cnt++; | |
ac8c2081 DDP |
3302 | } |
3303 | } | |
623540e4 | 3304 | } |
84d6d5eb | 3305 | |
8cbf4f47 | 3306 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3307 | struct dp_packet *packet = packets[i]; |
84d6d5eb | 3308 | struct dp_netdev_flow *flow; |
8cbf4f47 | 3309 | |
0de8783a | 3310 | if (OVS_UNLIKELY(!rules[i])) { |
84d6d5eb EJ |
3311 | continue; |
3312 | } | |
3313 | ||
84d6d5eb | 3314 | flow = dp_netdev_flow_cast(rules[i]); |
0de8783a | 3315 | |
0de8783a | 3316 | emc_insert(flow_cache, &keys[i], flow); |
8aaa125d | 3317 | dp_netdev_queue_batches(packet, flow, &keys[i].mf, batches, n_batches); |
8cbf4f47 DDP |
3318 | } |
3319 | ||
8aaa125d DDP |
3320 | dp_netdev_count_packet(pmd, DP_STAT_MASKED_HIT, cnt - miss_cnt); |
3321 | dp_netdev_count_packet(pmd, DP_STAT_MISS, miss_cnt); | |
3322 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); | |
72865317 BP |
3323 | } |
3324 | ||
adcf00ba | 3325 | static void |
65f13b50 | 3326 | dp_netdev_input(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3327 | struct dp_packet **packets, int cnt) |
9bbf1c3d | 3328 | { |
1a0d5831 | 3329 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3330 | const size_t PKT_ARRAY_SIZE = cnt; |
3331 | #else | |
1a0d5831 | 3332 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3333 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d DDP |
3334 | #endif |
3335 | struct netdev_flow_key keys[PKT_ARRAY_SIZE]; | |
8aaa125d | 3336 | struct packet_batch batches[PKT_ARRAY_SIZE]; |
11bfdadd | 3337 | long long now = time_msec(); |
8aaa125d | 3338 | size_t newcnt, n_batches, i; |
9bbf1c3d | 3339 | |
8aaa125d DDP |
3340 | n_batches = 0; |
3341 | newcnt = emc_processing(pmd, packets, cnt, keys, batches, &n_batches); | |
9bbf1c3d | 3342 | if (OVS_UNLIKELY(newcnt)) { |
8aaa125d DDP |
3343 | fast_path_processing(pmd, packets, newcnt, keys, batches, &n_batches); |
3344 | } | |
3345 | ||
603f2ce0 EJ |
3346 | for (i = 0; i < n_batches; i++) { |
3347 | batches[i].flow->batch = NULL; | |
3348 | } | |
3349 | ||
8aaa125d DDP |
3350 | for (i = 0; i < n_batches; i++) { |
3351 | packet_batch_execute(&batches[i], pmd, now); | |
9bbf1c3d DDP |
3352 | } |
3353 | } | |
3354 | ||
9080a111 | 3355 | struct dp_netdev_execute_aux { |
65f13b50 | 3356 | struct dp_netdev_pmd_thread *pmd; |
9080a111 JR |
3357 | }; |
3358 | ||
6b31e073 | 3359 | static void |
623540e4 EJ |
3360 | dpif_netdev_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, |
3361 | void *aux) | |
6b31e073 RW |
3362 | { |
3363 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
623540e4 | 3364 | dp->upcall_aux = aux; |
6b31e073 RW |
3365 | dp->upcall_cb = cb; |
3366 | } | |
3367 | ||
ac8c2081 | 3368 | static void |
3bcc10c0 | 3369 | dp_netdev_drop_packets(struct dp_packet **packets, int cnt, bool may_steal) |
ac8c2081 | 3370 | { |
ac8c2081 | 3371 | if (may_steal) { |
a36de779 PS |
3372 | int i; |
3373 | ||
ac8c2081 | 3374 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3375 | dp_packet_delete(packets[i]); |
ac8c2081 DDP |
3376 | } |
3377 | } | |
3378 | } | |
3379 | ||
a36de779 PS |
3380 | static int |
3381 | push_tnl_action(const struct dp_netdev *dp, | |
3382 | const struct nlattr *attr, | |
e14deea0 | 3383 | struct dp_packet **packets, int cnt) |
a36de779 PS |
3384 | { |
3385 | struct dp_netdev_port *tun_port; | |
3386 | const struct ovs_action_push_tnl *data; | |
3387 | ||
3388 | data = nl_attr_get(attr); | |
3389 | ||
3390 | tun_port = dp_netdev_lookup_port(dp, u32_to_odp(data->tnl_port)); | |
3391 | if (!tun_port) { | |
3392 | return -EINVAL; | |
3393 | } | |
3394 | netdev_push_header(tun_port->netdev, packets, cnt, data); | |
3395 | ||
3396 | return 0; | |
3397 | } | |
3398 | ||
3399 | static void | |
28e2fa02 DDP |
3400 | dp_netdev_clone_pkt_batch(struct dp_packet **dst_pkts, |
3401 | struct dp_packet **src_pkts, int cnt) | |
a36de779 PS |
3402 | { |
3403 | int i; | |
3404 | ||
3405 | for (i = 0; i < cnt; i++) { | |
28e2fa02 | 3406 | dst_pkts[i] = dp_packet_clone(src_pkts[i]); |
a36de779 PS |
3407 | } |
3408 | } | |
3409 | ||
9080a111 | 3410 | static void |
e14deea0 | 3411 | dp_execute_cb(void *aux_, struct dp_packet **packets, int cnt, |
09f9da0b | 3412 | const struct nlattr *a, bool may_steal) |
8a4e3a85 | 3413 | OVS_NO_THREAD_SAFETY_ANALYSIS |
9080a111 JR |
3414 | { |
3415 | struct dp_netdev_execute_aux *aux = aux_; | |
623540e4 | 3416 | uint32_t *depth = recirc_depth_get(); |
28e2fa02 DDP |
3417 | struct dp_netdev_pmd_thread *pmd = aux->pmd; |
3418 | struct dp_netdev *dp = pmd->dp; | |
09f9da0b | 3419 | int type = nl_attr_type(a); |
8a4e3a85 | 3420 | struct dp_netdev_port *p; |
8cbf4f47 | 3421 | int i; |
9080a111 | 3422 | |
09f9da0b JR |
3423 | switch ((enum ovs_action_attr)type) { |
3424 | case OVS_ACTION_ATTR_OUTPUT: | |
623540e4 | 3425 | p = dp_netdev_lookup_port(dp, u32_to_odp(nl_attr_get_u32(a))); |
26a5075b | 3426 | if (OVS_LIKELY(p)) { |
3bcc10c0 | 3427 | netdev_send(p->netdev, pmd->tx_qid, packets, cnt, may_steal); |
ac8c2081 | 3428 | return; |
8a4e3a85 | 3429 | } |
09f9da0b JR |
3430 | break; |
3431 | ||
a36de779 PS |
3432 | case OVS_ACTION_ATTR_TUNNEL_PUSH: |
3433 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3434 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3435 | int err; |
3436 | ||
3437 | if (!may_steal) { | |
3438 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3439 | packets = tnl_pkt; | |
3440 | } | |
3441 | ||
3442 | err = push_tnl_action(dp, a, packets, cnt); | |
3443 | if (!err) { | |
3444 | (*depth)++; | |
3445 | dp_netdev_input(pmd, packets, cnt); | |
3446 | (*depth)--; | |
3447 | } else { | |
3448 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3449 | } | |
3450 | return; | |
3451 | } | |
3452 | break; | |
3453 | ||
3454 | case OVS_ACTION_ATTR_TUNNEL_POP: | |
3455 | if (*depth < MAX_RECIRC_DEPTH) { | |
3456 | odp_port_t portno = u32_to_odp(nl_attr_get_u32(a)); | |
3457 | ||
3458 | p = dp_netdev_lookup_port(dp, portno); | |
3459 | if (p) { | |
cd159f1a | 3460 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3461 | int err; |
3462 | ||
3463 | if (!may_steal) { | |
3464 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3465 | packets = tnl_pkt; | |
3466 | } | |
3467 | ||
3468 | err = netdev_pop_header(p->netdev, packets, cnt); | |
3469 | if (!err) { | |
3470 | ||
3471 | for (i = 0; i < cnt; i++) { | |
3472 | packets[i]->md.in_port.odp_port = portno; | |
3473 | } | |
3474 | ||
3475 | (*depth)++; | |
3476 | dp_netdev_input(pmd, packets, cnt); | |
3477 | (*depth)--; | |
3478 | } else { | |
3479 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3480 | } | |
3481 | return; | |
3482 | } | |
3483 | } | |
3484 | break; | |
3485 | ||
623540e4 EJ |
3486 | case OVS_ACTION_ATTR_USERSPACE: |
3487 | if (!fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { | |
3488 | const struct nlattr *userdata; | |
3489 | struct ofpbuf actions; | |
3490 | struct flow flow; | |
7af12bd7 | 3491 | ovs_u128 ufid; |
4fc65926 | 3492 | |
623540e4 EJ |
3493 | userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); |
3494 | ofpbuf_init(&actions, 0); | |
8cbf4f47 | 3495 | |
623540e4 EJ |
3496 | for (i = 0; i < cnt; i++) { |
3497 | int error; | |
3498 | ||
3499 | ofpbuf_clear(&actions); | |
3500 | ||
cf62fa4c | 3501 | flow_extract(packets[i], &flow); |
7af12bd7 | 3502 | dpif_flow_hash(dp->dpif, &flow, sizeof flow, &ufid); |
1c1e46ed | 3503 | error = dp_netdev_upcall(pmd, packets[i], &flow, NULL, &ufid, |
7af12bd7 | 3504 | DPIF_UC_ACTION, userdata,&actions, |
623540e4 EJ |
3505 | NULL); |
3506 | if (!error || error == ENOSPC) { | |
ac8c2081 | 3507 | dp_netdev_execute_actions(pmd, &packets[i], 1, may_steal, |
6fd6ed71 | 3508 | actions.data, actions.size); |
ac8c2081 | 3509 | } else if (may_steal) { |
e14deea0 | 3510 | dp_packet_delete(packets[i]); |
623540e4 | 3511 | } |
db73f716 | 3512 | } |
623540e4 EJ |
3513 | ofpbuf_uninit(&actions); |
3514 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
6b31e073 | 3515 | |
ac8c2081 DDP |
3516 | return; |
3517 | } | |
09f9da0b | 3518 | break; |
572f732a | 3519 | |
adcf00ba AZ |
3520 | case OVS_ACTION_ATTR_RECIRC: |
3521 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3522 | struct dp_packet *recirc_pkts[NETDEV_MAX_BURST]; |
572f732a | 3523 | |
28e2fa02 DDP |
3524 | if (!may_steal) { |
3525 | dp_netdev_clone_pkt_batch(recirc_pkts, packets, cnt); | |
3526 | packets = recirc_pkts; | |
3527 | } | |
8cbf4f47 | 3528 | |
28e2fa02 DDP |
3529 | for (i = 0; i < cnt; i++) { |
3530 | packets[i]->md.recirc_id = nl_attr_get_u32(a); | |
8cbf4f47 | 3531 | } |
28e2fa02 DDP |
3532 | |
3533 | (*depth)++; | |
3534 | dp_netdev_input(pmd, packets, cnt); | |
adcf00ba AZ |
3535 | (*depth)--; |
3536 | ||
ac8c2081 | 3537 | return; |
adcf00ba | 3538 | } |
ac8c2081 DDP |
3539 | |
3540 | VLOG_WARN("Packet dropped. Max recirculation depth exceeded."); | |
572f732a | 3541 | break; |
572f732a | 3542 | |
09f9da0b JR |
3543 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3544 | case OVS_ACTION_ATTR_POP_VLAN: | |
3545 | case OVS_ACTION_ATTR_PUSH_MPLS: | |
3546 | case OVS_ACTION_ATTR_POP_MPLS: | |
3547 | case OVS_ACTION_ATTR_SET: | |
6d670e7f | 3548 | case OVS_ACTION_ATTR_SET_MASKED: |
09f9da0b | 3549 | case OVS_ACTION_ATTR_SAMPLE: |
53e1d6f1 | 3550 | case OVS_ACTION_ATTR_HASH: |
09f9da0b JR |
3551 | case OVS_ACTION_ATTR_UNSPEC: |
3552 | case __OVS_ACTION_ATTR_MAX: | |
3553 | OVS_NOT_REACHED(); | |
da546e07 | 3554 | } |
ac8c2081 DDP |
3555 | |
3556 | dp_netdev_drop_packets(packets, cnt, may_steal); | |
98403001 BP |
3557 | } |
3558 | ||
4edb9ae9 | 3559 | static void |
65f13b50 | 3560 | dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3561 | struct dp_packet **packets, int cnt, |
41ccaa24 | 3562 | bool may_steal, |
9080a111 | 3563 | const struct nlattr *actions, size_t actions_len) |
72865317 | 3564 | { |
41ccaa24 | 3565 | struct dp_netdev_execute_aux aux = { pmd }; |
9080a111 | 3566 | |
41ccaa24 | 3567 | odp_execute_actions(&aux, packets, cnt, may_steal, actions, |
8cbf4f47 | 3568 | actions_len, dp_execute_cb); |
72865317 BP |
3569 | } |
3570 | ||
3571 | const struct dpif_class dpif_netdev_class = { | |
72865317 | 3572 | "netdev", |
6553d06b | 3573 | dpif_netdev_init, |
2197d7ab | 3574 | dpif_netdev_enumerate, |
0aeaabc8 | 3575 | dpif_netdev_port_open_type, |
72865317 BP |
3576 | dpif_netdev_open, |
3577 | dpif_netdev_close, | |
7dab847a | 3578 | dpif_netdev_destroy, |
e4cfed38 PS |
3579 | dpif_netdev_run, |
3580 | dpif_netdev_wait, | |
72865317 | 3581 | dpif_netdev_get_stats, |
72865317 BP |
3582 | dpif_netdev_port_add, |
3583 | dpif_netdev_port_del, | |
3584 | dpif_netdev_port_query_by_number, | |
3585 | dpif_netdev_port_query_by_name, | |
98403001 | 3586 | NULL, /* port_get_pid */ |
b0ec0f27 BP |
3587 | dpif_netdev_port_dump_start, |
3588 | dpif_netdev_port_dump_next, | |
3589 | dpif_netdev_port_dump_done, | |
72865317 BP |
3590 | dpif_netdev_port_poll, |
3591 | dpif_netdev_port_poll_wait, | |
72865317 | 3592 | dpif_netdev_flow_flush, |
ac64794a BP |
3593 | dpif_netdev_flow_dump_create, |
3594 | dpif_netdev_flow_dump_destroy, | |
3595 | dpif_netdev_flow_dump_thread_create, | |
3596 | dpif_netdev_flow_dump_thread_destroy, | |
704a1e09 | 3597 | dpif_netdev_flow_dump_next, |
1a0c894a | 3598 | dpif_netdev_operate, |
6b31e073 RW |
3599 | NULL, /* recv_set */ |
3600 | NULL, /* handlers_set */ | |
f2eee189 | 3601 | dpif_netdev_pmd_set, |
5bf93d67 | 3602 | dpif_netdev_queue_to_priority, |
6b31e073 RW |
3603 | NULL, /* recv */ |
3604 | NULL, /* recv_wait */ | |
3605 | NULL, /* recv_purge */ | |
3606 | dpif_netdev_register_upcall_cb, | |
3607 | dpif_netdev_enable_upcall, | |
3608 | dpif_netdev_disable_upcall, | |
b5cbbcf6 | 3609 | dpif_netdev_get_datapath_version, |
72865317 | 3610 | }; |
614c4892 | 3611 | |
74cc3969 BP |
3612 | static void |
3613 | dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3614 | const char *argv[], void *aux OVS_UNUSED) | |
3615 | { | |
59e6d833 BP |
3616 | struct dp_netdev_port *old_port; |
3617 | struct dp_netdev_port *new_port; | |
74cc3969 | 3618 | struct dp_netdev *dp; |
ff073a71 | 3619 | odp_port_t port_no; |
74cc3969 | 3620 | |
8a4e3a85 | 3621 | ovs_mutex_lock(&dp_netdev_mutex); |
74cc3969 BP |
3622 | dp = shash_find_data(&dp_netdevs, argv[1]); |
3623 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
8a4e3a85 | 3624 | ovs_mutex_unlock(&dp_netdev_mutex); |
74cc3969 BP |
3625 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); |
3626 | return; | |
3627 | } | |
8a4e3a85 BP |
3628 | ovs_refcount_ref(&dp->ref_cnt); |
3629 | ovs_mutex_unlock(&dp_netdev_mutex); | |
74cc3969 | 3630 | |
59e6d833 BP |
3631 | ovs_mutex_lock(&dp->port_mutex); |
3632 | if (get_port_by_name(dp, argv[2], &old_port)) { | |
74cc3969 | 3633 | unixctl_command_reply_error(conn, "unknown port"); |
8a4e3a85 | 3634 | goto exit; |
74cc3969 BP |
3635 | } |
3636 | ||
ff073a71 BP |
3637 | port_no = u32_to_odp(atoi(argv[3])); |
3638 | if (!port_no || port_no == ODPP_NONE) { | |
74cc3969 | 3639 | unixctl_command_reply_error(conn, "bad port number"); |
8a4e3a85 | 3640 | goto exit; |
74cc3969 | 3641 | } |
ff073a71 | 3642 | if (dp_netdev_lookup_port(dp, port_no)) { |
74cc3969 | 3643 | unixctl_command_reply_error(conn, "port number already in use"); |
8a4e3a85 | 3644 | goto exit; |
74cc3969 | 3645 | } |
59e6d833 BP |
3646 | |
3647 | /* Remove old port. */ | |
efa2bcbb | 3648 | cmap_remove(&dp->ports, &old_port->node, hash_port_no(old_port->md.in_port.odp_port)); |
59e6d833 BP |
3649 | ovsrcu_postpone(free, old_port); |
3650 | ||
3651 | /* Insert new port (cmap semantics mean we cannot re-insert 'old_port'). */ | |
3652 | new_port = xmemdup(old_port, sizeof *old_port); | |
efa2bcbb | 3653 | new_port->md.in_port.odp_port = port_no; |
59e6d833 BP |
3654 | cmap_insert(&dp->ports, &new_port->node, hash_port_no(port_no)); |
3655 | ||
d33ed218 | 3656 | seq_change(dp->port_seq); |
74cc3969 | 3657 | unixctl_command_reply(conn, NULL); |
8a4e3a85 BP |
3658 | |
3659 | exit: | |
59e6d833 | 3660 | ovs_mutex_unlock(&dp->port_mutex); |
8a4e3a85 | 3661 | dp_netdev_unref(dp); |
74cc3969 BP |
3662 | } |
3663 | ||
c40b890f BP |
3664 | static void |
3665 | dpif_dummy_delete_port(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3666 | const char *argv[], void *aux OVS_UNUSED) | |
3667 | { | |
3668 | struct dp_netdev_port *port; | |
3669 | struct dp_netdev *dp; | |
3670 | ||
3671 | ovs_mutex_lock(&dp_netdev_mutex); | |
3672 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
3673 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
3674 | ovs_mutex_unlock(&dp_netdev_mutex); | |
3675 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); | |
3676 | return; | |
3677 | } | |
3678 | ovs_refcount_ref(&dp->ref_cnt); | |
3679 | ovs_mutex_unlock(&dp_netdev_mutex); | |
3680 | ||
3681 | ovs_mutex_lock(&dp->port_mutex); | |
3682 | if (get_port_by_name(dp, argv[2], &port)) { | |
3683 | unixctl_command_reply_error(conn, "unknown port"); | |
efa2bcbb | 3684 | } else if (port->md.in_port.odp_port == ODPP_LOCAL) { |
c40b890f BP |
3685 | unixctl_command_reply_error(conn, "can't delete local port"); |
3686 | } else { | |
3687 | do_del_port(dp, port); | |
3688 | unixctl_command_reply(conn, NULL); | |
3689 | } | |
3690 | ovs_mutex_unlock(&dp->port_mutex); | |
3691 | ||
3692 | dp_netdev_unref(dp); | |
3693 | } | |
3694 | ||
0cbfe35d BP |
3695 | static void |
3696 | dpif_dummy_register__(const char *type) | |
3697 | { | |
3698 | struct dpif_class *class; | |
3699 | ||
3700 | class = xmalloc(sizeof *class); | |
3701 | *class = dpif_netdev_class; | |
3702 | class->type = xstrdup(type); | |
3703 | dp_register_provider(class); | |
3704 | } | |
3705 | ||
8420c7ad BP |
3706 | static void |
3707 | dpif_dummy_override(const char *type) | |
3708 | { | |
3709 | if (!dp_unregister_provider(type)) { | |
3710 | dpif_dummy_register__(type); | |
3711 | } | |
3712 | } | |
3713 | ||
614c4892 | 3714 | void |
8420c7ad | 3715 | dpif_dummy_register(enum dummy_level level) |
614c4892 | 3716 | { |
8420c7ad | 3717 | if (level == DUMMY_OVERRIDE_ALL) { |
0cbfe35d BP |
3718 | struct sset types; |
3719 | const char *type; | |
3720 | ||
3721 | sset_init(&types); | |
3722 | dp_enumerate_types(&types); | |
3723 | SSET_FOR_EACH (type, &types) { | |
8420c7ad | 3724 | dpif_dummy_override(type); |
0cbfe35d BP |
3725 | } |
3726 | sset_destroy(&types); | |
8420c7ad BP |
3727 | } else if (level == DUMMY_OVERRIDE_SYSTEM) { |
3728 | dpif_dummy_override("system"); | |
614c4892 | 3729 | } |
0cbfe35d BP |
3730 | |
3731 | dpif_dummy_register__("dummy"); | |
74cc3969 BP |
3732 | |
3733 | unixctl_command_register("dpif-dummy/change-port-number", | |
74467d5c | 3734 | "dp port new-number", |
74cc3969 | 3735 | 3, 3, dpif_dummy_change_port_number, NULL); |
74467d5c | 3736 | unixctl_command_register("dpif-dummy/delete-port", "dp port", |
c40b890f | 3737 | 2, 2, dpif_dummy_delete_port, NULL); |
614c4892 | 3738 | } |
0de8783a JR |
3739 | \f |
3740 | /* Datapath Classifier. */ | |
3741 | ||
3742 | /* A set of rules that all have the same fields wildcarded. */ | |
3743 | struct dpcls_subtable { | |
3744 | /* The fields are only used by writers. */ | |
3745 | struct cmap_node cmap_node OVS_GUARDED; /* Within dpcls 'subtables_map'. */ | |
3746 | ||
3747 | /* These fields are accessed by readers. */ | |
3748 | struct cmap rules; /* Contains "struct dpcls_rule"s. */ | |
3749 | struct netdev_flow_key mask; /* Wildcards for fields (const). */ | |
3750 | /* 'mask' must be the last field, additional space is allocated here. */ | |
3751 | }; | |
3752 | ||
3753 | /* Initializes 'cls' as a classifier that initially contains no classification | |
3754 | * rules. */ | |
3755 | static void | |
3756 | dpcls_init(struct dpcls *cls) | |
3757 | { | |
3758 | cmap_init(&cls->subtables_map); | |
3759 | pvector_init(&cls->subtables); | |
3760 | } | |
3761 | ||
3762 | static void | |
3763 | dpcls_destroy_subtable(struct dpcls *cls, struct dpcls_subtable *subtable) | |
3764 | { | |
3765 | pvector_remove(&cls->subtables, subtable); | |
3766 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, | |
3767 | subtable->mask.hash); | |
3768 | cmap_destroy(&subtable->rules); | |
3769 | ovsrcu_postpone(free, subtable); | |
3770 | } | |
3771 | ||
3772 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
3773 | * caller's responsibility. | |
3774 | * May only be called after all the readers have been terminated. */ | |
3775 | static void | |
3776 | dpcls_destroy(struct dpcls *cls) | |
3777 | { | |
3778 | if (cls) { | |
3779 | struct dpcls_subtable *subtable; | |
3780 | ||
3781 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
3782 | dpcls_destroy_subtable(cls, subtable); | |
3783 | } | |
3784 | cmap_destroy(&cls->subtables_map); | |
3785 | pvector_destroy(&cls->subtables); | |
3786 | } | |
3787 | } | |
3788 | ||
3789 | static struct dpcls_subtable * | |
3790 | dpcls_create_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
3791 | { | |
3792 | struct dpcls_subtable *subtable; | |
3793 | ||
3794 | /* Need to add one. */ | |
caeb4906 JR |
3795 | subtable = xmalloc(sizeof *subtable |
3796 | - sizeof subtable->mask.mf + mask->len); | |
0de8783a JR |
3797 | cmap_init(&subtable->rules); |
3798 | netdev_flow_key_clone(&subtable->mask, mask); | |
3799 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, mask->hash); | |
3800 | pvector_insert(&cls->subtables, subtable, 0); | |
802f84ff | 3801 | pvector_publish(&cls->subtables); |
0de8783a JR |
3802 | |
3803 | return subtable; | |
3804 | } | |
3805 | ||
3806 | static inline struct dpcls_subtable * | |
3807 | dpcls_find_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
3808 | { | |
3809 | struct dpcls_subtable *subtable; | |
3810 | ||
3811 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, mask->hash, | |
3812 | &cls->subtables_map) { | |
3813 | if (netdev_flow_key_equal(&subtable->mask, mask)) { | |
3814 | return subtable; | |
3815 | } | |
3816 | } | |
3817 | return dpcls_create_subtable(cls, mask); | |
3818 | } | |
3819 | ||
3820 | /* Insert 'rule' into 'cls'. */ | |
3821 | static void | |
3822 | dpcls_insert(struct dpcls *cls, struct dpcls_rule *rule, | |
3823 | const struct netdev_flow_key *mask) | |
3824 | { | |
3825 | struct dpcls_subtable *subtable = dpcls_find_subtable(cls, mask); | |
3826 | ||
3827 | rule->mask = &subtable->mask; | |
3828 | cmap_insert(&subtable->rules, &rule->cmap_node, rule->flow.hash); | |
3829 | } | |
3830 | ||
3831 | /* Removes 'rule' from 'cls', also destructing the 'rule'. */ | |
3832 | static void | |
3833 | dpcls_remove(struct dpcls *cls, struct dpcls_rule *rule) | |
3834 | { | |
3835 | struct dpcls_subtable *subtable; | |
3836 | ||
3837 | ovs_assert(rule->mask); | |
3838 | ||
3839 | INIT_CONTAINER(subtable, rule->mask, mask); | |
3840 | ||
3841 | if (cmap_remove(&subtable->rules, &rule->cmap_node, rule->flow.hash) | |
3842 | == 0) { | |
3843 | dpcls_destroy_subtable(cls, subtable); | |
802f84ff | 3844 | pvector_publish(&cls->subtables); |
0de8783a JR |
3845 | } |
3846 | } | |
3847 | ||
3848 | /* Returns true if 'target' satisifies 'key' in 'mask', that is, if each 1-bit | |
3849 | * in 'mask' the values in 'key' and 'target' are the same. | |
3850 | * | |
3851 | * Note: 'key' and 'mask' have the same mask, and 'key' is already masked. */ | |
3852 | static inline bool | |
3853 | dpcls_rule_matches_key(const struct dpcls_rule *rule, | |
3854 | const struct netdev_flow_key *target) | |
3855 | { | |
d70e8c28 JR |
3856 | const uint64_t *keyp = rule->flow.mf.inline_values; |
3857 | const uint64_t *maskp = rule->mask->mf.inline_values; | |
3858 | uint64_t target_u64; | |
0de8783a | 3859 | |
d70e8c28 JR |
3860 | NETDEV_FLOW_KEY_FOR_EACH_IN_MAP(target_u64, target, rule->flow.mf.map) { |
3861 | if (OVS_UNLIKELY((target_u64 & *maskp++) != *keyp++)) { | |
0de8783a JR |
3862 | return false; |
3863 | } | |
3864 | } | |
3865 | return true; | |
3866 | } | |
3867 | ||
3868 | /* For each miniflow in 'flows' performs a classifier lookup writing the result | |
3869 | * into the corresponding slot in 'rules'. If a particular entry in 'flows' is | |
3870 | * NULL it is skipped. | |
3871 | * | |
3872 | * This function is optimized for use in the userspace datapath and therefore | |
3873 | * does not implement a lot of features available in the standard | |
3874 | * classifier_lookup() function. Specifically, it does not implement | |
3875 | * priorities, instead returning any rule which matches the flow. | |
3876 | * | |
3877 | * Returns true if all flows found a corresponding rule. */ | |
3878 | static bool | |
3879 | dpcls_lookup(const struct dpcls *cls, const struct netdev_flow_key keys[], | |
3880 | struct dpcls_rule **rules, const size_t cnt) | |
3881 | { | |
3882 | /* The batch size 16 was experimentally found faster than 8 or 32. */ | |
3883 | typedef uint16_t map_type; | |
3884 | #define MAP_BITS (sizeof(map_type) * CHAR_BIT) | |
3885 | ||
3886 | #if !defined(__CHECKER__) && !defined(_WIN32) | |
3887 | const int N_MAPS = DIV_ROUND_UP(cnt, MAP_BITS); | |
3888 | #else | |
cd159f1a | 3889 | enum { N_MAPS = DIV_ROUND_UP(NETDEV_MAX_BURST, MAP_BITS) }; |
0de8783a JR |
3890 | #endif |
3891 | map_type maps[N_MAPS]; | |
3892 | struct dpcls_subtable *subtable; | |
3893 | ||
3894 | memset(maps, 0xff, sizeof maps); | |
3895 | if (cnt % MAP_BITS) { | |
3896 | maps[N_MAPS - 1] >>= MAP_BITS - cnt % MAP_BITS; /* Clear extra bits. */ | |
3897 | } | |
3898 | memset(rules, 0, cnt * sizeof *rules); | |
3899 | ||
3900 | PVECTOR_FOR_EACH (subtable, &cls->subtables) { | |
3901 | const struct netdev_flow_key *mkeys = keys; | |
3902 | struct dpcls_rule **mrules = rules; | |
3903 | map_type remains = 0; | |
3904 | int m; | |
3905 | ||
3906 | BUILD_ASSERT_DECL(sizeof remains == sizeof *maps); | |
3907 | ||
3908 | for (m = 0; m < N_MAPS; m++, mkeys += MAP_BITS, mrules += MAP_BITS) { | |
3909 | uint32_t hashes[MAP_BITS]; | |
3910 | const struct cmap_node *nodes[MAP_BITS]; | |
3911 | unsigned long map = maps[m]; | |
3912 | int i; | |
3913 | ||
3914 | if (!map) { | |
3915 | continue; /* Skip empty maps. */ | |
3916 | } | |
3917 | ||
3918 | /* Compute hashes for the remaining keys. */ | |
3919 | ULONG_FOR_EACH_1(i, map) { | |
3920 | hashes[i] = netdev_flow_key_hash_in_mask(&mkeys[i], | |
3921 | &subtable->mask); | |
3922 | } | |
3923 | /* Lookup. */ | |
3924 | map = cmap_find_batch(&subtable->rules, map, hashes, nodes); | |
3925 | /* Check results. */ | |
3926 | ULONG_FOR_EACH_1(i, map) { | |
3927 | struct dpcls_rule *rule; | |
3928 | ||
3929 | CMAP_NODE_FOR_EACH (rule, cmap_node, nodes[i]) { | |
3930 | if (OVS_LIKELY(dpcls_rule_matches_key(rule, &mkeys[i]))) { | |
3931 | mrules[i] = rule; | |
3932 | goto next; | |
3933 | } | |
3934 | } | |
3935 | ULONG_SET0(map, i); /* Did not match. */ | |
3936 | next: | |
3937 | ; /* Keep Sparse happy. */ | |
3938 | } | |
3939 | maps[m] &= ~map; /* Clear the found rules. */ | |
3940 | remains |= maps[m]; | |
3941 | } | |
3942 | if (!remains) { | |
3943 | return true; /* All found. */ | |
3944 | } | |
3945 | } | |
3946 | return false; /* Some misses. */ | |
3947 | } |