]>
Commit | Line | Data |
---|---|---|
72865317 | 1 | /* |
d262ac2c | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2016 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 | ||
9f861c91 | 34 | #include "bitmap.h" |
59e6d833 | 35 | #include "cmap.h" |
72865317 | 36 | #include "csum.h" |
e14deea0 | 37 | #include "dp-packet.h" |
614c4892 | 38 | #include "dpif.h" |
72865317 | 39 | #include "dpif-provider.h" |
614c4892 | 40 | #include "dummy.h" |
36956a7d | 41 | #include "dynamic-string.h" |
afae68b1 | 42 | #include "fat-rwlock.h" |
72865317 | 43 | #include "flow.h" |
9f361d6b | 44 | #include "cmap.h" |
fbe0962b | 45 | #include "coverage.h" |
762d146a | 46 | #include "hmapx.h" |
6c3eee82 | 47 | #include "latch.h" |
72865317 | 48 | #include "list.h" |
0de8783a | 49 | #include "match.h" |
72865317 | 50 | #include "netdev.h" |
8617afff | 51 | #include "netdev-dpdk.h" |
de281153 | 52 | #include "netdev-vport.h" |
cdee00fd | 53 | #include "netlink.h" |
f094af7b | 54 | #include "odp-execute.h" |
72865317 BP |
55 | #include "odp-util.h" |
56 | #include "ofp-print.h" | |
57 | #include "ofpbuf.h" | |
5a034064 | 58 | #include "ovs-numa.h" |
61e7deb1 | 59 | #include "ovs-rcu.h" |
72865317 BP |
60 | #include "packets.h" |
61 | #include "poll-loop.h" | |
0de8783a | 62 | #include "pvector.h" |
26c6b6cd | 63 | #include "random.h" |
d33ed218 | 64 | #include "seq.h" |
462278db | 65 | #include "shash.h" |
0cbfe35d | 66 | #include "sset.h" |
72865317 | 67 | #include "timeval.h" |
53902038 | 68 | #include "tnl-neigh-cache.h" |
7f9b8504 | 69 | #include "tnl-ports.h" |
74cc3969 | 70 | #include "unixctl.h" |
72865317 | 71 | #include "util.h" |
e6211adc | 72 | #include "openvswitch/vlog.h" |
5136ce49 | 73 | |
d98e6007 | 74 | VLOG_DEFINE_THIS_MODULE(dpif_netdev); |
72865317 | 75 | |
8bb113da | 76 | #define FLOW_DUMP_MAX_BATCH 50 |
adcf00ba AZ |
77 | /* Use per thread recirc_depth to prevent recirculation loop. */ |
78 | #define MAX_RECIRC_DEPTH 5 | |
79 | DEFINE_STATIC_PER_THREAD_DATA(uint32_t, recirc_depth, 0) | |
e4cfed38 | 80 | |
72865317 | 81 | /* Configuration parameters. */ |
72865317 BP |
82 | enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ |
83 | ||
8a4e3a85 BP |
84 | /* Protects against changes to 'dp_netdevs'. */ |
85 | static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER; | |
86 | ||
87 | /* Contains all 'struct dp_netdev's. */ | |
88 | static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex) | |
89 | = SHASH_INITIALIZER(&dp_netdevs); | |
90 | ||
623540e4 | 91 | static struct vlog_rate_limit upcall_rl = VLOG_RATE_LIMIT_INIT(600, 600); |
6b31e073 | 92 | |
2494ccd7 JS |
93 | static struct odp_support dp_netdev_support = { |
94 | .max_mpls_depth = SIZE_MAX, | |
95 | .recirc = true, | |
96 | }; | |
97 | ||
79df317f | 98 | /* Stores a miniflow with inline values */ |
9bbf1c3d | 99 | |
9bbf1c3d | 100 | struct netdev_flow_key { |
caeb4906 JR |
101 | uint32_t hash; /* Hash function differs for different users. */ |
102 | uint32_t len; /* Length of the following miniflow (incl. map). */ | |
0de8783a | 103 | struct miniflow mf; |
8fd47924 | 104 | uint64_t buf[FLOW_MAX_PACKET_U64S]; |
9bbf1c3d DDP |
105 | }; |
106 | ||
107 | /* Exact match cache for frequently used flows | |
108 | * | |
109 | * The cache uses a 32-bit hash of the packet (which can be the RSS hash) to | |
110 | * search its entries for a miniflow that matches exactly the miniflow of the | |
0de8783a | 111 | * packet. It stores the 'dpcls_rule' (rule) that matches the miniflow. |
9bbf1c3d DDP |
112 | * |
113 | * A cache entry holds a reference to its 'dp_netdev_flow'. | |
114 | * | |
115 | * A miniflow with a given hash can be in one of EM_FLOW_HASH_SEGS different | |
116 | * entries. The 32-bit hash is split into EM_FLOW_HASH_SEGS values (each of | |
117 | * them is EM_FLOW_HASH_SHIFT bits wide and the remainder is thrown away). Each | |
118 | * value is the index of a cache entry where the miniflow could be. | |
119 | * | |
120 | * | |
121 | * Thread-safety | |
122 | * ============= | |
123 | * | |
124 | * Each pmd_thread has its own private exact match cache. | |
125 | * If dp_netdev_input is not called from a pmd thread, a mutex is used. | |
126 | */ | |
127 | ||
fc82e877 | 128 | #define EM_FLOW_HASH_SHIFT 13 |
9bbf1c3d DDP |
129 | #define EM_FLOW_HASH_ENTRIES (1u << EM_FLOW_HASH_SHIFT) |
130 | #define EM_FLOW_HASH_MASK (EM_FLOW_HASH_ENTRIES - 1) | |
131 | #define EM_FLOW_HASH_SEGS 2 | |
132 | ||
133 | struct emc_entry { | |
9bbf1c3d | 134 | struct dp_netdev_flow *flow; |
0de8783a | 135 | struct netdev_flow_key key; /* key.hash used for emc hash value. */ |
9bbf1c3d DDP |
136 | }; |
137 | ||
138 | struct emc_cache { | |
139 | struct emc_entry entries[EM_FLOW_HASH_ENTRIES]; | |
67ad54cb | 140 | int sweep_idx; /* For emc_cache_slow_sweep(). */ |
9bbf1c3d DDP |
141 | }; |
142 | ||
143 | /* Iterate in the exact match cache through every entry that might contain a | |
144 | * miniflow with hash 'HASH'. */ | |
145 | #define EMC_FOR_EACH_POS_WITH_HASH(EMC, CURRENT_ENTRY, HASH) \ | |
146 | for (uint32_t i__ = 0, srch_hash__ = (HASH); \ | |
147 | (CURRENT_ENTRY) = &(EMC)->entries[srch_hash__ & EM_FLOW_HASH_MASK], \ | |
148 | i__ < EM_FLOW_HASH_SEGS; \ | |
149 | i__++, srch_hash__ >>= EM_FLOW_HASH_SHIFT) | |
0de8783a JR |
150 | \f |
151 | /* Simple non-wildcarding single-priority classifier. */ | |
152 | ||
153 | struct dpcls { | |
154 | struct cmap subtables_map; | |
155 | struct pvector subtables; | |
156 | }; | |
9bbf1c3d | 157 | |
0de8783a JR |
158 | /* A rule to be inserted to the classifier. */ |
159 | struct dpcls_rule { | |
160 | struct cmap_node cmap_node; /* Within struct dpcls_subtable 'rules'. */ | |
161 | struct netdev_flow_key *mask; /* Subtable's mask. */ | |
162 | struct netdev_flow_key flow; /* Matching key. */ | |
163 | /* 'flow' must be the last field, additional space is allocated here. */ | |
164 | }; | |
165 | ||
166 | static void dpcls_init(struct dpcls *); | |
167 | static void dpcls_destroy(struct dpcls *); | |
168 | static void dpcls_insert(struct dpcls *, struct dpcls_rule *, | |
169 | const struct netdev_flow_key *mask); | |
170 | static void dpcls_remove(struct dpcls *, struct dpcls_rule *); | |
171 | static bool dpcls_lookup(const struct dpcls *cls, | |
172 | const struct netdev_flow_key keys[], | |
173 | struct dpcls_rule **rules, size_t cnt); | |
174 | \f | |
8a4e3a85 BP |
175 | /* Datapath based on the network device interface from netdev.h. |
176 | * | |
177 | * | |
178 | * Thread-safety | |
179 | * ============= | |
180 | * | |
181 | * Some members, marked 'const', are immutable. Accessing other members | |
182 | * requires synchronization, as noted in more detail below. | |
183 | * | |
184 | * Acquisition order is, from outermost to innermost: | |
185 | * | |
186 | * dp_netdev_mutex (global) | |
59e6d833 | 187 | * port_mutex |
8a4e3a85 | 188 | */ |
72865317 | 189 | struct dp_netdev { |
8a4e3a85 BP |
190 | const struct dpif_class *const class; |
191 | const char *const name; | |
6b31e073 | 192 | struct dpif *dpif; |
6a8267c5 BP |
193 | struct ovs_refcount ref_cnt; |
194 | atomic_flag destroyed; | |
72865317 | 195 | |
8a4e3a85 BP |
196 | /* Ports. |
197 | * | |
59e6d833 BP |
198 | * Protected by RCU. Take the mutex to add or remove ports. */ |
199 | struct ovs_mutex port_mutex; | |
200 | struct cmap ports; | |
d33ed218 | 201 | struct seq *port_seq; /* Incremented whenever a port changes. */ |
6c3eee82 | 202 | |
6b31e073 RW |
203 | /* Protects access to ofproto-dpif-upcall interface during revalidator |
204 | * thread synchronization. */ | |
205 | struct fat_rwlock upcall_rwlock; | |
623540e4 EJ |
206 | upcall_callback *upcall_cb; /* Callback function for executing upcalls. */ |
207 | void *upcall_aux; | |
6b31e073 | 208 | |
e4e74c3a AW |
209 | /* Callback function for notifying the purging of dp flows (during |
210 | * reseting pmd deletion). */ | |
211 | dp_purge_callback *dp_purge_cb; | |
212 | void *dp_purge_aux; | |
213 | ||
65f13b50 AW |
214 | /* Stores all 'struct dp_netdev_pmd_thread's. */ |
215 | struct cmap poll_threads; | |
216 | ||
217 | /* Protects the access of the 'struct dp_netdev_pmd_thread' | |
218 | * instance for non-pmd thread. */ | |
219 | struct ovs_mutex non_pmd_mutex; | |
220 | ||
221 | /* Each pmd thread will store its pointer to | |
222 | * 'struct dp_netdev_pmd_thread' in 'per_pmd_key'. */ | |
223 | ovsthread_key_t per_pmd_key; | |
f2eee189 | 224 | |
a14b8947 | 225 | /* Cpu mask for pin of pmd threads. */ |
f2eee189 | 226 | char *pmd_cmask; |
a36de779 | 227 | uint64_t last_tnl_conf_seq; |
72865317 BP |
228 | }; |
229 | ||
8a4e3a85 | 230 | static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp, |
59e6d833 | 231 | odp_port_t); |
ff073a71 | 232 | |
51852a57 | 233 | enum dp_stat_type { |
abcf3ef4 DDP |
234 | DP_STAT_EXACT_HIT, /* Packets that had an exact match (emc). */ |
235 | DP_STAT_MASKED_HIT, /* Packets that matched in the flow table. */ | |
51852a57 BP |
236 | DP_STAT_MISS, /* Packets that did not match. */ |
237 | DP_STAT_LOST, /* Packets not passed up to the client. */ | |
238 | DP_N_STATS | |
239 | }; | |
240 | ||
55e3ca97 DDP |
241 | enum pmd_cycles_counter_type { |
242 | PMD_CYCLES_POLLING, /* Cycles spent polling NICs. */ | |
243 | PMD_CYCLES_PROCESSING, /* Cycles spent processing packets */ | |
244 | PMD_N_CYCLES | |
245 | }; | |
246 | ||
72865317 BP |
247 | /* A port in a netdev-based datapath. */ |
248 | struct dp_netdev_port { | |
35303d71 | 249 | odp_port_t port_no; |
72865317 | 250 | struct netdev *netdev; |
efa2bcbb | 251 | struct cmap_node node; /* Node in dp_netdev's 'ports'. */ |
4b609110 | 252 | struct netdev_saved_flags *sf; |
55c955bd | 253 | struct netdev_rxq **rxq; |
b284085e | 254 | struct ovs_refcount ref_cnt; |
0cbfe35d | 255 | char *type; /* Port type as requested by user. */ |
a14b8947 IM |
256 | int latest_requested_n_rxq; /* Latest requested from netdev number |
257 | of rx queues. */ | |
72865317 BP |
258 | }; |
259 | ||
1c1e46ed AW |
260 | /* Contained by struct dp_netdev_flow's 'stats' member. */ |
261 | struct dp_netdev_flow_stats { | |
eb94da30 DDP |
262 | atomic_llong used; /* Last used time, in monotonic msecs. */ |
263 | atomic_ullong packet_count; /* Number of packets matched. */ | |
264 | atomic_ullong byte_count; /* Number of bytes matched. */ | |
265 | atomic_uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ | |
1c1e46ed AW |
266 | }; |
267 | ||
268 | /* A flow in 'dp_netdev_pmd_thread's 'flow_table'. | |
8a4e3a85 BP |
269 | * |
270 | * | |
271 | * Thread-safety | |
272 | * ============= | |
273 | * | |
274 | * Except near the beginning or ending of its lifespan, rule 'rule' belongs to | |
1c1e46ed | 275 | * its pmd thread's classifier. The text below calls this classifier 'cls'. |
8a4e3a85 BP |
276 | * |
277 | * Motivation | |
278 | * ---------- | |
279 | * | |
280 | * The thread safety rules described here for "struct dp_netdev_flow" are | |
281 | * motivated by two goals: | |
282 | * | |
283 | * - Prevent threads that read members of "struct dp_netdev_flow" from | |
284 | * reading bad data due to changes by some thread concurrently modifying | |
285 | * those members. | |
286 | * | |
287 | * - Prevent two threads making changes to members of a given "struct | |
288 | * dp_netdev_flow" from interfering with each other. | |
289 | * | |
290 | * | |
291 | * Rules | |
292 | * ----- | |
293 | * | |
ed79f89a DDP |
294 | * A flow 'flow' may be accessed without a risk of being freed during an RCU |
295 | * grace period. Code that needs to hold onto a flow for a while | |
296 | * should try incrementing 'flow->ref_cnt' with dp_netdev_flow_ref(). | |
8a4e3a85 BP |
297 | * |
298 | * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the | |
ed79f89a DDP |
299 | * flow from being deleted from 'cls' and it doesn't protect members of 'flow' |
300 | * from modification. | |
8a4e3a85 BP |
301 | * |
302 | * Some members, marked 'const', are immutable. Accessing other members | |
303 | * requires synchronization, as noted in more detail below. | |
304 | */ | |
72865317 | 305 | struct dp_netdev_flow { |
11e5cf1f | 306 | const struct flow flow; /* Unmasked flow that created this entry. */ |
8a4e3a85 | 307 | /* Hash table index by unmasked flow. */ |
1c1e46ed AW |
308 | const struct cmap_node node; /* In owning dp_netdev_pmd_thread's */ |
309 | /* 'flow_table'. */ | |
70e5ed6f | 310 | const ovs_u128 ufid; /* Unique flow identifier. */ |
bd5131ba | 311 | const unsigned pmd_id; /* The 'core_id' of pmd thread owning this */ |
1c1e46ed | 312 | /* flow. */ |
72865317 | 313 | |
ed79f89a DDP |
314 | /* Number of references. |
315 | * The classifier owns one reference. | |
316 | * Any thread trying to keep a rule from being freed should hold its own | |
317 | * reference. */ | |
318 | struct ovs_refcount ref_cnt; | |
319 | ||
11e5cf1f DDP |
320 | bool dead; |
321 | ||
1c1e46ed AW |
322 | /* Statistics. */ |
323 | struct dp_netdev_flow_stats stats; | |
8a4e3a85 | 324 | |
45c626a3 | 325 | /* Actions. */ |
61e7deb1 | 326 | OVSRCU_TYPE(struct dp_netdev_actions *) actions; |
0de8783a | 327 | |
11e5cf1f DDP |
328 | /* While processing a group of input packets, the datapath uses the next |
329 | * member to store a pointer to the output batch for the flow. It is | |
330 | * reset after the batch has been sent out (See dp_netdev_queue_batches(), | |
331 | * packet_batch_init() and packet_batch_execute()). */ | |
332 | struct packet_batch *batch; | |
333 | ||
0de8783a JR |
334 | /* Packet classification. */ |
335 | struct dpcls_rule cr; /* In owning dp_netdev's 'cls'. */ | |
336 | /* 'cr' must be the last member. */ | |
72865317 BP |
337 | }; |
338 | ||
ed79f89a | 339 | static void dp_netdev_flow_unref(struct dp_netdev_flow *); |
9bbf1c3d | 340 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *); |
70e5ed6f JS |
341 | static int dpif_netdev_flow_from_nlattrs(const struct nlattr *, uint32_t, |
342 | struct flow *); | |
8a4e3a85 | 343 | |
a84cb64a BP |
344 | /* A set of datapath actions within a "struct dp_netdev_flow". |
345 | * | |
346 | * | |
347 | * Thread-safety | |
348 | * ============= | |
349 | * | |
45c626a3 | 350 | * A struct dp_netdev_actions 'actions' is protected with RCU. */ |
a84cb64a | 351 | struct dp_netdev_actions { |
a84cb64a BP |
352 | /* These members are immutable: they do not change during the struct's |
353 | * lifetime. */ | |
a84cb64a | 354 | unsigned int size; /* Size of 'actions', in bytes. */ |
9ff55ae2 | 355 | struct nlattr actions[]; /* Sequence of OVS_ACTION_ATTR_* attributes. */ |
a84cb64a BP |
356 | }; |
357 | ||
358 | struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *, | |
359 | size_t); | |
61e7deb1 BP |
360 | struct dp_netdev_actions *dp_netdev_flow_get_actions( |
361 | const struct dp_netdev_flow *); | |
362 | static void dp_netdev_actions_free(struct dp_netdev_actions *); | |
a84cb64a | 363 | |
1c1e46ed AW |
364 | /* Contained by struct dp_netdev_pmd_thread's 'stats' member. */ |
365 | struct dp_netdev_pmd_stats { | |
366 | /* Indexed by DP_STAT_*. */ | |
eb94da30 | 367 | atomic_ullong n[DP_N_STATS]; |
1c1e46ed AW |
368 | }; |
369 | ||
55e3ca97 DDP |
370 | /* Contained by struct dp_netdev_pmd_thread's 'cycle' member. */ |
371 | struct dp_netdev_pmd_cycles { | |
372 | /* Indexed by PMD_CYCLES_*. */ | |
373 | atomic_ullong n[PMD_N_CYCLES]; | |
374 | }; | |
375 | ||
ae7ad0a1 IM |
376 | /* Contained by struct dp_netdev_pmd_thread's 'poll_list' member. */ |
377 | struct rxq_poll { | |
378 | struct dp_netdev_port *port; | |
379 | struct netdev_rxq *rx; | |
380 | struct ovs_list node; | |
381 | }; | |
382 | ||
e4cfed38 PS |
383 | /* PMD: Poll modes drivers. PMD accesses devices via polling to eliminate |
384 | * the performance overhead of interrupt processing. Therefore netdev can | |
385 | * not implement rx-wait for these devices. dpif-netdev needs to poll | |
386 | * these device to check for recv buffer. pmd-thread does polling for | |
1c1e46ed | 387 | * devices assigned to itself. |
e4cfed38 PS |
388 | * |
389 | * DPDK used PMD for accessing NIC. | |
390 | * | |
65f13b50 AW |
391 | * Note, instance with cpu core id NON_PMD_CORE_ID will be reserved for |
392 | * I/O of all non-pmd threads. There will be no actual thread created | |
393 | * for the instance. | |
1c1e46ed AW |
394 | * |
395 | * Each struct has its own flow table and classifier. Packets received | |
396 | * from managed ports are looked up in the corresponding pmd thread's | |
397 | * flow table, and are executed with the found actions. | |
398 | * */ | |
65f13b50 | 399 | struct dp_netdev_pmd_thread { |
6c3eee82 | 400 | struct dp_netdev *dp; |
1c1e46ed | 401 | struct ovs_refcount ref_cnt; /* Every reference must be refcount'ed. */ |
65f13b50 | 402 | struct cmap_node node; /* In 'dp->poll_threads'. */ |
accf8626 AW |
403 | |
404 | pthread_cond_t cond; /* For synchronizing pmd thread reload. */ | |
405 | struct ovs_mutex cond_mutex; /* Mutex for condition variable. */ | |
406 | ||
65f13b50 AW |
407 | /* Per thread exact-match cache. Note, the instance for cpu core |
408 | * NON_PMD_CORE_ID can be accessed by multiple threads, and thusly | |
409 | * need to be protected (e.g. by 'dp_netdev_mutex'). All other | |
410 | * instances will only be accessed by its own pmd thread. */ | |
9bbf1c3d | 411 | struct emc_cache flow_cache; |
1c1e46ed AW |
412 | |
413 | /* Classifier and Flow-Table. | |
414 | * | |
415 | * Writers of 'flow_table' must take the 'flow_mutex'. Corresponding | |
416 | * changes to 'cls' must be made while still holding the 'flow_mutex'. | |
417 | */ | |
418 | struct ovs_mutex flow_mutex; | |
419 | struct dpcls cls; | |
420 | struct cmap flow_table OVS_GUARDED; /* Flow table. */ | |
421 | ||
422 | /* Statistics. */ | |
423 | struct dp_netdev_pmd_stats stats; | |
424 | ||
55e3ca97 DDP |
425 | /* Cycles counters */ |
426 | struct dp_netdev_pmd_cycles cycles; | |
427 | ||
428 | /* Used to count cicles. See 'cycles_counter_end()' */ | |
429 | unsigned long long last_cycles; | |
430 | ||
65f13b50 AW |
431 | struct latch exit_latch; /* For terminating the pmd thread. */ |
432 | atomic_uint change_seq; /* For reloading pmd ports. */ | |
6c3eee82 | 433 | pthread_t thread; |
65f13b50 AW |
434 | int index; /* Idx of this pmd thread among pmd*/ |
435 | /* threads on same numa node. */ | |
bd5131ba | 436 | unsigned core_id; /* CPU core id of this pmd thread. */ |
65f13b50 | 437 | int numa_id; /* numa node id of this pmd thread. */ |
347ba9bb | 438 | atomic_int tx_qid; /* Queue id used by this pmd thread to |
3bcc10c0 | 439 | * send packets on all netdevs */ |
6553d06b | 440 | |
ae7ad0a1 IM |
441 | struct ovs_mutex poll_mutex; /* Mutex for poll_list. */ |
442 | /* List of rx queues to poll. */ | |
443 | struct ovs_list poll_list OVS_GUARDED; | |
444 | int poll_cnt; /* Number of elemints in poll_list. */ | |
445 | ||
6553d06b DDP |
446 | /* Only a pmd thread can write on its own 'cycles' and 'stats'. |
447 | * The main thread keeps 'stats_zero' and 'cycles_zero' as base | |
448 | * values and subtracts them from 'stats' and 'cycles' before | |
449 | * reporting to the user */ | |
450 | unsigned long long stats_zero[DP_N_STATS]; | |
451 | uint64_t cycles_zero[PMD_N_CYCLES]; | |
6c3eee82 BP |
452 | }; |
453 | ||
84067a4c JR |
454 | #define PMD_INITIAL_SEQ 1 |
455 | ||
72865317 BP |
456 | /* Interface to netdev-based datapath. */ |
457 | struct dpif_netdev { | |
458 | struct dpif dpif; | |
459 | struct dp_netdev *dp; | |
d33ed218 | 460 | uint64_t last_port_seq; |
72865317 BP |
461 | }; |
462 | ||
8a4e3a85 | 463 | static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no, |
59e6d833 | 464 | struct dp_netdev_port **portp); |
8a4e3a85 | 465 | static int get_port_by_name(struct dp_netdev *dp, const char *devname, |
59e6d833 | 466 | struct dp_netdev_port **portp); |
8a4e3a85 BP |
467 | static void dp_netdev_free(struct dp_netdev *) |
468 | OVS_REQUIRES(dp_netdev_mutex); | |
8a4e3a85 BP |
469 | static int do_add_port(struct dp_netdev *dp, const char *devname, |
470 | const char *type, odp_port_t port_no) | |
59e6d833 | 471 | OVS_REQUIRES(dp->port_mutex); |
c40b890f | 472 | static void do_del_port(struct dp_netdev *dp, struct dp_netdev_port *) |
59e6d833 | 473 | OVS_REQUIRES(dp->port_mutex); |
614c4892 BP |
474 | static int dpif_netdev_open(const struct dpif_class *, const char *name, |
475 | bool create, struct dpif **); | |
65f13b50 | 476 | static void dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 477 | struct dp_packet **, int c, |
41ccaa24 | 478 | bool may_steal, |
4edb9ae9 | 479 | const struct nlattr *actions, |
e4cfed38 | 480 | size_t actions_len); |
65f13b50 | 481 | static void dp_netdev_input(struct dp_netdev_pmd_thread *, |
a90ed026 DDP |
482 | struct dp_packet **, int cnt, odp_port_t port_no); |
483 | static void dp_netdev_recirculate(struct dp_netdev_pmd_thread *, | |
484 | struct dp_packet **, int cnt); | |
41ccaa24 | 485 | |
6b31e073 | 486 | static void dp_netdev_disable_upcall(struct dp_netdev *); |
ae7ad0a1 | 487 | static void dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd); |
65f13b50 AW |
488 | static void dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, |
489 | struct dp_netdev *dp, int index, | |
bd5131ba | 490 | unsigned core_id, int numa_id); |
1c1e46ed | 491 | static void dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd); |
f2eee189 | 492 | static void dp_netdev_set_nonpmd(struct dp_netdev *dp); |
b19befae | 493 | static struct dp_netdev_pmd_thread *dp_netdev_get_pmd(struct dp_netdev *dp, |
bd5131ba | 494 | unsigned core_id); |
1c1e46ed AW |
495 | static struct dp_netdev_pmd_thread * |
496 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos); | |
65f13b50 AW |
497 | static void dp_netdev_destroy_all_pmds(struct dp_netdev *dp); |
498 | static void dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
499 | static void dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
cc245ce8 IM |
500 | static void dp_netdev_pmd_clear_poll_list(struct dp_netdev_pmd_thread *pmd); |
501 | static void dp_netdev_del_port_from_pmd(struct dp_netdev_port *port, | |
502 | struct dp_netdev_pmd_thread *pmd); | |
503 | static void dp_netdev_del_port_from_all_pmds(struct dp_netdev *dp, | |
504 | struct dp_netdev_port *port); | |
505 | static void | |
506 | dp_netdev_add_port_to_pmds(struct dp_netdev *dp, struct dp_netdev_port *port); | |
ae7ad0a1 IM |
507 | static void |
508 | dp_netdev_add_rxq_to_pmd(struct dp_netdev_pmd_thread *pmd, | |
509 | struct dp_netdev_port *port, struct netdev_rxq *rx); | |
510 | static struct dp_netdev_pmd_thread * | |
511 | dp_netdev_less_loaded_pmd_on_numa(struct dp_netdev *dp, int numa_id); | |
f2eee189 | 512 | static void dp_netdev_reset_pmd_threads(struct dp_netdev *dp); |
1c1e46ed AW |
513 | static bool dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd); |
514 | static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd); | |
515 | static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd); | |
72865317 | 516 | |
67ad54cb | 517 | static inline bool emc_entry_alive(struct emc_entry *ce); |
9bbf1c3d DDP |
518 | static void emc_clear_entry(struct emc_entry *ce); |
519 | ||
520 | static void | |
521 | emc_cache_init(struct emc_cache *flow_cache) | |
522 | { | |
523 | int i; | |
524 | ||
67ad54cb | 525 | flow_cache->sweep_idx = 0; |
9bbf1c3d DDP |
526 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { |
527 | flow_cache->entries[i].flow = NULL; | |
0de8783a | 528 | flow_cache->entries[i].key.hash = 0; |
09b0fa9c | 529 | flow_cache->entries[i].key.len = sizeof(struct miniflow); |
5fcff47b | 530 | flowmap_init(&flow_cache->entries[i].key.mf.map); |
9bbf1c3d DDP |
531 | } |
532 | } | |
533 | ||
534 | static void | |
535 | emc_cache_uninit(struct emc_cache *flow_cache) | |
536 | { | |
537 | int i; | |
538 | ||
539 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { | |
540 | emc_clear_entry(&flow_cache->entries[i]); | |
541 | } | |
542 | } | |
543 | ||
67ad54cb AW |
544 | /* Check and clear dead flow references slowly (one entry at each |
545 | * invocation). */ | |
546 | static void | |
547 | emc_cache_slow_sweep(struct emc_cache *flow_cache) | |
548 | { | |
549 | struct emc_entry *entry = &flow_cache->entries[flow_cache->sweep_idx]; | |
550 | ||
551 | if (!emc_entry_alive(entry)) { | |
552 | emc_clear_entry(entry); | |
553 | } | |
554 | flow_cache->sweep_idx = (flow_cache->sweep_idx + 1) & EM_FLOW_HASH_MASK; | |
555 | } | |
556 | ||
c4ea7529 BP |
557 | /* Returns true if 'dpif' is a netdev or dummy dpif, false otherwise. */ |
558 | bool | |
559 | dpif_is_netdev(const struct dpif *dpif) | |
560 | { | |
561 | return dpif->dpif_class->open == dpif_netdev_open; | |
562 | } | |
563 | ||
72865317 BP |
564 | static struct dpif_netdev * |
565 | dpif_netdev_cast(const struct dpif *dpif) | |
566 | { | |
c4ea7529 | 567 | ovs_assert(dpif_is_netdev(dpif)); |
72865317 BP |
568 | return CONTAINER_OF(dpif, struct dpif_netdev, dpif); |
569 | } | |
570 | ||
571 | static struct dp_netdev * | |
572 | get_dp_netdev(const struct dpif *dpif) | |
573 | { | |
574 | return dpif_netdev_cast(dpif)->dp; | |
575 | } | |
6553d06b DDP |
576 | \f |
577 | enum pmd_info_type { | |
ce179f11 IM |
578 | PMD_INFO_SHOW_STATS, /* Show how cpu cycles are spent. */ |
579 | PMD_INFO_CLEAR_STATS, /* Set the cycles count to 0. */ | |
580 | PMD_INFO_SHOW_RXQ /* Show poll-lists of pmd threads. */ | |
6553d06b DDP |
581 | }; |
582 | ||
583 | static void | |
584 | pmd_info_show_stats(struct ds *reply, | |
585 | struct dp_netdev_pmd_thread *pmd, | |
586 | unsigned long long stats[DP_N_STATS], | |
587 | uint64_t cycles[PMD_N_CYCLES]) | |
588 | { | |
589 | unsigned long long total_packets = 0; | |
590 | uint64_t total_cycles = 0; | |
591 | int i; | |
592 | ||
593 | /* These loops subtracts reference values ('*_zero') from the counters. | |
594 | * Since loads and stores are relaxed, it might be possible for a '*_zero' | |
595 | * value to be more recent than the current value we're reading from the | |
596 | * counter. This is not a big problem, since these numbers are not | |
597 | * supposed to be too accurate, but we should at least make sure that | |
598 | * the result is not negative. */ | |
599 | for (i = 0; i < DP_N_STATS; i++) { | |
600 | if (stats[i] > pmd->stats_zero[i]) { | |
601 | stats[i] -= pmd->stats_zero[i]; | |
602 | } else { | |
603 | stats[i] = 0; | |
604 | } | |
605 | ||
606 | if (i != DP_STAT_LOST) { | |
607 | /* Lost packets are already included in DP_STAT_MISS */ | |
608 | total_packets += stats[i]; | |
609 | } | |
610 | } | |
611 | ||
612 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
613 | if (cycles[i] > pmd->cycles_zero[i]) { | |
614 | cycles[i] -= pmd->cycles_zero[i]; | |
615 | } else { | |
616 | cycles[i] = 0; | |
617 | } | |
618 | ||
619 | total_cycles += cycles[i]; | |
620 | } | |
621 | ||
622 | ds_put_cstr(reply, (pmd->core_id == NON_PMD_CORE_ID) | |
623 | ? "main thread" : "pmd thread"); | |
624 | ||
625 | if (pmd->numa_id != OVS_NUMA_UNSPEC) { | |
626 | ds_put_format(reply, " numa_id %d", pmd->numa_id); | |
627 | } | |
d5c199ea | 628 | if (pmd->core_id != OVS_CORE_UNSPEC && pmd->core_id != NON_PMD_CORE_ID) { |
bd5131ba | 629 | ds_put_format(reply, " core_id %u", pmd->core_id); |
6553d06b DDP |
630 | } |
631 | ds_put_cstr(reply, ":\n"); | |
632 | ||
633 | ds_put_format(reply, | |
634 | "\temc hits:%llu\n\tmegaflow hits:%llu\n" | |
635 | "\tmiss:%llu\n\tlost:%llu\n", | |
636 | stats[DP_STAT_EXACT_HIT], stats[DP_STAT_MASKED_HIT], | |
637 | stats[DP_STAT_MISS], stats[DP_STAT_LOST]); | |
638 | ||
639 | if (total_cycles == 0) { | |
640 | return; | |
641 | } | |
642 | ||
643 | ds_put_format(reply, | |
644 | "\tpolling cycles:%"PRIu64" (%.02f%%)\n" | |
645 | "\tprocessing cycles:%"PRIu64" (%.02f%%)\n", | |
646 | cycles[PMD_CYCLES_POLLING], | |
647 | cycles[PMD_CYCLES_POLLING] / (double)total_cycles * 100, | |
648 | cycles[PMD_CYCLES_PROCESSING], | |
649 | cycles[PMD_CYCLES_PROCESSING] / (double)total_cycles * 100); | |
650 | ||
651 | if (total_packets == 0) { | |
652 | return; | |
653 | } | |
654 | ||
655 | ds_put_format(reply, | |
656 | "\tavg cycles per packet: %.02f (%"PRIu64"/%llu)\n", | |
657 | total_cycles / (double)total_packets, | |
658 | total_cycles, total_packets); | |
659 | ||
660 | ds_put_format(reply, | |
661 | "\tavg processing cycles per packet: " | |
662 | "%.02f (%"PRIu64"/%llu)\n", | |
663 | cycles[PMD_CYCLES_PROCESSING] / (double)total_packets, | |
664 | cycles[PMD_CYCLES_PROCESSING], total_packets); | |
665 | } | |
666 | ||
667 | static void | |
668 | pmd_info_clear_stats(struct ds *reply OVS_UNUSED, | |
669 | struct dp_netdev_pmd_thread *pmd, | |
670 | unsigned long long stats[DP_N_STATS], | |
671 | uint64_t cycles[PMD_N_CYCLES]) | |
672 | { | |
673 | int i; | |
674 | ||
675 | /* We cannot write 'stats' and 'cycles' (because they're written by other | |
676 | * threads) and we shouldn't change 'stats' (because they're used to count | |
677 | * datapath stats, which must not be cleared here). Instead, we save the | |
678 | * current values and subtract them from the values to be displayed in the | |
679 | * future */ | |
680 | for (i = 0; i < DP_N_STATS; i++) { | |
681 | pmd->stats_zero[i] = stats[i]; | |
682 | } | |
683 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
684 | pmd->cycles_zero[i] = cycles[i]; | |
685 | } | |
686 | } | |
687 | ||
ce179f11 IM |
688 | static void |
689 | pmd_info_show_rxq(struct ds *reply, struct dp_netdev_pmd_thread *pmd) | |
690 | { | |
691 | if (pmd->core_id != NON_PMD_CORE_ID) { | |
692 | struct rxq_poll *poll; | |
693 | const char *prev_name = NULL; | |
694 | ||
695 | ds_put_format(reply, "pmd thread numa_id %d core_id %u:\n", | |
696 | pmd->numa_id, pmd->core_id); | |
697 | ||
698 | ovs_mutex_lock(&pmd->poll_mutex); | |
699 | LIST_FOR_EACH (poll, node, &pmd->poll_list) { | |
700 | const char *name = netdev_get_name(poll->port->netdev); | |
701 | ||
702 | if (!prev_name || strcmp(name, prev_name)) { | |
703 | if (prev_name) { | |
704 | ds_put_cstr(reply, "\n"); | |
705 | } | |
706 | ds_put_format(reply, "\tport: %s\tqueue-id:", | |
707 | netdev_get_name(poll->port->netdev)); | |
708 | } | |
709 | ds_put_format(reply, " %d", netdev_rxq_get_queue_id(poll->rx)); | |
710 | prev_name = name; | |
711 | } | |
712 | ovs_mutex_unlock(&pmd->poll_mutex); | |
713 | ds_put_cstr(reply, "\n"); | |
714 | } | |
715 | } | |
716 | ||
6553d06b DDP |
717 | static void |
718 | dpif_netdev_pmd_info(struct unixctl_conn *conn, int argc, const char *argv[], | |
719 | void *aux) | |
720 | { | |
721 | struct ds reply = DS_EMPTY_INITIALIZER; | |
722 | struct dp_netdev_pmd_thread *pmd; | |
723 | struct dp_netdev *dp = NULL; | |
724 | enum pmd_info_type type = *(enum pmd_info_type *) aux; | |
725 | ||
726 | ovs_mutex_lock(&dp_netdev_mutex); | |
727 | ||
728 | if (argc == 2) { | |
729 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
730 | } else if (shash_count(&dp_netdevs) == 1) { | |
731 | /* There's only one datapath */ | |
732 | dp = shash_first(&dp_netdevs)->data; | |
733 | } | |
734 | ||
735 | if (!dp) { | |
736 | ovs_mutex_unlock(&dp_netdev_mutex); | |
737 | unixctl_command_reply_error(conn, | |
738 | "please specify an existing datapath"); | |
739 | return; | |
740 | } | |
741 | ||
742 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
ce179f11 IM |
743 | if (type == PMD_INFO_SHOW_RXQ) { |
744 | pmd_info_show_rxq(&reply, pmd); | |
745 | } else { | |
746 | unsigned long long stats[DP_N_STATS]; | |
747 | uint64_t cycles[PMD_N_CYCLES]; | |
748 | int i; | |
6553d06b | 749 | |
ce179f11 IM |
750 | /* Read current stats and cycle counters */ |
751 | for (i = 0; i < ARRAY_SIZE(stats); i++) { | |
752 | atomic_read_relaxed(&pmd->stats.n[i], &stats[i]); | |
753 | } | |
754 | for (i = 0; i < ARRAY_SIZE(cycles); i++) { | |
755 | atomic_read_relaxed(&pmd->cycles.n[i], &cycles[i]); | |
756 | } | |
6553d06b | 757 | |
ce179f11 IM |
758 | if (type == PMD_INFO_CLEAR_STATS) { |
759 | pmd_info_clear_stats(&reply, pmd, stats, cycles); | |
760 | } else if (type == PMD_INFO_SHOW_STATS) { | |
761 | pmd_info_show_stats(&reply, pmd, stats, cycles); | |
762 | } | |
6553d06b DDP |
763 | } |
764 | } | |
765 | ||
766 | ovs_mutex_unlock(&dp_netdev_mutex); | |
767 | ||
768 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
769 | ds_destroy(&reply); | |
770 | } | |
771 | \f | |
772 | static int | |
773 | dpif_netdev_init(void) | |
774 | { | |
775 | static enum pmd_info_type show_aux = PMD_INFO_SHOW_STATS, | |
ce179f11 IM |
776 | clear_aux = PMD_INFO_CLEAR_STATS, |
777 | poll_aux = PMD_INFO_SHOW_RXQ; | |
6553d06b DDP |
778 | |
779 | unixctl_command_register("dpif-netdev/pmd-stats-show", "[dp]", | |
780 | 0, 1, dpif_netdev_pmd_info, | |
781 | (void *)&show_aux); | |
782 | unixctl_command_register("dpif-netdev/pmd-stats-clear", "[dp]", | |
783 | 0, 1, dpif_netdev_pmd_info, | |
784 | (void *)&clear_aux); | |
ce179f11 IM |
785 | unixctl_command_register("dpif-netdev/pmd-rxq-show", "[dp]", |
786 | 0, 1, dpif_netdev_pmd_info, | |
787 | (void *)&poll_aux); | |
6553d06b DDP |
788 | return 0; |
789 | } | |
72865317 | 790 | |
2197d7ab | 791 | static int |
2240af25 DDP |
792 | dpif_netdev_enumerate(struct sset *all_dps, |
793 | const struct dpif_class *dpif_class) | |
2197d7ab GL |
794 | { |
795 | struct shash_node *node; | |
796 | ||
97be1538 | 797 | ovs_mutex_lock(&dp_netdev_mutex); |
2197d7ab | 798 | SHASH_FOR_EACH(node, &dp_netdevs) { |
2240af25 DDP |
799 | struct dp_netdev *dp = node->data; |
800 | if (dpif_class != dp->class) { | |
801 | /* 'dp_netdevs' contains both "netdev" and "dummy" dpifs. | |
802 | * If the class doesn't match, skip this dpif. */ | |
803 | continue; | |
804 | } | |
2197d7ab GL |
805 | sset_add(all_dps, node->name); |
806 | } | |
97be1538 | 807 | ovs_mutex_unlock(&dp_netdev_mutex); |
5279f8fd | 808 | |
2197d7ab GL |
809 | return 0; |
810 | } | |
811 | ||
add90f6f EJ |
812 | static bool |
813 | dpif_netdev_class_is_dummy(const struct dpif_class *class) | |
814 | { | |
815 | return class != &dpif_netdev_class; | |
816 | } | |
817 | ||
0aeaabc8 JP |
818 | static const char * |
819 | dpif_netdev_port_open_type(const struct dpif_class *class, const char *type) | |
820 | { | |
821 | return strcmp(type, "internal") ? type | |
add90f6f | 822 | : dpif_netdev_class_is_dummy(class) ? "dummy" |
0aeaabc8 JP |
823 | : "tap"; |
824 | } | |
825 | ||
72865317 BP |
826 | static struct dpif * |
827 | create_dpif_netdev(struct dp_netdev *dp) | |
828 | { | |
462278db | 829 | uint16_t netflow_id = hash_string(dp->name, 0); |
72865317 | 830 | struct dpif_netdev *dpif; |
72865317 | 831 | |
6a8267c5 | 832 | ovs_refcount_ref(&dp->ref_cnt); |
72865317 | 833 | |
72865317 | 834 | dpif = xmalloc(sizeof *dpif); |
614c4892 | 835 | dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id); |
72865317 | 836 | dpif->dp = dp; |
d33ed218 | 837 | dpif->last_port_seq = seq_read(dp->port_seq); |
72865317 BP |
838 | |
839 | return &dpif->dpif; | |
840 | } | |
841 | ||
4e022ec0 AW |
842 | /* Choose an unused, non-zero port number and return it on success. |
843 | * Return ODPP_NONE on failure. */ | |
844 | static odp_port_t | |
e44768b7 | 845 | choose_port(struct dp_netdev *dp, const char *name) |
59e6d833 | 846 | OVS_REQUIRES(dp->port_mutex) |
e44768b7 | 847 | { |
4e022ec0 | 848 | uint32_t port_no; |
e44768b7 JP |
849 | |
850 | if (dp->class != &dpif_netdev_class) { | |
851 | const char *p; | |
852 | int start_no = 0; | |
853 | ||
854 | /* If the port name begins with "br", start the number search at | |
855 | * 100 to make writing tests easier. */ | |
856 | if (!strncmp(name, "br", 2)) { | |
857 | start_no = 100; | |
858 | } | |
859 | ||
860 | /* If the port name contains a number, try to assign that port number. | |
861 | * This can make writing unit tests easier because port numbers are | |
862 | * predictable. */ | |
863 | for (p = name; *p != '\0'; p++) { | |
864 | if (isdigit((unsigned char) *p)) { | |
865 | port_no = start_no + strtol(p, NULL, 10); | |
ff073a71 BP |
866 | if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE) |
867 | && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 868 | return u32_to_odp(port_no); |
e44768b7 JP |
869 | } |
870 | break; | |
871 | } | |
872 | } | |
873 | } | |
874 | ||
ff073a71 BP |
875 | for (port_no = 1; port_no <= UINT16_MAX; port_no++) { |
876 | if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 877 | return u32_to_odp(port_no); |
e44768b7 JP |
878 | } |
879 | } | |
880 | ||
4e022ec0 | 881 | return ODPP_NONE; |
e44768b7 JP |
882 | } |
883 | ||
72865317 | 884 | static int |
614c4892 BP |
885 | create_dp_netdev(const char *name, const struct dpif_class *class, |
886 | struct dp_netdev **dpp) | |
8a4e3a85 | 887 | OVS_REQUIRES(dp_netdev_mutex) |
72865317 BP |
888 | { |
889 | struct dp_netdev *dp; | |
890 | int error; | |
72865317 | 891 | |
462278db | 892 | dp = xzalloc(sizeof *dp); |
8a4e3a85 BP |
893 | shash_add(&dp_netdevs, name, dp); |
894 | ||
895 | *CONST_CAST(const struct dpif_class **, &dp->class) = class; | |
896 | *CONST_CAST(const char **, &dp->name) = xstrdup(name); | |
6a8267c5 | 897 | ovs_refcount_init(&dp->ref_cnt); |
1a65ba85 | 898 | atomic_flag_clear(&dp->destroyed); |
8a4e3a85 | 899 | |
59e6d833 BP |
900 | ovs_mutex_init(&dp->port_mutex); |
901 | cmap_init(&dp->ports); | |
d33ed218 | 902 | dp->port_seq = seq_create(); |
6b31e073 RW |
903 | fat_rwlock_init(&dp->upcall_rwlock); |
904 | ||
905 | /* Disable upcalls by default. */ | |
906 | dp_netdev_disable_upcall(dp); | |
623540e4 | 907 | dp->upcall_aux = NULL; |
6b31e073 | 908 | dp->upcall_cb = NULL; |
e44768b7 | 909 | |
65f13b50 AW |
910 | cmap_init(&dp->poll_threads); |
911 | ovs_mutex_init_recursive(&dp->non_pmd_mutex); | |
912 | ovsthread_key_create(&dp->per_pmd_key, NULL); | |
913 | ||
f2eee189 | 914 | dp_netdev_set_nonpmd(dp); |
65f13b50 | 915 | |
59e6d833 | 916 | ovs_mutex_lock(&dp->port_mutex); |
4e022ec0 | 917 | error = do_add_port(dp, name, "internal", ODPP_LOCAL); |
59e6d833 | 918 | ovs_mutex_unlock(&dp->port_mutex); |
72865317 BP |
919 | if (error) { |
920 | dp_netdev_free(dp); | |
462278db | 921 | return error; |
72865317 BP |
922 | } |
923 | ||
a36de779 | 924 | dp->last_tnl_conf_seq = seq_read(tnl_conf_seq); |
462278db | 925 | *dpp = dp; |
72865317 BP |
926 | return 0; |
927 | } | |
928 | ||
929 | static int | |
614c4892 | 930 | dpif_netdev_open(const struct dpif_class *class, const char *name, |
4a387741 | 931 | bool create, struct dpif **dpifp) |
72865317 | 932 | { |
462278db | 933 | struct dp_netdev *dp; |
5279f8fd | 934 | int error; |
462278db | 935 | |
97be1538 | 936 | ovs_mutex_lock(&dp_netdev_mutex); |
462278db BP |
937 | dp = shash_find_data(&dp_netdevs, name); |
938 | if (!dp) { | |
5279f8fd | 939 | error = create ? create_dp_netdev(name, class, &dp) : ENODEV; |
72865317 | 940 | } else { |
5279f8fd BP |
941 | error = (dp->class != class ? EINVAL |
942 | : create ? EEXIST | |
943 | : 0); | |
944 | } | |
945 | if (!error) { | |
946 | *dpifp = create_dpif_netdev(dp); | |
6b31e073 | 947 | dp->dpif = *dpifp; |
72865317 | 948 | } |
97be1538 | 949 | ovs_mutex_unlock(&dp_netdev_mutex); |
462278db | 950 | |
5279f8fd | 951 | return error; |
72865317 BP |
952 | } |
953 | ||
88ace79b DDP |
954 | static void |
955 | dp_netdev_destroy_upcall_lock(struct dp_netdev *dp) | |
956 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
957 | { | |
958 | /* Check that upcalls are disabled, i.e. that the rwlock is taken */ | |
959 | ovs_assert(fat_rwlock_tryrdlock(&dp->upcall_rwlock)); | |
960 | ||
961 | /* Before freeing a lock we should release it */ | |
962 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
963 | fat_rwlock_destroy(&dp->upcall_rwlock); | |
964 | } | |
965 | ||
8a4e3a85 BP |
966 | /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp' |
967 | * through the 'dp_netdevs' shash while freeing 'dp'. */ | |
1ba530f4 BP |
968 | static void |
969 | dp_netdev_free(struct dp_netdev *dp) | |
8a4e3a85 | 970 | OVS_REQUIRES(dp_netdev_mutex) |
1ba530f4 | 971 | { |
59e6d833 | 972 | struct dp_netdev_port *port; |
4ad28026 | 973 | |
8a4e3a85 BP |
974 | shash_find_and_delete(&dp_netdevs, dp->name); |
975 | ||
65f13b50 AW |
976 | dp_netdev_destroy_all_pmds(dp); |
977 | ovs_mutex_destroy(&dp->non_pmd_mutex); | |
978 | ovsthread_key_delete(dp->per_pmd_key); | |
6c3eee82 | 979 | |
59e6d833 | 980 | ovs_mutex_lock(&dp->port_mutex); |
a532e683 | 981 | CMAP_FOR_EACH (port, node, &dp->ports) { |
d916785c | 982 | /* PMD threads are destroyed here. do_del_port() cannot quiesce */ |
c40b890f | 983 | do_del_port(dp, port); |
1ba530f4 | 984 | } |
59e6d833 | 985 | ovs_mutex_unlock(&dp->port_mutex); |
d916785c | 986 | cmap_destroy(&dp->poll_threads); |
51852a57 | 987 | |
d33ed218 | 988 | seq_destroy(dp->port_seq); |
59e6d833 | 989 | cmap_destroy(&dp->ports); |
88ace79b DDP |
990 | |
991 | /* Upcalls must be disabled at this point */ | |
992 | dp_netdev_destroy_upcall_lock(dp); | |
9bbf1c3d | 993 | |
f2eee189 | 994 | free(dp->pmd_cmask); |
8a4e3a85 | 995 | free(CONST_CAST(char *, dp->name)); |
72865317 BP |
996 | free(dp); |
997 | } | |
998 | ||
8a4e3a85 BP |
999 | static void |
1000 | dp_netdev_unref(struct dp_netdev *dp) | |
1001 | { | |
1002 | if (dp) { | |
1003 | /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't | |
1004 | * get a new reference to 'dp' through the 'dp_netdevs' shash. */ | |
1005 | ovs_mutex_lock(&dp_netdev_mutex); | |
24f83812 | 1006 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
8a4e3a85 BP |
1007 | dp_netdev_free(dp); |
1008 | } | |
1009 | ovs_mutex_unlock(&dp_netdev_mutex); | |
1010 | } | |
1011 | } | |
1012 | ||
72865317 BP |
1013 | static void |
1014 | dpif_netdev_close(struct dpif *dpif) | |
1015 | { | |
1016 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 1017 | |
8a4e3a85 | 1018 | dp_netdev_unref(dp); |
72865317 BP |
1019 | free(dpif); |
1020 | } | |
1021 | ||
1022 | static int | |
7dab847a | 1023 | dpif_netdev_destroy(struct dpif *dpif) |
72865317 BP |
1024 | { |
1025 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 1026 | |
6a8267c5 | 1027 | if (!atomic_flag_test_and_set(&dp->destroyed)) { |
24f83812 | 1028 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
6a8267c5 BP |
1029 | /* Can't happen: 'dpif' still owns a reference to 'dp'. */ |
1030 | OVS_NOT_REACHED(); | |
1031 | } | |
1032 | } | |
5279f8fd | 1033 | |
72865317 BP |
1034 | return 0; |
1035 | } | |
1036 | ||
eb94da30 DDP |
1037 | /* Add 'n' to the atomic variable 'var' non-atomically and using relaxed |
1038 | * load/store semantics. While the increment is not atomic, the load and | |
1039 | * store operations are, making it impossible to read inconsistent values. | |
1040 | * | |
1041 | * This is used to update thread local stats counters. */ | |
1042 | static void | |
1043 | non_atomic_ullong_add(atomic_ullong *var, unsigned long long n) | |
1044 | { | |
1045 | unsigned long long tmp; | |
1046 | ||
1047 | atomic_read_relaxed(var, &tmp); | |
1048 | tmp += n; | |
1049 | atomic_store_relaxed(var, tmp); | |
1050 | } | |
1051 | ||
72865317 | 1052 | static int |
a8d9304d | 1053 | dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) |
72865317 BP |
1054 | { |
1055 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed | 1056 | struct dp_netdev_pmd_thread *pmd; |
8a4e3a85 | 1057 | |
1c1e46ed AW |
1058 | stats->n_flows = stats->n_hit = stats->n_missed = stats->n_lost = 0; |
1059 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
eb94da30 | 1060 | unsigned long long n; |
1c1e46ed | 1061 | stats->n_flows += cmap_count(&pmd->flow_table); |
eb94da30 | 1062 | |
abcf3ef4 DDP |
1063 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MASKED_HIT], &n); |
1064 | stats->n_hit += n; | |
1065 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_EXACT_HIT], &n); | |
eb94da30 DDP |
1066 | stats->n_hit += n; |
1067 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MISS], &n); | |
1068 | stats->n_missed += n; | |
1069 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_LOST], &n); | |
1070 | stats->n_lost += n; | |
51852a57 | 1071 | } |
1ce3fa06 | 1072 | stats->n_masks = UINT32_MAX; |
847108dc | 1073 | stats->n_mask_hit = UINT64_MAX; |
5279f8fd | 1074 | |
72865317 BP |
1075 | return 0; |
1076 | } | |
1077 | ||
e4cfed38 | 1078 | static void |
65f13b50 | 1079 | dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd) |
e4cfed38 | 1080 | { |
65f13b50 AW |
1081 | int old_seq; |
1082 | ||
accf8626 AW |
1083 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1084 | return; | |
1085 | } | |
1086 | ||
1087 | ovs_mutex_lock(&pmd->cond_mutex); | |
65f13b50 | 1088 | atomic_add_relaxed(&pmd->change_seq, 1, &old_seq); |
accf8626 AW |
1089 | ovs_mutex_cond_wait(&pmd->cond, &pmd->cond_mutex); |
1090 | ovs_mutex_unlock(&pmd->cond_mutex); | |
65f13b50 | 1091 | } |
e4cfed38 | 1092 | |
59e6d833 BP |
1093 | static uint32_t |
1094 | hash_port_no(odp_port_t port_no) | |
1095 | { | |
1096 | return hash_int(odp_to_u32(port_no), 0); | |
1097 | } | |
1098 | ||
72865317 | 1099 | static int |
c3827f61 | 1100 | do_add_port(struct dp_netdev *dp, const char *devname, const char *type, |
4e022ec0 | 1101 | odp_port_t port_no) |
59e6d833 | 1102 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1103 | { |
4b609110 | 1104 | struct netdev_saved_flags *sf; |
72865317 BP |
1105 | struct dp_netdev_port *port; |
1106 | struct netdev *netdev; | |
2499a8ce | 1107 | enum netdev_flags flags; |
0cbfe35d | 1108 | const char *open_type; |
72865317 | 1109 | int error; |
55c955bd | 1110 | int i; |
72865317 | 1111 | |
17050610 BP |
1112 | /* Reject devices already in 'dp'. */ |
1113 | if (!get_port_by_name(dp, devname, &port)) { | |
1114 | return EEXIST; | |
1115 | } | |
72865317 BP |
1116 | |
1117 | /* Open and validate network device. */ | |
0aeaabc8 | 1118 | open_type = dpif_netdev_port_open_type(dp->class, type); |
0cbfe35d | 1119 | error = netdev_open(devname, open_type, &netdev); |
72865317 BP |
1120 | if (error) { |
1121 | return error; | |
1122 | } | |
72865317 BP |
1123 | /* XXX reject non-Ethernet devices */ |
1124 | ||
2499a8ce AC |
1125 | netdev_get_flags(netdev, &flags); |
1126 | if (flags & NETDEV_LOOPBACK) { | |
1127 | VLOG_ERR("%s: cannot add a loopback device", devname); | |
1128 | netdev_close(netdev); | |
1129 | return EINVAL; | |
1130 | } | |
1131 | ||
5a034064 AW |
1132 | if (netdev_is_pmd(netdev)) { |
1133 | int n_cores = ovs_numa_get_n_cores(); | |
1134 | ||
1135 | if (n_cores == OVS_CORE_UNSPEC) { | |
1136 | VLOG_ERR("%s, cannot get cpu core info", devname); | |
1137 | return ENOENT; | |
1138 | } | |
1139 | /* There can only be ovs_numa_get_n_cores() pmd threads, | |
3bcc10c0 DDP |
1140 | * so creates a txq for each, and one extra for the non |
1141 | * pmd threads. */ | |
a14b8947 IM |
1142 | error = netdev_set_multiq(netdev, n_cores + 1, |
1143 | netdev_requested_n_rxq(netdev)); | |
7251515e | 1144 | if (error && (error != EOPNOTSUPP)) { |
5a034064 AW |
1145 | VLOG_ERR("%s, cannot set multiq", devname); |
1146 | return errno; | |
1147 | } | |
1148 | } | |
e4cfed38 | 1149 | port = xzalloc(sizeof *port); |
35303d71 | 1150 | port->port_no = port_no; |
e4cfed38 | 1151 | port->netdev = netdev; |
55c955bd | 1152 | port->rxq = xmalloc(sizeof *port->rxq * netdev_n_rxq(netdev)); |
e4cfed38 | 1153 | port->type = xstrdup(type); |
a14b8947 | 1154 | port->latest_requested_n_rxq = netdev_requested_n_rxq(netdev); |
55c955bd PS |
1155 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1156 | error = netdev_rxq_open(netdev, &port->rxq[i], i); | |
1157 | if (error | |
1158 | && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) { | |
1159 | VLOG_ERR("%s: cannot receive packets on this network device (%s)", | |
1160 | devname, ovs_strerror(errno)); | |
1161 | netdev_close(netdev); | |
16bea12c TG |
1162 | free(port->type); |
1163 | free(port->rxq); | |
1164 | free(port); | |
55c955bd PS |
1165 | return error; |
1166 | } | |
7b6b0ef4 BP |
1167 | } |
1168 | ||
4b609110 | 1169 | error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); |
72865317 | 1170 | if (error) { |
55c955bd PS |
1171 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1172 | netdev_rxq_close(port->rxq[i]); | |
1173 | } | |
72865317 | 1174 | netdev_close(netdev); |
16bea12c | 1175 | free(port->type); |
f7791740 | 1176 | free(port->rxq); |
e4cfed38 | 1177 | free(port); |
72865317 BP |
1178 | return error; |
1179 | } | |
4b609110 | 1180 | port->sf = sf; |
e4cfed38 | 1181 | |
f7d63652 AW |
1182 | ovs_refcount_init(&port->ref_cnt); |
1183 | cmap_insert(&dp->ports, &port->node, hash_port_no(port_no)); | |
1184 | ||
e4cfed38 | 1185 | if (netdev_is_pmd(netdev)) { |
cc245ce8 | 1186 | dp_netdev_add_port_to_pmds(dp, port); |
e4cfed38 | 1187 | } |
d33ed218 | 1188 | seq_change(dp->port_seq); |
72865317 BP |
1189 | |
1190 | return 0; | |
1191 | } | |
1192 | ||
247527db BP |
1193 | static int |
1194 | dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, | |
4e022ec0 | 1195 | odp_port_t *port_nop) |
247527db BP |
1196 | { |
1197 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3aa30359 BP |
1198 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1199 | const char *dpif_port; | |
4e022ec0 | 1200 | odp_port_t port_no; |
5279f8fd | 1201 | int error; |
247527db | 1202 | |
59e6d833 | 1203 | ovs_mutex_lock(&dp->port_mutex); |
3aa30359 | 1204 | dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
4e022ec0 | 1205 | if (*port_nop != ODPP_NONE) { |
ff073a71 BP |
1206 | port_no = *port_nop; |
1207 | error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0; | |
232dfa4a | 1208 | } else { |
3aa30359 | 1209 | port_no = choose_port(dp, dpif_port); |
5279f8fd | 1210 | error = port_no == ODPP_NONE ? EFBIG : 0; |
232dfa4a | 1211 | } |
5279f8fd | 1212 | if (!error) { |
247527db | 1213 | *port_nop = port_no; |
5279f8fd | 1214 | error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); |
247527db | 1215 | } |
59e6d833 | 1216 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1217 | |
1218 | return error; | |
72865317 BP |
1219 | } |
1220 | ||
1221 | static int | |
4e022ec0 | 1222 | dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no) |
72865317 BP |
1223 | { |
1224 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd BP |
1225 | int error; |
1226 | ||
59e6d833 | 1227 | ovs_mutex_lock(&dp->port_mutex); |
c40b890f BP |
1228 | if (port_no == ODPP_LOCAL) { |
1229 | error = EINVAL; | |
1230 | } else { | |
1231 | struct dp_netdev_port *port; | |
1232 | ||
1233 | error = get_port_by_number(dp, port_no, &port); | |
1234 | if (!error) { | |
1235 | do_del_port(dp, port); | |
1236 | } | |
1237 | } | |
59e6d833 | 1238 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1239 | |
1240 | return error; | |
72865317 BP |
1241 | } |
1242 | ||
1243 | static bool | |
4e022ec0 | 1244 | is_valid_port_number(odp_port_t port_no) |
72865317 | 1245 | { |
ff073a71 BP |
1246 | return port_no != ODPP_NONE; |
1247 | } | |
1248 | ||
1249 | static struct dp_netdev_port * | |
1250 | dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no) | |
1251 | { | |
1252 | struct dp_netdev_port *port; | |
1253 | ||
59e6d833 | 1254 | CMAP_FOR_EACH_WITH_HASH (port, node, hash_port_no(port_no), &dp->ports) { |
35303d71 | 1255 | if (port->port_no == port_no) { |
ff073a71 BP |
1256 | return port; |
1257 | } | |
1258 | } | |
1259 | return NULL; | |
72865317 BP |
1260 | } |
1261 | ||
1262 | static int | |
1263 | get_port_by_number(struct dp_netdev *dp, | |
4e022ec0 | 1264 | odp_port_t port_no, struct dp_netdev_port **portp) |
72865317 BP |
1265 | { |
1266 | if (!is_valid_port_number(port_no)) { | |
1267 | *portp = NULL; | |
1268 | return EINVAL; | |
1269 | } else { | |
ff073a71 | 1270 | *portp = dp_netdev_lookup_port(dp, port_no); |
72865317 BP |
1271 | return *portp ? 0 : ENOENT; |
1272 | } | |
1273 | } | |
1274 | ||
b284085e PS |
1275 | static void |
1276 | port_ref(struct dp_netdev_port *port) | |
1277 | { | |
1278 | if (port) { | |
1279 | ovs_refcount_ref(&port->ref_cnt); | |
1280 | } | |
1281 | } | |
1282 | ||
59e6d833 BP |
1283 | static void |
1284 | port_unref(struct dp_netdev_port *port) | |
1285 | { | |
24f83812 | 1286 | if (port && ovs_refcount_unref_relaxed(&port->ref_cnt) == 1) { |
accf8626 AW |
1287 | int n_rxq = netdev_n_rxq(port->netdev); |
1288 | int i; | |
1289 | ||
1290 | netdev_close(port->netdev); | |
1291 | netdev_restore_flags(port->sf); | |
1292 | ||
1293 | for (i = 0; i < n_rxq; i++) { | |
1294 | netdev_rxq_close(port->rxq[i]); | |
1295 | } | |
1296 | free(port->rxq); | |
1297 | free(port->type); | |
1298 | free(port); | |
b284085e PS |
1299 | } |
1300 | } | |
1301 | ||
72865317 BP |
1302 | static int |
1303 | get_port_by_name(struct dp_netdev *dp, | |
1304 | const char *devname, struct dp_netdev_port **portp) | |
59e6d833 | 1305 | OVS_REQUIRES(dp->port_mutex) |
72865317 BP |
1306 | { |
1307 | struct dp_netdev_port *port; | |
1308 | ||
a532e683 | 1309 | CMAP_FOR_EACH (port, node, &dp->ports) { |
3efb6063 | 1310 | if (!strcmp(netdev_get_name(port->netdev), devname)) { |
72865317 BP |
1311 | *portp = port; |
1312 | return 0; | |
1313 | } | |
1314 | } | |
1315 | return ENOENT; | |
1316 | } | |
1317 | ||
347ba9bb IM |
1318 | static int |
1319 | get_n_pmd_threads(struct dp_netdev *dp) | |
1320 | { | |
1321 | /* There is one non pmd thread in dp->poll_threads */ | |
1322 | return cmap_count(&dp->poll_threads) - 1; | |
1323 | } | |
1324 | ||
65f13b50 AW |
1325 | static int |
1326 | get_n_pmd_threads_on_numa(struct dp_netdev *dp, int numa_id) | |
1327 | { | |
1328 | struct dp_netdev_pmd_thread *pmd; | |
1329 | int n_pmds = 0; | |
1330 | ||
1331 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1332 | if (pmd->numa_id == numa_id) { | |
1333 | n_pmds++; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | return n_pmds; | |
1338 | } | |
1339 | ||
1340 | /* Returns 'true' if there is a port with pmd netdev and the netdev | |
1341 | * is on numa node 'numa_id'. */ | |
1342 | static bool | |
1343 | has_pmd_port_for_numa(struct dp_netdev *dp, int numa_id) | |
1344 | { | |
1345 | struct dp_netdev_port *port; | |
1346 | ||
1347 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
1348 | if (netdev_is_pmd(port->netdev) | |
1349 | && netdev_get_numa_id(port->netdev) == numa_id) { | |
1350 | return true; | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | return false; | |
1355 | } | |
1356 | ||
1357 | ||
c40b890f BP |
1358 | static void |
1359 | do_del_port(struct dp_netdev *dp, struct dp_netdev_port *port) | |
59e6d833 | 1360 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1361 | { |
35303d71 | 1362 | cmap_remove(&dp->ports, &port->node, hash_odp_port(port->port_no)); |
d33ed218 | 1363 | seq_change(dp->port_seq); |
e4cfed38 | 1364 | if (netdev_is_pmd(port->netdev)) { |
65f13b50 AW |
1365 | int numa_id = netdev_get_numa_id(port->netdev); |
1366 | ||
ae7ad0a1 IM |
1367 | /* PMD threads can not be on invalid numa node. */ |
1368 | ovs_assert(ovs_numa_numa_id_is_valid(numa_id)); | |
65f13b50 | 1369 | /* If there is no netdev on the numa node, deletes the pmd threads |
ae7ad0a1 | 1370 | * for that numa. Else, deletes the queues from polling lists. */ |
65f13b50 AW |
1371 | if (!has_pmd_port_for_numa(dp, numa_id)) { |
1372 | dp_netdev_del_pmds_on_numa(dp, numa_id); | |
ae7ad0a1 | 1373 | } else { |
cc245ce8 | 1374 | dp_netdev_del_port_from_all_pmds(dp, port); |
65f13b50 | 1375 | } |
e4cfed38 | 1376 | } |
72865317 | 1377 | |
b284085e | 1378 | port_unref(port); |
72865317 BP |
1379 | } |
1380 | ||
1381 | static void | |
4c738a8d BP |
1382 | answer_port_query(const struct dp_netdev_port *port, |
1383 | struct dpif_port *dpif_port) | |
72865317 | 1384 | { |
3efb6063 | 1385 | dpif_port->name = xstrdup(netdev_get_name(port->netdev)); |
0cbfe35d | 1386 | dpif_port->type = xstrdup(port->type); |
35303d71 | 1387 | dpif_port->port_no = port->port_no; |
72865317 BP |
1388 | } |
1389 | ||
1390 | static int | |
4e022ec0 | 1391 | dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, |
4c738a8d | 1392 | struct dpif_port *dpif_port) |
72865317 BP |
1393 | { |
1394 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1395 | struct dp_netdev_port *port; | |
1396 | int error; | |
1397 | ||
1398 | error = get_port_by_number(dp, port_no, &port); | |
4afba28d | 1399 | if (!error && dpif_port) { |
4c738a8d | 1400 | answer_port_query(port, dpif_port); |
72865317 | 1401 | } |
5279f8fd | 1402 | |
72865317 BP |
1403 | return error; |
1404 | } | |
1405 | ||
1406 | static int | |
1407 | dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, | |
4c738a8d | 1408 | struct dpif_port *dpif_port) |
72865317 BP |
1409 | { |
1410 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1411 | struct dp_netdev_port *port; | |
1412 | int error; | |
1413 | ||
59e6d833 | 1414 | ovs_mutex_lock(&dp->port_mutex); |
72865317 | 1415 | error = get_port_by_name(dp, devname, &port); |
4afba28d | 1416 | if (!error && dpif_port) { |
4c738a8d | 1417 | answer_port_query(port, dpif_port); |
72865317 | 1418 | } |
59e6d833 | 1419 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1420 | |
72865317 BP |
1421 | return error; |
1422 | } | |
1423 | ||
61e7deb1 BP |
1424 | static void |
1425 | dp_netdev_flow_free(struct dp_netdev_flow *flow) | |
1426 | { | |
61e7deb1 | 1427 | dp_netdev_actions_free(dp_netdev_flow_get_actions(flow)); |
61e7deb1 BP |
1428 | free(flow); |
1429 | } | |
1430 | ||
ed79f89a DDP |
1431 | static void dp_netdev_flow_unref(struct dp_netdev_flow *flow) |
1432 | { | |
1433 | if (ovs_refcount_unref_relaxed(&flow->ref_cnt) == 1) { | |
1434 | ovsrcu_postpone(dp_netdev_flow_free, flow); | |
1435 | } | |
1436 | } | |
1437 | ||
70e5ed6f JS |
1438 | static uint32_t |
1439 | dp_netdev_flow_hash(const ovs_u128 *ufid) | |
1440 | { | |
1441 | return ufid->u32[0]; | |
1442 | } | |
1443 | ||
72865317 | 1444 | static void |
1c1e46ed AW |
1445 | dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, |
1446 | struct dp_netdev_flow *flow) | |
1447 | OVS_REQUIRES(pmd->flow_mutex) | |
72865317 | 1448 | { |
9f361d6b | 1449 | struct cmap_node *node = CONST_CAST(struct cmap_node *, &flow->node); |
2c0ea78f | 1450 | |
1c1e46ed | 1451 | dpcls_remove(&pmd->cls, &flow->cr); |
361d808d JR |
1452 | flow->cr.mask = NULL; /* Accessing rule's mask after this is not safe. */ |
1453 | ||
1c1e46ed | 1454 | cmap_remove(&pmd->flow_table, node, dp_netdev_flow_hash(&flow->ufid)); |
9bbf1c3d | 1455 | flow->dead = true; |
ed79f89a DDP |
1456 | |
1457 | dp_netdev_flow_unref(flow); | |
72865317 BP |
1458 | } |
1459 | ||
1460 | static void | |
1c1e46ed | 1461 | dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd) |
72865317 | 1462 | { |
78c8df12 | 1463 | struct dp_netdev_flow *netdev_flow; |
72865317 | 1464 | |
1c1e46ed AW |
1465 | ovs_mutex_lock(&pmd->flow_mutex); |
1466 | CMAP_FOR_EACH (netdev_flow, node, &pmd->flow_table) { | |
1467 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); | |
72865317 | 1468 | } |
1c1e46ed | 1469 | ovs_mutex_unlock(&pmd->flow_mutex); |
72865317 BP |
1470 | } |
1471 | ||
1472 | static int | |
1473 | dpif_netdev_flow_flush(struct dpif *dpif) | |
1474 | { | |
1475 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed AW |
1476 | struct dp_netdev_pmd_thread *pmd; |
1477 | ||
1478 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1479 | dp_netdev_pmd_flow_flush(pmd); | |
1480 | } | |
5279f8fd | 1481 | |
72865317 BP |
1482 | return 0; |
1483 | } | |
1484 | ||
b0ec0f27 | 1485 | struct dp_netdev_port_state { |
59e6d833 | 1486 | struct cmap_position position; |
4c738a8d | 1487 | char *name; |
b0ec0f27 BP |
1488 | }; |
1489 | ||
1490 | static int | |
1491 | dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) | |
1492 | { | |
1493 | *statep = xzalloc(sizeof(struct dp_netdev_port_state)); | |
1494 | return 0; | |
1495 | } | |
1496 | ||
72865317 | 1497 | static int |
b0ec0f27 | 1498 | dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, |
4c738a8d | 1499 | struct dpif_port *dpif_port) |
72865317 | 1500 | { |
b0ec0f27 | 1501 | struct dp_netdev_port_state *state = state_; |
72865317 | 1502 | struct dp_netdev *dp = get_dp_netdev(dpif); |
59e6d833 | 1503 | struct cmap_node *node; |
ff073a71 | 1504 | int retval; |
72865317 | 1505 | |
59e6d833 | 1506 | node = cmap_next_position(&dp->ports, &state->position); |
ff073a71 BP |
1507 | if (node) { |
1508 | struct dp_netdev_port *port; | |
5279f8fd | 1509 | |
ff073a71 BP |
1510 | port = CONTAINER_OF(node, struct dp_netdev_port, node); |
1511 | ||
1512 | free(state->name); | |
1513 | state->name = xstrdup(netdev_get_name(port->netdev)); | |
1514 | dpif_port->name = state->name; | |
1515 | dpif_port->type = port->type; | |
35303d71 | 1516 | dpif_port->port_no = port->port_no; |
ff073a71 BP |
1517 | |
1518 | retval = 0; | |
1519 | } else { | |
1520 | retval = EOF; | |
72865317 | 1521 | } |
5279f8fd | 1522 | |
ff073a71 | 1523 | return retval; |
b0ec0f27 BP |
1524 | } |
1525 | ||
1526 | static int | |
4c738a8d | 1527 | dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) |
b0ec0f27 | 1528 | { |
4c738a8d BP |
1529 | struct dp_netdev_port_state *state = state_; |
1530 | free(state->name); | |
b0ec0f27 BP |
1531 | free(state); |
1532 | return 0; | |
72865317 BP |
1533 | } |
1534 | ||
1535 | static int | |
67a4917b | 1536 | dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) |
72865317 BP |
1537 | { |
1538 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
d33ed218 | 1539 | uint64_t new_port_seq; |
5279f8fd BP |
1540 | int error; |
1541 | ||
d33ed218 BP |
1542 | new_port_seq = seq_read(dpif->dp->port_seq); |
1543 | if (dpif->last_port_seq != new_port_seq) { | |
1544 | dpif->last_port_seq = new_port_seq; | |
5279f8fd | 1545 | error = ENOBUFS; |
72865317 | 1546 | } else { |
5279f8fd | 1547 | error = EAGAIN; |
72865317 | 1548 | } |
5279f8fd BP |
1549 | |
1550 | return error; | |
72865317 BP |
1551 | } |
1552 | ||
1553 | static void | |
1554 | dpif_netdev_port_poll_wait(const struct dpif *dpif_) | |
1555 | { | |
1556 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
5279f8fd | 1557 | |
d33ed218 | 1558 | seq_wait(dpif->dp->port_seq, dpif->last_port_seq); |
8a4e3a85 BP |
1559 | } |
1560 | ||
1561 | static struct dp_netdev_flow * | |
0de8783a | 1562 | dp_netdev_flow_cast(const struct dpcls_rule *cr) |
8a4e3a85 BP |
1563 | { |
1564 | return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL; | |
72865317 BP |
1565 | } |
1566 | ||
9bbf1c3d DDP |
1567 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *flow) |
1568 | { | |
1569 | return ovs_refcount_try_ref_rcu(&flow->ref_cnt); | |
1570 | } | |
1571 | ||
79df317f DDP |
1572 | /* netdev_flow_key utilities. |
1573 | * | |
1574 | * netdev_flow_key is basically a miniflow. We use these functions | |
1575 | * (netdev_flow_key_clone, netdev_flow_key_equal, ...) instead of the miniflow | |
1576 | * functions (miniflow_clone_inline, miniflow_equal, ...), because: | |
1577 | * | |
1578 | * - Since we are dealing exclusively with miniflows created by | |
1579 | * miniflow_extract(), if the map is different the miniflow is different. | |
1580 | * Therefore we can be faster by comparing the map and the miniflow in a | |
1581 | * single memcmp(). | |
5fcff47b | 1582 | * - These functions can be inlined by the compiler. */ |
79df317f | 1583 | |
361d808d | 1584 | /* Given the number of bits set in miniflow's maps, returns the size of the |
caeb4906 | 1585 | * 'netdev_flow_key.mf' */ |
361d808d JR |
1586 | static inline size_t |
1587 | netdev_flow_key_size(size_t flow_u64s) | |
79df317f | 1588 | { |
361d808d | 1589 | return sizeof(struct miniflow) + MINIFLOW_VALUES_SIZE(flow_u64s); |
79df317f DDP |
1590 | } |
1591 | ||
79df317f DDP |
1592 | static inline bool |
1593 | netdev_flow_key_equal(const struct netdev_flow_key *a, | |
0de8783a JR |
1594 | const struct netdev_flow_key *b) |
1595 | { | |
caeb4906 JR |
1596 | /* 'b->len' may be not set yet. */ |
1597 | return a->hash == b->hash && !memcmp(&a->mf, &b->mf, a->len); | |
0de8783a JR |
1598 | } |
1599 | ||
1600 | /* Used to compare 'netdev_flow_key' in the exact match cache to a miniflow. | |
1601 | * The maps are compared bitwise, so both 'key->mf' 'mf' must have been | |
1602 | * generated by miniflow_extract. */ | |
1603 | static inline bool | |
1604 | netdev_flow_key_equal_mf(const struct netdev_flow_key *key, | |
1605 | const struct miniflow *mf) | |
79df317f | 1606 | { |
caeb4906 | 1607 | return !memcmp(&key->mf, mf, key->len); |
79df317f DDP |
1608 | } |
1609 | ||
1610 | static inline void | |
1611 | netdev_flow_key_clone(struct netdev_flow_key *dst, | |
0de8783a JR |
1612 | const struct netdev_flow_key *src) |
1613 | { | |
caeb4906 JR |
1614 | memcpy(dst, src, |
1615 | offsetof(struct netdev_flow_key, mf) + src->len); | |
0de8783a JR |
1616 | } |
1617 | ||
1618 | /* Slow. */ | |
1619 | static void | |
1620 | netdev_flow_key_from_flow(struct netdev_flow_key *dst, | |
1621 | const struct flow *src) | |
1622 | { | |
cf62fa4c | 1623 | struct dp_packet packet; |
0de8783a | 1624 | uint64_t buf_stub[512 / 8]; |
0de8783a | 1625 | |
cf62fa4c PS |
1626 | dp_packet_use_stub(&packet, buf_stub, sizeof buf_stub); |
1627 | pkt_metadata_from_flow(&packet.md, src); | |
0de8783a | 1628 | flow_compose(&packet, src); |
cf62fa4c PS |
1629 | miniflow_extract(&packet, &dst->mf); |
1630 | dp_packet_uninit(&packet); | |
0de8783a | 1631 | |
361d808d | 1632 | dst->len = netdev_flow_key_size(miniflow_n_values(&dst->mf)); |
0de8783a JR |
1633 | dst->hash = 0; /* Not computed yet. */ |
1634 | } | |
1635 | ||
1636 | /* Initialize a netdev_flow_key 'mask' from 'match'. */ | |
1637 | static inline void | |
1638 | netdev_flow_mask_init(struct netdev_flow_key *mask, | |
1639 | const struct match *match) | |
1640 | { | |
09b0fa9c | 1641 | uint64_t *dst = miniflow_values(&mask->mf); |
5fcff47b | 1642 | struct flowmap fmap; |
0de8783a | 1643 | uint32_t hash = 0; |
5fcff47b | 1644 | size_t idx; |
0de8783a JR |
1645 | |
1646 | /* Only check masks that make sense for the flow. */ | |
5fcff47b JR |
1647 | flow_wc_map(&match->flow, &fmap); |
1648 | flowmap_init(&mask->mf.map); | |
0de8783a | 1649 | |
5fcff47b JR |
1650 | FLOWMAP_FOR_EACH_INDEX(idx, fmap) { |
1651 | uint64_t mask_u64 = flow_u64_value(&match->wc.masks, idx); | |
0de8783a | 1652 | |
5fcff47b JR |
1653 | if (mask_u64) { |
1654 | flowmap_set(&mask->mf.map, idx, 1); | |
1655 | *dst++ = mask_u64; | |
1656 | hash = hash_add64(hash, mask_u64); | |
0de8783a | 1657 | } |
0de8783a JR |
1658 | } |
1659 | ||
5fcff47b | 1660 | map_t map; |
0de8783a | 1661 | |
5fcff47b JR |
1662 | FLOWMAP_FOR_EACH_MAP (map, mask->mf.map) { |
1663 | hash = hash_add64(hash, map); | |
1664 | } | |
0de8783a | 1665 | |
5fcff47b | 1666 | size_t n = dst - miniflow_get_values(&mask->mf); |
0de8783a | 1667 | |
d70e8c28 | 1668 | mask->hash = hash_finish(hash, n * 8); |
0de8783a JR |
1669 | mask->len = netdev_flow_key_size(n); |
1670 | } | |
1671 | ||
361d808d | 1672 | /* Initializes 'dst' as a copy of 'flow' masked with 'mask'. */ |
0de8783a JR |
1673 | static inline void |
1674 | netdev_flow_key_init_masked(struct netdev_flow_key *dst, | |
1675 | const struct flow *flow, | |
1676 | const struct netdev_flow_key *mask) | |
79df317f | 1677 | { |
09b0fa9c JR |
1678 | uint64_t *dst_u64 = miniflow_values(&dst->mf); |
1679 | const uint64_t *mask_u64 = miniflow_get_values(&mask->mf); | |
0de8783a | 1680 | uint32_t hash = 0; |
d70e8c28 | 1681 | uint64_t value; |
0de8783a JR |
1682 | |
1683 | dst->len = mask->len; | |
361d808d | 1684 | dst->mf = mask->mf; /* Copy maps. */ |
0de8783a | 1685 | |
5fcff47b | 1686 | FLOW_FOR_EACH_IN_MAPS(value, flow, mask->mf.map) { |
d70e8c28 JR |
1687 | *dst_u64 = value & *mask_u64++; |
1688 | hash = hash_add64(hash, *dst_u64++); | |
0de8783a | 1689 | } |
09b0fa9c JR |
1690 | dst->hash = hash_finish(hash, |
1691 | (dst_u64 - miniflow_get_values(&dst->mf)) * 8); | |
0de8783a JR |
1692 | } |
1693 | ||
5fcff47b JR |
1694 | /* Iterate through netdev_flow_key TNL u64 values specified by 'FLOWMAP'. */ |
1695 | #define NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(VALUE, KEY, FLOWMAP) \ | |
1696 | MINIFLOW_FOR_EACH_IN_FLOWMAP(VALUE, &(KEY)->mf, FLOWMAP) | |
0de8783a JR |
1697 | |
1698 | /* Returns a hash value for the bits of 'key' where there are 1-bits in | |
1699 | * 'mask'. */ | |
1700 | static inline uint32_t | |
1701 | netdev_flow_key_hash_in_mask(const struct netdev_flow_key *key, | |
1702 | const struct netdev_flow_key *mask) | |
1703 | { | |
09b0fa9c | 1704 | const uint64_t *p = miniflow_get_values(&mask->mf); |
0de8783a | 1705 | uint32_t hash = 0; |
5fcff47b | 1706 | uint64_t value; |
0de8783a | 1707 | |
5fcff47b JR |
1708 | NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(value, key, mask->mf.map) { |
1709 | hash = hash_add64(hash, value & *p++); | |
0de8783a JR |
1710 | } |
1711 | ||
09b0fa9c | 1712 | return hash_finish(hash, (p - miniflow_get_values(&mask->mf)) * 8); |
79df317f DDP |
1713 | } |
1714 | ||
9bbf1c3d DDP |
1715 | static inline bool |
1716 | emc_entry_alive(struct emc_entry *ce) | |
1717 | { | |
1718 | return ce->flow && !ce->flow->dead; | |
1719 | } | |
1720 | ||
1721 | static void | |
1722 | emc_clear_entry(struct emc_entry *ce) | |
1723 | { | |
1724 | if (ce->flow) { | |
1725 | dp_netdev_flow_unref(ce->flow); | |
1726 | ce->flow = NULL; | |
1727 | } | |
1728 | } | |
1729 | ||
1730 | static inline void | |
1731 | emc_change_entry(struct emc_entry *ce, struct dp_netdev_flow *flow, | |
0de8783a | 1732 | const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1733 | { |
1734 | if (ce->flow != flow) { | |
1735 | if (ce->flow) { | |
1736 | dp_netdev_flow_unref(ce->flow); | |
1737 | } | |
1738 | ||
1739 | if (dp_netdev_flow_ref(flow)) { | |
1740 | ce->flow = flow; | |
1741 | } else { | |
1742 | ce->flow = NULL; | |
1743 | } | |
1744 | } | |
0de8783a JR |
1745 | if (key) { |
1746 | netdev_flow_key_clone(&ce->key, key); | |
9bbf1c3d DDP |
1747 | } |
1748 | } | |
1749 | ||
1750 | static inline void | |
0de8783a | 1751 | emc_insert(struct emc_cache *cache, const struct netdev_flow_key *key, |
9bbf1c3d DDP |
1752 | struct dp_netdev_flow *flow) |
1753 | { | |
1754 | struct emc_entry *to_be_replaced = NULL; | |
1755 | struct emc_entry *current_entry; | |
1756 | ||
0de8783a JR |
1757 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1758 | if (netdev_flow_key_equal(¤t_entry->key, key)) { | |
9bbf1c3d | 1759 | /* We found the entry with the 'mf' miniflow */ |
0de8783a | 1760 | emc_change_entry(current_entry, flow, NULL); |
9bbf1c3d DDP |
1761 | return; |
1762 | } | |
1763 | ||
1764 | /* Replacement policy: put the flow in an empty (not alive) entry, or | |
1765 | * in the first entry where it can be */ | |
1766 | if (!to_be_replaced | |
1767 | || (emc_entry_alive(to_be_replaced) | |
1768 | && !emc_entry_alive(current_entry)) | |
0de8783a | 1769 | || current_entry->key.hash < to_be_replaced->key.hash) { |
9bbf1c3d DDP |
1770 | to_be_replaced = current_entry; |
1771 | } | |
1772 | } | |
1773 | /* We didn't find the miniflow in the cache. | |
1774 | * The 'to_be_replaced' entry is where the new flow will be stored */ | |
1775 | ||
0de8783a | 1776 | emc_change_entry(to_be_replaced, flow, key); |
9bbf1c3d DDP |
1777 | } |
1778 | ||
1779 | static inline struct dp_netdev_flow * | |
0de8783a | 1780 | emc_lookup(struct emc_cache *cache, const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1781 | { |
1782 | struct emc_entry *current_entry; | |
1783 | ||
0de8783a JR |
1784 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1785 | if (current_entry->key.hash == key->hash | |
1786 | && emc_entry_alive(current_entry) | |
1787 | && netdev_flow_key_equal_mf(¤t_entry->key, &key->mf)) { | |
9bbf1c3d | 1788 | |
0de8783a | 1789 | /* We found the entry with the 'key->mf' miniflow */ |
9bbf1c3d DDP |
1790 | return current_entry->flow; |
1791 | } | |
1792 | } | |
1793 | ||
1794 | return NULL; | |
1795 | } | |
1796 | ||
72865317 | 1797 | static struct dp_netdev_flow * |
1c1e46ed AW |
1798 | dp_netdev_pmd_lookup_flow(const struct dp_netdev_pmd_thread *pmd, |
1799 | const struct netdev_flow_key *key) | |
2c0ea78f | 1800 | { |
8a4e3a85 | 1801 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 1802 | struct dpcls_rule *rule; |
2c0ea78f | 1803 | |
1c1e46ed | 1804 | dpcls_lookup(&pmd->cls, key, &rule, 1); |
4f150744 | 1805 | netdev_flow = dp_netdev_flow_cast(rule); |
2c0ea78f | 1806 | |
8a4e3a85 | 1807 | return netdev_flow; |
2c0ea78f GS |
1808 | } |
1809 | ||
1810 | static struct dp_netdev_flow * | |
1c1e46ed AW |
1811 | dp_netdev_pmd_find_flow(const struct dp_netdev_pmd_thread *pmd, |
1812 | const ovs_u128 *ufidp, const struct nlattr *key, | |
1813 | size_t key_len) | |
72865317 | 1814 | { |
1763b4b8 | 1815 | struct dp_netdev_flow *netdev_flow; |
70e5ed6f JS |
1816 | struct flow flow; |
1817 | ovs_u128 ufid; | |
1818 | ||
1819 | /* If a UFID is not provided, determine one based on the key. */ | |
1820 | if (!ufidp && key && key_len | |
1821 | && !dpif_netdev_flow_from_nlattrs(key, key_len, &flow)) { | |
1c1e46ed | 1822 | dpif_flow_hash(pmd->dp->dpif, &flow, sizeof flow, &ufid); |
70e5ed6f JS |
1823 | ufidp = &ufid; |
1824 | } | |
72865317 | 1825 | |
70e5ed6f JS |
1826 | if (ufidp) { |
1827 | CMAP_FOR_EACH_WITH_HASH (netdev_flow, node, dp_netdev_flow_hash(ufidp), | |
1c1e46ed | 1828 | &pmd->flow_table) { |
bdd7ecf5 | 1829 | if (ovs_u128_equals(&netdev_flow->ufid, ufidp)) { |
70e5ed6f JS |
1830 | return netdev_flow; |
1831 | } | |
72865317 BP |
1832 | } |
1833 | } | |
8a4e3a85 | 1834 | |
72865317 BP |
1835 | return NULL; |
1836 | } | |
1837 | ||
1838 | static void | |
eb94da30 | 1839 | get_dpif_flow_stats(const struct dp_netdev_flow *netdev_flow_, |
1763b4b8 | 1840 | struct dpif_flow_stats *stats) |
feebdea2 | 1841 | { |
eb94da30 DDP |
1842 | struct dp_netdev_flow *netdev_flow; |
1843 | unsigned long long n; | |
1844 | long long used; | |
1845 | uint16_t flags; | |
1846 | ||
1847 | netdev_flow = CONST_CAST(struct dp_netdev_flow *, netdev_flow_); | |
1848 | ||
1849 | atomic_read_relaxed(&netdev_flow->stats.packet_count, &n); | |
1850 | stats->n_packets = n; | |
1851 | atomic_read_relaxed(&netdev_flow->stats.byte_count, &n); | |
1852 | stats->n_bytes = n; | |
1853 | atomic_read_relaxed(&netdev_flow->stats.used, &used); | |
1854 | stats->used = used; | |
1855 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
1856 | stats->tcp_flags = flags; | |
72865317 BP |
1857 | } |
1858 | ||
7af12bd7 JS |
1859 | /* Converts to the dpif_flow format, using 'key_buf' and 'mask_buf' for |
1860 | * storing the netlink-formatted key/mask. 'key_buf' may be the same as | |
1861 | * 'mask_buf'. Actions will be returned without copying, by relying on RCU to | |
1862 | * protect them. */ | |
6fe09f8c | 1863 | static void |
70e5ed6f | 1864 | dp_netdev_flow_to_dpif_flow(const struct dp_netdev_flow *netdev_flow, |
7af12bd7 | 1865 | struct ofpbuf *key_buf, struct ofpbuf *mask_buf, |
64bb477f | 1866 | struct dpif_flow *flow, bool terse) |
6fe09f8c | 1867 | { |
64bb477f JS |
1868 | if (terse) { |
1869 | memset(flow, 0, sizeof *flow); | |
1870 | } else { | |
1871 | struct flow_wildcards wc; | |
1872 | struct dp_netdev_actions *actions; | |
1873 | size_t offset; | |
5262eea1 JG |
1874 | struct odp_flow_key_parms odp_parms = { |
1875 | .flow = &netdev_flow->flow, | |
1876 | .mask = &wc.masks, | |
2494ccd7 | 1877 | .support = dp_netdev_support, |
5262eea1 | 1878 | }; |
64bb477f JS |
1879 | |
1880 | miniflow_expand(&netdev_flow->cr.mask->mf, &wc.masks); | |
1881 | ||
1882 | /* Key */ | |
6fd6ed71 | 1883 | offset = key_buf->size; |
64bb477f | 1884 | flow->key = ofpbuf_tail(key_buf); |
5262eea1 JG |
1885 | odp_parms.odp_in_port = netdev_flow->flow.in_port.odp_port; |
1886 | odp_flow_key_from_flow(&odp_parms, key_buf); | |
6fd6ed71 | 1887 | flow->key_len = key_buf->size - offset; |
64bb477f JS |
1888 | |
1889 | /* Mask */ | |
6fd6ed71 | 1890 | offset = mask_buf->size; |
64bb477f | 1891 | flow->mask = ofpbuf_tail(mask_buf); |
5262eea1 | 1892 | odp_parms.odp_in_port = wc.masks.in_port.odp_port; |
ec1f6f32 | 1893 | odp_parms.key_buf = key_buf; |
5262eea1 | 1894 | odp_flow_key_from_mask(&odp_parms, mask_buf); |
6fd6ed71 | 1895 | flow->mask_len = mask_buf->size - offset; |
64bb477f JS |
1896 | |
1897 | /* Actions */ | |
1898 | actions = dp_netdev_flow_get_actions(netdev_flow); | |
1899 | flow->actions = actions->actions; | |
1900 | flow->actions_len = actions->size; | |
1901 | } | |
6fe09f8c | 1902 | |
70e5ed6f JS |
1903 | flow->ufid = netdev_flow->ufid; |
1904 | flow->ufid_present = true; | |
1c1e46ed | 1905 | flow->pmd_id = netdev_flow->pmd_id; |
6fe09f8c JS |
1906 | get_dpif_flow_stats(netdev_flow, &flow->stats); |
1907 | } | |
1908 | ||
36956a7d | 1909 | static int |
8c301900 JR |
1910 | dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len, |
1911 | const struct nlattr *mask_key, | |
1912 | uint32_t mask_key_len, const struct flow *flow, | |
9f861c91 | 1913 | struct flow_wildcards *wc) |
8c301900 | 1914 | { |
ca8d3442 DDP |
1915 | enum odp_key_fitness fitness; |
1916 | ||
1917 | fitness = odp_flow_key_to_mask_udpif(mask_key, mask_key_len, key, | |
1918 | key_len, wc, flow); | |
1919 | if (fitness) { | |
1920 | /* This should not happen: it indicates that | |
1921 | * odp_flow_key_from_mask() and odp_flow_key_to_mask() | |
1922 | * disagree on the acceptable form of a mask. Log the problem | |
1923 | * as an error, with enough details to enable debugging. */ | |
1924 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1925 | ||
1926 | if (!VLOG_DROP_ERR(&rl)) { | |
1927 | struct ds s; | |
8c301900 | 1928 | |
ca8d3442 DDP |
1929 | ds_init(&s); |
1930 | odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s, | |
1931 | true); | |
1932 | VLOG_ERR("internal error parsing flow mask %s (%s)", | |
1933 | ds_cstr(&s), odp_key_fitness_to_string(fitness)); | |
1934 | ds_destroy(&s); | |
8c301900 | 1935 | } |
ca8d3442 DDP |
1936 | |
1937 | return EINVAL; | |
8c301900 JR |
1938 | } |
1939 | ||
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | static int | |
1944 | dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, | |
1945 | struct flow *flow) | |
36956a7d | 1946 | { |
586ddea5 BP |
1947 | odp_port_t in_port; |
1948 | ||
6728d578 | 1949 | if (odp_flow_key_to_flow_udpif(key, key_len, flow)) { |
36956a7d | 1950 | /* This should not happen: it indicates that odp_flow_key_from_flow() |
8c301900 JR |
1951 | * and odp_flow_key_to_flow() disagree on the acceptable form of a |
1952 | * flow. Log the problem as an error, with enough details to enable | |
1953 | * debugging. */ | |
36956a7d BP |
1954 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
1955 | ||
1956 | if (!VLOG_DROP_ERR(&rl)) { | |
1957 | struct ds s; | |
1958 | ||
1959 | ds_init(&s); | |
8c301900 | 1960 | odp_flow_format(key, key_len, NULL, 0, NULL, &s, true); |
36956a7d BP |
1961 | VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s)); |
1962 | ds_destroy(&s); | |
1963 | } | |
1964 | ||
1965 | return EINVAL; | |
1966 | } | |
1967 | ||
586ddea5 BP |
1968 | in_port = flow->in_port.odp_port; |
1969 | if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) { | |
18886b60 BP |
1970 | return EINVAL; |
1971 | } | |
1972 | ||
07659514 | 1973 | /* Userspace datapath doesn't support conntrack. */ |
9daf2348 JS |
1974 | if (flow->ct_state || flow->ct_zone || flow->ct_mark |
1975 | || !ovs_u128_is_zero(&flow->ct_label)) { | |
07659514 JS |
1976 | return EINVAL; |
1977 | } | |
1978 | ||
36956a7d BP |
1979 | return 0; |
1980 | } | |
1981 | ||
72865317 | 1982 | static int |
6fe09f8c | 1983 | dpif_netdev_flow_get(const struct dpif *dpif, const struct dpif_flow_get *get) |
72865317 BP |
1984 | { |
1985 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 1986 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 1987 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
1988 | unsigned pmd_id = get->pmd_id == PMD_ID_NULL |
1989 | ? NON_PMD_CORE_ID : get->pmd_id; | |
70e5ed6f | 1990 | int error = 0; |
8a4e3a85 | 1991 | |
1c1e46ed AW |
1992 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
1993 | if (!pmd) { | |
1994 | return EINVAL; | |
1995 | } | |
1996 | ||
1997 | netdev_flow = dp_netdev_pmd_find_flow(pmd, get->ufid, get->key, | |
1998 | get->key_len); | |
1763b4b8 | 1999 | if (netdev_flow) { |
70e5ed6f | 2000 | dp_netdev_flow_to_dpif_flow(netdev_flow, get->buffer, get->buffer, |
64bb477f | 2001 | get->flow, false); |
70e5ed6f | 2002 | } else { |
5279f8fd | 2003 | error = ENOENT; |
72865317 | 2004 | } |
1c1e46ed AW |
2005 | dp_netdev_pmd_unref(pmd); |
2006 | ||
bc4a05c6 | 2007 | |
5279f8fd | 2008 | return error; |
72865317 BP |
2009 | } |
2010 | ||
0de8783a | 2011 | static struct dp_netdev_flow * |
1c1e46ed AW |
2012 | dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd, |
2013 | struct match *match, const ovs_u128 *ufid, | |
ae2ceebd | 2014 | const struct nlattr *actions, size_t actions_len) |
1c1e46ed | 2015 | OVS_REQUIRES(pmd->flow_mutex) |
72865317 | 2016 | { |
0de8783a JR |
2017 | struct dp_netdev_flow *flow; |
2018 | struct netdev_flow_key mask; | |
ed79f89a | 2019 | |
0de8783a JR |
2020 | netdev_flow_mask_init(&mask, match); |
2021 | /* Make sure wc does not have metadata. */ | |
5fcff47b JR |
2022 | ovs_assert(!FLOWMAP_HAS_FIELD(&mask.mf.map, metadata) |
2023 | && !FLOWMAP_HAS_FIELD(&mask.mf.map, regs)); | |
679ba04c | 2024 | |
0de8783a | 2025 | /* Do not allocate extra space. */ |
caeb4906 | 2026 | flow = xmalloc(sizeof *flow - sizeof flow->cr.flow.mf + mask.len); |
1c1e46ed | 2027 | memset(&flow->stats, 0, sizeof flow->stats); |
0de8783a | 2028 | flow->dead = false; |
11e5cf1f | 2029 | flow->batch = NULL; |
bd5131ba | 2030 | *CONST_CAST(unsigned *, &flow->pmd_id) = pmd->core_id; |
0de8783a | 2031 | *CONST_CAST(struct flow *, &flow->flow) = match->flow; |
70e5ed6f | 2032 | *CONST_CAST(ovs_u128 *, &flow->ufid) = *ufid; |
0de8783a | 2033 | ovs_refcount_init(&flow->ref_cnt); |
0de8783a | 2034 | ovsrcu_set(&flow->actions, dp_netdev_actions_create(actions, actions_len)); |
2c0ea78f | 2035 | |
0de8783a | 2036 | netdev_flow_key_init_masked(&flow->cr.flow, &match->flow, &mask); |
1c1e46ed | 2037 | dpcls_insert(&pmd->cls, &flow->cr, &mask); |
72865317 | 2038 | |
4c75aaab EJ |
2039 | cmap_insert(&pmd->flow_table, CONST_CAST(struct cmap_node *, &flow->node), |
2040 | dp_netdev_flow_hash(&flow->ufid)); | |
2041 | ||
623540e4 | 2042 | if (OVS_UNLIKELY(VLOG_IS_DBG_ENABLED())) { |
0de8783a | 2043 | struct match match; |
623540e4 EJ |
2044 | struct ds ds = DS_EMPTY_INITIALIZER; |
2045 | ||
4d8f90b1 | 2046 | match.tun_md.valid = false; |
0de8783a JR |
2047 | match.flow = flow->flow; |
2048 | miniflow_expand(&flow->cr.mask->mf, &match.wc.masks); | |
2049 | ||
623540e4 | 2050 | ds_put_cstr(&ds, "flow_add: "); |
70e5ed6f JS |
2051 | odp_format_ufid(ufid, &ds); |
2052 | ds_put_cstr(&ds, " "); | |
0de8783a | 2053 | match_format(&match, &ds, OFP_DEFAULT_PRIORITY); |
623540e4 EJ |
2054 | ds_put_cstr(&ds, ", actions:"); |
2055 | format_odp_actions(&ds, actions, actions_len); | |
2056 | ||
2057 | VLOG_DBG_RL(&upcall_rl, "%s", ds_cstr(&ds)); | |
2058 | ||
2059 | ds_destroy(&ds); | |
2060 | } | |
2061 | ||
0de8783a | 2062 | return flow; |
72865317 BP |
2063 | } |
2064 | ||
72865317 | 2065 | static int |
89625d1e | 2066 | dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put) |
72865317 BP |
2067 | { |
2068 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2069 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 2070 | struct netdev_flow_key key; |
1c1e46ed | 2071 | struct dp_netdev_pmd_thread *pmd; |
ae2ceebd | 2072 | struct match match; |
70e5ed6f | 2073 | ovs_u128 ufid; |
bd5131ba DDP |
2074 | unsigned pmd_id = put->pmd_id == PMD_ID_NULL |
2075 | ? NON_PMD_CORE_ID : put->pmd_id; | |
36956a7d BP |
2076 | int error; |
2077 | ||
ae2ceebd | 2078 | error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &match.flow); |
8c301900 JR |
2079 | if (error) { |
2080 | return error; | |
2081 | } | |
2082 | error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len, | |
2083 | put->mask, put->mask_len, | |
9f861c91 | 2084 | &match.flow, &match.wc); |
36956a7d BP |
2085 | if (error) { |
2086 | return error; | |
2087 | } | |
0de8783a | 2088 | |
1c1e46ed AW |
2089 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2090 | if (!pmd) { | |
2091 | return EINVAL; | |
2092 | } | |
2093 | ||
0de8783a JR |
2094 | /* Must produce a netdev_flow_key for lookup. |
2095 | * This interface is no longer performance critical, since it is not used | |
2096 | * for upcall processing any more. */ | |
2097 | netdev_flow_key_from_flow(&key, &match.flow); | |
72865317 | 2098 | |
70e5ed6f JS |
2099 | if (put->ufid) { |
2100 | ufid = *put->ufid; | |
2101 | } else { | |
2102 | dpif_flow_hash(dpif, &match.flow, sizeof match.flow, &ufid); | |
2103 | } | |
2104 | ||
1c1e46ed AW |
2105 | ovs_mutex_lock(&pmd->flow_mutex); |
2106 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &key); | |
1763b4b8 | 2107 | if (!netdev_flow) { |
89625d1e | 2108 | if (put->flags & DPIF_FP_CREATE) { |
1c1e46ed | 2109 | if (cmap_count(&pmd->flow_table) < MAX_FLOWS) { |
89625d1e BP |
2110 | if (put->stats) { |
2111 | memset(put->stats, 0, sizeof *put->stats); | |
feebdea2 | 2112 | } |
1c1e46ed | 2113 | dp_netdev_flow_add(pmd, &match, &ufid, put->actions, |
70e5ed6f | 2114 | put->actions_len); |
0de8783a | 2115 | error = 0; |
72865317 | 2116 | } else { |
5279f8fd | 2117 | error = EFBIG; |
72865317 BP |
2118 | } |
2119 | } else { | |
5279f8fd | 2120 | error = ENOENT; |
72865317 BP |
2121 | } |
2122 | } else { | |
2c0ea78f | 2123 | if (put->flags & DPIF_FP_MODIFY |
ae2ceebd | 2124 | && flow_equal(&match.flow, &netdev_flow->flow)) { |
8a4e3a85 BP |
2125 | struct dp_netdev_actions *new_actions; |
2126 | struct dp_netdev_actions *old_actions; | |
2127 | ||
2128 | new_actions = dp_netdev_actions_create(put->actions, | |
2129 | put->actions_len); | |
2130 | ||
61e7deb1 BP |
2131 | old_actions = dp_netdev_flow_get_actions(netdev_flow); |
2132 | ovsrcu_set(&netdev_flow->actions, new_actions); | |
679ba04c | 2133 | |
a84cb64a BP |
2134 | if (put->stats) { |
2135 | get_dpif_flow_stats(netdev_flow, put->stats); | |
2136 | } | |
2137 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
97447f55 DDP |
2138 | /* XXX: The userspace datapath uses thread local statistics |
2139 | * (for flows), which should be updated only by the owning | |
2140 | * thread. Since we cannot write on stats memory here, | |
2141 | * we choose not to support this flag. Please note: | |
2142 | * - This feature is currently used only by dpctl commands with | |
2143 | * option --clear. | |
2144 | * - Should the need arise, this operation can be implemented | |
2145 | * by keeping a base value (to be update here) for each | |
2146 | * counter, and subtracting it before outputting the stats */ | |
2147 | error = EOPNOTSUPP; | |
72865317 | 2148 | } |
8a4e3a85 | 2149 | |
61e7deb1 | 2150 | ovsrcu_postpone(dp_netdev_actions_free, old_actions); |
2c0ea78f | 2151 | } else if (put->flags & DPIF_FP_CREATE) { |
5279f8fd | 2152 | error = EEXIST; |
2c0ea78f GS |
2153 | } else { |
2154 | /* Overlapping flow. */ | |
2155 | error = EINVAL; | |
72865317 BP |
2156 | } |
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 | ||
72865317 | 2164 | static int |
b99d3cee | 2165 | dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del) |
72865317 BP |
2166 | { |
2167 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2168 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 2169 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
2170 | unsigned pmd_id = del->pmd_id == PMD_ID_NULL |
2171 | ? NON_PMD_CORE_ID : del->pmd_id; | |
70e5ed6f | 2172 | int error = 0; |
72865317 | 2173 | |
1c1e46ed AW |
2174 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2175 | if (!pmd) { | |
2176 | return EINVAL; | |
2177 | } | |
2178 | ||
2179 | ovs_mutex_lock(&pmd->flow_mutex); | |
2180 | netdev_flow = dp_netdev_pmd_find_flow(pmd, del->ufid, del->key, | |
2181 | del->key_len); | |
1763b4b8 | 2182 | if (netdev_flow) { |
b99d3cee | 2183 | if (del->stats) { |
1763b4b8 | 2184 | get_dpif_flow_stats(netdev_flow, del->stats); |
feebdea2 | 2185 | } |
1c1e46ed | 2186 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); |
72865317 | 2187 | } else { |
5279f8fd | 2188 | error = ENOENT; |
72865317 | 2189 | } |
1c1e46ed AW |
2190 | ovs_mutex_unlock(&pmd->flow_mutex); |
2191 | dp_netdev_pmd_unref(pmd); | |
5279f8fd BP |
2192 | |
2193 | return error; | |
72865317 BP |
2194 | } |
2195 | ||
ac64794a BP |
2196 | struct dpif_netdev_flow_dump { |
2197 | struct dpif_flow_dump up; | |
1c1e46ed AW |
2198 | struct cmap_position poll_thread_pos; |
2199 | struct cmap_position flow_pos; | |
2200 | struct dp_netdev_pmd_thread *cur_pmd; | |
d2ad7ef1 JS |
2201 | int status; |
2202 | struct ovs_mutex mutex; | |
e723fd32 JS |
2203 | }; |
2204 | ||
ac64794a BP |
2205 | static struct dpif_netdev_flow_dump * |
2206 | dpif_netdev_flow_dump_cast(struct dpif_flow_dump *dump) | |
72865317 | 2207 | { |
ac64794a | 2208 | return CONTAINER_OF(dump, struct dpif_netdev_flow_dump, up); |
e723fd32 JS |
2209 | } |
2210 | ||
ac64794a | 2211 | static struct dpif_flow_dump * |
64bb477f | 2212 | dpif_netdev_flow_dump_create(const struct dpif *dpif_, bool terse) |
e723fd32 | 2213 | { |
ac64794a | 2214 | struct dpif_netdev_flow_dump *dump; |
e723fd32 | 2215 | |
1c1e46ed | 2216 | dump = xzalloc(sizeof *dump); |
ac64794a | 2217 | dpif_flow_dump_init(&dump->up, dpif_); |
64bb477f | 2218 | dump->up.terse = terse; |
ac64794a BP |
2219 | ovs_mutex_init(&dump->mutex); |
2220 | ||
2221 | return &dump->up; | |
e723fd32 JS |
2222 | } |
2223 | ||
2224 | static int | |
ac64794a | 2225 | dpif_netdev_flow_dump_destroy(struct dpif_flow_dump *dump_) |
e723fd32 | 2226 | { |
ac64794a | 2227 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); |
e723fd32 | 2228 | |
ac64794a BP |
2229 | ovs_mutex_destroy(&dump->mutex); |
2230 | free(dump); | |
704a1e09 BP |
2231 | return 0; |
2232 | } | |
2233 | ||
ac64794a BP |
2234 | struct dpif_netdev_flow_dump_thread { |
2235 | struct dpif_flow_dump_thread up; | |
2236 | struct dpif_netdev_flow_dump *dump; | |
8bb113da RW |
2237 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; |
2238 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
2239 | }; |
2240 | ||
2241 | static struct dpif_netdev_flow_dump_thread * | |
2242 | dpif_netdev_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
2243 | { | |
2244 | return CONTAINER_OF(thread, struct dpif_netdev_flow_dump_thread, up); | |
2245 | } | |
2246 | ||
2247 | static struct dpif_flow_dump_thread * | |
2248 | dpif_netdev_flow_dump_thread_create(struct dpif_flow_dump *dump_) | |
2249 | { | |
2250 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); | |
2251 | struct dpif_netdev_flow_dump_thread *thread; | |
2252 | ||
2253 | thread = xmalloc(sizeof *thread); | |
2254 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
2255 | thread->dump = dump; | |
2256 | return &thread->up; | |
2257 | } | |
2258 | ||
2259 | static void | |
2260 | dpif_netdev_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) | |
2261 | { | |
2262 | struct dpif_netdev_flow_dump_thread *thread | |
2263 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2264 | ||
2265 | free(thread); | |
2266 | } | |
2267 | ||
704a1e09 | 2268 | static int |
ac64794a | 2269 | dpif_netdev_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
8bb113da | 2270 | struct dpif_flow *flows, int max_flows) |
ac64794a BP |
2271 | { |
2272 | struct dpif_netdev_flow_dump_thread *thread | |
2273 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2274 | struct dpif_netdev_flow_dump *dump = thread->dump; | |
8bb113da | 2275 | struct dp_netdev_flow *netdev_flows[FLOW_DUMP_MAX_BATCH]; |
8bb113da RW |
2276 | int n_flows = 0; |
2277 | int i; | |
14608a15 | 2278 | |
ac64794a | 2279 | ovs_mutex_lock(&dump->mutex); |
8bb113da | 2280 | if (!dump->status) { |
1c1e46ed AW |
2281 | struct dpif_netdev *dpif = dpif_netdev_cast(thread->up.dpif); |
2282 | struct dp_netdev *dp = get_dp_netdev(&dpif->dpif); | |
2283 | struct dp_netdev_pmd_thread *pmd = dump->cur_pmd; | |
2284 | int flow_limit = MIN(max_flows, FLOW_DUMP_MAX_BATCH); | |
2285 | ||
2286 | /* First call to dump_next(), extracts the first pmd thread. | |
2287 | * If there is no pmd thread, returns immediately. */ | |
2288 | if (!pmd) { | |
2289 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2290 | if (!pmd) { | |
2291 | ovs_mutex_unlock(&dump->mutex); | |
2292 | return n_flows; | |
8bb113da | 2293 | |
8bb113da | 2294 | } |
d2ad7ef1 | 2295 | } |
1c1e46ed AW |
2296 | |
2297 | do { | |
2298 | for (n_flows = 0; n_flows < flow_limit; n_flows++) { | |
2299 | struct cmap_node *node; | |
2300 | ||
2301 | node = cmap_next_position(&pmd->flow_table, &dump->flow_pos); | |
2302 | if (!node) { | |
2303 | break; | |
2304 | } | |
2305 | netdev_flows[n_flows] = CONTAINER_OF(node, | |
2306 | struct dp_netdev_flow, | |
2307 | node); | |
2308 | } | |
2309 | /* When finishing dumping the current pmd thread, moves to | |
2310 | * the next. */ | |
2311 | if (n_flows < flow_limit) { | |
2312 | memset(&dump->flow_pos, 0, sizeof dump->flow_pos); | |
2313 | dp_netdev_pmd_unref(pmd); | |
2314 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2315 | if (!pmd) { | |
2316 | dump->status = EOF; | |
2317 | break; | |
2318 | } | |
2319 | } | |
2320 | /* Keeps the reference to next caller. */ | |
2321 | dump->cur_pmd = pmd; | |
2322 | ||
2323 | /* If the current dump is empty, do not exit the loop, since the | |
2324 | * remaining pmds could have flows to be dumped. Just dumps again | |
2325 | * on the new 'pmd'. */ | |
2326 | } while (!n_flows); | |
8a4e3a85 | 2327 | } |
ac64794a | 2328 | ovs_mutex_unlock(&dump->mutex); |
ac64794a | 2329 | |
8bb113da RW |
2330 | for (i = 0; i < n_flows; i++) { |
2331 | struct odputil_keybuf *maskbuf = &thread->maskbuf[i]; | |
2332 | struct odputil_keybuf *keybuf = &thread->keybuf[i]; | |
2333 | struct dp_netdev_flow *netdev_flow = netdev_flows[i]; | |
2334 | struct dpif_flow *f = &flows[i]; | |
7af12bd7 | 2335 | struct ofpbuf key, mask; |
8bb113da | 2336 | |
7af12bd7 JS |
2337 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); |
2338 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
64bb477f JS |
2339 | dp_netdev_flow_to_dpif_flow(netdev_flow, &key, &mask, f, |
2340 | dump->up.terse); | |
8bb113da | 2341 | } |
feebdea2 | 2342 | |
8bb113da | 2343 | return n_flows; |
72865317 BP |
2344 | } |
2345 | ||
2346 | static int | |
758c456d | 2347 | dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute) |
65f13b50 | 2348 | OVS_NO_THREAD_SAFETY_ANALYSIS |
72865317 BP |
2349 | { |
2350 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
65f13b50 | 2351 | struct dp_netdev_pmd_thread *pmd; |
cf62fa4c | 2352 | struct dp_packet *pp; |
72865317 | 2353 | |
cf62fa4c PS |
2354 | if (dp_packet_size(execute->packet) < ETH_HEADER_LEN || |
2355 | dp_packet_size(execute->packet) > UINT16_MAX) { | |
72865317 BP |
2356 | return EINVAL; |
2357 | } | |
2358 | ||
65f13b50 AW |
2359 | /* Tries finding the 'pmd'. If NULL is returned, that means |
2360 | * the current thread is a non-pmd thread and should use | |
b19befae | 2361 | * dp_netdev_get_pmd(dp, NON_PMD_CORE_ID). */ |
65f13b50 AW |
2362 | pmd = ovsthread_getspecific(dp->per_pmd_key); |
2363 | if (!pmd) { | |
b19befae | 2364 | pmd = dp_netdev_get_pmd(dp, NON_PMD_CORE_ID); |
65f13b50 AW |
2365 | } |
2366 | ||
2367 | /* If the current thread is non-pmd thread, acquires | |
2368 | * the 'non_pmd_mutex'. */ | |
2369 | if (pmd->core_id == NON_PMD_CORE_ID) { | |
2370 | ovs_mutex_lock(&dp->non_pmd_mutex); | |
433330a8 | 2371 | ovs_mutex_lock(&dp->port_mutex); |
65f13b50 | 2372 | } |
1c1e46ed | 2373 | |
cf62fa4c | 2374 | pp = execute->packet; |
41ccaa24 | 2375 | dp_netdev_execute_actions(pmd, &pp, 1, false, execute->actions, |
9bbf1c3d | 2376 | execute->actions_len); |
65f13b50 | 2377 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1c1e46ed | 2378 | dp_netdev_pmd_unref(pmd); |
433330a8 | 2379 | ovs_mutex_unlock(&dp->port_mutex); |
65f13b50 AW |
2380 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
2381 | } | |
8a4e3a85 | 2382 | |
758c456d | 2383 | return 0; |
72865317 BP |
2384 | } |
2385 | ||
1a0c894a BP |
2386 | static void |
2387 | dpif_netdev_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) | |
2388 | { | |
2389 | size_t i; | |
2390 | ||
2391 | for (i = 0; i < n_ops; i++) { | |
2392 | struct dpif_op *op = ops[i]; | |
2393 | ||
2394 | switch (op->type) { | |
2395 | case DPIF_OP_FLOW_PUT: | |
2396 | op->error = dpif_netdev_flow_put(dpif, &op->u.flow_put); | |
2397 | break; | |
2398 | ||
2399 | case DPIF_OP_FLOW_DEL: | |
2400 | op->error = dpif_netdev_flow_del(dpif, &op->u.flow_del); | |
2401 | break; | |
2402 | ||
2403 | case DPIF_OP_EXECUTE: | |
2404 | op->error = dpif_netdev_execute(dpif, &op->u.execute); | |
2405 | break; | |
6fe09f8c JS |
2406 | |
2407 | case DPIF_OP_FLOW_GET: | |
2408 | op->error = dpif_netdev_flow_get(dpif, &op->u.flow_get); | |
2409 | break; | |
1a0c894a BP |
2410 | } |
2411 | } | |
2412 | } | |
2413 | ||
f2eee189 AW |
2414 | /* Returns true if the configuration for rx queues or cpu mask |
2415 | * is changed. */ | |
2416 | static bool | |
a14b8947 | 2417 | pmd_config_changed(const struct dp_netdev *dp, const char *cmask) |
f2eee189 | 2418 | { |
a14b8947 IM |
2419 | struct dp_netdev_port *port; |
2420 | ||
2421 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
2422 | struct netdev *netdev = port->netdev; | |
2423 | int requested_n_rxq = netdev_requested_n_rxq(netdev); | |
2424 | if (netdev_is_pmd(netdev) | |
2425 | && port->latest_requested_n_rxq != requested_n_rxq) { | |
2426 | return true; | |
f2eee189 AW |
2427 | } |
2428 | } | |
a14b8947 IM |
2429 | |
2430 | if (dp->pmd_cmask != NULL && cmask != NULL) { | |
2431 | return strcmp(dp->pmd_cmask, cmask); | |
2432 | } else { | |
2433 | return (dp->pmd_cmask != NULL || cmask != NULL); | |
2434 | } | |
f2eee189 AW |
2435 | } |
2436 | ||
2437 | /* Resets pmd threads if the configuration for 'rxq's or cpu mask changes. */ | |
2438 | static int | |
a14b8947 | 2439 | dpif_netdev_pmd_set(struct dpif *dpif, const char *cmask) |
f2eee189 AW |
2440 | { |
2441 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2442 | ||
a14b8947 | 2443 | if (pmd_config_changed(dp, cmask)) { |
f2eee189 AW |
2444 | struct dp_netdev_port *port; |
2445 | ||
2446 | dp_netdev_destroy_all_pmds(dp); | |
2447 | ||
2448 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
a14b8947 IM |
2449 | struct netdev *netdev = port->netdev; |
2450 | int requested_n_rxq = netdev_requested_n_rxq(netdev); | |
2451 | if (netdev_is_pmd(port->netdev) | |
2452 | && port->latest_requested_n_rxq != requested_n_rxq) { | |
f2eee189 AW |
2453 | int i, err; |
2454 | ||
2455 | /* Closes the existing 'rxq's. */ | |
2456 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2457 | netdev_rxq_close(port->rxq[i]); | |
2458 | port->rxq[i] = NULL; | |
2459 | } | |
2460 | ||
2461 | /* Sets the new rx queue config. */ | |
3bcc10c0 DDP |
2462 | err = netdev_set_multiq(port->netdev, |
2463 | ovs_numa_get_n_cores() + 1, | |
a14b8947 | 2464 | requested_n_rxq); |
7251515e | 2465 | if (err && (err != EOPNOTSUPP)) { |
f2eee189 AW |
2466 | VLOG_ERR("Failed to set dpdk interface %s rx_queue to:" |
2467 | " %u", netdev_get_name(port->netdev), | |
a14b8947 | 2468 | requested_n_rxq); |
f2eee189 AW |
2469 | return err; |
2470 | } | |
a14b8947 | 2471 | port->latest_requested_n_rxq = requested_n_rxq; |
f2eee189 AW |
2472 | /* If the set_multiq() above succeeds, reopens the 'rxq's. */ |
2473 | port->rxq = xrealloc(port->rxq, sizeof *port->rxq | |
2474 | * netdev_n_rxq(port->netdev)); | |
2475 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2476 | netdev_rxq_open(port->netdev, &port->rxq[i], i); | |
2477 | } | |
2478 | } | |
2479 | } | |
f2eee189 AW |
2480 | /* Reconfigures the cpu mask. */ |
2481 | ovs_numa_set_cpu_mask(cmask); | |
2482 | free(dp->pmd_cmask); | |
2483 | dp->pmd_cmask = cmask ? xstrdup(cmask) : NULL; | |
2484 | ||
2485 | /* Restores the non-pmd. */ | |
2486 | dp_netdev_set_nonpmd(dp); | |
2487 | /* Restores all pmd threads. */ | |
2488 | dp_netdev_reset_pmd_threads(dp); | |
2489 | } | |
2490 | ||
2491 | return 0; | |
2492 | } | |
2493 | ||
5bf93d67 EJ |
2494 | static int |
2495 | dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED, | |
2496 | uint32_t queue_id, uint32_t *priority) | |
2497 | { | |
2498 | *priority = queue_id; | |
2499 | return 0; | |
2500 | } | |
2501 | ||
72865317 | 2502 | \f |
9ff55ae2 DDP |
2503 | /* Creates and returns a new 'struct dp_netdev_actions', whose actions are |
2504 | * a copy of the 'ofpacts_len' bytes of 'ofpacts'. */ | |
a84cb64a BP |
2505 | struct dp_netdev_actions * |
2506 | dp_netdev_actions_create(const struct nlattr *actions, size_t size) | |
2507 | { | |
2508 | struct dp_netdev_actions *netdev_actions; | |
2509 | ||
9ff55ae2 DDP |
2510 | netdev_actions = xmalloc(sizeof *netdev_actions + size); |
2511 | memcpy(netdev_actions->actions, actions, size); | |
a84cb64a BP |
2512 | netdev_actions->size = size; |
2513 | ||
2514 | return netdev_actions; | |
2515 | } | |
2516 | ||
a84cb64a | 2517 | struct dp_netdev_actions * |
61e7deb1 | 2518 | dp_netdev_flow_get_actions(const struct dp_netdev_flow *flow) |
a84cb64a | 2519 | { |
61e7deb1 | 2520 | return ovsrcu_get(struct dp_netdev_actions *, &flow->actions); |
a84cb64a BP |
2521 | } |
2522 | ||
61e7deb1 BP |
2523 | static void |
2524 | dp_netdev_actions_free(struct dp_netdev_actions *actions) | |
a84cb64a | 2525 | { |
61e7deb1 | 2526 | free(actions); |
a84cb64a BP |
2527 | } |
2528 | \f | |
55e3ca97 DDP |
2529 | static inline unsigned long long |
2530 | cycles_counter(void) | |
2531 | { | |
2532 | #ifdef DPDK_NETDEV | |
2533 | return rte_get_tsc_cycles(); | |
2534 | #else | |
2535 | return 0; | |
2536 | #endif | |
2537 | } | |
2538 | ||
2539 | /* Fake mutex to make sure that the calls to cycles_count_* are balanced */ | |
2540 | extern struct ovs_mutex cycles_counter_fake_mutex; | |
2541 | ||
2542 | /* Start counting cycles. Must be followed by 'cycles_count_end()' */ | |
2543 | static inline void | |
2544 | cycles_count_start(struct dp_netdev_pmd_thread *pmd) | |
2545 | OVS_ACQUIRES(&cycles_counter_fake_mutex) | |
2546 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2547 | { | |
2548 | pmd->last_cycles = cycles_counter(); | |
2549 | } | |
2550 | ||
2551 | /* Stop counting cycles and add them to the counter 'type' */ | |
2552 | static inline void | |
2553 | cycles_count_end(struct dp_netdev_pmd_thread *pmd, | |
2554 | enum pmd_cycles_counter_type type) | |
2555 | OVS_RELEASES(&cycles_counter_fake_mutex) | |
2556 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2557 | { | |
2558 | unsigned long long interval = cycles_counter() - pmd->last_cycles; | |
2559 | ||
2560 | non_atomic_ullong_add(&pmd->cycles.n[type], interval); | |
2561 | } | |
e4cfed38 | 2562 | |
5794e276 | 2563 | static void |
65f13b50 | 2564 | dp_netdev_process_rxq_port(struct dp_netdev_pmd_thread *pmd, |
9bbf1c3d DDP |
2565 | struct dp_netdev_port *port, |
2566 | struct netdev_rxq *rxq) | |
e4cfed38 | 2567 | { |
cd159f1a | 2568 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 | 2569 | int error, cnt; |
e4cfed38 | 2570 | |
55e3ca97 | 2571 | cycles_count_start(pmd); |
8cbf4f47 | 2572 | error = netdev_rxq_recv(rxq, packets, &cnt); |
55e3ca97 | 2573 | cycles_count_end(pmd, PMD_CYCLES_POLLING); |
e4cfed38 | 2574 | if (!error) { |
3c33f0ff | 2575 | *recirc_depth_get() = 0; |
41ccaa24 | 2576 | |
55e3ca97 | 2577 | cycles_count_start(pmd); |
a90ed026 | 2578 | dp_netdev_input(pmd, packets, cnt, port->port_no); |
55e3ca97 | 2579 | cycles_count_end(pmd, PMD_CYCLES_PROCESSING); |
e4cfed38 | 2580 | } else if (error != EAGAIN && error != EOPNOTSUPP) { |
3c33f0ff | 2581 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
e4cfed38 PS |
2582 | |
2583 | VLOG_ERR_RL(&rl, "error receiving data from %s: %s", | |
3c33f0ff | 2584 | netdev_get_name(port->netdev), ovs_strerror(error)); |
e4cfed38 PS |
2585 | } |
2586 | } | |
2587 | ||
a36de779 PS |
2588 | /* Return true if needs to revalidate datapath flows. */ |
2589 | static bool | |
e4cfed38 PS |
2590 | dpif_netdev_run(struct dpif *dpif) |
2591 | { | |
2592 | struct dp_netdev_port *port; | |
2593 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
b19befae AW |
2594 | struct dp_netdev_pmd_thread *non_pmd = dp_netdev_get_pmd(dp, |
2595 | NON_PMD_CORE_ID); | |
a36de779 | 2596 | uint64_t new_tnl_seq; |
e4cfed38 | 2597 | |
65f13b50 | 2598 | ovs_mutex_lock(&dp->non_pmd_mutex); |
a532e683 | 2599 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2600 | if (!netdev_is_pmd(port->netdev)) { |
2601 | int i; | |
2602 | ||
2603 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
65f13b50 | 2604 | dp_netdev_process_rxq_port(non_pmd, port, port->rxq[i]); |
55c955bd | 2605 | } |
e4cfed38 PS |
2606 | } |
2607 | } | |
65f13b50 | 2608 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
1c1e46ed AW |
2609 | dp_netdev_pmd_unref(non_pmd); |
2610 | ||
53902038 | 2611 | tnl_neigh_cache_run(); |
7f9b8504 | 2612 | tnl_port_map_run(); |
a36de779 PS |
2613 | new_tnl_seq = seq_read(tnl_conf_seq); |
2614 | ||
2615 | if (dp->last_tnl_conf_seq != new_tnl_seq) { | |
2616 | dp->last_tnl_conf_seq = new_tnl_seq; | |
2617 | return true; | |
2618 | } | |
2619 | return false; | |
e4cfed38 PS |
2620 | } |
2621 | ||
2622 | static void | |
2623 | dpif_netdev_wait(struct dpif *dpif) | |
2624 | { | |
2625 | struct dp_netdev_port *port; | |
2626 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2627 | ||
59e6d833 | 2628 | ovs_mutex_lock(&dp_netdev_mutex); |
a532e683 | 2629 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2630 | if (!netdev_is_pmd(port->netdev)) { |
2631 | int i; | |
2632 | ||
2633 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2634 | netdev_rxq_wait(port->rxq[i]); | |
2635 | } | |
e4cfed38 PS |
2636 | } |
2637 | } | |
59e6d833 | 2638 | ovs_mutex_unlock(&dp_netdev_mutex); |
a36de779 | 2639 | seq_wait(tnl_conf_seq, dp->last_tnl_conf_seq); |
e4cfed38 PS |
2640 | } |
2641 | ||
e4cfed38 | 2642 | static int |
65f13b50 | 2643 | pmd_load_queues(struct dp_netdev_pmd_thread *pmd, |
f7791740 | 2644 | struct rxq_poll **ppoll_list, int poll_cnt) |
ae7ad0a1 | 2645 | OVS_REQUIRES(pmd->poll_mutex) |
e4cfed38 | 2646 | { |
f7791740 | 2647 | struct rxq_poll *poll_list = *ppoll_list; |
ae7ad0a1 IM |
2648 | struct rxq_poll *poll; |
2649 | int i; | |
e4cfed38 | 2650 | |
e4cfed38 | 2651 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2652 | port_unref(poll_list[i].port); |
e4cfed38 PS |
2653 | } |
2654 | ||
ae7ad0a1 | 2655 | poll_list = xrealloc(poll_list, pmd->poll_cnt * sizeof *poll_list); |
a1fdee13 | 2656 | |
ae7ad0a1 IM |
2657 | i = 0; |
2658 | LIST_FOR_EACH (poll, node, &pmd->poll_list) { | |
2659 | port_ref(poll->port); | |
2660 | poll_list[i++] = *poll; | |
e4cfed38 PS |
2661 | } |
2662 | ||
e4cfed38 | 2663 | *ppoll_list = poll_list; |
ae7ad0a1 | 2664 | return pmd->poll_cnt; |
e4cfed38 PS |
2665 | } |
2666 | ||
6c3eee82 | 2667 | static void * |
e4cfed38 | 2668 | pmd_thread_main(void *f_) |
6c3eee82 | 2669 | { |
65f13b50 | 2670 | struct dp_netdev_pmd_thread *pmd = f_; |
e4cfed38 | 2671 | unsigned int lc = 0; |
f7791740 | 2672 | struct rxq_poll *poll_list; |
84067a4c | 2673 | unsigned int port_seq = PMD_INITIAL_SEQ; |
e4cfed38 PS |
2674 | int poll_cnt; |
2675 | int i; | |
6c3eee82 | 2676 | |
e4cfed38 PS |
2677 | poll_cnt = 0; |
2678 | poll_list = NULL; | |
2679 | ||
65f13b50 AW |
2680 | /* Stores the pmd thread's 'pmd' to 'per_pmd_key'. */ |
2681 | ovsthread_setspecific(pmd->dp->per_pmd_key, pmd); | |
2682 | pmd_thread_setaffinity_cpu(pmd->core_id); | |
e4cfed38 | 2683 | reload: |
65f13b50 | 2684 | emc_cache_init(&pmd->flow_cache); |
ae7ad0a1 IM |
2685 | |
2686 | ovs_mutex_lock(&pmd->poll_mutex); | |
65f13b50 | 2687 | poll_cnt = pmd_load_queues(pmd, &poll_list, poll_cnt); |
ae7ad0a1 | 2688 | ovs_mutex_unlock(&pmd->poll_mutex); |
6c3eee82 | 2689 | |
7dd671f0 MK |
2690 | /* List port/core affinity */ |
2691 | for (i = 0; i < poll_cnt; i++) { | |
ce179f11 IM |
2692 | VLOG_DBG("Core %d processing port \'%s\' with queue-id %d\n", |
2693 | pmd->core_id, netdev_get_name(poll_list[i].port->netdev), | |
2694 | netdev_rxq_get_queue_id(poll_list[i].rx)); | |
7dd671f0 MK |
2695 | } |
2696 | ||
accf8626 AW |
2697 | /* Signal here to make sure the pmd finishes |
2698 | * reloading the updated configuration. */ | |
2699 | dp_netdev_pmd_reload_done(pmd); | |
2700 | ||
e4cfed38 | 2701 | for (;;) { |
e4cfed38 | 2702 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2703 | dp_netdev_process_rxq_port(pmd, poll_list[i].port, poll_list[i].rx); |
e4cfed38 PS |
2704 | } |
2705 | ||
2706 | if (lc++ > 1024) { | |
84067a4c | 2707 | unsigned int seq; |
6c3eee82 | 2708 | |
e4cfed38 | 2709 | lc = 0; |
84067a4c | 2710 | |
67ad54cb | 2711 | emc_cache_slow_sweep(&pmd->flow_cache); |
fbe0962b | 2712 | coverage_try_clear(); |
84067a4c JR |
2713 | ovsrcu_quiesce(); |
2714 | ||
65f13b50 | 2715 | atomic_read_relaxed(&pmd->change_seq, &seq); |
84067a4c JR |
2716 | if (seq != port_seq) { |
2717 | port_seq = seq; | |
6c3eee82 BP |
2718 | break; |
2719 | } | |
2720 | } | |
e4cfed38 | 2721 | } |
6c3eee82 | 2722 | |
65f13b50 | 2723 | emc_cache_uninit(&pmd->flow_cache); |
9bbf1c3d | 2724 | |
65f13b50 | 2725 | if (!latch_is_set(&pmd->exit_latch)){ |
e4cfed38 PS |
2726 | goto reload; |
2727 | } | |
6c3eee82 | 2728 | |
e4cfed38 | 2729 | for (i = 0; i < poll_cnt; i++) { |
ae7ad0a1 | 2730 | port_unref(poll_list[i].port); |
6c3eee82 | 2731 | } |
6c3eee82 | 2732 | |
accf8626 AW |
2733 | dp_netdev_pmd_reload_done(pmd); |
2734 | ||
e4cfed38 | 2735 | free(poll_list); |
6c3eee82 BP |
2736 | return NULL; |
2737 | } | |
2738 | ||
6b31e073 RW |
2739 | static void |
2740 | dp_netdev_disable_upcall(struct dp_netdev *dp) | |
2741 | OVS_ACQUIRES(dp->upcall_rwlock) | |
2742 | { | |
2743 | fat_rwlock_wrlock(&dp->upcall_rwlock); | |
2744 | } | |
2745 | ||
2746 | static void | |
2747 | dpif_netdev_disable_upcall(struct dpif *dpif) | |
2748 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2749 | { | |
2750 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2751 | dp_netdev_disable_upcall(dp); | |
2752 | } | |
2753 | ||
2754 | static void | |
2755 | dp_netdev_enable_upcall(struct dp_netdev *dp) | |
2756 | OVS_RELEASES(dp->upcall_rwlock) | |
2757 | { | |
2758 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
2759 | } | |
2760 | ||
2761 | static void | |
2762 | dpif_netdev_enable_upcall(struct dpif *dpif) | |
2763 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2764 | { | |
2765 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2766 | dp_netdev_enable_upcall(dp); | |
2767 | } | |
2768 | ||
ae7ad0a1 | 2769 | static void |
accf8626 AW |
2770 | dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd) |
2771 | { | |
2772 | ovs_mutex_lock(&pmd->cond_mutex); | |
2773 | xpthread_cond_signal(&pmd->cond); | |
2774 | ovs_mutex_unlock(&pmd->cond_mutex); | |
2775 | } | |
2776 | ||
1c1e46ed AW |
2777 | /* Finds and refs the dp_netdev_pmd_thread on core 'core_id'. Returns |
2778 | * the pointer if succeeds, otherwise, NULL. | |
2779 | * | |
2780 | * Caller must unrefs the returned reference. */ | |
65f13b50 | 2781 | static struct dp_netdev_pmd_thread * |
bd5131ba | 2782 | dp_netdev_get_pmd(struct dp_netdev *dp, unsigned core_id) |
65f13b50 AW |
2783 | { |
2784 | struct dp_netdev_pmd_thread *pmd; | |
55847abe | 2785 | const struct cmap_node *pnode; |
65f13b50 | 2786 | |
b19befae | 2787 | pnode = cmap_find(&dp->poll_threads, hash_int(core_id, 0)); |
1c1e46ed AW |
2788 | if (!pnode) { |
2789 | return NULL; | |
2790 | } | |
65f13b50 AW |
2791 | pmd = CONTAINER_OF(pnode, struct dp_netdev_pmd_thread, node); |
2792 | ||
1c1e46ed | 2793 | return dp_netdev_pmd_try_ref(pmd) ? pmd : NULL; |
65f13b50 AW |
2794 | } |
2795 | ||
f2eee189 AW |
2796 | /* Sets the 'struct dp_netdev_pmd_thread' for non-pmd threads. */ |
2797 | static void | |
2798 | dp_netdev_set_nonpmd(struct dp_netdev *dp) | |
2799 | { | |
2800 | struct dp_netdev_pmd_thread *non_pmd; | |
2801 | ||
2802 | non_pmd = xzalloc(sizeof *non_pmd); | |
2803 | dp_netdev_configure_pmd(non_pmd, dp, 0, NON_PMD_CORE_ID, | |
2804 | OVS_NUMA_UNSPEC); | |
2805 | } | |
2806 | ||
1c1e46ed AW |
2807 | /* Caller must have valid pointer to 'pmd'. */ |
2808 | static bool | |
2809 | dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd) | |
2810 | { | |
2811 | return ovs_refcount_try_ref_rcu(&pmd->ref_cnt); | |
2812 | } | |
2813 | ||
2814 | static void | |
2815 | dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd) | |
2816 | { | |
2817 | if (pmd && ovs_refcount_unref(&pmd->ref_cnt) == 1) { | |
2818 | ovsrcu_postpone(dp_netdev_destroy_pmd, pmd); | |
2819 | } | |
2820 | } | |
2821 | ||
2822 | /* Given cmap position 'pos', tries to ref the next node. If try_ref() | |
2823 | * fails, keeps checking for next node until reaching the end of cmap. | |
2824 | * | |
2825 | * Caller must unrefs the returned reference. */ | |
2826 | static struct dp_netdev_pmd_thread * | |
2827 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos) | |
2828 | { | |
2829 | struct dp_netdev_pmd_thread *next; | |
2830 | ||
2831 | do { | |
2832 | struct cmap_node *node; | |
2833 | ||
2834 | node = cmap_next_position(&dp->poll_threads, pos); | |
2835 | next = node ? CONTAINER_OF(node, struct dp_netdev_pmd_thread, node) | |
2836 | : NULL; | |
2837 | } while (next && !dp_netdev_pmd_try_ref(next)); | |
2838 | ||
2839 | return next; | |
2840 | } | |
2841 | ||
65f13b50 | 2842 | /* Configures the 'pmd' based on the input argument. */ |
6c3eee82 | 2843 | static void |
65f13b50 | 2844 | dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp, |
bd5131ba | 2845 | int index, unsigned core_id, int numa_id) |
65f13b50 AW |
2846 | { |
2847 | pmd->dp = dp; | |
2848 | pmd->index = index; | |
2849 | pmd->core_id = core_id; | |
2850 | pmd->numa_id = numa_id; | |
ae7ad0a1 | 2851 | pmd->poll_cnt = 0; |
1c1e46ed | 2852 | |
347ba9bb IM |
2853 | atomic_init(&pmd->tx_qid, |
2854 | (core_id == NON_PMD_CORE_ID) | |
2855 | ? ovs_numa_get_n_cores() | |
2856 | : get_n_pmd_threads(dp)); | |
2857 | ||
1c1e46ed | 2858 | ovs_refcount_init(&pmd->ref_cnt); |
65f13b50 AW |
2859 | latch_init(&pmd->exit_latch); |
2860 | atomic_init(&pmd->change_seq, PMD_INITIAL_SEQ); | |
accf8626 AW |
2861 | xpthread_cond_init(&pmd->cond, NULL); |
2862 | ovs_mutex_init(&pmd->cond_mutex); | |
1c1e46ed | 2863 | ovs_mutex_init(&pmd->flow_mutex); |
ae7ad0a1 | 2864 | ovs_mutex_init(&pmd->poll_mutex); |
1c1e46ed AW |
2865 | dpcls_init(&pmd->cls); |
2866 | cmap_init(&pmd->flow_table); | |
ae7ad0a1 | 2867 | list_init(&pmd->poll_list); |
65f13b50 AW |
2868 | /* init the 'flow_cache' since there is no |
2869 | * actual thread created for NON_PMD_CORE_ID. */ | |
2870 | if (core_id == NON_PMD_CORE_ID) { | |
2871 | emc_cache_init(&pmd->flow_cache); | |
2872 | } | |
2873 | cmap_insert(&dp->poll_threads, CONST_CAST(struct cmap_node *, &pmd->node), | |
2874 | hash_int(core_id, 0)); | |
2875 | } | |
2876 | ||
1c1e46ed AW |
2877 | static void |
2878 | dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd) | |
2879 | { | |
2880 | dp_netdev_pmd_flow_flush(pmd); | |
2881 | dpcls_destroy(&pmd->cls); | |
2882 | cmap_destroy(&pmd->flow_table); | |
2883 | ovs_mutex_destroy(&pmd->flow_mutex); | |
2884 | latch_destroy(&pmd->exit_latch); | |
2885 | xpthread_cond_destroy(&pmd->cond); | |
2886 | ovs_mutex_destroy(&pmd->cond_mutex); | |
ae7ad0a1 | 2887 | ovs_mutex_destroy(&pmd->poll_mutex); |
1c1e46ed AW |
2888 | free(pmd); |
2889 | } | |
2890 | ||
2891 | /* Stops the pmd thread, removes it from the 'dp->poll_threads', | |
2892 | * and unrefs the struct. */ | |
65f13b50 | 2893 | static void |
e4e74c3a | 2894 | dp_netdev_del_pmd(struct dp_netdev *dp, struct dp_netdev_pmd_thread *pmd) |
6c3eee82 | 2895 | { |
65f13b50 | 2896 | /* Uninit the 'flow_cache' since there is |
1c1e46ed | 2897 | * no actual thread uninit it for NON_PMD_CORE_ID. */ |
65f13b50 AW |
2898 | if (pmd->core_id == NON_PMD_CORE_ID) { |
2899 | emc_cache_uninit(&pmd->flow_cache); | |
2900 | } else { | |
2901 | latch_set(&pmd->exit_latch); | |
2902 | dp_netdev_reload_pmd__(pmd); | |
2903 | ovs_numa_unpin_core(pmd->core_id); | |
2904 | xpthread_join(pmd->thread, NULL); | |
2905 | } | |
ae7ad0a1 IM |
2906 | |
2907 | /* Unref all ports and free poll_list. */ | |
cc245ce8 | 2908 | dp_netdev_pmd_clear_poll_list(pmd); |
ae7ad0a1 | 2909 | |
e4e74c3a AW |
2910 | /* Purges the 'pmd''s flows after stopping the thread, but before |
2911 | * destroying the flows, so that the flow stats can be collected. */ | |
2912 | if (dp->dp_purge_cb) { | |
2913 | dp->dp_purge_cb(dp->dp_purge_aux, pmd->core_id); | |
2914 | } | |
65f13b50 | 2915 | cmap_remove(&pmd->dp->poll_threads, &pmd->node, hash_int(pmd->core_id, 0)); |
1c1e46ed | 2916 | dp_netdev_pmd_unref(pmd); |
65f13b50 | 2917 | } |
6c3eee82 | 2918 | |
65f13b50 AW |
2919 | /* Destroys all pmd threads. */ |
2920 | static void | |
2921 | dp_netdev_destroy_all_pmds(struct dp_netdev *dp) | |
2922 | { | |
2923 | struct dp_netdev_pmd_thread *pmd; | |
d916785c DDP |
2924 | struct dp_netdev_pmd_thread **pmd_list; |
2925 | size_t k = 0, n_pmds; | |
2926 | ||
2927 | n_pmds = cmap_count(&dp->poll_threads); | |
2928 | pmd_list = xcalloc(n_pmds, sizeof *pmd_list); | |
65f13b50 AW |
2929 | |
2930 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
d916785c DDP |
2931 | /* We cannot call dp_netdev_del_pmd(), since it alters |
2932 | * 'dp->poll_threads' (while we're iterating it) and it | |
2933 | * might quiesce. */ | |
2934 | ovs_assert(k < n_pmds); | |
2935 | pmd_list[k++] = pmd; | |
6c3eee82 | 2936 | } |
d916785c DDP |
2937 | |
2938 | for (size_t i = 0; i < k; i++) { | |
2939 | dp_netdev_del_pmd(dp, pmd_list[i]); | |
2940 | } | |
2941 | free(pmd_list); | |
65f13b50 | 2942 | } |
6c3eee82 | 2943 | |
347ba9bb IM |
2944 | /* Deletes all pmd threads on numa node 'numa_id' and |
2945 | * fixes tx_qids of other threads to keep them sequential. */ | |
65f13b50 AW |
2946 | static void |
2947 | dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
2948 | { | |
2949 | struct dp_netdev_pmd_thread *pmd; | |
347ba9bb IM |
2950 | int n_pmds_on_numa, n_pmds; |
2951 | int *free_idx, k = 0; | |
d916785c | 2952 | struct dp_netdev_pmd_thread **pmd_list; |
347ba9bb IM |
2953 | |
2954 | n_pmds_on_numa = get_n_pmd_threads_on_numa(dp, numa_id); | |
d916785c DDP |
2955 | free_idx = xcalloc(n_pmds_on_numa, sizeof *free_idx); |
2956 | pmd_list = xcalloc(n_pmds_on_numa, sizeof *pmd_list); | |
6c3eee82 | 2957 | |
65f13b50 | 2958 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { |
d916785c DDP |
2959 | /* We cannot call dp_netdev_del_pmd(), since it alters |
2960 | * 'dp->poll_threads' (while we're iterating it) and it | |
2961 | * might quiesce. */ | |
65f13b50 | 2962 | if (pmd->numa_id == numa_id) { |
347ba9bb | 2963 | atomic_read_relaxed(&pmd->tx_qid, &free_idx[k]); |
d916785c DDP |
2964 | pmd_list[k] = pmd; |
2965 | ovs_assert(k < n_pmds_on_numa); | |
347ba9bb | 2966 | k++; |
65f13b50 | 2967 | } |
6c3eee82 | 2968 | } |
347ba9bb | 2969 | |
d916785c DDP |
2970 | for (int i = 0; i < k; i++) { |
2971 | dp_netdev_del_pmd(dp, pmd_list[i]); | |
2972 | } | |
2973 | ||
347ba9bb IM |
2974 | n_pmds = get_n_pmd_threads(dp); |
2975 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2976 | int old_tx_qid; | |
2977 | ||
2978 | atomic_read_relaxed(&pmd->tx_qid, &old_tx_qid); | |
2979 | ||
2980 | if (old_tx_qid >= n_pmds) { | |
2981 | int new_tx_qid = free_idx[--k]; | |
2982 | ||
2983 | atomic_store_relaxed(&pmd->tx_qid, new_tx_qid); | |
2984 | } | |
2985 | } | |
2986 | ||
d916785c | 2987 | free(pmd_list); |
347ba9bb | 2988 | free(free_idx); |
65f13b50 | 2989 | } |
6c3eee82 | 2990 | |
cc245ce8 IM |
2991 | /* Deletes all rx queues from pmd->poll_list. */ |
2992 | static void | |
2993 | dp_netdev_pmd_clear_poll_list(struct dp_netdev_pmd_thread *pmd) | |
2994 | { | |
2995 | struct rxq_poll *poll; | |
2996 | ||
2997 | ovs_mutex_lock(&pmd->poll_mutex); | |
2998 | LIST_FOR_EACH_POP (poll, node, &pmd->poll_list) { | |
2999 | port_unref(poll->port); | |
3000 | free(poll); | |
3001 | } | |
3002 | pmd->poll_cnt = 0; | |
3003 | ovs_mutex_unlock(&pmd->poll_mutex); | |
3004 | } | |
3005 | ||
3006 | /* Deletes all rx queues of 'port' from poll_list of pmd thread and | |
3007 | * reloads it if poll_list was changed. */ | |
3008 | static void | |
3009 | dp_netdev_del_port_from_pmd(struct dp_netdev_port *port, | |
3010 | struct dp_netdev_pmd_thread *pmd) | |
3011 | { | |
3012 | struct rxq_poll *poll, *next; | |
3013 | bool found = false; | |
3014 | ||
3015 | ovs_mutex_lock(&pmd->poll_mutex); | |
3016 | LIST_FOR_EACH_SAFE (poll, next, node, &pmd->poll_list) { | |
3017 | if (poll->port == port) { | |
3018 | found = true; | |
3019 | port_unref(poll->port); | |
3020 | list_remove(&poll->node); | |
3021 | pmd->poll_cnt--; | |
3022 | free(poll); | |
3023 | } | |
3024 | } | |
3025 | ovs_mutex_unlock(&pmd->poll_mutex); | |
3026 | if (found) { | |
3027 | dp_netdev_reload_pmd__(pmd); | |
3028 | } | |
3029 | } | |
3030 | ||
3031 | /* Deletes all rx queues of 'port' from all pmd threads of dp and | |
3032 | * reloads them if needed. */ | |
3033 | static void | |
3034 | dp_netdev_del_port_from_all_pmds(struct dp_netdev *dp, | |
3035 | struct dp_netdev_port *port) | |
3036 | { | |
3037 | int numa_id = netdev_get_numa_id(port->netdev); | |
3038 | struct dp_netdev_pmd_thread *pmd; | |
3039 | ||
3040 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3041 | if (pmd->numa_id == numa_id) { | |
3042 | dp_netdev_del_port_from_pmd(port, pmd); | |
3043 | } | |
3044 | } | |
3045 | } | |
3046 | ||
ae7ad0a1 IM |
3047 | /* Returns PMD thread from this numa node with fewer rx queues to poll. |
3048 | * Returns NULL if there is no PMD threads on this numa node. | |
3049 | * Can be called safely only by main thread. */ | |
3050 | static struct dp_netdev_pmd_thread * | |
3051 | dp_netdev_less_loaded_pmd_on_numa(struct dp_netdev *dp, int numa_id) | |
3052 | { | |
3053 | int min_cnt = -1; | |
3054 | struct dp_netdev_pmd_thread *pmd, *res = NULL; | |
3055 | ||
3056 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3057 | if (pmd->numa_id == numa_id | |
3058 | && (min_cnt > pmd->poll_cnt || res == NULL)) { | |
3059 | min_cnt = pmd->poll_cnt; | |
3060 | res = pmd; | |
3061 | } | |
3062 | } | |
3063 | ||
3064 | return res; | |
3065 | } | |
3066 | ||
3067 | /* Adds rx queue to poll_list of PMD thread. */ | |
3068 | static void | |
3069 | dp_netdev_add_rxq_to_pmd(struct dp_netdev_pmd_thread *pmd, | |
3070 | struct dp_netdev_port *port, struct netdev_rxq *rx) | |
3071 | OVS_REQUIRES(pmd->poll_mutex) | |
3072 | { | |
3073 | struct rxq_poll *poll = xmalloc(sizeof *poll); | |
3074 | ||
3075 | port_ref(port); | |
3076 | poll->port = port; | |
3077 | poll->rx = rx; | |
3078 | ||
3079 | list_push_back(&pmd->poll_list, &poll->node); | |
3080 | pmd->poll_cnt++; | |
3081 | } | |
3082 | ||
cc245ce8 IM |
3083 | /* Distributes all rx queues of 'port' between all PMD threads and reloads |
3084 | * them if needed. */ | |
3085 | static void | |
3086 | dp_netdev_add_port_to_pmds(struct dp_netdev *dp, struct dp_netdev_port *port) | |
3087 | { | |
3088 | int numa_id = netdev_get_numa_id(port->netdev); | |
3089 | struct dp_netdev_pmd_thread *pmd; | |
3090 | struct hmapx to_reload; | |
3091 | struct hmapx_node *node; | |
3092 | int i; | |
3093 | ||
3094 | hmapx_init(&to_reload); | |
3095 | /* Cannot create pmd threads for invalid numa node. */ | |
3096 | ovs_assert(ovs_numa_numa_id_is_valid(numa_id)); | |
3097 | ||
3098 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
3099 | pmd = dp_netdev_less_loaded_pmd_on_numa(dp, numa_id); | |
3100 | if (!pmd) { | |
3101 | /* There is no pmd threads on this numa node. */ | |
3102 | dp_netdev_set_pmds_on_numa(dp, numa_id); | |
3103 | /* Assigning of rx queues done. */ | |
3104 | break; | |
3105 | } | |
3106 | ||
3107 | ovs_mutex_lock(&pmd->poll_mutex); | |
3108 | dp_netdev_add_rxq_to_pmd(pmd, port, port->rxq[i]); | |
3109 | ovs_mutex_unlock(&pmd->poll_mutex); | |
3110 | ||
3111 | hmapx_add(&to_reload, pmd); | |
3112 | } | |
3113 | ||
3114 | HMAPX_FOR_EACH (node, &to_reload) { | |
3115 | pmd = (struct dp_netdev_pmd_thread *) node->data; | |
3116 | dp_netdev_reload_pmd__(pmd); | |
3117 | } | |
3118 | ||
3119 | hmapx_destroy(&to_reload); | |
3120 | } | |
3121 | ||
65f13b50 AW |
3122 | /* Checks the numa node id of 'netdev' and starts pmd threads for |
3123 | * the numa node. */ | |
3124 | static void | |
3125 | dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
3126 | { | |
3127 | int n_pmds; | |
e4cfed38 | 3128 | |
65f13b50 AW |
3129 | if (!ovs_numa_numa_id_is_valid(numa_id)) { |
3130 | VLOG_ERR("Cannot create pmd threads due to numa id (%d)" | |
3131 | "invalid", numa_id); | |
3132 | return ; | |
3133 | } | |
3134 | ||
3135 | n_pmds = get_n_pmd_threads_on_numa(dp, numa_id); | |
3136 | ||
3137 | /* If there are already pmd threads created for the numa node | |
3138 | * in which 'netdev' is on, do nothing. Else, creates the | |
3139 | * pmd threads for the numa node. */ | |
3140 | if (!n_pmds) { | |
ae7ad0a1 | 3141 | int can_have, n_unpinned, i, index = 0; |
2aca813c | 3142 | struct dp_netdev_pmd_thread **pmds; |
ae7ad0a1 | 3143 | struct dp_netdev_port *port; |
65f13b50 AW |
3144 | |
3145 | n_unpinned = ovs_numa_get_n_unpinned_cores_on_numa(numa_id); | |
3146 | if (!n_unpinned) { | |
3147 | VLOG_ERR("Cannot create pmd threads due to out of unpinned " | |
3148 | "cores on numa node"); | |
3149 | return; | |
3150 | } | |
6c3eee82 | 3151 | |
f2eee189 AW |
3152 | /* If cpu mask is specified, uses all unpinned cores, otherwise |
3153 | * tries creating NR_PMD_THREADS pmd threads. */ | |
3154 | can_have = dp->pmd_cmask ? n_unpinned : MIN(n_unpinned, NR_PMD_THREADS); | |
2aca813c | 3155 | pmds = xzalloc(can_have * sizeof *pmds); |
65f13b50 | 3156 | for (i = 0; i < can_have; i++) { |
bd5131ba | 3157 | unsigned core_id = ovs_numa_get_unpinned_core_on_numa(numa_id); |
2aca813c IM |
3158 | pmds[i] = xzalloc(sizeof **pmds); |
3159 | dp_netdev_configure_pmd(pmds[i], dp, i, core_id, numa_id); | |
3160 | } | |
ae7ad0a1 IM |
3161 | |
3162 | /* Distributes rx queues of this numa node between new pmd threads. */ | |
3163 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
3164 | if (netdev_is_pmd(port->netdev) | |
3165 | && netdev_get_numa_id(port->netdev) == numa_id) { | |
3166 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
3167 | /* Make thread-safety analyser happy. */ | |
3168 | ovs_mutex_lock(&pmds[index]->poll_mutex); | |
3169 | dp_netdev_add_rxq_to_pmd(pmds[index], port, port->rxq[i]); | |
3170 | ovs_mutex_unlock(&pmds[index]->poll_mutex); | |
3171 | index = (index + 1) % can_have; | |
3172 | } | |
3173 | } | |
3174 | } | |
3175 | ||
3176 | /* Actual start of pmd threads. */ | |
2aca813c | 3177 | for (i = 0; i < can_have; i++) { |
2aca813c | 3178 | pmds[i]->thread = ovs_thread_create("pmd", pmd_thread_main, pmds[i]); |
65f13b50 | 3179 | } |
2aca813c | 3180 | free(pmds); |
65f13b50 | 3181 | VLOG_INFO("Created %d pmd threads on numa node %d", can_have, numa_id); |
6c3eee82 BP |
3182 | } |
3183 | } | |
e4cfed38 | 3184 | |
6c3eee82 | 3185 | \f |
f2eee189 AW |
3186 | /* Called after pmd threads config change. Restarts pmd threads with |
3187 | * new configuration. */ | |
3188 | static void | |
3189 | dp_netdev_reset_pmd_threads(struct dp_netdev *dp) | |
3190 | { | |
3191 | struct dp_netdev_port *port; | |
3192 | ||
3193 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
3194 | if (netdev_is_pmd(port->netdev)) { | |
3195 | int numa_id = netdev_get_numa_id(port->netdev); | |
3196 | ||
3197 | dp_netdev_set_pmds_on_numa(dp, numa_id); | |
3198 | } | |
3199 | } | |
3200 | } | |
3201 | ||
b5cbbcf6 AZ |
3202 | static char * |
3203 | dpif_netdev_get_datapath_version(void) | |
3204 | { | |
3205 | return xstrdup("<built-in>"); | |
3206 | } | |
3207 | ||
72865317 | 3208 | static void |
1c1e46ed | 3209 | dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow, int cnt, int size, |
11bfdadd | 3210 | uint16_t tcp_flags, long long now) |
72865317 | 3211 | { |
eb94da30 | 3212 | uint16_t flags; |
72865317 | 3213 | |
eb94da30 DDP |
3214 | atomic_store_relaxed(&netdev_flow->stats.used, now); |
3215 | non_atomic_ullong_add(&netdev_flow->stats.packet_count, cnt); | |
3216 | non_atomic_ullong_add(&netdev_flow->stats.byte_count, size); | |
3217 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
3218 | flags |= tcp_flags; | |
3219 | atomic_store_relaxed(&netdev_flow->stats.tcp_flags, flags); | |
51852a57 BP |
3220 | } |
3221 | ||
3222 | static void | |
1c1e46ed AW |
3223 | dp_netdev_count_packet(struct dp_netdev_pmd_thread *pmd, |
3224 | enum dp_stat_type type, int cnt) | |
51852a57 | 3225 | { |
eb94da30 | 3226 | non_atomic_ullong_add(&pmd->stats.n[type], cnt); |
51852a57 BP |
3227 | } |
3228 | ||
623540e4 | 3229 | static int |
e14deea0 | 3230 | dp_netdev_upcall(struct dp_netdev_pmd_thread *pmd, struct dp_packet *packet_, |
7af12bd7 | 3231 | struct flow *flow, struct flow_wildcards *wc, ovs_u128 *ufid, |
623540e4 EJ |
3232 | enum dpif_upcall_type type, const struct nlattr *userdata, |
3233 | struct ofpbuf *actions, struct ofpbuf *put_actions) | |
3234 | { | |
1c1e46ed | 3235 | struct dp_netdev *dp = pmd->dp; |
6728d578 JG |
3236 | struct flow_tnl orig_tunnel; |
3237 | int err; | |
623540e4 | 3238 | |
623540e4 EJ |
3239 | if (OVS_UNLIKELY(!dp->upcall_cb)) { |
3240 | return ENODEV; | |
3241 | } | |
3242 | ||
6728d578 JG |
3243 | /* Upcall processing expects the Geneve options to be in the translated |
3244 | * format but we need to retain the raw format for datapath use. */ | |
3245 | orig_tunnel.flags = flow->tunnel.flags; | |
3246 | if (flow->tunnel.flags & FLOW_TNL_F_UDPIF) { | |
3247 | orig_tunnel.metadata.present.len = flow->tunnel.metadata.present.len; | |
3248 | memcpy(orig_tunnel.metadata.opts.gnv, flow->tunnel.metadata.opts.gnv, | |
3249 | flow->tunnel.metadata.present.len); | |
3250 | err = tun_metadata_from_geneve_udpif(&orig_tunnel, &orig_tunnel, | |
3251 | &flow->tunnel); | |
3252 | if (err) { | |
3253 | return err; | |
3254 | } | |
3255 | } | |
3256 | ||
623540e4 EJ |
3257 | if (OVS_UNLIKELY(!VLOG_DROP_DBG(&upcall_rl))) { |
3258 | struct ds ds = DS_EMPTY_INITIALIZER; | |
623540e4 | 3259 | char *packet_str; |
cf62fa4c | 3260 | struct ofpbuf key; |
5262eea1 JG |
3261 | struct odp_flow_key_parms odp_parms = { |
3262 | .flow = flow, | |
3263 | .mask = &wc->masks, | |
3264 | .odp_in_port = flow->in_port.odp_port, | |
2494ccd7 | 3265 | .support = dp_netdev_support, |
5262eea1 | 3266 | }; |
623540e4 EJ |
3267 | |
3268 | ofpbuf_init(&key, 0); | |
5262eea1 | 3269 | odp_flow_key_from_flow(&odp_parms, &key); |
cf62fa4c PS |
3270 | packet_str = ofp_packet_to_string(dp_packet_data(packet_), |
3271 | dp_packet_size(packet_)); | |
623540e4 | 3272 | |
6fd6ed71 | 3273 | odp_flow_key_format(key.data, key.size, &ds); |
623540e4 EJ |
3274 | |
3275 | VLOG_DBG("%s: %s upcall:\n%s\n%s", dp->name, | |
3276 | dpif_upcall_type_to_string(type), ds_cstr(&ds), packet_str); | |
3277 | ||
3278 | ofpbuf_uninit(&key); | |
3279 | free(packet_str); | |
6fd6ed71 | 3280 | |
623540e4 EJ |
3281 | ds_destroy(&ds); |
3282 | } | |
3283 | ||
6728d578 JG |
3284 | err = dp->upcall_cb(packet_, flow, ufid, pmd->core_id, type, userdata, |
3285 | actions, wc, put_actions, dp->upcall_aux); | |
3286 | if (err && err != ENOSPC) { | |
3287 | return err; | |
3288 | } | |
3289 | ||
3290 | /* Translate tunnel metadata masks to datapath format. */ | |
3291 | if (wc) { | |
3292 | if (wc->masks.tunnel.metadata.present.map) { | |
4e548ad9 | 3293 | struct geneve_opt opts[TLV_TOT_OPT_SIZE / |
6728d578 JG |
3294 | sizeof(struct geneve_opt)]; |
3295 | ||
3f32cfeb JG |
3296 | if (orig_tunnel.flags & FLOW_TNL_F_UDPIF) { |
3297 | tun_metadata_to_geneve_udpif_mask(&flow->tunnel, | |
3298 | &wc->masks.tunnel, | |
3299 | orig_tunnel.metadata.opts.gnv, | |
3300 | orig_tunnel.metadata.present.len, | |
3301 | opts); | |
3302 | } else { | |
3303 | orig_tunnel.metadata.present.len = 0; | |
3304 | } | |
6728d578 JG |
3305 | |
3306 | memset(&wc->masks.tunnel.metadata, 0, | |
3307 | sizeof wc->masks.tunnel.metadata); | |
3308 | memcpy(&wc->masks.tunnel.metadata.opts.gnv, opts, | |
3309 | orig_tunnel.metadata.present.len); | |
3310 | } | |
3311 | wc->masks.tunnel.metadata.present.len = 0xff; | |
3312 | } | |
3313 | ||
3314 | /* Restore tunnel metadata. We need to use the saved options to ensure | |
3315 | * that any unknown options are not lost. The generated mask will have | |
3316 | * the same structure, matching on types and lengths but wildcarding | |
3317 | * option data we don't care about. */ | |
3318 | if (orig_tunnel.flags & FLOW_TNL_F_UDPIF) { | |
3319 | memcpy(&flow->tunnel.metadata.opts.gnv, orig_tunnel.metadata.opts.gnv, | |
3320 | orig_tunnel.metadata.present.len); | |
3321 | flow->tunnel.metadata.present.len = orig_tunnel.metadata.present.len; | |
3322 | flow->tunnel.flags |= FLOW_TNL_F_UDPIF; | |
3323 | } | |
3324 | ||
3325 | return err; | |
623540e4 EJ |
3326 | } |
3327 | ||
9bbf1c3d | 3328 | static inline uint32_t |
048963aa DDP |
3329 | dpif_netdev_packet_get_rss_hash(struct dp_packet *packet, |
3330 | const struct miniflow *mf) | |
9bbf1c3d | 3331 | { |
048963aa | 3332 | uint32_t hash, recirc_depth; |
9bbf1c3d | 3333 | |
f2f44f5d DDP |
3334 | if (OVS_LIKELY(dp_packet_rss_valid(packet))) { |
3335 | hash = dp_packet_get_rss_hash(packet); | |
3336 | } else { | |
9bbf1c3d | 3337 | hash = miniflow_hash_5tuple(mf, 0); |
2bc1bbd2 | 3338 | dp_packet_set_rss_hash(packet, hash); |
9bbf1c3d | 3339 | } |
048963aa DDP |
3340 | |
3341 | /* The RSS hash must account for the recirculation depth to avoid | |
3342 | * collisions in the exact match cache */ | |
3343 | recirc_depth = *recirc_depth_get_unsafe(); | |
3344 | if (OVS_UNLIKELY(recirc_depth)) { | |
3345 | hash = hash_finish(hash, recirc_depth); | |
3346 | dp_packet_set_rss_hash(packet, hash); | |
3347 | } | |
9bbf1c3d DDP |
3348 | return hash; |
3349 | } | |
3350 | ||
567bbb2e | 3351 | struct packet_batch { |
8cbf4f47 DDP |
3352 | unsigned int packet_count; |
3353 | unsigned int byte_count; | |
3354 | uint16_t tcp_flags; | |
3355 | ||
3356 | struct dp_netdev_flow *flow; | |
3357 | ||
cd159f1a | 3358 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 DDP |
3359 | }; |
3360 | ||
3361 | static inline void | |
e14deea0 | 3362 | packet_batch_update(struct packet_batch *batch, struct dp_packet *packet, |
9bbf1c3d | 3363 | const struct miniflow *mf) |
8cbf4f47 DDP |
3364 | { |
3365 | batch->tcp_flags |= miniflow_get_tcp_flags(mf); | |
3366 | batch->packets[batch->packet_count++] = packet; | |
cf62fa4c | 3367 | batch->byte_count += dp_packet_size(packet); |
8cbf4f47 DDP |
3368 | } |
3369 | ||
3370 | static inline void | |
41ccaa24 | 3371 | packet_batch_init(struct packet_batch *batch, struct dp_netdev_flow *flow) |
8cbf4f47 | 3372 | { |
11e5cf1f | 3373 | flow->batch = batch; |
8cbf4f47 | 3374 | |
11e5cf1f | 3375 | batch->flow = flow; |
8cbf4f47 DDP |
3376 | batch->packet_count = 0; |
3377 | batch->byte_count = 0; | |
3378 | batch->tcp_flags = 0; | |
8cbf4f47 DDP |
3379 | } |
3380 | ||
3381 | static inline void | |
65f13b50 | 3382 | packet_batch_execute(struct packet_batch *batch, |
abcf3ef4 | 3383 | struct dp_netdev_pmd_thread *pmd, |
11bfdadd | 3384 | long long now) |
8cbf4f47 DDP |
3385 | { |
3386 | struct dp_netdev_actions *actions; | |
3387 | struct dp_netdev_flow *flow = batch->flow; | |
3388 | ||
11e5cf1f | 3389 | dp_netdev_flow_used(flow, batch->packet_count, batch->byte_count, |
11bfdadd | 3390 | batch->tcp_flags, now); |
8cbf4f47 DDP |
3391 | |
3392 | actions = dp_netdev_flow_get_actions(flow); | |
3393 | ||
65f13b50 | 3394 | dp_netdev_execute_actions(pmd, batch->packets, batch->packet_count, true, |
41ccaa24 | 3395 | actions->actions, actions->size); |
8cbf4f47 DDP |
3396 | } |
3397 | ||
8aaa125d | 3398 | static inline void |
e14deea0 | 3399 | dp_netdev_queue_batches(struct dp_packet *pkt, |
9bbf1c3d | 3400 | struct dp_netdev_flow *flow, const struct miniflow *mf, |
8aaa125d | 3401 | struct packet_batch *batches, size_t *n_batches) |
9bbf1c3d | 3402 | { |
8aaa125d | 3403 | struct packet_batch *batch = flow->batch; |
11e5cf1f | 3404 | |
f9fe365b AZ |
3405 | if (OVS_UNLIKELY(!batch)) { |
3406 | batch = &batches[(*n_batches)++]; | |
3407 | packet_batch_init(batch, flow); | |
9bbf1c3d DDP |
3408 | } |
3409 | ||
9bbf1c3d | 3410 | packet_batch_update(batch, pkt, mf); |
9bbf1c3d DDP |
3411 | } |
3412 | ||
9bbf1c3d | 3413 | /* Try to process all ('cnt') the 'packets' using only the exact match cache |
a90ed026 | 3414 | * 'pmd->flow_cache'. If a flow is not found for a packet 'packets[i]', the |
8aaa125d DDP |
3415 | * miniflow is copied into 'keys' and the packet pointer is moved at the |
3416 | * beginning of the 'packets' array. | |
9bbf1c3d DDP |
3417 | * |
3418 | * The function returns the number of packets that needs to be processed in the | |
3419 | * 'packets' array (they have been moved to the beginning of the vector). | |
a90ed026 DDP |
3420 | * |
3421 | * If 'md_is_valid' is false, the metadata in 'packets' is not valid and must be | |
3422 | * initialized by this function using 'port_no'. | |
9bbf1c3d DDP |
3423 | */ |
3424 | static inline size_t | |
e14deea0 | 3425 | emc_processing(struct dp_netdev_pmd_thread *pmd, struct dp_packet **packets, |
8aaa125d | 3426 | size_t cnt, struct netdev_flow_key *keys, |
a90ed026 DDP |
3427 | struct packet_batch batches[], size_t *n_batches, |
3428 | bool md_is_valid, odp_port_t port_no) | |
72865317 | 3429 | { |
65f13b50 | 3430 | struct emc_cache *flow_cache = &pmd->flow_cache; |
b89c678b | 3431 | struct netdev_flow_key *key = &keys[0]; |
3d88a620 | 3432 | size_t i, n_missed = 0, n_dropped = 0; |
8cbf4f47 | 3433 | |
84d6d5eb | 3434 | for (i = 0; i < cnt; i++) { |
9bbf1c3d | 3435 | struct dp_netdev_flow *flow; |
5a2fed48 | 3436 | struct dp_packet *packet = packets[i]; |
9bbf1c3d | 3437 | |
5a2fed48 AZ |
3438 | if (OVS_UNLIKELY(dp_packet_size(packet) < ETH_HEADER_LEN)) { |
3439 | dp_packet_delete(packet); | |
3d88a620 | 3440 | n_dropped++; |
84d6d5eb EJ |
3441 | continue; |
3442 | } | |
8cbf4f47 | 3443 | |
72a5e2b8 | 3444 | if (i != cnt - 1) { |
a90ed026 | 3445 | /* Prefetch next packet data and metadata. */ |
72a5e2b8 | 3446 | OVS_PREFETCH(dp_packet_data(packets[i+1])); |
a90ed026 | 3447 | pkt_metadata_prefetch_init(&packets[i+1]->md); |
72a5e2b8 DDP |
3448 | } |
3449 | ||
a90ed026 DDP |
3450 | if (!md_is_valid) { |
3451 | pkt_metadata_init(&packet->md, port_no); | |
3452 | } | |
5a2fed48 | 3453 | miniflow_extract(packet, &key->mf); |
d262ac2c | 3454 | key->len = 0; /* Not computed yet. */ |
5a2fed48 | 3455 | key->hash = dpif_netdev_packet_get_rss_hash(packet, &key->mf); |
9bbf1c3d | 3456 | |
d262ac2c | 3457 | flow = emc_lookup(flow_cache, key); |
8aaa125d | 3458 | if (OVS_LIKELY(flow)) { |
5a2fed48 | 3459 | dp_netdev_queue_batches(packet, flow, &key->mf, batches, |
8aaa125d DDP |
3460 | n_batches); |
3461 | } else { | |
d1aa0b94 AZ |
3462 | /* Exact match cache missed. Group missed packets together at |
3463 | * the beginning of the 'packets' array. */ | |
b89c678b | 3464 | packets[n_missed] = packet; |
400486f7 DDP |
3465 | /* 'key[n_missed]' contains the key of the current packet and it |
3466 | * must be returned to the caller. The next key should be extracted | |
3467 | * to 'keys[n_missed + 1]'. */ | |
3468 | key = &keys[++n_missed]; | |
9bbf1c3d DDP |
3469 | } |
3470 | } | |
3471 | ||
3d88a620 | 3472 | dp_netdev_count_packet(pmd, DP_STAT_EXACT_HIT, cnt - n_dropped - n_missed); |
4f150744 | 3473 | |
3d88a620 | 3474 | return n_missed; |
9bbf1c3d DDP |
3475 | } |
3476 | ||
3477 | static inline void | |
65f13b50 | 3478 | fast_path_processing(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3479 | struct dp_packet **packets, size_t cnt, |
8aaa125d DDP |
3480 | struct netdev_flow_key *keys, |
3481 | struct packet_batch batches[], size_t *n_batches) | |
9bbf1c3d | 3482 | { |
1a0d5831 | 3483 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3484 | const size_t PKT_ARRAY_SIZE = cnt; |
3485 | #else | |
1a0d5831 | 3486 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3487 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d | 3488 | #endif |
0de8783a | 3489 | struct dpcls_rule *rules[PKT_ARRAY_SIZE]; |
65f13b50 AW |
3490 | struct dp_netdev *dp = pmd->dp; |
3491 | struct emc_cache *flow_cache = &pmd->flow_cache; | |
8aaa125d | 3492 | int miss_cnt = 0, lost_cnt = 0; |
9bbf1c3d | 3493 | bool any_miss; |
8aaa125d | 3494 | size_t i; |
9bbf1c3d DDP |
3495 | |
3496 | for (i = 0; i < cnt; i++) { | |
0de8783a | 3497 | /* Key length is needed in all the cases, hash computed on demand. */ |
361d808d | 3498 | keys[i].len = netdev_flow_key_size(miniflow_n_values(&keys[i].mf)); |
9bbf1c3d | 3499 | } |
1c1e46ed | 3500 | any_miss = !dpcls_lookup(&pmd->cls, keys, rules, cnt); |
623540e4 EJ |
3501 | if (OVS_UNLIKELY(any_miss) && !fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { |
3502 | uint64_t actions_stub[512 / 8], slow_stub[512 / 8]; | |
3503 | struct ofpbuf actions, put_actions; | |
7af12bd7 | 3504 | ovs_u128 ufid; |
623540e4 EJ |
3505 | |
3506 | ofpbuf_use_stub(&actions, actions_stub, sizeof actions_stub); | |
3507 | ofpbuf_use_stub(&put_actions, slow_stub, sizeof slow_stub); | |
3508 | ||
3509 | for (i = 0; i < cnt; i++) { | |
0de8783a | 3510 | struct dp_netdev_flow *netdev_flow; |
623540e4 | 3511 | struct ofpbuf *add_actions; |
0de8783a | 3512 | struct match match; |
623540e4 EJ |
3513 | int error; |
3514 | ||
0de8783a | 3515 | if (OVS_LIKELY(rules[i])) { |
623540e4 EJ |
3516 | continue; |
3517 | } | |
3518 | ||
3519 | /* It's possible that an earlier slow path execution installed | |
0de8783a | 3520 | * a rule covering this flow. In this case, it's a lot cheaper |
623540e4 | 3521 | * to catch it here than execute a miss. */ |
1c1e46ed | 3522 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); |
623540e4 | 3523 | if (netdev_flow) { |
0de8783a | 3524 | rules[i] = &netdev_flow->cr; |
623540e4 EJ |
3525 | continue; |
3526 | } | |
3527 | ||
60fc3b7b DDP |
3528 | miss_cnt++; |
3529 | ||
4d8f90b1 | 3530 | match.tun_md.valid = false; |
0de8783a | 3531 | miniflow_expand(&keys[i].mf, &match.flow); |
623540e4 EJ |
3532 | |
3533 | ofpbuf_clear(&actions); | |
3534 | ofpbuf_clear(&put_actions); | |
3535 | ||
7af12bd7 | 3536 | dpif_flow_hash(dp->dpif, &match.flow, sizeof match.flow, &ufid); |
1c1e46ed | 3537 | error = dp_netdev_upcall(pmd, packets[i], &match.flow, &match.wc, |
7af12bd7 | 3538 | &ufid, DPIF_UC_MISS, NULL, &actions, |
0de8783a | 3539 | &put_actions); |
623540e4 | 3540 | if (OVS_UNLIKELY(error && error != ENOSPC)) { |
7ad20cbd | 3541 | dp_packet_delete(packets[i]); |
60fc3b7b | 3542 | lost_cnt++; |
623540e4 EJ |
3543 | continue; |
3544 | } | |
3545 | ||
449b8131 JR |
3546 | /* The Netlink encoding of datapath flow keys cannot express |
3547 | * wildcarding the presence of a VLAN tag. Instead, a missing VLAN | |
3548 | * tag is interpreted as exact match on the fact that there is no | |
3549 | * VLAN. Unless we refactor a lot of code that translates between | |
3550 | * Netlink and struct flow representations, we have to do the same | |
3551 | * here. */ | |
3552 | if (!match.wc.masks.vlan_tci) { | |
3553 | match.wc.masks.vlan_tci = htons(0xffff); | |
3554 | } | |
3555 | ||
623540e4 EJ |
3556 | /* We can't allow the packet batching in the next loop to execute |
3557 | * the actions. Otherwise, if there are any slow path actions, | |
3558 | * we'll send the packet up twice. */ | |
41ccaa24 | 3559 | dp_netdev_execute_actions(pmd, &packets[i], 1, true, |
6fd6ed71 | 3560 | actions.data, actions.size); |
623540e4 | 3561 | |
6fd6ed71 | 3562 | add_actions = put_actions.size ? &put_actions : &actions; |
0de8783a JR |
3563 | if (OVS_LIKELY(error != ENOSPC)) { |
3564 | /* XXX: There's a race window where a flow covering this packet | |
3565 | * could have already been installed since we last did the flow | |
3566 | * lookup before upcall. This could be solved by moving the | |
3567 | * mutex lock outside the loop, but that's an awful long time | |
3568 | * to be locking everyone out of making flow installs. If we | |
3569 | * move to a per-core classifier, it would be reasonable. */ | |
1c1e46ed AW |
3570 | ovs_mutex_lock(&pmd->flow_mutex); |
3571 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); | |
0de8783a | 3572 | if (OVS_LIKELY(!netdev_flow)) { |
1c1e46ed | 3573 | netdev_flow = dp_netdev_flow_add(pmd, &match, &ufid, |
6fd6ed71 PS |
3574 | add_actions->data, |
3575 | add_actions->size); | |
0de8783a | 3576 | } |
1c1e46ed | 3577 | ovs_mutex_unlock(&pmd->flow_mutex); |
0de8783a | 3578 | |
0de8783a | 3579 | emc_insert(flow_cache, &keys[i], netdev_flow); |
623540e4 | 3580 | } |
623540e4 EJ |
3581 | } |
3582 | ||
3583 | ofpbuf_uninit(&actions); | |
3584 | ofpbuf_uninit(&put_actions); | |
3585 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
60fc3b7b | 3586 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); |
ac8c2081 | 3587 | } else if (OVS_UNLIKELY(any_miss)) { |
ac8c2081 | 3588 | for (i = 0; i < cnt; i++) { |
0de8783a | 3589 | if (OVS_UNLIKELY(!rules[i])) { |
e14deea0 | 3590 | dp_packet_delete(packets[i]); |
8aaa125d DDP |
3591 | lost_cnt++; |
3592 | miss_cnt++; | |
ac8c2081 DDP |
3593 | } |
3594 | } | |
623540e4 | 3595 | } |
84d6d5eb | 3596 | |
8cbf4f47 | 3597 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3598 | struct dp_packet *packet = packets[i]; |
84d6d5eb | 3599 | struct dp_netdev_flow *flow; |
8cbf4f47 | 3600 | |
0de8783a | 3601 | if (OVS_UNLIKELY(!rules[i])) { |
84d6d5eb EJ |
3602 | continue; |
3603 | } | |
3604 | ||
84d6d5eb | 3605 | flow = dp_netdev_flow_cast(rules[i]); |
0de8783a | 3606 | |
0de8783a | 3607 | emc_insert(flow_cache, &keys[i], flow); |
8aaa125d | 3608 | dp_netdev_queue_batches(packet, flow, &keys[i].mf, batches, n_batches); |
8cbf4f47 DDP |
3609 | } |
3610 | ||
8aaa125d DDP |
3611 | dp_netdev_count_packet(pmd, DP_STAT_MASKED_HIT, cnt - miss_cnt); |
3612 | dp_netdev_count_packet(pmd, DP_STAT_MISS, miss_cnt); | |
3613 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); | |
72865317 BP |
3614 | } |
3615 | ||
a90ed026 DDP |
3616 | /* Packets enter the datapath from a port (or from recirculation) here. |
3617 | * | |
3618 | * For performance reasons a caller may choose not to initialize the metadata | |
3619 | * in 'packets': in this case 'mdinit' is false and this function needs to | |
3620 | * initialize it using 'port_no'. If the metadata in 'packets' is already | |
3621 | * valid, 'md_is_valid' must be true and 'port_no' will be ignored. */ | |
adcf00ba | 3622 | static void |
a90ed026 DDP |
3623 | dp_netdev_input__(struct dp_netdev_pmd_thread *pmd, |
3624 | struct dp_packet **packets, int cnt, | |
3625 | bool md_is_valid, odp_port_t port_no) | |
9bbf1c3d | 3626 | { |
1a0d5831 | 3627 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3628 | const size_t PKT_ARRAY_SIZE = cnt; |
3629 | #else | |
1a0d5831 | 3630 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3631 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d DDP |
3632 | #endif |
3633 | struct netdev_flow_key keys[PKT_ARRAY_SIZE]; | |
8aaa125d | 3634 | struct packet_batch batches[PKT_ARRAY_SIZE]; |
11bfdadd | 3635 | long long now = time_msec(); |
8aaa125d | 3636 | size_t newcnt, n_batches, i; |
9bbf1c3d | 3637 | |
8aaa125d | 3638 | n_batches = 0; |
a90ed026 DDP |
3639 | newcnt = emc_processing(pmd, packets, cnt, keys, batches, &n_batches, |
3640 | md_is_valid, port_no); | |
9bbf1c3d | 3641 | if (OVS_UNLIKELY(newcnt)) { |
8aaa125d DDP |
3642 | fast_path_processing(pmd, packets, newcnt, keys, batches, &n_batches); |
3643 | } | |
3644 | ||
603f2ce0 EJ |
3645 | for (i = 0; i < n_batches; i++) { |
3646 | batches[i].flow->batch = NULL; | |
3647 | } | |
3648 | ||
8aaa125d DDP |
3649 | for (i = 0; i < n_batches; i++) { |
3650 | packet_batch_execute(&batches[i], pmd, now); | |
9bbf1c3d DDP |
3651 | } |
3652 | } | |
3653 | ||
a90ed026 DDP |
3654 | static void |
3655 | dp_netdev_input(struct dp_netdev_pmd_thread *pmd, | |
3656 | struct dp_packet **packets, int cnt, | |
3657 | odp_port_t port_no) | |
3658 | { | |
3659 | dp_netdev_input__(pmd, packets, cnt, false, port_no); | |
3660 | } | |
3661 | ||
3662 | static void | |
3663 | dp_netdev_recirculate(struct dp_netdev_pmd_thread *pmd, | |
3664 | struct dp_packet **packets, int cnt) | |
3665 | { | |
3666 | dp_netdev_input__(pmd, packets, cnt, true, 0); | |
3667 | } | |
3668 | ||
9080a111 | 3669 | struct dp_netdev_execute_aux { |
65f13b50 | 3670 | struct dp_netdev_pmd_thread *pmd; |
9080a111 JR |
3671 | }; |
3672 | ||
e4e74c3a AW |
3673 | static void |
3674 | dpif_netdev_register_dp_purge_cb(struct dpif *dpif, dp_purge_callback *cb, | |
3675 | void *aux) | |
3676 | { | |
3677 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3678 | dp->dp_purge_aux = aux; | |
3679 | dp->dp_purge_cb = cb; | |
3680 | } | |
3681 | ||
6b31e073 | 3682 | static void |
623540e4 EJ |
3683 | dpif_netdev_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, |
3684 | void *aux) | |
6b31e073 RW |
3685 | { |
3686 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
623540e4 | 3687 | dp->upcall_aux = aux; |
6b31e073 RW |
3688 | dp->upcall_cb = cb; |
3689 | } | |
3690 | ||
ac8c2081 | 3691 | static void |
3bcc10c0 | 3692 | dp_netdev_drop_packets(struct dp_packet **packets, int cnt, bool may_steal) |
ac8c2081 | 3693 | { |
ac8c2081 | 3694 | if (may_steal) { |
a36de779 PS |
3695 | int i; |
3696 | ||
ac8c2081 | 3697 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3698 | dp_packet_delete(packets[i]); |
ac8c2081 DDP |
3699 | } |
3700 | } | |
3701 | } | |
3702 | ||
a36de779 PS |
3703 | static int |
3704 | push_tnl_action(const struct dp_netdev *dp, | |
3705 | const struct nlattr *attr, | |
e14deea0 | 3706 | struct dp_packet **packets, int cnt) |
a36de779 PS |
3707 | { |
3708 | struct dp_netdev_port *tun_port; | |
3709 | const struct ovs_action_push_tnl *data; | |
3710 | ||
3711 | data = nl_attr_get(attr); | |
3712 | ||
3713 | tun_port = dp_netdev_lookup_port(dp, u32_to_odp(data->tnl_port)); | |
3714 | if (!tun_port) { | |
3715 | return -EINVAL; | |
3716 | } | |
3717 | netdev_push_header(tun_port->netdev, packets, cnt, data); | |
3718 | ||
3719 | return 0; | |
3720 | } | |
3721 | ||
3722 | static void | |
28e2fa02 DDP |
3723 | dp_netdev_clone_pkt_batch(struct dp_packet **dst_pkts, |
3724 | struct dp_packet **src_pkts, int cnt) | |
a36de779 PS |
3725 | { |
3726 | int i; | |
3727 | ||
3728 | for (i = 0; i < cnt; i++) { | |
28e2fa02 | 3729 | dst_pkts[i] = dp_packet_clone(src_pkts[i]); |
a36de779 PS |
3730 | } |
3731 | } | |
3732 | ||
9080a111 | 3733 | static void |
e14deea0 | 3734 | dp_execute_cb(void *aux_, struct dp_packet **packets, int cnt, |
09f9da0b | 3735 | const struct nlattr *a, bool may_steal) |
8a4e3a85 | 3736 | OVS_NO_THREAD_SAFETY_ANALYSIS |
9080a111 JR |
3737 | { |
3738 | struct dp_netdev_execute_aux *aux = aux_; | |
623540e4 | 3739 | uint32_t *depth = recirc_depth_get(); |
28e2fa02 DDP |
3740 | struct dp_netdev_pmd_thread *pmd = aux->pmd; |
3741 | struct dp_netdev *dp = pmd->dp; | |
09f9da0b | 3742 | int type = nl_attr_type(a); |
8a4e3a85 | 3743 | struct dp_netdev_port *p; |
8cbf4f47 | 3744 | int i; |
9080a111 | 3745 | |
09f9da0b JR |
3746 | switch ((enum ovs_action_attr)type) { |
3747 | case OVS_ACTION_ATTR_OUTPUT: | |
623540e4 | 3748 | p = dp_netdev_lookup_port(dp, u32_to_odp(nl_attr_get_u32(a))); |
26a5075b | 3749 | if (OVS_LIKELY(p)) { |
347ba9bb IM |
3750 | int tx_qid; |
3751 | ||
3752 | atomic_read_relaxed(&pmd->tx_qid, &tx_qid); | |
3753 | ||
3754 | netdev_send(p->netdev, tx_qid, packets, cnt, may_steal); | |
ac8c2081 | 3755 | return; |
8a4e3a85 | 3756 | } |
09f9da0b JR |
3757 | break; |
3758 | ||
a36de779 PS |
3759 | case OVS_ACTION_ATTR_TUNNEL_PUSH: |
3760 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3761 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3762 | int err; |
3763 | ||
3764 | if (!may_steal) { | |
3765 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3766 | packets = tnl_pkt; | |
3767 | } | |
3768 | ||
3769 | err = push_tnl_action(dp, a, packets, cnt); | |
3770 | if (!err) { | |
3771 | (*depth)++; | |
a90ed026 | 3772 | dp_netdev_recirculate(pmd, packets, cnt); |
a36de779 PS |
3773 | (*depth)--; |
3774 | } else { | |
3775 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3776 | } | |
3777 | return; | |
3778 | } | |
3779 | break; | |
3780 | ||
3781 | case OVS_ACTION_ATTR_TUNNEL_POP: | |
3782 | if (*depth < MAX_RECIRC_DEPTH) { | |
3783 | odp_port_t portno = u32_to_odp(nl_attr_get_u32(a)); | |
3784 | ||
3785 | p = dp_netdev_lookup_port(dp, portno); | |
3786 | if (p) { | |
cd159f1a | 3787 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3788 | int err; |
3789 | ||
3790 | if (!may_steal) { | |
3791 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3792 | packets = tnl_pkt; | |
3793 | } | |
3794 | ||
3795 | err = netdev_pop_header(p->netdev, packets, cnt); | |
3796 | if (!err) { | |
3797 | ||
3798 | for (i = 0; i < cnt; i++) { | |
3799 | packets[i]->md.in_port.odp_port = portno; | |
3800 | } | |
3801 | ||
3802 | (*depth)++; | |
a90ed026 | 3803 | dp_netdev_recirculate(pmd, packets, cnt); |
a36de779 PS |
3804 | (*depth)--; |
3805 | } else { | |
3806 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3807 | } | |
3808 | return; | |
3809 | } | |
3810 | } | |
3811 | break; | |
3812 | ||
623540e4 EJ |
3813 | case OVS_ACTION_ATTR_USERSPACE: |
3814 | if (!fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { | |
3815 | const struct nlattr *userdata; | |
3816 | struct ofpbuf actions; | |
3817 | struct flow flow; | |
7af12bd7 | 3818 | ovs_u128 ufid; |
4fc65926 | 3819 | |
623540e4 EJ |
3820 | userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); |
3821 | ofpbuf_init(&actions, 0); | |
8cbf4f47 | 3822 | |
623540e4 EJ |
3823 | for (i = 0; i < cnt; i++) { |
3824 | int error; | |
3825 | ||
3826 | ofpbuf_clear(&actions); | |
3827 | ||
cf62fa4c | 3828 | flow_extract(packets[i], &flow); |
7af12bd7 | 3829 | dpif_flow_hash(dp->dpif, &flow, sizeof flow, &ufid); |
1c1e46ed | 3830 | error = dp_netdev_upcall(pmd, packets[i], &flow, NULL, &ufid, |
7af12bd7 | 3831 | DPIF_UC_ACTION, userdata,&actions, |
623540e4 EJ |
3832 | NULL); |
3833 | if (!error || error == ENOSPC) { | |
ac8c2081 | 3834 | dp_netdev_execute_actions(pmd, &packets[i], 1, may_steal, |
6fd6ed71 | 3835 | actions.data, actions.size); |
ac8c2081 | 3836 | } else if (may_steal) { |
e14deea0 | 3837 | dp_packet_delete(packets[i]); |
623540e4 | 3838 | } |
db73f716 | 3839 | } |
623540e4 EJ |
3840 | ofpbuf_uninit(&actions); |
3841 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
6b31e073 | 3842 | |
ac8c2081 DDP |
3843 | return; |
3844 | } | |
09f9da0b | 3845 | break; |
572f732a | 3846 | |
adcf00ba AZ |
3847 | case OVS_ACTION_ATTR_RECIRC: |
3848 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3849 | struct dp_packet *recirc_pkts[NETDEV_MAX_BURST]; |
572f732a | 3850 | |
28e2fa02 DDP |
3851 | if (!may_steal) { |
3852 | dp_netdev_clone_pkt_batch(recirc_pkts, packets, cnt); | |
3853 | packets = recirc_pkts; | |
3854 | } | |
8cbf4f47 | 3855 | |
28e2fa02 DDP |
3856 | for (i = 0; i < cnt; i++) { |
3857 | packets[i]->md.recirc_id = nl_attr_get_u32(a); | |
8cbf4f47 | 3858 | } |
28e2fa02 DDP |
3859 | |
3860 | (*depth)++; | |
a90ed026 | 3861 | dp_netdev_recirculate(pmd, packets, cnt); |
adcf00ba AZ |
3862 | (*depth)--; |
3863 | ||
ac8c2081 | 3864 | return; |
adcf00ba | 3865 | } |
ac8c2081 DDP |
3866 | |
3867 | VLOG_WARN("Packet dropped. Max recirculation depth exceeded."); | |
572f732a | 3868 | break; |
572f732a | 3869 | |
07659514 JS |
3870 | case OVS_ACTION_ATTR_CT: |
3871 | /* If a flow with this action is slow-pathed, datapath assistance is | |
3872 | * required to implement it. However, we don't support this action | |
3873 | * in the userspace datapath. */ | |
3874 | VLOG_WARN("Cannot execute conntrack action in userspace."); | |
3875 | break; | |
3876 | ||
09f9da0b JR |
3877 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3878 | case OVS_ACTION_ATTR_POP_VLAN: | |
3879 | case OVS_ACTION_ATTR_PUSH_MPLS: | |
3880 | case OVS_ACTION_ATTR_POP_MPLS: | |
3881 | case OVS_ACTION_ATTR_SET: | |
6d670e7f | 3882 | case OVS_ACTION_ATTR_SET_MASKED: |
09f9da0b | 3883 | case OVS_ACTION_ATTR_SAMPLE: |
53e1d6f1 | 3884 | case OVS_ACTION_ATTR_HASH: |
09f9da0b JR |
3885 | case OVS_ACTION_ATTR_UNSPEC: |
3886 | case __OVS_ACTION_ATTR_MAX: | |
3887 | OVS_NOT_REACHED(); | |
da546e07 | 3888 | } |
ac8c2081 DDP |
3889 | |
3890 | dp_netdev_drop_packets(packets, cnt, may_steal); | |
98403001 BP |
3891 | } |
3892 | ||
4edb9ae9 | 3893 | static void |
65f13b50 | 3894 | dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3895 | struct dp_packet **packets, int cnt, |
41ccaa24 | 3896 | bool may_steal, |
9080a111 | 3897 | const struct nlattr *actions, size_t actions_len) |
72865317 | 3898 | { |
41ccaa24 | 3899 | struct dp_netdev_execute_aux aux = { pmd }; |
9080a111 | 3900 | |
41ccaa24 | 3901 | odp_execute_actions(&aux, packets, cnt, may_steal, actions, |
8cbf4f47 | 3902 | actions_len, dp_execute_cb); |
72865317 BP |
3903 | } |
3904 | ||
3905 | const struct dpif_class dpif_netdev_class = { | |
72865317 | 3906 | "netdev", |
6553d06b | 3907 | dpif_netdev_init, |
2197d7ab | 3908 | dpif_netdev_enumerate, |
0aeaabc8 | 3909 | dpif_netdev_port_open_type, |
72865317 BP |
3910 | dpif_netdev_open, |
3911 | dpif_netdev_close, | |
7dab847a | 3912 | dpif_netdev_destroy, |
e4cfed38 PS |
3913 | dpif_netdev_run, |
3914 | dpif_netdev_wait, | |
72865317 | 3915 | dpif_netdev_get_stats, |
72865317 BP |
3916 | dpif_netdev_port_add, |
3917 | dpif_netdev_port_del, | |
3918 | dpif_netdev_port_query_by_number, | |
3919 | dpif_netdev_port_query_by_name, | |
98403001 | 3920 | NULL, /* port_get_pid */ |
b0ec0f27 BP |
3921 | dpif_netdev_port_dump_start, |
3922 | dpif_netdev_port_dump_next, | |
3923 | dpif_netdev_port_dump_done, | |
72865317 BP |
3924 | dpif_netdev_port_poll, |
3925 | dpif_netdev_port_poll_wait, | |
72865317 | 3926 | dpif_netdev_flow_flush, |
ac64794a BP |
3927 | dpif_netdev_flow_dump_create, |
3928 | dpif_netdev_flow_dump_destroy, | |
3929 | dpif_netdev_flow_dump_thread_create, | |
3930 | dpif_netdev_flow_dump_thread_destroy, | |
704a1e09 | 3931 | dpif_netdev_flow_dump_next, |
1a0c894a | 3932 | dpif_netdev_operate, |
6b31e073 RW |
3933 | NULL, /* recv_set */ |
3934 | NULL, /* handlers_set */ | |
f2eee189 | 3935 | dpif_netdev_pmd_set, |
5bf93d67 | 3936 | dpif_netdev_queue_to_priority, |
6b31e073 RW |
3937 | NULL, /* recv */ |
3938 | NULL, /* recv_wait */ | |
3939 | NULL, /* recv_purge */ | |
e4e74c3a | 3940 | dpif_netdev_register_dp_purge_cb, |
6b31e073 RW |
3941 | dpif_netdev_register_upcall_cb, |
3942 | dpif_netdev_enable_upcall, | |
3943 | dpif_netdev_disable_upcall, | |
b5cbbcf6 | 3944 | dpif_netdev_get_datapath_version, |
b77d9629 DDP |
3945 | NULL, /* ct_dump_start */ |
3946 | NULL, /* ct_dump_next */ | |
3947 | NULL, /* ct_dump_done */ | |
a0f7b6d5 | 3948 | NULL, /* ct_flush */ |
72865317 | 3949 | }; |
614c4892 | 3950 | |
74cc3969 BP |
3951 | static void |
3952 | dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3953 | const char *argv[], void *aux OVS_UNUSED) | |
3954 | { | |
59e6d833 BP |
3955 | struct dp_netdev_port *old_port; |
3956 | struct dp_netdev_port *new_port; | |
74cc3969 | 3957 | struct dp_netdev *dp; |
ff073a71 | 3958 | odp_port_t port_no; |
74cc3969 | 3959 | |
8a4e3a85 | 3960 | ovs_mutex_lock(&dp_netdev_mutex); |
74cc3969 BP |
3961 | dp = shash_find_data(&dp_netdevs, argv[1]); |
3962 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
8a4e3a85 | 3963 | ovs_mutex_unlock(&dp_netdev_mutex); |
74cc3969 BP |
3964 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); |
3965 | return; | |
3966 | } | |
8a4e3a85 BP |
3967 | ovs_refcount_ref(&dp->ref_cnt); |
3968 | ovs_mutex_unlock(&dp_netdev_mutex); | |
74cc3969 | 3969 | |
59e6d833 BP |
3970 | ovs_mutex_lock(&dp->port_mutex); |
3971 | if (get_port_by_name(dp, argv[2], &old_port)) { | |
74cc3969 | 3972 | unixctl_command_reply_error(conn, "unknown port"); |
8a4e3a85 | 3973 | goto exit; |
74cc3969 BP |
3974 | } |
3975 | ||
ff073a71 BP |
3976 | port_no = u32_to_odp(atoi(argv[3])); |
3977 | if (!port_no || port_no == ODPP_NONE) { | |
74cc3969 | 3978 | unixctl_command_reply_error(conn, "bad port number"); |
8a4e3a85 | 3979 | goto exit; |
74cc3969 | 3980 | } |
ff073a71 | 3981 | if (dp_netdev_lookup_port(dp, port_no)) { |
74cc3969 | 3982 | unixctl_command_reply_error(conn, "port number already in use"); |
8a4e3a85 | 3983 | goto exit; |
74cc3969 | 3984 | } |
59e6d833 BP |
3985 | |
3986 | /* Remove old port. */ | |
35303d71 | 3987 | cmap_remove(&dp->ports, &old_port->node, hash_port_no(old_port->port_no)); |
59e6d833 BP |
3988 | ovsrcu_postpone(free, old_port); |
3989 | ||
3990 | /* Insert new port (cmap semantics mean we cannot re-insert 'old_port'). */ | |
3991 | new_port = xmemdup(old_port, sizeof *old_port); | |
35303d71 | 3992 | new_port->port_no = port_no; |
59e6d833 BP |
3993 | cmap_insert(&dp->ports, &new_port->node, hash_port_no(port_no)); |
3994 | ||
d33ed218 | 3995 | seq_change(dp->port_seq); |
74cc3969 | 3996 | unixctl_command_reply(conn, NULL); |
8a4e3a85 BP |
3997 | |
3998 | exit: | |
59e6d833 | 3999 | ovs_mutex_unlock(&dp->port_mutex); |
8a4e3a85 | 4000 | dp_netdev_unref(dp); |
74cc3969 BP |
4001 | } |
4002 | ||
c40b890f BP |
4003 | static void |
4004 | dpif_dummy_delete_port(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
4005 | const char *argv[], void *aux OVS_UNUSED) | |
4006 | { | |
4007 | struct dp_netdev_port *port; | |
4008 | struct dp_netdev *dp; | |
4009 | ||
4010 | ovs_mutex_lock(&dp_netdev_mutex); | |
4011 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
4012 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
4013 | ovs_mutex_unlock(&dp_netdev_mutex); | |
4014 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); | |
4015 | return; | |
4016 | } | |
4017 | ovs_refcount_ref(&dp->ref_cnt); | |
4018 | ovs_mutex_unlock(&dp_netdev_mutex); | |
4019 | ||
4020 | ovs_mutex_lock(&dp->port_mutex); | |
4021 | if (get_port_by_name(dp, argv[2], &port)) { | |
4022 | unixctl_command_reply_error(conn, "unknown port"); | |
35303d71 | 4023 | } else if (port->port_no == ODPP_LOCAL) { |
c40b890f BP |
4024 | unixctl_command_reply_error(conn, "can't delete local port"); |
4025 | } else { | |
4026 | do_del_port(dp, port); | |
4027 | unixctl_command_reply(conn, NULL); | |
4028 | } | |
4029 | ovs_mutex_unlock(&dp->port_mutex); | |
4030 | ||
4031 | dp_netdev_unref(dp); | |
4032 | } | |
4033 | ||
0cbfe35d BP |
4034 | static void |
4035 | dpif_dummy_register__(const char *type) | |
4036 | { | |
4037 | struct dpif_class *class; | |
4038 | ||
4039 | class = xmalloc(sizeof *class); | |
4040 | *class = dpif_netdev_class; | |
4041 | class->type = xstrdup(type); | |
4042 | dp_register_provider(class); | |
4043 | } | |
4044 | ||
8420c7ad BP |
4045 | static void |
4046 | dpif_dummy_override(const char *type) | |
4047 | { | |
65d43fdc YT |
4048 | int error; |
4049 | ||
4050 | /* | |
4051 | * Ignore EAFNOSUPPORT to allow --enable-dummy=system with | |
4052 | * a userland-only build. It's useful for testsuite. | |
4053 | */ | |
4054 | error = dp_unregister_provider(type); | |
4055 | if (error == 0 || error == EAFNOSUPPORT) { | |
8420c7ad BP |
4056 | dpif_dummy_register__(type); |
4057 | } | |
4058 | } | |
4059 | ||
614c4892 | 4060 | void |
8420c7ad | 4061 | dpif_dummy_register(enum dummy_level level) |
614c4892 | 4062 | { |
8420c7ad | 4063 | if (level == DUMMY_OVERRIDE_ALL) { |
0cbfe35d BP |
4064 | struct sset types; |
4065 | const char *type; | |
4066 | ||
4067 | sset_init(&types); | |
4068 | dp_enumerate_types(&types); | |
4069 | SSET_FOR_EACH (type, &types) { | |
8420c7ad | 4070 | dpif_dummy_override(type); |
0cbfe35d BP |
4071 | } |
4072 | sset_destroy(&types); | |
8420c7ad BP |
4073 | } else if (level == DUMMY_OVERRIDE_SYSTEM) { |
4074 | dpif_dummy_override("system"); | |
614c4892 | 4075 | } |
0cbfe35d BP |
4076 | |
4077 | dpif_dummy_register__("dummy"); | |
74cc3969 BP |
4078 | |
4079 | unixctl_command_register("dpif-dummy/change-port-number", | |
74467d5c | 4080 | "dp port new-number", |
74cc3969 | 4081 | 3, 3, dpif_dummy_change_port_number, NULL); |
74467d5c | 4082 | unixctl_command_register("dpif-dummy/delete-port", "dp port", |
c40b890f | 4083 | 2, 2, dpif_dummy_delete_port, NULL); |
614c4892 | 4084 | } |
0de8783a JR |
4085 | \f |
4086 | /* Datapath Classifier. */ | |
4087 | ||
4088 | /* A set of rules that all have the same fields wildcarded. */ | |
4089 | struct dpcls_subtable { | |
4090 | /* The fields are only used by writers. */ | |
4091 | struct cmap_node cmap_node OVS_GUARDED; /* Within dpcls 'subtables_map'. */ | |
4092 | ||
4093 | /* These fields are accessed by readers. */ | |
4094 | struct cmap rules; /* Contains "struct dpcls_rule"s. */ | |
4095 | struct netdev_flow_key mask; /* Wildcards for fields (const). */ | |
4096 | /* 'mask' must be the last field, additional space is allocated here. */ | |
4097 | }; | |
4098 | ||
4099 | /* Initializes 'cls' as a classifier that initially contains no classification | |
4100 | * rules. */ | |
4101 | static void | |
4102 | dpcls_init(struct dpcls *cls) | |
4103 | { | |
4104 | cmap_init(&cls->subtables_map); | |
4105 | pvector_init(&cls->subtables); | |
4106 | } | |
4107 | ||
4108 | static void | |
4109 | dpcls_destroy_subtable(struct dpcls *cls, struct dpcls_subtable *subtable) | |
4110 | { | |
4111 | pvector_remove(&cls->subtables, subtable); | |
4112 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, | |
4113 | subtable->mask.hash); | |
4114 | cmap_destroy(&subtable->rules); | |
4115 | ovsrcu_postpone(free, subtable); | |
4116 | } | |
4117 | ||
4118 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
4119 | * caller's responsibility. | |
4120 | * May only be called after all the readers have been terminated. */ | |
4121 | static void | |
4122 | dpcls_destroy(struct dpcls *cls) | |
4123 | { | |
4124 | if (cls) { | |
4125 | struct dpcls_subtable *subtable; | |
4126 | ||
4127 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
361d808d | 4128 | ovs_assert(cmap_count(&subtable->rules) == 0); |
0de8783a JR |
4129 | dpcls_destroy_subtable(cls, subtable); |
4130 | } | |
4131 | cmap_destroy(&cls->subtables_map); | |
4132 | pvector_destroy(&cls->subtables); | |
4133 | } | |
4134 | } | |
4135 | ||
4136 | static struct dpcls_subtable * | |
4137 | dpcls_create_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
4138 | { | |
4139 | struct dpcls_subtable *subtable; | |
4140 | ||
4141 | /* Need to add one. */ | |
caeb4906 JR |
4142 | subtable = xmalloc(sizeof *subtable |
4143 | - sizeof subtable->mask.mf + mask->len); | |
0de8783a JR |
4144 | cmap_init(&subtable->rules); |
4145 | netdev_flow_key_clone(&subtable->mask, mask); | |
4146 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, mask->hash); | |
4147 | pvector_insert(&cls->subtables, subtable, 0); | |
802f84ff | 4148 | pvector_publish(&cls->subtables); |
0de8783a JR |
4149 | |
4150 | return subtable; | |
4151 | } | |
4152 | ||
4153 | static inline struct dpcls_subtable * | |
4154 | dpcls_find_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
4155 | { | |
4156 | struct dpcls_subtable *subtable; | |
4157 | ||
4158 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, mask->hash, | |
4159 | &cls->subtables_map) { | |
4160 | if (netdev_flow_key_equal(&subtable->mask, mask)) { | |
4161 | return subtable; | |
4162 | } | |
4163 | } | |
4164 | return dpcls_create_subtable(cls, mask); | |
4165 | } | |
4166 | ||
4167 | /* Insert 'rule' into 'cls'. */ | |
4168 | static void | |
4169 | dpcls_insert(struct dpcls *cls, struct dpcls_rule *rule, | |
4170 | const struct netdev_flow_key *mask) | |
4171 | { | |
4172 | struct dpcls_subtable *subtable = dpcls_find_subtable(cls, mask); | |
4173 | ||
4174 | rule->mask = &subtable->mask; | |
4175 | cmap_insert(&subtable->rules, &rule->cmap_node, rule->flow.hash); | |
4176 | } | |
4177 | ||
4178 | /* Removes 'rule' from 'cls', also destructing the 'rule'. */ | |
4179 | static void | |
4180 | dpcls_remove(struct dpcls *cls, struct dpcls_rule *rule) | |
4181 | { | |
4182 | struct dpcls_subtable *subtable; | |
4183 | ||
4184 | ovs_assert(rule->mask); | |
4185 | ||
4186 | INIT_CONTAINER(subtable, rule->mask, mask); | |
4187 | ||
4188 | if (cmap_remove(&subtable->rules, &rule->cmap_node, rule->flow.hash) | |
4189 | == 0) { | |
4190 | dpcls_destroy_subtable(cls, subtable); | |
802f84ff | 4191 | pvector_publish(&cls->subtables); |
0de8783a JR |
4192 | } |
4193 | } | |
4194 | ||
361d808d JR |
4195 | /* Returns true if 'target' satisfies 'key' in 'mask', that is, if each 1-bit |
4196 | * in 'mask' the values in 'key' and 'target' are the same. */ | |
0de8783a JR |
4197 | static inline bool |
4198 | dpcls_rule_matches_key(const struct dpcls_rule *rule, | |
4199 | const struct netdev_flow_key *target) | |
4200 | { | |
09b0fa9c JR |
4201 | const uint64_t *keyp = miniflow_get_values(&rule->flow.mf); |
4202 | const uint64_t *maskp = miniflow_get_values(&rule->mask->mf); | |
5fcff47b | 4203 | uint64_t value; |
0de8783a | 4204 | |
5fcff47b JR |
4205 | NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(value, target, rule->flow.mf.map) { |
4206 | if (OVS_UNLIKELY((value & *maskp++) != *keyp++)) { | |
0de8783a JR |
4207 | return false; |
4208 | } | |
4209 | } | |
4210 | return true; | |
4211 | } | |
4212 | ||
4213 | /* For each miniflow in 'flows' performs a classifier lookup writing the result | |
4214 | * into the corresponding slot in 'rules'. If a particular entry in 'flows' is | |
4215 | * NULL it is skipped. | |
4216 | * | |
4217 | * This function is optimized for use in the userspace datapath and therefore | |
4218 | * does not implement a lot of features available in the standard | |
4219 | * classifier_lookup() function. Specifically, it does not implement | |
4220 | * priorities, instead returning any rule which matches the flow. | |
4221 | * | |
4222 | * Returns true if all flows found a corresponding rule. */ | |
4223 | static bool | |
4224 | dpcls_lookup(const struct dpcls *cls, const struct netdev_flow_key keys[], | |
4225 | struct dpcls_rule **rules, const size_t cnt) | |
4226 | { | |
4227 | /* The batch size 16 was experimentally found faster than 8 or 32. */ | |
4228 | typedef uint16_t map_type; | |
4229 | #define MAP_BITS (sizeof(map_type) * CHAR_BIT) | |
4230 | ||
4231 | #if !defined(__CHECKER__) && !defined(_WIN32) | |
4232 | const int N_MAPS = DIV_ROUND_UP(cnt, MAP_BITS); | |
4233 | #else | |
cd159f1a | 4234 | enum { N_MAPS = DIV_ROUND_UP(NETDEV_MAX_BURST, MAP_BITS) }; |
0de8783a JR |
4235 | #endif |
4236 | map_type maps[N_MAPS]; | |
4237 | struct dpcls_subtable *subtable; | |
4238 | ||
4239 | memset(maps, 0xff, sizeof maps); | |
4240 | if (cnt % MAP_BITS) { | |
4241 | maps[N_MAPS - 1] >>= MAP_BITS - cnt % MAP_BITS; /* Clear extra bits. */ | |
4242 | } | |
4243 | memset(rules, 0, cnt * sizeof *rules); | |
4244 | ||
4245 | PVECTOR_FOR_EACH (subtable, &cls->subtables) { | |
4246 | const struct netdev_flow_key *mkeys = keys; | |
4247 | struct dpcls_rule **mrules = rules; | |
4248 | map_type remains = 0; | |
4249 | int m; | |
4250 | ||
4251 | BUILD_ASSERT_DECL(sizeof remains == sizeof *maps); | |
4252 | ||
4253 | for (m = 0; m < N_MAPS; m++, mkeys += MAP_BITS, mrules += MAP_BITS) { | |
4254 | uint32_t hashes[MAP_BITS]; | |
4255 | const struct cmap_node *nodes[MAP_BITS]; | |
4256 | unsigned long map = maps[m]; | |
4257 | int i; | |
4258 | ||
4259 | if (!map) { | |
4260 | continue; /* Skip empty maps. */ | |
4261 | } | |
4262 | ||
4263 | /* Compute hashes for the remaining keys. */ | |
3ee6026a | 4264 | ULLONG_FOR_EACH_1(i, map) { |
0de8783a JR |
4265 | hashes[i] = netdev_flow_key_hash_in_mask(&mkeys[i], |
4266 | &subtable->mask); | |
4267 | } | |
4268 | /* Lookup. */ | |
4269 | map = cmap_find_batch(&subtable->rules, map, hashes, nodes); | |
4270 | /* Check results. */ | |
3ee6026a | 4271 | ULLONG_FOR_EACH_1(i, map) { |
0de8783a JR |
4272 | struct dpcls_rule *rule; |
4273 | ||
4274 | CMAP_NODE_FOR_EACH (rule, cmap_node, nodes[i]) { | |
4275 | if (OVS_LIKELY(dpcls_rule_matches_key(rule, &mkeys[i]))) { | |
4276 | mrules[i] = rule; | |
4277 | goto next; | |
4278 | } | |
4279 | } | |
3ee6026a | 4280 | ULLONG_SET0(map, i); /* Did not match. */ |
0de8783a JR |
4281 | next: |
4282 | ; /* Keep Sparse happy. */ | |
4283 | } | |
4284 | maps[m] &= ~map; /* Clear the found rules. */ | |
4285 | remains |= maps[m]; | |
4286 | } | |
4287 | if (!remains) { | |
4288 | return true; /* All found. */ | |
4289 | } | |
4290 | } | |
4291 | return false; /* Some misses. */ | |
4292 | } |