]>
Commit | Line | Data |
---|---|---|
72865317 | 1 | /* |
ff073a71 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. |
72865317 BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
db73f716 | 18 | #include "dpif-netdev.h" |
72865317 | 19 | |
72865317 BP |
20 | #include <ctype.h> |
21 | #include <errno.h> | |
22 | #include <fcntl.h> | |
23 | #include <inttypes.h> | |
72865317 | 24 | #include <netinet/in.h> |
9d82ec47 | 25 | #include <sys/socket.h> |
7f3adc00 | 26 | #include <net/if.h> |
cdee00fd | 27 | #include <stdint.h> |
72865317 BP |
28 | #include <stdlib.h> |
29 | #include <string.h> | |
30 | #include <sys/ioctl.h> | |
31 | #include <sys/stat.h> | |
72865317 BP |
32 | #include <unistd.h> |
33 | ||
59e6d833 | 34 | #include "cmap.h" |
72865317 | 35 | #include "csum.h" |
e14deea0 | 36 | #include "dp-packet.h" |
614c4892 | 37 | #include "dpif.h" |
72865317 | 38 | #include "dpif-provider.h" |
614c4892 | 39 | #include "dummy.h" |
36956a7d | 40 | #include "dynamic-string.h" |
afae68b1 | 41 | #include "fat-rwlock.h" |
72865317 | 42 | #include "flow.h" |
9f361d6b | 43 | #include "cmap.h" |
6c3eee82 | 44 | #include "latch.h" |
72865317 | 45 | #include "list.h" |
0de8783a | 46 | #include "match.h" |
8c301900 | 47 | #include "meta-flow.h" |
72865317 | 48 | #include "netdev.h" |
8617afff | 49 | #include "netdev-dpdk.h" |
de281153 | 50 | #include "netdev-vport.h" |
cdee00fd | 51 | #include "netlink.h" |
f094af7b | 52 | #include "odp-execute.h" |
72865317 BP |
53 | #include "odp-util.h" |
54 | #include "ofp-print.h" | |
55 | #include "ofpbuf.h" | |
5a034064 | 56 | #include "ovs-numa.h" |
61e7deb1 | 57 | #include "ovs-rcu.h" |
72865317 BP |
58 | #include "packets.h" |
59 | #include "poll-loop.h" | |
0de8783a | 60 | #include "pvector.h" |
26c6b6cd | 61 | #include "random.h" |
d33ed218 | 62 | #include "seq.h" |
462278db | 63 | #include "shash.h" |
0cbfe35d | 64 | #include "sset.h" |
72865317 | 65 | #include "timeval.h" |
a36de779 | 66 | #include "tnl-arp-cache.h" |
74cc3969 | 67 | #include "unixctl.h" |
72865317 | 68 | #include "util.h" |
e6211adc | 69 | #include "openvswitch/vlog.h" |
5136ce49 | 70 | |
d98e6007 | 71 | VLOG_DEFINE_THIS_MODULE(dpif_netdev); |
72865317 | 72 | |
8bb113da | 73 | #define FLOW_DUMP_MAX_BATCH 50 |
adcf00ba AZ |
74 | /* Use per thread recirc_depth to prevent recirculation loop. */ |
75 | #define MAX_RECIRC_DEPTH 5 | |
76 | DEFINE_STATIC_PER_THREAD_DATA(uint32_t, recirc_depth, 0) | |
e4cfed38 | 77 | |
72865317 | 78 | /* Configuration parameters. */ |
72865317 BP |
79 | enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ |
80 | ||
8a4e3a85 BP |
81 | /* Protects against changes to 'dp_netdevs'. */ |
82 | static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER; | |
83 | ||
84 | /* Contains all 'struct dp_netdev's. */ | |
85 | static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex) | |
86 | = SHASH_INITIALIZER(&dp_netdevs); | |
87 | ||
623540e4 | 88 | static struct vlog_rate_limit upcall_rl = VLOG_RATE_LIMIT_INIT(600, 600); |
6b31e073 | 89 | |
2494ccd7 JS |
90 | static struct odp_support dp_netdev_support = { |
91 | .max_mpls_depth = SIZE_MAX, | |
92 | .recirc = true, | |
93 | }; | |
94 | ||
79df317f | 95 | /* Stores a miniflow with inline values */ |
9bbf1c3d | 96 | |
9bbf1c3d | 97 | struct netdev_flow_key { |
caeb4906 JR |
98 | uint32_t hash; /* Hash function differs for different users. */ |
99 | uint32_t len; /* Length of the following miniflow (incl. map). */ | |
0de8783a | 100 | struct miniflow mf; |
8fd47924 | 101 | uint64_t buf[FLOW_MAX_PACKET_U64S]; |
9bbf1c3d DDP |
102 | }; |
103 | ||
104 | /* Exact match cache for frequently used flows | |
105 | * | |
106 | * The cache uses a 32-bit hash of the packet (which can be the RSS hash) to | |
107 | * search its entries for a miniflow that matches exactly the miniflow of the | |
0de8783a | 108 | * packet. It stores the 'dpcls_rule' (rule) that matches the miniflow. |
9bbf1c3d DDP |
109 | * |
110 | * A cache entry holds a reference to its 'dp_netdev_flow'. | |
111 | * | |
112 | * A miniflow with a given hash can be in one of EM_FLOW_HASH_SEGS different | |
113 | * entries. The 32-bit hash is split into EM_FLOW_HASH_SEGS values (each of | |
114 | * them is EM_FLOW_HASH_SHIFT bits wide and the remainder is thrown away). Each | |
115 | * value is the index of a cache entry where the miniflow could be. | |
116 | * | |
117 | * | |
118 | * Thread-safety | |
119 | * ============= | |
120 | * | |
121 | * Each pmd_thread has its own private exact match cache. | |
122 | * If dp_netdev_input is not called from a pmd thread, a mutex is used. | |
123 | */ | |
124 | ||
fc82e877 | 125 | #define EM_FLOW_HASH_SHIFT 13 |
9bbf1c3d DDP |
126 | #define EM_FLOW_HASH_ENTRIES (1u << EM_FLOW_HASH_SHIFT) |
127 | #define EM_FLOW_HASH_MASK (EM_FLOW_HASH_ENTRIES - 1) | |
128 | #define EM_FLOW_HASH_SEGS 2 | |
129 | ||
130 | struct emc_entry { | |
9bbf1c3d | 131 | struct dp_netdev_flow *flow; |
0de8783a | 132 | struct netdev_flow_key key; /* key.hash used for emc hash value. */ |
9bbf1c3d DDP |
133 | }; |
134 | ||
135 | struct emc_cache { | |
136 | struct emc_entry entries[EM_FLOW_HASH_ENTRIES]; | |
67ad54cb | 137 | int sweep_idx; /* For emc_cache_slow_sweep(). */ |
9bbf1c3d DDP |
138 | }; |
139 | ||
140 | /* Iterate in the exact match cache through every entry that might contain a | |
141 | * miniflow with hash 'HASH'. */ | |
142 | #define EMC_FOR_EACH_POS_WITH_HASH(EMC, CURRENT_ENTRY, HASH) \ | |
143 | for (uint32_t i__ = 0, srch_hash__ = (HASH); \ | |
144 | (CURRENT_ENTRY) = &(EMC)->entries[srch_hash__ & EM_FLOW_HASH_MASK], \ | |
145 | i__ < EM_FLOW_HASH_SEGS; \ | |
146 | i__++, srch_hash__ >>= EM_FLOW_HASH_SHIFT) | |
0de8783a JR |
147 | \f |
148 | /* Simple non-wildcarding single-priority classifier. */ | |
149 | ||
150 | struct dpcls { | |
151 | struct cmap subtables_map; | |
152 | struct pvector subtables; | |
153 | }; | |
9bbf1c3d | 154 | |
0de8783a JR |
155 | /* A rule to be inserted to the classifier. */ |
156 | struct dpcls_rule { | |
157 | struct cmap_node cmap_node; /* Within struct dpcls_subtable 'rules'. */ | |
158 | struct netdev_flow_key *mask; /* Subtable's mask. */ | |
159 | struct netdev_flow_key flow; /* Matching key. */ | |
160 | /* 'flow' must be the last field, additional space is allocated here. */ | |
161 | }; | |
162 | ||
163 | static void dpcls_init(struct dpcls *); | |
164 | static void dpcls_destroy(struct dpcls *); | |
165 | static void dpcls_insert(struct dpcls *, struct dpcls_rule *, | |
166 | const struct netdev_flow_key *mask); | |
167 | static void dpcls_remove(struct dpcls *, struct dpcls_rule *); | |
168 | static bool dpcls_lookup(const struct dpcls *cls, | |
169 | const struct netdev_flow_key keys[], | |
170 | struct dpcls_rule **rules, size_t cnt); | |
171 | \f | |
8a4e3a85 BP |
172 | /* Datapath based on the network device interface from netdev.h. |
173 | * | |
174 | * | |
175 | * Thread-safety | |
176 | * ============= | |
177 | * | |
178 | * Some members, marked 'const', are immutable. Accessing other members | |
179 | * requires synchronization, as noted in more detail below. | |
180 | * | |
181 | * Acquisition order is, from outermost to innermost: | |
182 | * | |
183 | * dp_netdev_mutex (global) | |
59e6d833 | 184 | * port_mutex |
8a4e3a85 | 185 | */ |
72865317 | 186 | struct dp_netdev { |
8a4e3a85 BP |
187 | const struct dpif_class *const class; |
188 | const char *const name; | |
6b31e073 | 189 | struct dpif *dpif; |
6a8267c5 BP |
190 | struct ovs_refcount ref_cnt; |
191 | atomic_flag destroyed; | |
72865317 | 192 | |
8a4e3a85 BP |
193 | /* Ports. |
194 | * | |
59e6d833 BP |
195 | * Protected by RCU. Take the mutex to add or remove ports. */ |
196 | struct ovs_mutex port_mutex; | |
197 | struct cmap ports; | |
d33ed218 | 198 | struct seq *port_seq; /* Incremented whenever a port changes. */ |
6c3eee82 | 199 | |
6b31e073 RW |
200 | /* Protects access to ofproto-dpif-upcall interface during revalidator |
201 | * thread synchronization. */ | |
202 | struct fat_rwlock upcall_rwlock; | |
623540e4 EJ |
203 | upcall_callback *upcall_cb; /* Callback function for executing upcalls. */ |
204 | void *upcall_aux; | |
6b31e073 | 205 | |
65f13b50 AW |
206 | /* Stores all 'struct dp_netdev_pmd_thread's. */ |
207 | struct cmap poll_threads; | |
208 | ||
209 | /* Protects the access of the 'struct dp_netdev_pmd_thread' | |
210 | * instance for non-pmd thread. */ | |
211 | struct ovs_mutex non_pmd_mutex; | |
212 | ||
213 | /* Each pmd thread will store its pointer to | |
214 | * 'struct dp_netdev_pmd_thread' in 'per_pmd_key'. */ | |
215 | ovsthread_key_t per_pmd_key; | |
f2eee189 AW |
216 | |
217 | /* Number of rx queues for each dpdk interface and the cpu mask | |
218 | * for pin of pmd threads. */ | |
219 | size_t n_dpdk_rxqs; | |
220 | char *pmd_cmask; | |
a36de779 | 221 | uint64_t last_tnl_conf_seq; |
72865317 BP |
222 | }; |
223 | ||
8a4e3a85 | 224 | static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp, |
59e6d833 | 225 | odp_port_t); |
ff073a71 | 226 | |
51852a57 | 227 | enum dp_stat_type { |
abcf3ef4 DDP |
228 | DP_STAT_EXACT_HIT, /* Packets that had an exact match (emc). */ |
229 | DP_STAT_MASKED_HIT, /* Packets that matched in the flow table. */ | |
51852a57 BP |
230 | DP_STAT_MISS, /* Packets that did not match. */ |
231 | DP_STAT_LOST, /* Packets not passed up to the client. */ | |
232 | DP_N_STATS | |
233 | }; | |
234 | ||
55e3ca97 DDP |
235 | enum pmd_cycles_counter_type { |
236 | PMD_CYCLES_POLLING, /* Cycles spent polling NICs. */ | |
237 | PMD_CYCLES_PROCESSING, /* Cycles spent processing packets */ | |
238 | PMD_N_CYCLES | |
239 | }; | |
240 | ||
72865317 BP |
241 | /* A port in a netdev-based datapath. */ |
242 | struct dp_netdev_port { | |
35303d71 | 243 | odp_port_t port_no; |
72865317 | 244 | struct netdev *netdev; |
efa2bcbb | 245 | struct cmap_node node; /* Node in dp_netdev's 'ports'. */ |
4b609110 | 246 | struct netdev_saved_flags *sf; |
55c955bd | 247 | struct netdev_rxq **rxq; |
b284085e | 248 | struct ovs_refcount ref_cnt; |
0cbfe35d | 249 | char *type; /* Port type as requested by user. */ |
72865317 BP |
250 | }; |
251 | ||
1c1e46ed AW |
252 | /* Contained by struct dp_netdev_flow's 'stats' member. */ |
253 | struct dp_netdev_flow_stats { | |
eb94da30 DDP |
254 | atomic_llong used; /* Last used time, in monotonic msecs. */ |
255 | atomic_ullong packet_count; /* Number of packets matched. */ | |
256 | atomic_ullong byte_count; /* Number of bytes matched. */ | |
257 | atomic_uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ | |
1c1e46ed AW |
258 | }; |
259 | ||
260 | /* A flow in 'dp_netdev_pmd_thread's 'flow_table'. | |
8a4e3a85 BP |
261 | * |
262 | * | |
263 | * Thread-safety | |
264 | * ============= | |
265 | * | |
266 | * Except near the beginning or ending of its lifespan, rule 'rule' belongs to | |
1c1e46ed | 267 | * its pmd thread's classifier. The text below calls this classifier 'cls'. |
8a4e3a85 BP |
268 | * |
269 | * Motivation | |
270 | * ---------- | |
271 | * | |
272 | * The thread safety rules described here for "struct dp_netdev_flow" are | |
273 | * motivated by two goals: | |
274 | * | |
275 | * - Prevent threads that read members of "struct dp_netdev_flow" from | |
276 | * reading bad data due to changes by some thread concurrently modifying | |
277 | * those members. | |
278 | * | |
279 | * - Prevent two threads making changes to members of a given "struct | |
280 | * dp_netdev_flow" from interfering with each other. | |
281 | * | |
282 | * | |
283 | * Rules | |
284 | * ----- | |
285 | * | |
ed79f89a DDP |
286 | * A flow 'flow' may be accessed without a risk of being freed during an RCU |
287 | * grace period. Code that needs to hold onto a flow for a while | |
288 | * should try incrementing 'flow->ref_cnt' with dp_netdev_flow_ref(). | |
8a4e3a85 BP |
289 | * |
290 | * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the | |
ed79f89a DDP |
291 | * flow from being deleted from 'cls' and it doesn't protect members of 'flow' |
292 | * from modification. | |
8a4e3a85 BP |
293 | * |
294 | * Some members, marked 'const', are immutable. Accessing other members | |
295 | * requires synchronization, as noted in more detail below. | |
296 | */ | |
72865317 | 297 | struct dp_netdev_flow { |
11e5cf1f | 298 | const struct flow flow; /* Unmasked flow that created this entry. */ |
8a4e3a85 | 299 | /* Hash table index by unmasked flow. */ |
1c1e46ed AW |
300 | const struct cmap_node node; /* In owning dp_netdev_pmd_thread's */ |
301 | /* 'flow_table'. */ | |
70e5ed6f | 302 | const ovs_u128 ufid; /* Unique flow identifier. */ |
bd5131ba | 303 | const unsigned pmd_id; /* The 'core_id' of pmd thread owning this */ |
1c1e46ed | 304 | /* flow. */ |
72865317 | 305 | |
ed79f89a DDP |
306 | /* Number of references. |
307 | * The classifier owns one reference. | |
308 | * Any thread trying to keep a rule from being freed should hold its own | |
309 | * reference. */ | |
310 | struct ovs_refcount ref_cnt; | |
311 | ||
11e5cf1f DDP |
312 | bool dead; |
313 | ||
1c1e46ed AW |
314 | /* Statistics. */ |
315 | struct dp_netdev_flow_stats stats; | |
8a4e3a85 | 316 | |
45c626a3 | 317 | /* Actions. */ |
61e7deb1 | 318 | OVSRCU_TYPE(struct dp_netdev_actions *) actions; |
0de8783a | 319 | |
11e5cf1f DDP |
320 | /* While processing a group of input packets, the datapath uses the next |
321 | * member to store a pointer to the output batch for the flow. It is | |
322 | * reset after the batch has been sent out (See dp_netdev_queue_batches(), | |
323 | * packet_batch_init() and packet_batch_execute()). */ | |
324 | struct packet_batch *batch; | |
325 | ||
0de8783a JR |
326 | /* Packet classification. */ |
327 | struct dpcls_rule cr; /* In owning dp_netdev's 'cls'. */ | |
328 | /* 'cr' must be the last member. */ | |
72865317 BP |
329 | }; |
330 | ||
ed79f89a | 331 | static void dp_netdev_flow_unref(struct dp_netdev_flow *); |
9bbf1c3d | 332 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *); |
70e5ed6f JS |
333 | static int dpif_netdev_flow_from_nlattrs(const struct nlattr *, uint32_t, |
334 | struct flow *); | |
8a4e3a85 | 335 | |
a84cb64a BP |
336 | /* A set of datapath actions within a "struct dp_netdev_flow". |
337 | * | |
338 | * | |
339 | * Thread-safety | |
340 | * ============= | |
341 | * | |
45c626a3 | 342 | * A struct dp_netdev_actions 'actions' is protected with RCU. */ |
a84cb64a | 343 | struct dp_netdev_actions { |
a84cb64a BP |
344 | /* These members are immutable: they do not change during the struct's |
345 | * lifetime. */ | |
a84cb64a | 346 | unsigned int size; /* Size of 'actions', in bytes. */ |
9ff55ae2 | 347 | struct nlattr actions[]; /* Sequence of OVS_ACTION_ATTR_* attributes. */ |
a84cb64a BP |
348 | }; |
349 | ||
350 | struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *, | |
351 | size_t); | |
61e7deb1 BP |
352 | struct dp_netdev_actions *dp_netdev_flow_get_actions( |
353 | const struct dp_netdev_flow *); | |
354 | static void dp_netdev_actions_free(struct dp_netdev_actions *); | |
a84cb64a | 355 | |
1c1e46ed AW |
356 | /* Contained by struct dp_netdev_pmd_thread's 'stats' member. */ |
357 | struct dp_netdev_pmd_stats { | |
358 | /* Indexed by DP_STAT_*. */ | |
eb94da30 | 359 | atomic_ullong n[DP_N_STATS]; |
1c1e46ed AW |
360 | }; |
361 | ||
55e3ca97 DDP |
362 | /* Contained by struct dp_netdev_pmd_thread's 'cycle' member. */ |
363 | struct dp_netdev_pmd_cycles { | |
364 | /* Indexed by PMD_CYCLES_*. */ | |
365 | atomic_ullong n[PMD_N_CYCLES]; | |
366 | }; | |
367 | ||
e4cfed38 PS |
368 | /* PMD: Poll modes drivers. PMD accesses devices via polling to eliminate |
369 | * the performance overhead of interrupt processing. Therefore netdev can | |
370 | * not implement rx-wait for these devices. dpif-netdev needs to poll | |
371 | * these device to check for recv buffer. pmd-thread does polling for | |
1c1e46ed | 372 | * devices assigned to itself. |
e4cfed38 PS |
373 | * |
374 | * DPDK used PMD for accessing NIC. | |
375 | * | |
65f13b50 AW |
376 | * Note, instance with cpu core id NON_PMD_CORE_ID will be reserved for |
377 | * I/O of all non-pmd threads. There will be no actual thread created | |
378 | * for the instance. | |
1c1e46ed AW |
379 | * |
380 | * Each struct has its own flow table and classifier. Packets received | |
381 | * from managed ports are looked up in the corresponding pmd thread's | |
382 | * flow table, and are executed with the found actions. | |
383 | * */ | |
65f13b50 | 384 | struct dp_netdev_pmd_thread { |
6c3eee82 | 385 | struct dp_netdev *dp; |
1c1e46ed | 386 | struct ovs_refcount ref_cnt; /* Every reference must be refcount'ed. */ |
65f13b50 | 387 | struct cmap_node node; /* In 'dp->poll_threads'. */ |
accf8626 AW |
388 | |
389 | pthread_cond_t cond; /* For synchronizing pmd thread reload. */ | |
390 | struct ovs_mutex cond_mutex; /* Mutex for condition variable. */ | |
391 | ||
65f13b50 AW |
392 | /* Per thread exact-match cache. Note, the instance for cpu core |
393 | * NON_PMD_CORE_ID can be accessed by multiple threads, and thusly | |
394 | * need to be protected (e.g. by 'dp_netdev_mutex'). All other | |
395 | * instances will only be accessed by its own pmd thread. */ | |
9bbf1c3d | 396 | struct emc_cache flow_cache; |
1c1e46ed AW |
397 | |
398 | /* Classifier and Flow-Table. | |
399 | * | |
400 | * Writers of 'flow_table' must take the 'flow_mutex'. Corresponding | |
401 | * changes to 'cls' must be made while still holding the 'flow_mutex'. | |
402 | */ | |
403 | struct ovs_mutex flow_mutex; | |
404 | struct dpcls cls; | |
405 | struct cmap flow_table OVS_GUARDED; /* Flow table. */ | |
406 | ||
407 | /* Statistics. */ | |
408 | struct dp_netdev_pmd_stats stats; | |
409 | ||
55e3ca97 DDP |
410 | /* Cycles counters */ |
411 | struct dp_netdev_pmd_cycles cycles; | |
412 | ||
413 | /* Used to count cicles. See 'cycles_counter_end()' */ | |
414 | unsigned long long last_cycles; | |
415 | ||
65f13b50 AW |
416 | struct latch exit_latch; /* For terminating the pmd thread. */ |
417 | atomic_uint change_seq; /* For reloading pmd ports. */ | |
6c3eee82 | 418 | pthread_t thread; |
65f13b50 AW |
419 | int index; /* Idx of this pmd thread among pmd*/ |
420 | /* threads on same numa node. */ | |
bd5131ba | 421 | unsigned core_id; /* CPU core id of this pmd thread. */ |
65f13b50 | 422 | int numa_id; /* numa node id of this pmd thread. */ |
3bcc10c0 DDP |
423 | int tx_qid; /* Queue id used by this pmd thread to |
424 | * send packets on all netdevs */ | |
6553d06b DDP |
425 | |
426 | /* Only a pmd thread can write on its own 'cycles' and 'stats'. | |
427 | * The main thread keeps 'stats_zero' and 'cycles_zero' as base | |
428 | * values and subtracts them from 'stats' and 'cycles' before | |
429 | * reporting to the user */ | |
430 | unsigned long long stats_zero[DP_N_STATS]; | |
431 | uint64_t cycles_zero[PMD_N_CYCLES]; | |
6c3eee82 BP |
432 | }; |
433 | ||
84067a4c JR |
434 | #define PMD_INITIAL_SEQ 1 |
435 | ||
72865317 BP |
436 | /* Interface to netdev-based datapath. */ |
437 | struct dpif_netdev { | |
438 | struct dpif dpif; | |
439 | struct dp_netdev *dp; | |
d33ed218 | 440 | uint64_t last_port_seq; |
72865317 BP |
441 | }; |
442 | ||
8a4e3a85 | 443 | static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no, |
59e6d833 | 444 | struct dp_netdev_port **portp); |
8a4e3a85 | 445 | static int get_port_by_name(struct dp_netdev *dp, const char *devname, |
59e6d833 | 446 | struct dp_netdev_port **portp); |
8a4e3a85 BP |
447 | static void dp_netdev_free(struct dp_netdev *) |
448 | OVS_REQUIRES(dp_netdev_mutex); | |
8a4e3a85 BP |
449 | static int do_add_port(struct dp_netdev *dp, const char *devname, |
450 | const char *type, odp_port_t port_no) | |
59e6d833 | 451 | OVS_REQUIRES(dp->port_mutex); |
c40b890f | 452 | static void do_del_port(struct dp_netdev *dp, struct dp_netdev_port *) |
59e6d833 | 453 | OVS_REQUIRES(dp->port_mutex); |
614c4892 BP |
454 | static int dpif_netdev_open(const struct dpif_class *, const char *name, |
455 | bool create, struct dpif **); | |
65f13b50 | 456 | static void dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 457 | struct dp_packet **, int c, |
41ccaa24 | 458 | bool may_steal, |
4edb9ae9 | 459 | const struct nlattr *actions, |
e4cfed38 | 460 | size_t actions_len); |
65f13b50 | 461 | static void dp_netdev_input(struct dp_netdev_pmd_thread *, |
e14deea0 | 462 | struct dp_packet **, int cnt); |
41ccaa24 | 463 | |
6b31e073 | 464 | static void dp_netdev_disable_upcall(struct dp_netdev *); |
accf8626 | 465 | void dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd); |
65f13b50 AW |
466 | static void dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, |
467 | struct dp_netdev *dp, int index, | |
bd5131ba | 468 | unsigned core_id, int numa_id); |
1c1e46ed | 469 | static void dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd); |
f2eee189 | 470 | static void dp_netdev_set_nonpmd(struct dp_netdev *dp); |
b19befae | 471 | static struct dp_netdev_pmd_thread *dp_netdev_get_pmd(struct dp_netdev *dp, |
bd5131ba | 472 | unsigned core_id); |
1c1e46ed AW |
473 | static struct dp_netdev_pmd_thread * |
474 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos); | |
65f13b50 AW |
475 | static void dp_netdev_destroy_all_pmds(struct dp_netdev *dp); |
476 | static void dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
477 | static void dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id); | |
f2eee189 | 478 | static void dp_netdev_reset_pmd_threads(struct dp_netdev *dp); |
1c1e46ed AW |
479 | static bool dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd); |
480 | static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd); | |
481 | static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd); | |
72865317 | 482 | |
67ad54cb | 483 | static inline bool emc_entry_alive(struct emc_entry *ce); |
9bbf1c3d DDP |
484 | static void emc_clear_entry(struct emc_entry *ce); |
485 | ||
486 | static void | |
487 | emc_cache_init(struct emc_cache *flow_cache) | |
488 | { | |
489 | int i; | |
490 | ||
361d808d | 491 | BUILD_ASSERT(sizeof(struct miniflow) == 2 * sizeof(uint64_t)); |
caeb4906 | 492 | |
67ad54cb | 493 | flow_cache->sweep_idx = 0; |
9bbf1c3d DDP |
494 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { |
495 | flow_cache->entries[i].flow = NULL; | |
0de8783a | 496 | flow_cache->entries[i].key.hash = 0; |
09b0fa9c | 497 | flow_cache->entries[i].key.len = sizeof(struct miniflow); |
361d808d JR |
498 | flow_cache->entries[i].key.mf.tnl_map = 0; |
499 | flow_cache->entries[i].key.mf.pkt_map = 0; | |
9bbf1c3d DDP |
500 | } |
501 | } | |
502 | ||
503 | static void | |
504 | emc_cache_uninit(struct emc_cache *flow_cache) | |
505 | { | |
506 | int i; | |
507 | ||
508 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { | |
509 | emc_clear_entry(&flow_cache->entries[i]); | |
510 | } | |
511 | } | |
512 | ||
67ad54cb AW |
513 | /* Check and clear dead flow references slowly (one entry at each |
514 | * invocation). */ | |
515 | static void | |
516 | emc_cache_slow_sweep(struct emc_cache *flow_cache) | |
517 | { | |
518 | struct emc_entry *entry = &flow_cache->entries[flow_cache->sweep_idx]; | |
519 | ||
520 | if (!emc_entry_alive(entry)) { | |
521 | emc_clear_entry(entry); | |
522 | } | |
523 | flow_cache->sweep_idx = (flow_cache->sweep_idx + 1) & EM_FLOW_HASH_MASK; | |
524 | } | |
525 | ||
c4ea7529 BP |
526 | /* Returns true if 'dpif' is a netdev or dummy dpif, false otherwise. */ |
527 | bool | |
528 | dpif_is_netdev(const struct dpif *dpif) | |
529 | { | |
530 | return dpif->dpif_class->open == dpif_netdev_open; | |
531 | } | |
532 | ||
72865317 BP |
533 | static struct dpif_netdev * |
534 | dpif_netdev_cast(const struct dpif *dpif) | |
535 | { | |
c4ea7529 | 536 | ovs_assert(dpif_is_netdev(dpif)); |
72865317 BP |
537 | return CONTAINER_OF(dpif, struct dpif_netdev, dpif); |
538 | } | |
539 | ||
540 | static struct dp_netdev * | |
541 | get_dp_netdev(const struct dpif *dpif) | |
542 | { | |
543 | return dpif_netdev_cast(dpif)->dp; | |
544 | } | |
6553d06b DDP |
545 | \f |
546 | enum pmd_info_type { | |
547 | PMD_INFO_SHOW_STATS, /* show how cpu cycles are spent */ | |
548 | PMD_INFO_CLEAR_STATS /* set the cycles count to 0 */ | |
549 | }; | |
550 | ||
551 | static void | |
552 | pmd_info_show_stats(struct ds *reply, | |
553 | struct dp_netdev_pmd_thread *pmd, | |
554 | unsigned long long stats[DP_N_STATS], | |
555 | uint64_t cycles[PMD_N_CYCLES]) | |
556 | { | |
557 | unsigned long long total_packets = 0; | |
558 | uint64_t total_cycles = 0; | |
559 | int i; | |
560 | ||
561 | /* These loops subtracts reference values ('*_zero') from the counters. | |
562 | * Since loads and stores are relaxed, it might be possible for a '*_zero' | |
563 | * value to be more recent than the current value we're reading from the | |
564 | * counter. This is not a big problem, since these numbers are not | |
565 | * supposed to be too accurate, but we should at least make sure that | |
566 | * the result is not negative. */ | |
567 | for (i = 0; i < DP_N_STATS; i++) { | |
568 | if (stats[i] > pmd->stats_zero[i]) { | |
569 | stats[i] -= pmd->stats_zero[i]; | |
570 | } else { | |
571 | stats[i] = 0; | |
572 | } | |
573 | ||
574 | if (i != DP_STAT_LOST) { | |
575 | /* Lost packets are already included in DP_STAT_MISS */ | |
576 | total_packets += stats[i]; | |
577 | } | |
578 | } | |
579 | ||
580 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
581 | if (cycles[i] > pmd->cycles_zero[i]) { | |
582 | cycles[i] -= pmd->cycles_zero[i]; | |
583 | } else { | |
584 | cycles[i] = 0; | |
585 | } | |
586 | ||
587 | total_cycles += cycles[i]; | |
588 | } | |
589 | ||
590 | ds_put_cstr(reply, (pmd->core_id == NON_PMD_CORE_ID) | |
591 | ? "main thread" : "pmd thread"); | |
592 | ||
593 | if (pmd->numa_id != OVS_NUMA_UNSPEC) { | |
594 | ds_put_format(reply, " numa_id %d", pmd->numa_id); | |
595 | } | |
d5c199ea | 596 | if (pmd->core_id != OVS_CORE_UNSPEC && pmd->core_id != NON_PMD_CORE_ID) { |
bd5131ba | 597 | ds_put_format(reply, " core_id %u", pmd->core_id); |
6553d06b DDP |
598 | } |
599 | ds_put_cstr(reply, ":\n"); | |
600 | ||
601 | ds_put_format(reply, | |
602 | "\temc hits:%llu\n\tmegaflow hits:%llu\n" | |
603 | "\tmiss:%llu\n\tlost:%llu\n", | |
604 | stats[DP_STAT_EXACT_HIT], stats[DP_STAT_MASKED_HIT], | |
605 | stats[DP_STAT_MISS], stats[DP_STAT_LOST]); | |
606 | ||
607 | if (total_cycles == 0) { | |
608 | return; | |
609 | } | |
610 | ||
611 | ds_put_format(reply, | |
612 | "\tpolling cycles:%"PRIu64" (%.02f%%)\n" | |
613 | "\tprocessing cycles:%"PRIu64" (%.02f%%)\n", | |
614 | cycles[PMD_CYCLES_POLLING], | |
615 | cycles[PMD_CYCLES_POLLING] / (double)total_cycles * 100, | |
616 | cycles[PMD_CYCLES_PROCESSING], | |
617 | cycles[PMD_CYCLES_PROCESSING] / (double)total_cycles * 100); | |
618 | ||
619 | if (total_packets == 0) { | |
620 | return; | |
621 | } | |
622 | ||
623 | ds_put_format(reply, | |
624 | "\tavg cycles per packet: %.02f (%"PRIu64"/%llu)\n", | |
625 | total_cycles / (double)total_packets, | |
626 | total_cycles, total_packets); | |
627 | ||
628 | ds_put_format(reply, | |
629 | "\tavg processing cycles per packet: " | |
630 | "%.02f (%"PRIu64"/%llu)\n", | |
631 | cycles[PMD_CYCLES_PROCESSING] / (double)total_packets, | |
632 | cycles[PMD_CYCLES_PROCESSING], total_packets); | |
633 | } | |
634 | ||
635 | static void | |
636 | pmd_info_clear_stats(struct ds *reply OVS_UNUSED, | |
637 | struct dp_netdev_pmd_thread *pmd, | |
638 | unsigned long long stats[DP_N_STATS], | |
639 | uint64_t cycles[PMD_N_CYCLES]) | |
640 | { | |
641 | int i; | |
642 | ||
643 | /* We cannot write 'stats' and 'cycles' (because they're written by other | |
644 | * threads) and we shouldn't change 'stats' (because they're used to count | |
645 | * datapath stats, which must not be cleared here). Instead, we save the | |
646 | * current values and subtract them from the values to be displayed in the | |
647 | * future */ | |
648 | for (i = 0; i < DP_N_STATS; i++) { | |
649 | pmd->stats_zero[i] = stats[i]; | |
650 | } | |
651 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
652 | pmd->cycles_zero[i] = cycles[i]; | |
653 | } | |
654 | } | |
655 | ||
656 | static void | |
657 | dpif_netdev_pmd_info(struct unixctl_conn *conn, int argc, const char *argv[], | |
658 | void *aux) | |
659 | { | |
660 | struct ds reply = DS_EMPTY_INITIALIZER; | |
661 | struct dp_netdev_pmd_thread *pmd; | |
662 | struct dp_netdev *dp = NULL; | |
663 | enum pmd_info_type type = *(enum pmd_info_type *) aux; | |
664 | ||
665 | ovs_mutex_lock(&dp_netdev_mutex); | |
666 | ||
667 | if (argc == 2) { | |
668 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
669 | } else if (shash_count(&dp_netdevs) == 1) { | |
670 | /* There's only one datapath */ | |
671 | dp = shash_first(&dp_netdevs)->data; | |
672 | } | |
673 | ||
674 | if (!dp) { | |
675 | ovs_mutex_unlock(&dp_netdev_mutex); | |
676 | unixctl_command_reply_error(conn, | |
677 | "please specify an existing datapath"); | |
678 | return; | |
679 | } | |
680 | ||
681 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
682 | unsigned long long stats[DP_N_STATS]; | |
683 | uint64_t cycles[PMD_N_CYCLES]; | |
684 | int i; | |
685 | ||
686 | /* Read current stats and cycle counters */ | |
687 | for (i = 0; i < ARRAY_SIZE(stats); i++) { | |
688 | atomic_read_relaxed(&pmd->stats.n[i], &stats[i]); | |
689 | } | |
690 | for (i = 0; i < ARRAY_SIZE(cycles); i++) { | |
691 | atomic_read_relaxed(&pmd->cycles.n[i], &cycles[i]); | |
692 | } | |
693 | ||
694 | if (type == PMD_INFO_CLEAR_STATS) { | |
695 | pmd_info_clear_stats(&reply, pmd, stats, cycles); | |
696 | } else if (type == PMD_INFO_SHOW_STATS) { | |
697 | pmd_info_show_stats(&reply, pmd, stats, cycles); | |
698 | } | |
699 | } | |
700 | ||
701 | ovs_mutex_unlock(&dp_netdev_mutex); | |
702 | ||
703 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
704 | ds_destroy(&reply); | |
705 | } | |
706 | \f | |
707 | static int | |
708 | dpif_netdev_init(void) | |
709 | { | |
710 | static enum pmd_info_type show_aux = PMD_INFO_SHOW_STATS, | |
711 | clear_aux = PMD_INFO_CLEAR_STATS; | |
712 | ||
713 | unixctl_command_register("dpif-netdev/pmd-stats-show", "[dp]", | |
714 | 0, 1, dpif_netdev_pmd_info, | |
715 | (void *)&show_aux); | |
716 | unixctl_command_register("dpif-netdev/pmd-stats-clear", "[dp]", | |
717 | 0, 1, dpif_netdev_pmd_info, | |
718 | (void *)&clear_aux); | |
719 | return 0; | |
720 | } | |
72865317 | 721 | |
2197d7ab | 722 | static int |
2240af25 DDP |
723 | dpif_netdev_enumerate(struct sset *all_dps, |
724 | const struct dpif_class *dpif_class) | |
2197d7ab GL |
725 | { |
726 | struct shash_node *node; | |
727 | ||
97be1538 | 728 | ovs_mutex_lock(&dp_netdev_mutex); |
2197d7ab | 729 | SHASH_FOR_EACH(node, &dp_netdevs) { |
2240af25 DDP |
730 | struct dp_netdev *dp = node->data; |
731 | if (dpif_class != dp->class) { | |
732 | /* 'dp_netdevs' contains both "netdev" and "dummy" dpifs. | |
733 | * If the class doesn't match, skip this dpif. */ | |
734 | continue; | |
735 | } | |
2197d7ab GL |
736 | sset_add(all_dps, node->name); |
737 | } | |
97be1538 | 738 | ovs_mutex_unlock(&dp_netdev_mutex); |
5279f8fd | 739 | |
2197d7ab GL |
740 | return 0; |
741 | } | |
742 | ||
add90f6f EJ |
743 | static bool |
744 | dpif_netdev_class_is_dummy(const struct dpif_class *class) | |
745 | { | |
746 | return class != &dpif_netdev_class; | |
747 | } | |
748 | ||
0aeaabc8 JP |
749 | static const char * |
750 | dpif_netdev_port_open_type(const struct dpif_class *class, const char *type) | |
751 | { | |
752 | return strcmp(type, "internal") ? type | |
add90f6f | 753 | : dpif_netdev_class_is_dummy(class) ? "dummy" |
0aeaabc8 JP |
754 | : "tap"; |
755 | } | |
756 | ||
72865317 BP |
757 | static struct dpif * |
758 | create_dpif_netdev(struct dp_netdev *dp) | |
759 | { | |
462278db | 760 | uint16_t netflow_id = hash_string(dp->name, 0); |
72865317 | 761 | struct dpif_netdev *dpif; |
72865317 | 762 | |
6a8267c5 | 763 | ovs_refcount_ref(&dp->ref_cnt); |
72865317 | 764 | |
72865317 | 765 | dpif = xmalloc(sizeof *dpif); |
614c4892 | 766 | dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id); |
72865317 | 767 | dpif->dp = dp; |
d33ed218 | 768 | dpif->last_port_seq = seq_read(dp->port_seq); |
72865317 BP |
769 | |
770 | return &dpif->dpif; | |
771 | } | |
772 | ||
4e022ec0 AW |
773 | /* Choose an unused, non-zero port number and return it on success. |
774 | * Return ODPP_NONE on failure. */ | |
775 | static odp_port_t | |
e44768b7 | 776 | choose_port(struct dp_netdev *dp, const char *name) |
59e6d833 | 777 | OVS_REQUIRES(dp->port_mutex) |
e44768b7 | 778 | { |
4e022ec0 | 779 | uint32_t port_no; |
e44768b7 JP |
780 | |
781 | if (dp->class != &dpif_netdev_class) { | |
782 | const char *p; | |
783 | int start_no = 0; | |
784 | ||
785 | /* If the port name begins with "br", start the number search at | |
786 | * 100 to make writing tests easier. */ | |
787 | if (!strncmp(name, "br", 2)) { | |
788 | start_no = 100; | |
789 | } | |
790 | ||
791 | /* If the port name contains a number, try to assign that port number. | |
792 | * This can make writing unit tests easier because port numbers are | |
793 | * predictable. */ | |
794 | for (p = name; *p != '\0'; p++) { | |
795 | if (isdigit((unsigned char) *p)) { | |
796 | port_no = start_no + strtol(p, NULL, 10); | |
ff073a71 BP |
797 | if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE) |
798 | && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 799 | return u32_to_odp(port_no); |
e44768b7 JP |
800 | } |
801 | break; | |
802 | } | |
803 | } | |
804 | } | |
805 | ||
ff073a71 BP |
806 | for (port_no = 1; port_no <= UINT16_MAX; port_no++) { |
807 | if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 808 | return u32_to_odp(port_no); |
e44768b7 JP |
809 | } |
810 | } | |
811 | ||
4e022ec0 | 812 | return ODPP_NONE; |
e44768b7 JP |
813 | } |
814 | ||
72865317 | 815 | static int |
614c4892 BP |
816 | create_dp_netdev(const char *name, const struct dpif_class *class, |
817 | struct dp_netdev **dpp) | |
8a4e3a85 | 818 | OVS_REQUIRES(dp_netdev_mutex) |
72865317 BP |
819 | { |
820 | struct dp_netdev *dp; | |
821 | int error; | |
72865317 | 822 | |
462278db | 823 | dp = xzalloc(sizeof *dp); |
8a4e3a85 BP |
824 | shash_add(&dp_netdevs, name, dp); |
825 | ||
826 | *CONST_CAST(const struct dpif_class **, &dp->class) = class; | |
827 | *CONST_CAST(const char **, &dp->name) = xstrdup(name); | |
6a8267c5 | 828 | ovs_refcount_init(&dp->ref_cnt); |
1a65ba85 | 829 | atomic_flag_clear(&dp->destroyed); |
8a4e3a85 | 830 | |
59e6d833 BP |
831 | ovs_mutex_init(&dp->port_mutex); |
832 | cmap_init(&dp->ports); | |
d33ed218 | 833 | dp->port_seq = seq_create(); |
6b31e073 RW |
834 | fat_rwlock_init(&dp->upcall_rwlock); |
835 | ||
836 | /* Disable upcalls by default. */ | |
837 | dp_netdev_disable_upcall(dp); | |
623540e4 | 838 | dp->upcall_aux = NULL; |
6b31e073 | 839 | dp->upcall_cb = NULL; |
e44768b7 | 840 | |
65f13b50 AW |
841 | cmap_init(&dp->poll_threads); |
842 | ovs_mutex_init_recursive(&dp->non_pmd_mutex); | |
843 | ovsthread_key_create(&dp->per_pmd_key, NULL); | |
844 | ||
f2eee189 AW |
845 | dp_netdev_set_nonpmd(dp); |
846 | dp->n_dpdk_rxqs = NR_QUEUE; | |
65f13b50 | 847 | |
59e6d833 | 848 | ovs_mutex_lock(&dp->port_mutex); |
4e022ec0 | 849 | error = do_add_port(dp, name, "internal", ODPP_LOCAL); |
59e6d833 | 850 | ovs_mutex_unlock(&dp->port_mutex); |
72865317 BP |
851 | if (error) { |
852 | dp_netdev_free(dp); | |
462278db | 853 | return error; |
72865317 BP |
854 | } |
855 | ||
a36de779 | 856 | dp->last_tnl_conf_seq = seq_read(tnl_conf_seq); |
462278db | 857 | *dpp = dp; |
72865317 BP |
858 | return 0; |
859 | } | |
860 | ||
861 | static int | |
614c4892 | 862 | dpif_netdev_open(const struct dpif_class *class, const char *name, |
4a387741 | 863 | bool create, struct dpif **dpifp) |
72865317 | 864 | { |
462278db | 865 | struct dp_netdev *dp; |
5279f8fd | 866 | int error; |
462278db | 867 | |
97be1538 | 868 | ovs_mutex_lock(&dp_netdev_mutex); |
462278db BP |
869 | dp = shash_find_data(&dp_netdevs, name); |
870 | if (!dp) { | |
5279f8fd | 871 | error = create ? create_dp_netdev(name, class, &dp) : ENODEV; |
72865317 | 872 | } else { |
5279f8fd BP |
873 | error = (dp->class != class ? EINVAL |
874 | : create ? EEXIST | |
875 | : 0); | |
876 | } | |
877 | if (!error) { | |
878 | *dpifp = create_dpif_netdev(dp); | |
6b31e073 | 879 | dp->dpif = *dpifp; |
72865317 | 880 | } |
97be1538 | 881 | ovs_mutex_unlock(&dp_netdev_mutex); |
462278db | 882 | |
5279f8fd | 883 | return error; |
72865317 BP |
884 | } |
885 | ||
88ace79b DDP |
886 | static void |
887 | dp_netdev_destroy_upcall_lock(struct dp_netdev *dp) | |
888 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
889 | { | |
890 | /* Check that upcalls are disabled, i.e. that the rwlock is taken */ | |
891 | ovs_assert(fat_rwlock_tryrdlock(&dp->upcall_rwlock)); | |
892 | ||
893 | /* Before freeing a lock we should release it */ | |
894 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
895 | fat_rwlock_destroy(&dp->upcall_rwlock); | |
896 | } | |
897 | ||
8a4e3a85 BP |
898 | /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp' |
899 | * through the 'dp_netdevs' shash while freeing 'dp'. */ | |
1ba530f4 BP |
900 | static void |
901 | dp_netdev_free(struct dp_netdev *dp) | |
8a4e3a85 | 902 | OVS_REQUIRES(dp_netdev_mutex) |
1ba530f4 | 903 | { |
59e6d833 | 904 | struct dp_netdev_port *port; |
4ad28026 | 905 | |
8a4e3a85 BP |
906 | shash_find_and_delete(&dp_netdevs, dp->name); |
907 | ||
65f13b50 | 908 | dp_netdev_destroy_all_pmds(dp); |
8bd89cdc | 909 | cmap_destroy(&dp->poll_threads); |
65f13b50 AW |
910 | ovs_mutex_destroy(&dp->non_pmd_mutex); |
911 | ovsthread_key_delete(dp->per_pmd_key); | |
6c3eee82 | 912 | |
59e6d833 | 913 | ovs_mutex_lock(&dp->port_mutex); |
a532e683 | 914 | CMAP_FOR_EACH (port, node, &dp->ports) { |
c40b890f | 915 | do_del_port(dp, port); |
1ba530f4 | 916 | } |
59e6d833 | 917 | ovs_mutex_unlock(&dp->port_mutex); |
51852a57 | 918 | |
d33ed218 | 919 | seq_destroy(dp->port_seq); |
59e6d833 | 920 | cmap_destroy(&dp->ports); |
88ace79b DDP |
921 | |
922 | /* Upcalls must be disabled at this point */ | |
923 | dp_netdev_destroy_upcall_lock(dp); | |
9bbf1c3d | 924 | |
f2eee189 | 925 | free(dp->pmd_cmask); |
8a4e3a85 | 926 | free(CONST_CAST(char *, dp->name)); |
72865317 BP |
927 | free(dp); |
928 | } | |
929 | ||
8a4e3a85 BP |
930 | static void |
931 | dp_netdev_unref(struct dp_netdev *dp) | |
932 | { | |
933 | if (dp) { | |
934 | /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't | |
935 | * get a new reference to 'dp' through the 'dp_netdevs' shash. */ | |
936 | ovs_mutex_lock(&dp_netdev_mutex); | |
24f83812 | 937 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
8a4e3a85 BP |
938 | dp_netdev_free(dp); |
939 | } | |
940 | ovs_mutex_unlock(&dp_netdev_mutex); | |
941 | } | |
942 | } | |
943 | ||
72865317 BP |
944 | static void |
945 | dpif_netdev_close(struct dpif *dpif) | |
946 | { | |
947 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 948 | |
8a4e3a85 | 949 | dp_netdev_unref(dp); |
72865317 BP |
950 | free(dpif); |
951 | } | |
952 | ||
953 | static int | |
7dab847a | 954 | dpif_netdev_destroy(struct dpif *dpif) |
72865317 BP |
955 | { |
956 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 957 | |
6a8267c5 | 958 | if (!atomic_flag_test_and_set(&dp->destroyed)) { |
24f83812 | 959 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
6a8267c5 BP |
960 | /* Can't happen: 'dpif' still owns a reference to 'dp'. */ |
961 | OVS_NOT_REACHED(); | |
962 | } | |
963 | } | |
5279f8fd | 964 | |
72865317 BP |
965 | return 0; |
966 | } | |
967 | ||
eb94da30 DDP |
968 | /* Add 'n' to the atomic variable 'var' non-atomically and using relaxed |
969 | * load/store semantics. While the increment is not atomic, the load and | |
970 | * store operations are, making it impossible to read inconsistent values. | |
971 | * | |
972 | * This is used to update thread local stats counters. */ | |
973 | static void | |
974 | non_atomic_ullong_add(atomic_ullong *var, unsigned long long n) | |
975 | { | |
976 | unsigned long long tmp; | |
977 | ||
978 | atomic_read_relaxed(var, &tmp); | |
979 | tmp += n; | |
980 | atomic_store_relaxed(var, tmp); | |
981 | } | |
982 | ||
72865317 | 983 | static int |
a8d9304d | 984 | dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) |
72865317 BP |
985 | { |
986 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed | 987 | struct dp_netdev_pmd_thread *pmd; |
8a4e3a85 | 988 | |
1c1e46ed AW |
989 | stats->n_flows = stats->n_hit = stats->n_missed = stats->n_lost = 0; |
990 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
eb94da30 | 991 | unsigned long long n; |
1c1e46ed | 992 | stats->n_flows += cmap_count(&pmd->flow_table); |
eb94da30 | 993 | |
abcf3ef4 DDP |
994 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MASKED_HIT], &n); |
995 | stats->n_hit += n; | |
996 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_EXACT_HIT], &n); | |
eb94da30 DDP |
997 | stats->n_hit += n; |
998 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MISS], &n); | |
999 | stats->n_missed += n; | |
1000 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_LOST], &n); | |
1001 | stats->n_lost += n; | |
51852a57 | 1002 | } |
1ce3fa06 | 1003 | stats->n_masks = UINT32_MAX; |
847108dc | 1004 | stats->n_mask_hit = UINT64_MAX; |
5279f8fd | 1005 | |
72865317 BP |
1006 | return 0; |
1007 | } | |
1008 | ||
e4cfed38 | 1009 | static void |
65f13b50 | 1010 | dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd) |
e4cfed38 | 1011 | { |
65f13b50 AW |
1012 | int old_seq; |
1013 | ||
accf8626 AW |
1014 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1015 | return; | |
1016 | } | |
1017 | ||
1018 | ovs_mutex_lock(&pmd->cond_mutex); | |
65f13b50 | 1019 | atomic_add_relaxed(&pmd->change_seq, 1, &old_seq); |
accf8626 AW |
1020 | ovs_mutex_cond_wait(&pmd->cond, &pmd->cond_mutex); |
1021 | ovs_mutex_unlock(&pmd->cond_mutex); | |
65f13b50 | 1022 | } |
e4cfed38 | 1023 | |
65f13b50 AW |
1024 | /* Causes all pmd threads to reload its tx/rx devices. |
1025 | * Must be called after adding/removing ports. */ | |
1026 | static void | |
1027 | dp_netdev_reload_pmds(struct dp_netdev *dp) | |
1028 | { | |
1029 | struct dp_netdev_pmd_thread *pmd; | |
e4cfed38 | 1030 | |
65f13b50 AW |
1031 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { |
1032 | dp_netdev_reload_pmd__(pmd); | |
84067a4c | 1033 | } |
e4cfed38 PS |
1034 | } |
1035 | ||
59e6d833 BP |
1036 | static uint32_t |
1037 | hash_port_no(odp_port_t port_no) | |
1038 | { | |
1039 | return hash_int(odp_to_u32(port_no), 0); | |
1040 | } | |
1041 | ||
72865317 | 1042 | static int |
c3827f61 | 1043 | do_add_port(struct dp_netdev *dp, const char *devname, const char *type, |
4e022ec0 | 1044 | odp_port_t port_no) |
59e6d833 | 1045 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1046 | { |
4b609110 | 1047 | struct netdev_saved_flags *sf; |
72865317 BP |
1048 | struct dp_netdev_port *port; |
1049 | struct netdev *netdev; | |
2499a8ce | 1050 | enum netdev_flags flags; |
0cbfe35d | 1051 | const char *open_type; |
72865317 | 1052 | int error; |
55c955bd | 1053 | int i; |
72865317 | 1054 | |
17050610 BP |
1055 | /* Reject devices already in 'dp'. */ |
1056 | if (!get_port_by_name(dp, devname, &port)) { | |
1057 | return EEXIST; | |
1058 | } | |
72865317 BP |
1059 | |
1060 | /* Open and validate network device. */ | |
0aeaabc8 | 1061 | open_type = dpif_netdev_port_open_type(dp->class, type); |
0cbfe35d | 1062 | error = netdev_open(devname, open_type, &netdev); |
72865317 BP |
1063 | if (error) { |
1064 | return error; | |
1065 | } | |
72865317 BP |
1066 | /* XXX reject non-Ethernet devices */ |
1067 | ||
2499a8ce AC |
1068 | netdev_get_flags(netdev, &flags); |
1069 | if (flags & NETDEV_LOOPBACK) { | |
1070 | VLOG_ERR("%s: cannot add a loopback device", devname); | |
1071 | netdev_close(netdev); | |
1072 | return EINVAL; | |
1073 | } | |
1074 | ||
5a034064 AW |
1075 | if (netdev_is_pmd(netdev)) { |
1076 | int n_cores = ovs_numa_get_n_cores(); | |
1077 | ||
1078 | if (n_cores == OVS_CORE_UNSPEC) { | |
1079 | VLOG_ERR("%s, cannot get cpu core info", devname); | |
1080 | return ENOENT; | |
1081 | } | |
1082 | /* There can only be ovs_numa_get_n_cores() pmd threads, | |
3bcc10c0 DDP |
1083 | * so creates a txq for each, and one extra for the non |
1084 | * pmd threads. */ | |
1085 | error = netdev_set_multiq(netdev, n_cores + 1, dp->n_dpdk_rxqs); | |
7251515e | 1086 | if (error && (error != EOPNOTSUPP)) { |
5a034064 AW |
1087 | VLOG_ERR("%s, cannot set multiq", devname); |
1088 | return errno; | |
1089 | } | |
1090 | } | |
e4cfed38 | 1091 | port = xzalloc(sizeof *port); |
35303d71 | 1092 | port->port_no = port_no; |
e4cfed38 | 1093 | port->netdev = netdev; |
55c955bd | 1094 | port->rxq = xmalloc(sizeof *port->rxq * netdev_n_rxq(netdev)); |
e4cfed38 | 1095 | port->type = xstrdup(type); |
55c955bd PS |
1096 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1097 | error = netdev_rxq_open(netdev, &port->rxq[i], i); | |
1098 | if (error | |
1099 | && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) { | |
1100 | VLOG_ERR("%s: cannot receive packets on this network device (%s)", | |
1101 | devname, ovs_strerror(errno)); | |
1102 | netdev_close(netdev); | |
16bea12c TG |
1103 | free(port->type); |
1104 | free(port->rxq); | |
1105 | free(port); | |
55c955bd PS |
1106 | return error; |
1107 | } | |
7b6b0ef4 BP |
1108 | } |
1109 | ||
4b609110 | 1110 | error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); |
72865317 | 1111 | if (error) { |
55c955bd PS |
1112 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
1113 | netdev_rxq_close(port->rxq[i]); | |
1114 | } | |
72865317 | 1115 | netdev_close(netdev); |
16bea12c | 1116 | free(port->type); |
f7791740 | 1117 | free(port->rxq); |
e4cfed38 | 1118 | free(port); |
72865317 BP |
1119 | return error; |
1120 | } | |
4b609110 | 1121 | port->sf = sf; |
e4cfed38 | 1122 | |
f7d63652 AW |
1123 | ovs_refcount_init(&port->ref_cnt); |
1124 | cmap_insert(&dp->ports, &port->node, hash_port_no(port_no)); | |
1125 | ||
e4cfed38 | 1126 | if (netdev_is_pmd(netdev)) { |
65f13b50 AW |
1127 | dp_netdev_set_pmds_on_numa(dp, netdev_get_numa_id(netdev)); |
1128 | dp_netdev_reload_pmds(dp); | |
e4cfed38 | 1129 | } |
d33ed218 | 1130 | seq_change(dp->port_seq); |
72865317 BP |
1131 | |
1132 | return 0; | |
1133 | } | |
1134 | ||
247527db BP |
1135 | static int |
1136 | dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, | |
4e022ec0 | 1137 | odp_port_t *port_nop) |
247527db BP |
1138 | { |
1139 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3aa30359 BP |
1140 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1141 | const char *dpif_port; | |
4e022ec0 | 1142 | odp_port_t port_no; |
5279f8fd | 1143 | int error; |
247527db | 1144 | |
59e6d833 | 1145 | ovs_mutex_lock(&dp->port_mutex); |
3aa30359 | 1146 | dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
4e022ec0 | 1147 | if (*port_nop != ODPP_NONE) { |
ff073a71 BP |
1148 | port_no = *port_nop; |
1149 | error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0; | |
232dfa4a | 1150 | } else { |
3aa30359 | 1151 | port_no = choose_port(dp, dpif_port); |
5279f8fd | 1152 | error = port_no == ODPP_NONE ? EFBIG : 0; |
232dfa4a | 1153 | } |
5279f8fd | 1154 | if (!error) { |
247527db | 1155 | *port_nop = port_no; |
5279f8fd | 1156 | error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); |
247527db | 1157 | } |
59e6d833 | 1158 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1159 | |
1160 | return error; | |
72865317 BP |
1161 | } |
1162 | ||
1163 | static int | |
4e022ec0 | 1164 | dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no) |
72865317 BP |
1165 | { |
1166 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd BP |
1167 | int error; |
1168 | ||
59e6d833 | 1169 | ovs_mutex_lock(&dp->port_mutex); |
c40b890f BP |
1170 | if (port_no == ODPP_LOCAL) { |
1171 | error = EINVAL; | |
1172 | } else { | |
1173 | struct dp_netdev_port *port; | |
1174 | ||
1175 | error = get_port_by_number(dp, port_no, &port); | |
1176 | if (!error) { | |
1177 | do_del_port(dp, port); | |
1178 | } | |
1179 | } | |
59e6d833 | 1180 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1181 | |
1182 | return error; | |
72865317 BP |
1183 | } |
1184 | ||
1185 | static bool | |
4e022ec0 | 1186 | is_valid_port_number(odp_port_t port_no) |
72865317 | 1187 | { |
ff073a71 BP |
1188 | return port_no != ODPP_NONE; |
1189 | } | |
1190 | ||
1191 | static struct dp_netdev_port * | |
1192 | dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no) | |
1193 | { | |
1194 | struct dp_netdev_port *port; | |
1195 | ||
59e6d833 | 1196 | CMAP_FOR_EACH_WITH_HASH (port, node, hash_port_no(port_no), &dp->ports) { |
35303d71 | 1197 | if (port->port_no == port_no) { |
ff073a71 BP |
1198 | return port; |
1199 | } | |
1200 | } | |
1201 | return NULL; | |
72865317 BP |
1202 | } |
1203 | ||
1204 | static int | |
1205 | get_port_by_number(struct dp_netdev *dp, | |
4e022ec0 | 1206 | odp_port_t port_no, struct dp_netdev_port **portp) |
72865317 BP |
1207 | { |
1208 | if (!is_valid_port_number(port_no)) { | |
1209 | *portp = NULL; | |
1210 | return EINVAL; | |
1211 | } else { | |
ff073a71 | 1212 | *portp = dp_netdev_lookup_port(dp, port_no); |
72865317 BP |
1213 | return *portp ? 0 : ENOENT; |
1214 | } | |
1215 | } | |
1216 | ||
b284085e PS |
1217 | static void |
1218 | port_ref(struct dp_netdev_port *port) | |
1219 | { | |
1220 | if (port) { | |
1221 | ovs_refcount_ref(&port->ref_cnt); | |
1222 | } | |
1223 | } | |
1224 | ||
a1fdee13 AW |
1225 | static bool |
1226 | port_try_ref(struct dp_netdev_port *port) | |
1227 | { | |
1228 | if (port) { | |
1229 | return ovs_refcount_try_ref_rcu(&port->ref_cnt); | |
1230 | } | |
1231 | ||
1232 | return false; | |
1233 | } | |
1234 | ||
59e6d833 BP |
1235 | static void |
1236 | port_unref(struct dp_netdev_port *port) | |
1237 | { | |
24f83812 | 1238 | if (port && ovs_refcount_unref_relaxed(&port->ref_cnt) == 1) { |
accf8626 AW |
1239 | int n_rxq = netdev_n_rxq(port->netdev); |
1240 | int i; | |
1241 | ||
1242 | netdev_close(port->netdev); | |
1243 | netdev_restore_flags(port->sf); | |
1244 | ||
1245 | for (i = 0; i < n_rxq; i++) { | |
1246 | netdev_rxq_close(port->rxq[i]); | |
1247 | } | |
1248 | free(port->rxq); | |
1249 | free(port->type); | |
1250 | free(port); | |
b284085e PS |
1251 | } |
1252 | } | |
1253 | ||
72865317 BP |
1254 | static int |
1255 | get_port_by_name(struct dp_netdev *dp, | |
1256 | const char *devname, struct dp_netdev_port **portp) | |
59e6d833 | 1257 | OVS_REQUIRES(dp->port_mutex) |
72865317 BP |
1258 | { |
1259 | struct dp_netdev_port *port; | |
1260 | ||
a532e683 | 1261 | CMAP_FOR_EACH (port, node, &dp->ports) { |
3efb6063 | 1262 | if (!strcmp(netdev_get_name(port->netdev), devname)) { |
72865317 BP |
1263 | *portp = port; |
1264 | return 0; | |
1265 | } | |
1266 | } | |
1267 | return ENOENT; | |
1268 | } | |
1269 | ||
65f13b50 AW |
1270 | static int |
1271 | get_n_pmd_threads_on_numa(struct dp_netdev *dp, int numa_id) | |
1272 | { | |
1273 | struct dp_netdev_pmd_thread *pmd; | |
1274 | int n_pmds = 0; | |
1275 | ||
1276 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1277 | if (pmd->numa_id == numa_id) { | |
1278 | n_pmds++; | |
1279 | } | |
1280 | } | |
1281 | ||
1282 | return n_pmds; | |
1283 | } | |
1284 | ||
1285 | /* Returns 'true' if there is a port with pmd netdev and the netdev | |
1286 | * is on numa node 'numa_id'. */ | |
1287 | static bool | |
1288 | has_pmd_port_for_numa(struct dp_netdev *dp, int numa_id) | |
1289 | { | |
1290 | struct dp_netdev_port *port; | |
1291 | ||
1292 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
1293 | if (netdev_is_pmd(port->netdev) | |
1294 | && netdev_get_numa_id(port->netdev) == numa_id) { | |
1295 | return true; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | return false; | |
1300 | } | |
1301 | ||
1302 | ||
c40b890f BP |
1303 | static void |
1304 | do_del_port(struct dp_netdev *dp, struct dp_netdev_port *port) | |
59e6d833 | 1305 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1306 | { |
35303d71 | 1307 | cmap_remove(&dp->ports, &port->node, hash_odp_port(port->port_no)); |
d33ed218 | 1308 | seq_change(dp->port_seq); |
e4cfed38 | 1309 | if (netdev_is_pmd(port->netdev)) { |
65f13b50 AW |
1310 | int numa_id = netdev_get_numa_id(port->netdev); |
1311 | ||
1312 | /* If there is no netdev on the numa node, deletes the pmd threads | |
1313 | * for that numa. Else, just reloads the queues. */ | |
1314 | if (!has_pmd_port_for_numa(dp, numa_id)) { | |
1315 | dp_netdev_del_pmds_on_numa(dp, numa_id); | |
1316 | } | |
1317 | dp_netdev_reload_pmds(dp); | |
e4cfed38 | 1318 | } |
72865317 | 1319 | |
b284085e | 1320 | port_unref(port); |
72865317 BP |
1321 | } |
1322 | ||
1323 | static void | |
4c738a8d BP |
1324 | answer_port_query(const struct dp_netdev_port *port, |
1325 | struct dpif_port *dpif_port) | |
72865317 | 1326 | { |
3efb6063 | 1327 | dpif_port->name = xstrdup(netdev_get_name(port->netdev)); |
0cbfe35d | 1328 | dpif_port->type = xstrdup(port->type); |
35303d71 | 1329 | dpif_port->port_no = port->port_no; |
72865317 BP |
1330 | } |
1331 | ||
1332 | static int | |
4e022ec0 | 1333 | dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, |
4c738a8d | 1334 | struct dpif_port *dpif_port) |
72865317 BP |
1335 | { |
1336 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1337 | struct dp_netdev_port *port; | |
1338 | int error; | |
1339 | ||
1340 | error = get_port_by_number(dp, port_no, &port); | |
4afba28d | 1341 | if (!error && dpif_port) { |
4c738a8d | 1342 | answer_port_query(port, dpif_port); |
72865317 | 1343 | } |
5279f8fd | 1344 | |
72865317 BP |
1345 | return error; |
1346 | } | |
1347 | ||
1348 | static int | |
1349 | dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, | |
4c738a8d | 1350 | struct dpif_port *dpif_port) |
72865317 BP |
1351 | { |
1352 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1353 | struct dp_netdev_port *port; | |
1354 | int error; | |
1355 | ||
59e6d833 | 1356 | ovs_mutex_lock(&dp->port_mutex); |
72865317 | 1357 | error = get_port_by_name(dp, devname, &port); |
4afba28d | 1358 | if (!error && dpif_port) { |
4c738a8d | 1359 | answer_port_query(port, dpif_port); |
72865317 | 1360 | } |
59e6d833 | 1361 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1362 | |
72865317 BP |
1363 | return error; |
1364 | } | |
1365 | ||
61e7deb1 BP |
1366 | static void |
1367 | dp_netdev_flow_free(struct dp_netdev_flow *flow) | |
1368 | { | |
61e7deb1 | 1369 | dp_netdev_actions_free(dp_netdev_flow_get_actions(flow)); |
61e7deb1 BP |
1370 | free(flow); |
1371 | } | |
1372 | ||
ed79f89a DDP |
1373 | static void dp_netdev_flow_unref(struct dp_netdev_flow *flow) |
1374 | { | |
1375 | if (ovs_refcount_unref_relaxed(&flow->ref_cnt) == 1) { | |
1376 | ovsrcu_postpone(dp_netdev_flow_free, flow); | |
1377 | } | |
1378 | } | |
1379 | ||
70e5ed6f JS |
1380 | static uint32_t |
1381 | dp_netdev_flow_hash(const ovs_u128 *ufid) | |
1382 | { | |
1383 | return ufid->u32[0]; | |
1384 | } | |
1385 | ||
72865317 | 1386 | static void |
1c1e46ed AW |
1387 | dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, |
1388 | struct dp_netdev_flow *flow) | |
1389 | OVS_REQUIRES(pmd->flow_mutex) | |
72865317 | 1390 | { |
9f361d6b | 1391 | struct cmap_node *node = CONST_CAST(struct cmap_node *, &flow->node); |
2c0ea78f | 1392 | |
1c1e46ed | 1393 | dpcls_remove(&pmd->cls, &flow->cr); |
361d808d JR |
1394 | flow->cr.mask = NULL; /* Accessing rule's mask after this is not safe. */ |
1395 | ||
1c1e46ed | 1396 | cmap_remove(&pmd->flow_table, node, dp_netdev_flow_hash(&flow->ufid)); |
9bbf1c3d | 1397 | flow->dead = true; |
ed79f89a DDP |
1398 | |
1399 | dp_netdev_flow_unref(flow); | |
72865317 BP |
1400 | } |
1401 | ||
1402 | static void | |
1c1e46ed | 1403 | dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd) |
72865317 | 1404 | { |
78c8df12 | 1405 | struct dp_netdev_flow *netdev_flow; |
72865317 | 1406 | |
1c1e46ed AW |
1407 | ovs_mutex_lock(&pmd->flow_mutex); |
1408 | CMAP_FOR_EACH (netdev_flow, node, &pmd->flow_table) { | |
1409 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); | |
72865317 | 1410 | } |
1c1e46ed | 1411 | ovs_mutex_unlock(&pmd->flow_mutex); |
72865317 BP |
1412 | } |
1413 | ||
1414 | static int | |
1415 | dpif_netdev_flow_flush(struct dpif *dpif) | |
1416 | { | |
1417 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed AW |
1418 | struct dp_netdev_pmd_thread *pmd; |
1419 | ||
1420 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1421 | dp_netdev_pmd_flow_flush(pmd); | |
1422 | } | |
5279f8fd | 1423 | |
72865317 BP |
1424 | return 0; |
1425 | } | |
1426 | ||
b0ec0f27 | 1427 | struct dp_netdev_port_state { |
59e6d833 | 1428 | struct cmap_position position; |
4c738a8d | 1429 | char *name; |
b0ec0f27 BP |
1430 | }; |
1431 | ||
1432 | static int | |
1433 | dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) | |
1434 | { | |
1435 | *statep = xzalloc(sizeof(struct dp_netdev_port_state)); | |
1436 | return 0; | |
1437 | } | |
1438 | ||
72865317 | 1439 | static int |
b0ec0f27 | 1440 | dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, |
4c738a8d | 1441 | struct dpif_port *dpif_port) |
72865317 | 1442 | { |
b0ec0f27 | 1443 | struct dp_netdev_port_state *state = state_; |
72865317 | 1444 | struct dp_netdev *dp = get_dp_netdev(dpif); |
59e6d833 | 1445 | struct cmap_node *node; |
ff073a71 | 1446 | int retval; |
72865317 | 1447 | |
59e6d833 | 1448 | node = cmap_next_position(&dp->ports, &state->position); |
ff073a71 BP |
1449 | if (node) { |
1450 | struct dp_netdev_port *port; | |
5279f8fd | 1451 | |
ff073a71 BP |
1452 | port = CONTAINER_OF(node, struct dp_netdev_port, node); |
1453 | ||
1454 | free(state->name); | |
1455 | state->name = xstrdup(netdev_get_name(port->netdev)); | |
1456 | dpif_port->name = state->name; | |
1457 | dpif_port->type = port->type; | |
35303d71 | 1458 | dpif_port->port_no = port->port_no; |
ff073a71 BP |
1459 | |
1460 | retval = 0; | |
1461 | } else { | |
1462 | retval = EOF; | |
72865317 | 1463 | } |
5279f8fd | 1464 | |
ff073a71 | 1465 | return retval; |
b0ec0f27 BP |
1466 | } |
1467 | ||
1468 | static int | |
4c738a8d | 1469 | dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) |
b0ec0f27 | 1470 | { |
4c738a8d BP |
1471 | struct dp_netdev_port_state *state = state_; |
1472 | free(state->name); | |
b0ec0f27 BP |
1473 | free(state); |
1474 | return 0; | |
72865317 BP |
1475 | } |
1476 | ||
1477 | static int | |
67a4917b | 1478 | dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) |
72865317 BP |
1479 | { |
1480 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
d33ed218 | 1481 | uint64_t new_port_seq; |
5279f8fd BP |
1482 | int error; |
1483 | ||
d33ed218 BP |
1484 | new_port_seq = seq_read(dpif->dp->port_seq); |
1485 | if (dpif->last_port_seq != new_port_seq) { | |
1486 | dpif->last_port_seq = new_port_seq; | |
5279f8fd | 1487 | error = ENOBUFS; |
72865317 | 1488 | } else { |
5279f8fd | 1489 | error = EAGAIN; |
72865317 | 1490 | } |
5279f8fd BP |
1491 | |
1492 | return error; | |
72865317 BP |
1493 | } |
1494 | ||
1495 | static void | |
1496 | dpif_netdev_port_poll_wait(const struct dpif *dpif_) | |
1497 | { | |
1498 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
5279f8fd | 1499 | |
d33ed218 | 1500 | seq_wait(dpif->dp->port_seq, dpif->last_port_seq); |
8a4e3a85 BP |
1501 | } |
1502 | ||
1503 | static struct dp_netdev_flow * | |
0de8783a | 1504 | dp_netdev_flow_cast(const struct dpcls_rule *cr) |
8a4e3a85 BP |
1505 | { |
1506 | return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL; | |
72865317 BP |
1507 | } |
1508 | ||
9bbf1c3d DDP |
1509 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *flow) |
1510 | { | |
1511 | return ovs_refcount_try_ref_rcu(&flow->ref_cnt); | |
1512 | } | |
1513 | ||
79df317f DDP |
1514 | /* netdev_flow_key utilities. |
1515 | * | |
1516 | * netdev_flow_key is basically a miniflow. We use these functions | |
1517 | * (netdev_flow_key_clone, netdev_flow_key_equal, ...) instead of the miniflow | |
1518 | * functions (miniflow_clone_inline, miniflow_equal, ...), because: | |
1519 | * | |
1520 | * - Since we are dealing exclusively with miniflows created by | |
1521 | * miniflow_extract(), if the map is different the miniflow is different. | |
1522 | * Therefore we can be faster by comparing the map and the miniflow in a | |
1523 | * single memcmp(). | |
79df317f DDP |
1524 | * - These functions can be inlined by the compiler. |
1525 | * | |
1526 | * The following assertions make sure that what we're doing with miniflow is | |
361d808d | 1527 | * safe. |
79df317f | 1528 | */ |
361d808d | 1529 | BUILD_ASSERT_DECL(sizeof(struct miniflow) == 2 * sizeof(uint64_t)); |
79df317f | 1530 | |
361d808d | 1531 | /* Given the number of bits set in miniflow's maps, returns the size of the |
caeb4906 | 1532 | * 'netdev_flow_key.mf' */ |
361d808d JR |
1533 | static inline size_t |
1534 | netdev_flow_key_size(size_t flow_u64s) | |
79df317f | 1535 | { |
361d808d | 1536 | return sizeof(struct miniflow) + MINIFLOW_VALUES_SIZE(flow_u64s); |
79df317f DDP |
1537 | } |
1538 | ||
79df317f DDP |
1539 | static inline bool |
1540 | netdev_flow_key_equal(const struct netdev_flow_key *a, | |
0de8783a JR |
1541 | const struct netdev_flow_key *b) |
1542 | { | |
caeb4906 JR |
1543 | /* 'b->len' may be not set yet. */ |
1544 | return a->hash == b->hash && !memcmp(&a->mf, &b->mf, a->len); | |
0de8783a JR |
1545 | } |
1546 | ||
1547 | /* Used to compare 'netdev_flow_key' in the exact match cache to a miniflow. | |
1548 | * The maps are compared bitwise, so both 'key->mf' 'mf' must have been | |
1549 | * generated by miniflow_extract. */ | |
1550 | static inline bool | |
1551 | netdev_flow_key_equal_mf(const struct netdev_flow_key *key, | |
1552 | const struct miniflow *mf) | |
79df317f | 1553 | { |
caeb4906 | 1554 | return !memcmp(&key->mf, mf, key->len); |
79df317f DDP |
1555 | } |
1556 | ||
1557 | static inline void | |
1558 | netdev_flow_key_clone(struct netdev_flow_key *dst, | |
0de8783a JR |
1559 | const struct netdev_flow_key *src) |
1560 | { | |
caeb4906 JR |
1561 | memcpy(dst, src, |
1562 | offsetof(struct netdev_flow_key, mf) + src->len); | |
0de8783a JR |
1563 | } |
1564 | ||
1565 | /* Slow. */ | |
1566 | static void | |
1567 | netdev_flow_key_from_flow(struct netdev_flow_key *dst, | |
1568 | const struct flow *src) | |
1569 | { | |
cf62fa4c | 1570 | struct dp_packet packet; |
0de8783a | 1571 | uint64_t buf_stub[512 / 8]; |
0de8783a | 1572 | |
cf62fa4c PS |
1573 | dp_packet_use_stub(&packet, buf_stub, sizeof buf_stub); |
1574 | pkt_metadata_from_flow(&packet.md, src); | |
0de8783a | 1575 | flow_compose(&packet, src); |
cf62fa4c PS |
1576 | miniflow_extract(&packet, &dst->mf); |
1577 | dp_packet_uninit(&packet); | |
0de8783a | 1578 | |
361d808d | 1579 | dst->len = netdev_flow_key_size(miniflow_n_values(&dst->mf)); |
0de8783a JR |
1580 | dst->hash = 0; /* Not computed yet. */ |
1581 | } | |
1582 | ||
1583 | /* Initialize a netdev_flow_key 'mask' from 'match'. */ | |
1584 | static inline void | |
1585 | netdev_flow_mask_init(struct netdev_flow_key *mask, | |
1586 | const struct match *match) | |
1587 | { | |
d70e8c28 | 1588 | const uint64_t *mask_u64 = (const uint64_t *) &match->wc.masks; |
09b0fa9c | 1589 | uint64_t *dst = miniflow_values(&mask->mf); |
361d808d JR |
1590 | struct miniflow maps; |
1591 | uint64_t map; | |
0de8783a JR |
1592 | uint32_t hash = 0; |
1593 | int n; | |
1594 | ||
1595 | /* Only check masks that make sense for the flow. */ | |
361d808d JR |
1596 | flow_wc_map(&match->flow, &maps); |
1597 | memset(&mask->mf, 0, sizeof mask->mf); /* Clear maps. */ | |
0de8783a | 1598 | |
361d808d | 1599 | map = maps.tnl_map; |
0de8783a JR |
1600 | while (map) { |
1601 | uint64_t rm1bit = rightmost_1bit(map); | |
1602 | int i = raw_ctz(map); | |
1603 | ||
d70e8c28 | 1604 | if (mask_u64[i]) { |
361d808d | 1605 | mask->mf.tnl_map |= rm1bit; |
d70e8c28 JR |
1606 | *dst++ = mask_u64[i]; |
1607 | hash = hash_add64(hash, mask_u64[i]); | |
0de8783a JR |
1608 | } |
1609 | map -= rm1bit; | |
1610 | } | |
361d808d JR |
1611 | mask_u64 += FLOW_TNL_U64S; |
1612 | map = maps.pkt_map; | |
1613 | while (map) { | |
1614 | uint64_t rm1bit = rightmost_1bit(map); | |
1615 | int i = raw_ctz(map); | |
0de8783a | 1616 | |
361d808d JR |
1617 | if (mask_u64[i]) { |
1618 | mask->mf.pkt_map |= rm1bit; | |
1619 | *dst++ = mask_u64[i]; | |
1620 | hash = hash_add64(hash, mask_u64[i]); | |
1621 | } | |
1622 | map -= rm1bit; | |
1623 | } | |
0de8783a | 1624 | |
361d808d JR |
1625 | hash = hash_add64(hash, mask->mf.tnl_map); |
1626 | hash = hash_add64(hash, mask->mf.pkt_map); | |
0de8783a | 1627 | |
09b0fa9c | 1628 | n = dst - miniflow_get_values(&mask->mf); |
0de8783a | 1629 | |
d70e8c28 | 1630 | mask->hash = hash_finish(hash, n * 8); |
0de8783a JR |
1631 | mask->len = netdev_flow_key_size(n); |
1632 | } | |
1633 | ||
361d808d | 1634 | /* Initializes 'dst' as a copy of 'flow' masked with 'mask'. */ |
0de8783a JR |
1635 | static inline void |
1636 | netdev_flow_key_init_masked(struct netdev_flow_key *dst, | |
1637 | const struct flow *flow, | |
1638 | const struct netdev_flow_key *mask) | |
79df317f | 1639 | { |
09b0fa9c JR |
1640 | uint64_t *dst_u64 = miniflow_values(&dst->mf); |
1641 | const uint64_t *mask_u64 = miniflow_get_values(&mask->mf); | |
0de8783a | 1642 | uint32_t hash = 0; |
d70e8c28 | 1643 | uint64_t value; |
0de8783a JR |
1644 | |
1645 | dst->len = mask->len; | |
361d808d | 1646 | dst->mf = mask->mf; /* Copy maps. */ |
0de8783a | 1647 | |
361d808d | 1648 | FLOW_FOR_EACH_IN_MAPS(value, flow, mask->mf) { |
d70e8c28 JR |
1649 | *dst_u64 = value & *mask_u64++; |
1650 | hash = hash_add64(hash, *dst_u64++); | |
0de8783a | 1651 | } |
09b0fa9c JR |
1652 | dst->hash = hash_finish(hash, |
1653 | (dst_u64 - miniflow_get_values(&dst->mf)) * 8); | |
0de8783a JR |
1654 | } |
1655 | ||
361d808d JR |
1656 | /* Iterate through netdev_flow_key TNL u64 values specified by 'MAPS'. */ |
1657 | #define NETDEV_FLOW_KEY_FOR_EACH_IN_TNL_MAP(VALUE, KEY, MAPS) \ | |
1658 | MINIFLOW_FOR_EACH_IN_TNL_MAP(VALUE, &(KEY)->mf, MAPS) | |
1659 | ||
1660 | /* Iterate through netdev_flow_key PKT u64 values specified by 'MAPS'. */ | |
1661 | #define NETDEV_FLOW_KEY_FOR_EACH_IN_PKT_MAP(VALUE, KEY, MAPS) \ | |
1662 | MINIFLOW_FOR_EACH_IN_PKT_MAP(VALUE, &(KEY)->mf, MAPS) | |
0de8783a JR |
1663 | |
1664 | /* Returns a hash value for the bits of 'key' where there are 1-bits in | |
1665 | * 'mask'. */ | |
1666 | static inline uint32_t | |
1667 | netdev_flow_key_hash_in_mask(const struct netdev_flow_key *key, | |
1668 | const struct netdev_flow_key *mask) | |
1669 | { | |
09b0fa9c | 1670 | const uint64_t *p = miniflow_get_values(&mask->mf); |
0de8783a | 1671 | uint32_t hash = 0; |
d70e8c28 | 1672 | uint64_t key_u64; |
0de8783a | 1673 | |
361d808d JR |
1674 | NETDEV_FLOW_KEY_FOR_EACH_IN_TNL_MAP(key_u64, key, mask->mf) { |
1675 | hash = hash_add64(hash, key_u64 & *p++); | |
1676 | } | |
1677 | NETDEV_FLOW_KEY_FOR_EACH_IN_PKT_MAP(key_u64, key, mask->mf) { | |
d70e8c28 | 1678 | hash = hash_add64(hash, key_u64 & *p++); |
0de8783a JR |
1679 | } |
1680 | ||
09b0fa9c | 1681 | return hash_finish(hash, (p - miniflow_get_values(&mask->mf)) * 8); |
79df317f DDP |
1682 | } |
1683 | ||
9bbf1c3d DDP |
1684 | static inline bool |
1685 | emc_entry_alive(struct emc_entry *ce) | |
1686 | { | |
1687 | return ce->flow && !ce->flow->dead; | |
1688 | } | |
1689 | ||
1690 | static void | |
1691 | emc_clear_entry(struct emc_entry *ce) | |
1692 | { | |
1693 | if (ce->flow) { | |
1694 | dp_netdev_flow_unref(ce->flow); | |
1695 | ce->flow = NULL; | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | static inline void | |
1700 | emc_change_entry(struct emc_entry *ce, struct dp_netdev_flow *flow, | |
0de8783a | 1701 | const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1702 | { |
1703 | if (ce->flow != flow) { | |
1704 | if (ce->flow) { | |
1705 | dp_netdev_flow_unref(ce->flow); | |
1706 | } | |
1707 | ||
1708 | if (dp_netdev_flow_ref(flow)) { | |
1709 | ce->flow = flow; | |
1710 | } else { | |
1711 | ce->flow = NULL; | |
1712 | } | |
1713 | } | |
0de8783a JR |
1714 | if (key) { |
1715 | netdev_flow_key_clone(&ce->key, key); | |
9bbf1c3d DDP |
1716 | } |
1717 | } | |
1718 | ||
1719 | static inline void | |
0de8783a | 1720 | emc_insert(struct emc_cache *cache, const struct netdev_flow_key *key, |
9bbf1c3d DDP |
1721 | struct dp_netdev_flow *flow) |
1722 | { | |
1723 | struct emc_entry *to_be_replaced = NULL; | |
1724 | struct emc_entry *current_entry; | |
1725 | ||
0de8783a JR |
1726 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1727 | if (netdev_flow_key_equal(¤t_entry->key, key)) { | |
9bbf1c3d | 1728 | /* We found the entry with the 'mf' miniflow */ |
0de8783a | 1729 | emc_change_entry(current_entry, flow, NULL); |
9bbf1c3d DDP |
1730 | return; |
1731 | } | |
1732 | ||
1733 | /* Replacement policy: put the flow in an empty (not alive) entry, or | |
1734 | * in the first entry where it can be */ | |
1735 | if (!to_be_replaced | |
1736 | || (emc_entry_alive(to_be_replaced) | |
1737 | && !emc_entry_alive(current_entry)) | |
0de8783a | 1738 | || current_entry->key.hash < to_be_replaced->key.hash) { |
9bbf1c3d DDP |
1739 | to_be_replaced = current_entry; |
1740 | } | |
1741 | } | |
1742 | /* We didn't find the miniflow in the cache. | |
1743 | * The 'to_be_replaced' entry is where the new flow will be stored */ | |
1744 | ||
0de8783a | 1745 | emc_change_entry(to_be_replaced, flow, key); |
9bbf1c3d DDP |
1746 | } |
1747 | ||
1748 | static inline struct dp_netdev_flow * | |
0de8783a | 1749 | emc_lookup(struct emc_cache *cache, const struct netdev_flow_key *key) |
9bbf1c3d DDP |
1750 | { |
1751 | struct emc_entry *current_entry; | |
1752 | ||
0de8783a JR |
1753 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
1754 | if (current_entry->key.hash == key->hash | |
1755 | && emc_entry_alive(current_entry) | |
1756 | && netdev_flow_key_equal_mf(¤t_entry->key, &key->mf)) { | |
9bbf1c3d | 1757 | |
0de8783a | 1758 | /* We found the entry with the 'key->mf' miniflow */ |
9bbf1c3d DDP |
1759 | return current_entry->flow; |
1760 | } | |
1761 | } | |
1762 | ||
1763 | return NULL; | |
1764 | } | |
1765 | ||
72865317 | 1766 | static struct dp_netdev_flow * |
1c1e46ed AW |
1767 | dp_netdev_pmd_lookup_flow(const struct dp_netdev_pmd_thread *pmd, |
1768 | const struct netdev_flow_key *key) | |
2c0ea78f | 1769 | { |
8a4e3a85 | 1770 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 1771 | struct dpcls_rule *rule; |
2c0ea78f | 1772 | |
1c1e46ed | 1773 | dpcls_lookup(&pmd->cls, key, &rule, 1); |
4f150744 | 1774 | netdev_flow = dp_netdev_flow_cast(rule); |
2c0ea78f | 1775 | |
8a4e3a85 | 1776 | return netdev_flow; |
2c0ea78f GS |
1777 | } |
1778 | ||
1779 | static struct dp_netdev_flow * | |
1c1e46ed AW |
1780 | dp_netdev_pmd_find_flow(const struct dp_netdev_pmd_thread *pmd, |
1781 | const ovs_u128 *ufidp, const struct nlattr *key, | |
1782 | size_t key_len) | |
72865317 | 1783 | { |
1763b4b8 | 1784 | struct dp_netdev_flow *netdev_flow; |
70e5ed6f JS |
1785 | struct flow flow; |
1786 | ovs_u128 ufid; | |
1787 | ||
1788 | /* If a UFID is not provided, determine one based on the key. */ | |
1789 | if (!ufidp && key && key_len | |
1790 | && !dpif_netdev_flow_from_nlattrs(key, key_len, &flow)) { | |
1c1e46ed | 1791 | dpif_flow_hash(pmd->dp->dpif, &flow, sizeof flow, &ufid); |
70e5ed6f JS |
1792 | ufidp = &ufid; |
1793 | } | |
72865317 | 1794 | |
70e5ed6f JS |
1795 | if (ufidp) { |
1796 | CMAP_FOR_EACH_WITH_HASH (netdev_flow, node, dp_netdev_flow_hash(ufidp), | |
1c1e46ed | 1797 | &pmd->flow_table) { |
bdd7ecf5 | 1798 | if (ovs_u128_equals(&netdev_flow->ufid, ufidp)) { |
70e5ed6f JS |
1799 | return netdev_flow; |
1800 | } | |
72865317 BP |
1801 | } |
1802 | } | |
8a4e3a85 | 1803 | |
72865317 BP |
1804 | return NULL; |
1805 | } | |
1806 | ||
1807 | static void | |
eb94da30 | 1808 | get_dpif_flow_stats(const struct dp_netdev_flow *netdev_flow_, |
1763b4b8 | 1809 | struct dpif_flow_stats *stats) |
feebdea2 | 1810 | { |
eb94da30 DDP |
1811 | struct dp_netdev_flow *netdev_flow; |
1812 | unsigned long long n; | |
1813 | long long used; | |
1814 | uint16_t flags; | |
1815 | ||
1816 | netdev_flow = CONST_CAST(struct dp_netdev_flow *, netdev_flow_); | |
1817 | ||
1818 | atomic_read_relaxed(&netdev_flow->stats.packet_count, &n); | |
1819 | stats->n_packets = n; | |
1820 | atomic_read_relaxed(&netdev_flow->stats.byte_count, &n); | |
1821 | stats->n_bytes = n; | |
1822 | atomic_read_relaxed(&netdev_flow->stats.used, &used); | |
1823 | stats->used = used; | |
1824 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
1825 | stats->tcp_flags = flags; | |
72865317 BP |
1826 | } |
1827 | ||
7af12bd7 JS |
1828 | /* Converts to the dpif_flow format, using 'key_buf' and 'mask_buf' for |
1829 | * storing the netlink-formatted key/mask. 'key_buf' may be the same as | |
1830 | * 'mask_buf'. Actions will be returned without copying, by relying on RCU to | |
1831 | * protect them. */ | |
6fe09f8c | 1832 | static void |
70e5ed6f | 1833 | dp_netdev_flow_to_dpif_flow(const struct dp_netdev_flow *netdev_flow, |
7af12bd7 | 1834 | struct ofpbuf *key_buf, struct ofpbuf *mask_buf, |
64bb477f | 1835 | struct dpif_flow *flow, bool terse) |
6fe09f8c | 1836 | { |
64bb477f JS |
1837 | if (terse) { |
1838 | memset(flow, 0, sizeof *flow); | |
1839 | } else { | |
1840 | struct flow_wildcards wc; | |
1841 | struct dp_netdev_actions *actions; | |
1842 | size_t offset; | |
5262eea1 JG |
1843 | struct odp_flow_key_parms odp_parms = { |
1844 | .flow = &netdev_flow->flow, | |
1845 | .mask = &wc.masks, | |
2494ccd7 | 1846 | .support = dp_netdev_support, |
5262eea1 | 1847 | }; |
64bb477f JS |
1848 | |
1849 | miniflow_expand(&netdev_flow->cr.mask->mf, &wc.masks); | |
1850 | ||
1851 | /* Key */ | |
6fd6ed71 | 1852 | offset = key_buf->size; |
64bb477f | 1853 | flow->key = ofpbuf_tail(key_buf); |
5262eea1 JG |
1854 | odp_parms.odp_in_port = netdev_flow->flow.in_port.odp_port; |
1855 | odp_flow_key_from_flow(&odp_parms, key_buf); | |
6fd6ed71 | 1856 | flow->key_len = key_buf->size - offset; |
64bb477f JS |
1857 | |
1858 | /* Mask */ | |
6fd6ed71 | 1859 | offset = mask_buf->size; |
64bb477f | 1860 | flow->mask = ofpbuf_tail(mask_buf); |
5262eea1 | 1861 | odp_parms.odp_in_port = wc.masks.in_port.odp_port; |
ec1f6f32 | 1862 | odp_parms.key_buf = key_buf; |
5262eea1 | 1863 | odp_flow_key_from_mask(&odp_parms, mask_buf); |
6fd6ed71 | 1864 | flow->mask_len = mask_buf->size - offset; |
64bb477f JS |
1865 | |
1866 | /* Actions */ | |
1867 | actions = dp_netdev_flow_get_actions(netdev_flow); | |
1868 | flow->actions = actions->actions; | |
1869 | flow->actions_len = actions->size; | |
1870 | } | |
6fe09f8c | 1871 | |
70e5ed6f JS |
1872 | flow->ufid = netdev_flow->ufid; |
1873 | flow->ufid_present = true; | |
1c1e46ed | 1874 | flow->pmd_id = netdev_flow->pmd_id; |
6fe09f8c JS |
1875 | get_dpif_flow_stats(netdev_flow, &flow->stats); |
1876 | } | |
1877 | ||
36956a7d | 1878 | static int |
8c301900 JR |
1879 | dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len, |
1880 | const struct nlattr *mask_key, | |
1881 | uint32_t mask_key_len, const struct flow *flow, | |
1882 | struct flow *mask) | |
1883 | { | |
1884 | if (mask_key_len) { | |
80e44883 BP |
1885 | enum odp_key_fitness fitness; |
1886 | ||
ec1f6f32 JG |
1887 | fitness = odp_flow_key_to_mask(mask_key, mask_key_len, key, key_len, |
1888 | mask, flow); | |
80e44883 | 1889 | if (fitness) { |
8c301900 JR |
1890 | /* This should not happen: it indicates that |
1891 | * odp_flow_key_from_mask() and odp_flow_key_to_mask() | |
1892 | * disagree on the acceptable form of a mask. Log the problem | |
1893 | * as an error, with enough details to enable debugging. */ | |
1894 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1895 | ||
1896 | if (!VLOG_DROP_ERR(&rl)) { | |
1897 | struct ds s; | |
1898 | ||
1899 | ds_init(&s); | |
1900 | odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s, | |
1901 | true); | |
80e44883 BP |
1902 | VLOG_ERR("internal error parsing flow mask %s (%s)", |
1903 | ds_cstr(&s), odp_key_fitness_to_string(fitness)); | |
8c301900 JR |
1904 | ds_destroy(&s); |
1905 | } | |
1906 | ||
1907 | return EINVAL; | |
1908 | } | |
8c301900 JR |
1909 | } else { |
1910 | enum mf_field_id id; | |
1911 | /* No mask key, unwildcard everything except fields whose | |
1912 | * prerequisities are not met. */ | |
1913 | memset(mask, 0x0, sizeof *mask); | |
1914 | ||
1915 | for (id = 0; id < MFF_N_IDS; ++id) { | |
1916 | /* Skip registers and metadata. */ | |
1917 | if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS) | |
8cd27cfd | 1918 | && !(id >= MFF_XREG0 && id < MFF_XREG0 + FLOW_N_XREGS) |
8c301900 JR |
1919 | && id != MFF_METADATA) { |
1920 | const struct mf_field *mf = mf_from_id(id); | |
1921 | if (mf_are_prereqs_ok(mf, flow)) { | |
1922 | mf_mask_field(mf, mask); | |
1923 | } | |
1924 | } | |
1925 | } | |
1926 | } | |
1927 | ||
f3f750e5 BP |
1928 | /* Force unwildcard the in_port. |
1929 | * | |
1930 | * We need to do this even in the case where we unwildcard "everything" | |
1931 | * above because "everything" only includes the 16-bit OpenFlow port number | |
1932 | * mask->in_port.ofp_port, which only covers half of the 32-bit datapath | |
1933 | * port number mask->in_port.odp_port. */ | |
1934 | mask->in_port.odp_port = u32_to_odp(UINT32_MAX); | |
8c301900 JR |
1935 | return 0; |
1936 | } | |
1937 | ||
1938 | static int | |
1939 | dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, | |
1940 | struct flow *flow) | |
36956a7d | 1941 | { |
586ddea5 BP |
1942 | odp_port_t in_port; |
1943 | ||
8c301900 | 1944 | if (odp_flow_key_to_flow(key, key_len, flow)) { |
36956a7d | 1945 | /* This should not happen: it indicates that odp_flow_key_from_flow() |
8c301900 JR |
1946 | * and odp_flow_key_to_flow() disagree on the acceptable form of a |
1947 | * flow. Log the problem as an error, with enough details to enable | |
1948 | * debugging. */ | |
36956a7d BP |
1949 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
1950 | ||
1951 | if (!VLOG_DROP_ERR(&rl)) { | |
1952 | struct ds s; | |
1953 | ||
1954 | ds_init(&s); | |
8c301900 | 1955 | odp_flow_format(key, key_len, NULL, 0, NULL, &s, true); |
36956a7d BP |
1956 | VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s)); |
1957 | ds_destroy(&s); | |
1958 | } | |
1959 | ||
1960 | return EINVAL; | |
1961 | } | |
1962 | ||
586ddea5 BP |
1963 | in_port = flow->in_port.odp_port; |
1964 | if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) { | |
18886b60 BP |
1965 | return EINVAL; |
1966 | } | |
1967 | ||
36956a7d BP |
1968 | return 0; |
1969 | } | |
1970 | ||
72865317 | 1971 | static int |
6fe09f8c | 1972 | dpif_netdev_flow_get(const struct dpif *dpif, const struct dpif_flow_get *get) |
72865317 BP |
1973 | { |
1974 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 1975 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 1976 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
1977 | unsigned pmd_id = get->pmd_id == PMD_ID_NULL |
1978 | ? NON_PMD_CORE_ID : get->pmd_id; | |
70e5ed6f | 1979 | int error = 0; |
8a4e3a85 | 1980 | |
1c1e46ed AW |
1981 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
1982 | if (!pmd) { | |
1983 | return EINVAL; | |
1984 | } | |
1985 | ||
1986 | netdev_flow = dp_netdev_pmd_find_flow(pmd, get->ufid, get->key, | |
1987 | get->key_len); | |
1763b4b8 | 1988 | if (netdev_flow) { |
70e5ed6f | 1989 | dp_netdev_flow_to_dpif_flow(netdev_flow, get->buffer, get->buffer, |
64bb477f | 1990 | get->flow, false); |
70e5ed6f | 1991 | } else { |
5279f8fd | 1992 | error = ENOENT; |
72865317 | 1993 | } |
1c1e46ed AW |
1994 | dp_netdev_pmd_unref(pmd); |
1995 | ||
bc4a05c6 | 1996 | |
5279f8fd | 1997 | return error; |
72865317 BP |
1998 | } |
1999 | ||
0de8783a | 2000 | static struct dp_netdev_flow * |
1c1e46ed AW |
2001 | dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd, |
2002 | struct match *match, const ovs_u128 *ufid, | |
ae2ceebd | 2003 | const struct nlattr *actions, size_t actions_len) |
1c1e46ed | 2004 | OVS_REQUIRES(pmd->flow_mutex) |
72865317 | 2005 | { |
0de8783a JR |
2006 | struct dp_netdev_flow *flow; |
2007 | struct netdev_flow_key mask; | |
ed79f89a | 2008 | |
0de8783a JR |
2009 | netdev_flow_mask_init(&mask, match); |
2010 | /* Make sure wc does not have metadata. */ | |
361d808d JR |
2011 | ovs_assert(!(mask.mf.pkt_map |
2012 | & (MINIFLOW_PKT_MAP(metadata) | MINIFLOW_PKT_MAP(regs)))); | |
679ba04c | 2013 | |
0de8783a | 2014 | /* Do not allocate extra space. */ |
caeb4906 | 2015 | flow = xmalloc(sizeof *flow - sizeof flow->cr.flow.mf + mask.len); |
1c1e46ed | 2016 | memset(&flow->stats, 0, sizeof flow->stats); |
0de8783a | 2017 | flow->dead = false; |
11e5cf1f | 2018 | flow->batch = NULL; |
bd5131ba | 2019 | *CONST_CAST(unsigned *, &flow->pmd_id) = pmd->core_id; |
0de8783a | 2020 | *CONST_CAST(struct flow *, &flow->flow) = match->flow; |
70e5ed6f | 2021 | *CONST_CAST(ovs_u128 *, &flow->ufid) = *ufid; |
0de8783a | 2022 | ovs_refcount_init(&flow->ref_cnt); |
0de8783a | 2023 | ovsrcu_set(&flow->actions, dp_netdev_actions_create(actions, actions_len)); |
2c0ea78f | 2024 | |
0de8783a | 2025 | netdev_flow_key_init_masked(&flow->cr.flow, &match->flow, &mask); |
1c1e46ed | 2026 | dpcls_insert(&pmd->cls, &flow->cr, &mask); |
72865317 | 2027 | |
4c75aaab EJ |
2028 | cmap_insert(&pmd->flow_table, CONST_CAST(struct cmap_node *, &flow->node), |
2029 | dp_netdev_flow_hash(&flow->ufid)); | |
2030 | ||
623540e4 | 2031 | if (OVS_UNLIKELY(VLOG_IS_DBG_ENABLED())) { |
0de8783a | 2032 | struct match match; |
623540e4 EJ |
2033 | struct ds ds = DS_EMPTY_INITIALIZER; |
2034 | ||
0de8783a JR |
2035 | match.flow = flow->flow; |
2036 | miniflow_expand(&flow->cr.mask->mf, &match.wc.masks); | |
2037 | ||
623540e4 | 2038 | ds_put_cstr(&ds, "flow_add: "); |
70e5ed6f JS |
2039 | odp_format_ufid(ufid, &ds); |
2040 | ds_put_cstr(&ds, " "); | |
0de8783a | 2041 | match_format(&match, &ds, OFP_DEFAULT_PRIORITY); |
623540e4 EJ |
2042 | ds_put_cstr(&ds, ", actions:"); |
2043 | format_odp_actions(&ds, actions, actions_len); | |
2044 | ||
2045 | VLOG_DBG_RL(&upcall_rl, "%s", ds_cstr(&ds)); | |
2046 | ||
2047 | ds_destroy(&ds); | |
2048 | } | |
2049 | ||
0de8783a | 2050 | return flow; |
72865317 BP |
2051 | } |
2052 | ||
72865317 | 2053 | static int |
89625d1e | 2054 | dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put) |
72865317 BP |
2055 | { |
2056 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2057 | struct dp_netdev_flow *netdev_flow; |
0de8783a | 2058 | struct netdev_flow_key key; |
1c1e46ed | 2059 | struct dp_netdev_pmd_thread *pmd; |
ae2ceebd | 2060 | struct match match; |
70e5ed6f | 2061 | ovs_u128 ufid; |
bd5131ba DDP |
2062 | unsigned pmd_id = put->pmd_id == PMD_ID_NULL |
2063 | ? NON_PMD_CORE_ID : put->pmd_id; | |
36956a7d BP |
2064 | int error; |
2065 | ||
ae2ceebd | 2066 | error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &match.flow); |
8c301900 JR |
2067 | if (error) { |
2068 | return error; | |
2069 | } | |
2070 | error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len, | |
2071 | put->mask, put->mask_len, | |
ae2ceebd | 2072 | &match.flow, &match.wc.masks); |
36956a7d BP |
2073 | if (error) { |
2074 | return error; | |
2075 | } | |
0de8783a | 2076 | |
1c1e46ed AW |
2077 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2078 | if (!pmd) { | |
2079 | return EINVAL; | |
2080 | } | |
2081 | ||
0de8783a JR |
2082 | /* Must produce a netdev_flow_key for lookup. |
2083 | * This interface is no longer performance critical, since it is not used | |
2084 | * for upcall processing any more. */ | |
2085 | netdev_flow_key_from_flow(&key, &match.flow); | |
72865317 | 2086 | |
70e5ed6f JS |
2087 | if (put->ufid) { |
2088 | ufid = *put->ufid; | |
2089 | } else { | |
2090 | dpif_flow_hash(dpif, &match.flow, sizeof match.flow, &ufid); | |
2091 | } | |
2092 | ||
1c1e46ed AW |
2093 | ovs_mutex_lock(&pmd->flow_mutex); |
2094 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &key); | |
1763b4b8 | 2095 | if (!netdev_flow) { |
89625d1e | 2096 | if (put->flags & DPIF_FP_CREATE) { |
1c1e46ed | 2097 | if (cmap_count(&pmd->flow_table) < MAX_FLOWS) { |
89625d1e BP |
2098 | if (put->stats) { |
2099 | memset(put->stats, 0, sizeof *put->stats); | |
feebdea2 | 2100 | } |
1c1e46ed | 2101 | dp_netdev_flow_add(pmd, &match, &ufid, put->actions, |
70e5ed6f | 2102 | put->actions_len); |
0de8783a | 2103 | error = 0; |
72865317 | 2104 | } else { |
5279f8fd | 2105 | error = EFBIG; |
72865317 BP |
2106 | } |
2107 | } else { | |
5279f8fd | 2108 | error = ENOENT; |
72865317 BP |
2109 | } |
2110 | } else { | |
2c0ea78f | 2111 | if (put->flags & DPIF_FP_MODIFY |
ae2ceebd | 2112 | && flow_equal(&match.flow, &netdev_flow->flow)) { |
8a4e3a85 BP |
2113 | struct dp_netdev_actions *new_actions; |
2114 | struct dp_netdev_actions *old_actions; | |
2115 | ||
2116 | new_actions = dp_netdev_actions_create(put->actions, | |
2117 | put->actions_len); | |
2118 | ||
61e7deb1 BP |
2119 | old_actions = dp_netdev_flow_get_actions(netdev_flow); |
2120 | ovsrcu_set(&netdev_flow->actions, new_actions); | |
679ba04c | 2121 | |
a84cb64a BP |
2122 | if (put->stats) { |
2123 | get_dpif_flow_stats(netdev_flow, put->stats); | |
2124 | } | |
2125 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
97447f55 DDP |
2126 | /* XXX: The userspace datapath uses thread local statistics |
2127 | * (for flows), which should be updated only by the owning | |
2128 | * thread. Since we cannot write on stats memory here, | |
2129 | * we choose not to support this flag. Please note: | |
2130 | * - This feature is currently used only by dpctl commands with | |
2131 | * option --clear. | |
2132 | * - Should the need arise, this operation can be implemented | |
2133 | * by keeping a base value (to be update here) for each | |
2134 | * counter, and subtracting it before outputting the stats */ | |
2135 | error = EOPNOTSUPP; | |
72865317 | 2136 | } |
8a4e3a85 | 2137 | |
61e7deb1 | 2138 | ovsrcu_postpone(dp_netdev_actions_free, old_actions); |
2c0ea78f | 2139 | } else if (put->flags & DPIF_FP_CREATE) { |
5279f8fd | 2140 | error = EEXIST; |
2c0ea78f GS |
2141 | } else { |
2142 | /* Overlapping flow. */ | |
2143 | error = EINVAL; | |
72865317 BP |
2144 | } |
2145 | } | |
1c1e46ed AW |
2146 | ovs_mutex_unlock(&pmd->flow_mutex); |
2147 | dp_netdev_pmd_unref(pmd); | |
5279f8fd BP |
2148 | |
2149 | return error; | |
72865317 BP |
2150 | } |
2151 | ||
72865317 | 2152 | static int |
b99d3cee | 2153 | dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del) |
72865317 BP |
2154 | { |
2155 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2156 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 2157 | struct dp_netdev_pmd_thread *pmd; |
bd5131ba DDP |
2158 | unsigned pmd_id = del->pmd_id == PMD_ID_NULL |
2159 | ? NON_PMD_CORE_ID : del->pmd_id; | |
70e5ed6f | 2160 | int error = 0; |
72865317 | 2161 | |
1c1e46ed AW |
2162 | pmd = dp_netdev_get_pmd(dp, pmd_id); |
2163 | if (!pmd) { | |
2164 | return EINVAL; | |
2165 | } | |
2166 | ||
2167 | ovs_mutex_lock(&pmd->flow_mutex); | |
2168 | netdev_flow = dp_netdev_pmd_find_flow(pmd, del->ufid, del->key, | |
2169 | del->key_len); | |
1763b4b8 | 2170 | if (netdev_flow) { |
b99d3cee | 2171 | if (del->stats) { |
1763b4b8 | 2172 | get_dpif_flow_stats(netdev_flow, del->stats); |
feebdea2 | 2173 | } |
1c1e46ed | 2174 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); |
72865317 | 2175 | } else { |
5279f8fd | 2176 | error = ENOENT; |
72865317 | 2177 | } |
1c1e46ed AW |
2178 | ovs_mutex_unlock(&pmd->flow_mutex); |
2179 | dp_netdev_pmd_unref(pmd); | |
5279f8fd BP |
2180 | |
2181 | return error; | |
72865317 BP |
2182 | } |
2183 | ||
ac64794a BP |
2184 | struct dpif_netdev_flow_dump { |
2185 | struct dpif_flow_dump up; | |
1c1e46ed AW |
2186 | struct cmap_position poll_thread_pos; |
2187 | struct cmap_position flow_pos; | |
2188 | struct dp_netdev_pmd_thread *cur_pmd; | |
d2ad7ef1 JS |
2189 | int status; |
2190 | struct ovs_mutex mutex; | |
e723fd32 JS |
2191 | }; |
2192 | ||
ac64794a BP |
2193 | static struct dpif_netdev_flow_dump * |
2194 | dpif_netdev_flow_dump_cast(struct dpif_flow_dump *dump) | |
72865317 | 2195 | { |
ac64794a | 2196 | return CONTAINER_OF(dump, struct dpif_netdev_flow_dump, up); |
e723fd32 JS |
2197 | } |
2198 | ||
ac64794a | 2199 | static struct dpif_flow_dump * |
64bb477f | 2200 | dpif_netdev_flow_dump_create(const struct dpif *dpif_, bool terse) |
e723fd32 | 2201 | { |
ac64794a | 2202 | struct dpif_netdev_flow_dump *dump; |
e723fd32 | 2203 | |
1c1e46ed | 2204 | dump = xzalloc(sizeof *dump); |
ac64794a | 2205 | dpif_flow_dump_init(&dump->up, dpif_); |
64bb477f | 2206 | dump->up.terse = terse; |
ac64794a BP |
2207 | ovs_mutex_init(&dump->mutex); |
2208 | ||
2209 | return &dump->up; | |
e723fd32 JS |
2210 | } |
2211 | ||
2212 | static int | |
ac64794a | 2213 | dpif_netdev_flow_dump_destroy(struct dpif_flow_dump *dump_) |
e723fd32 | 2214 | { |
ac64794a | 2215 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); |
e723fd32 | 2216 | |
ac64794a BP |
2217 | ovs_mutex_destroy(&dump->mutex); |
2218 | free(dump); | |
704a1e09 BP |
2219 | return 0; |
2220 | } | |
2221 | ||
ac64794a BP |
2222 | struct dpif_netdev_flow_dump_thread { |
2223 | struct dpif_flow_dump_thread up; | |
2224 | struct dpif_netdev_flow_dump *dump; | |
8bb113da RW |
2225 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; |
2226 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
2227 | }; |
2228 | ||
2229 | static struct dpif_netdev_flow_dump_thread * | |
2230 | dpif_netdev_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
2231 | { | |
2232 | return CONTAINER_OF(thread, struct dpif_netdev_flow_dump_thread, up); | |
2233 | } | |
2234 | ||
2235 | static struct dpif_flow_dump_thread * | |
2236 | dpif_netdev_flow_dump_thread_create(struct dpif_flow_dump *dump_) | |
2237 | { | |
2238 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); | |
2239 | struct dpif_netdev_flow_dump_thread *thread; | |
2240 | ||
2241 | thread = xmalloc(sizeof *thread); | |
2242 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
2243 | thread->dump = dump; | |
2244 | return &thread->up; | |
2245 | } | |
2246 | ||
2247 | static void | |
2248 | dpif_netdev_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) | |
2249 | { | |
2250 | struct dpif_netdev_flow_dump_thread *thread | |
2251 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2252 | ||
2253 | free(thread); | |
2254 | } | |
2255 | ||
704a1e09 | 2256 | static int |
ac64794a | 2257 | dpif_netdev_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
8bb113da | 2258 | struct dpif_flow *flows, int max_flows) |
ac64794a BP |
2259 | { |
2260 | struct dpif_netdev_flow_dump_thread *thread | |
2261 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2262 | struct dpif_netdev_flow_dump *dump = thread->dump; | |
8bb113da | 2263 | struct dp_netdev_flow *netdev_flows[FLOW_DUMP_MAX_BATCH]; |
8bb113da RW |
2264 | int n_flows = 0; |
2265 | int i; | |
14608a15 | 2266 | |
ac64794a | 2267 | ovs_mutex_lock(&dump->mutex); |
8bb113da | 2268 | if (!dump->status) { |
1c1e46ed AW |
2269 | struct dpif_netdev *dpif = dpif_netdev_cast(thread->up.dpif); |
2270 | struct dp_netdev *dp = get_dp_netdev(&dpif->dpif); | |
2271 | struct dp_netdev_pmd_thread *pmd = dump->cur_pmd; | |
2272 | int flow_limit = MIN(max_flows, FLOW_DUMP_MAX_BATCH); | |
2273 | ||
2274 | /* First call to dump_next(), extracts the first pmd thread. | |
2275 | * If there is no pmd thread, returns immediately. */ | |
2276 | if (!pmd) { | |
2277 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2278 | if (!pmd) { | |
2279 | ovs_mutex_unlock(&dump->mutex); | |
2280 | return n_flows; | |
8bb113da | 2281 | |
8bb113da | 2282 | } |
d2ad7ef1 | 2283 | } |
1c1e46ed AW |
2284 | |
2285 | do { | |
2286 | for (n_flows = 0; n_flows < flow_limit; n_flows++) { | |
2287 | struct cmap_node *node; | |
2288 | ||
2289 | node = cmap_next_position(&pmd->flow_table, &dump->flow_pos); | |
2290 | if (!node) { | |
2291 | break; | |
2292 | } | |
2293 | netdev_flows[n_flows] = CONTAINER_OF(node, | |
2294 | struct dp_netdev_flow, | |
2295 | node); | |
2296 | } | |
2297 | /* When finishing dumping the current pmd thread, moves to | |
2298 | * the next. */ | |
2299 | if (n_flows < flow_limit) { | |
2300 | memset(&dump->flow_pos, 0, sizeof dump->flow_pos); | |
2301 | dp_netdev_pmd_unref(pmd); | |
2302 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2303 | if (!pmd) { | |
2304 | dump->status = EOF; | |
2305 | break; | |
2306 | } | |
2307 | } | |
2308 | /* Keeps the reference to next caller. */ | |
2309 | dump->cur_pmd = pmd; | |
2310 | ||
2311 | /* If the current dump is empty, do not exit the loop, since the | |
2312 | * remaining pmds could have flows to be dumped. Just dumps again | |
2313 | * on the new 'pmd'. */ | |
2314 | } while (!n_flows); | |
8a4e3a85 | 2315 | } |
ac64794a | 2316 | ovs_mutex_unlock(&dump->mutex); |
ac64794a | 2317 | |
8bb113da RW |
2318 | for (i = 0; i < n_flows; i++) { |
2319 | struct odputil_keybuf *maskbuf = &thread->maskbuf[i]; | |
2320 | struct odputil_keybuf *keybuf = &thread->keybuf[i]; | |
2321 | struct dp_netdev_flow *netdev_flow = netdev_flows[i]; | |
2322 | struct dpif_flow *f = &flows[i]; | |
7af12bd7 | 2323 | struct ofpbuf key, mask; |
8bb113da | 2324 | |
7af12bd7 JS |
2325 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); |
2326 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
64bb477f JS |
2327 | dp_netdev_flow_to_dpif_flow(netdev_flow, &key, &mask, f, |
2328 | dump->up.terse); | |
8bb113da | 2329 | } |
feebdea2 | 2330 | |
8bb113da | 2331 | return n_flows; |
72865317 BP |
2332 | } |
2333 | ||
2334 | static int | |
758c456d | 2335 | dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute) |
65f13b50 | 2336 | OVS_NO_THREAD_SAFETY_ANALYSIS |
72865317 BP |
2337 | { |
2338 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
65f13b50 | 2339 | struct dp_netdev_pmd_thread *pmd; |
cf62fa4c | 2340 | struct dp_packet *pp; |
72865317 | 2341 | |
cf62fa4c PS |
2342 | if (dp_packet_size(execute->packet) < ETH_HEADER_LEN || |
2343 | dp_packet_size(execute->packet) > UINT16_MAX) { | |
72865317 BP |
2344 | return EINVAL; |
2345 | } | |
2346 | ||
65f13b50 AW |
2347 | /* Tries finding the 'pmd'. If NULL is returned, that means |
2348 | * the current thread is a non-pmd thread and should use | |
b19befae | 2349 | * dp_netdev_get_pmd(dp, NON_PMD_CORE_ID). */ |
65f13b50 AW |
2350 | pmd = ovsthread_getspecific(dp->per_pmd_key); |
2351 | if (!pmd) { | |
b19befae | 2352 | pmd = dp_netdev_get_pmd(dp, NON_PMD_CORE_ID); |
65f13b50 AW |
2353 | } |
2354 | ||
2355 | /* If the current thread is non-pmd thread, acquires | |
2356 | * the 'non_pmd_mutex'. */ | |
2357 | if (pmd->core_id == NON_PMD_CORE_ID) { | |
2358 | ovs_mutex_lock(&dp->non_pmd_mutex); | |
433330a8 | 2359 | ovs_mutex_lock(&dp->port_mutex); |
65f13b50 | 2360 | } |
1c1e46ed | 2361 | |
cf62fa4c | 2362 | pp = execute->packet; |
41ccaa24 | 2363 | dp_netdev_execute_actions(pmd, &pp, 1, false, execute->actions, |
9bbf1c3d | 2364 | execute->actions_len); |
65f13b50 | 2365 | if (pmd->core_id == NON_PMD_CORE_ID) { |
1c1e46ed | 2366 | dp_netdev_pmd_unref(pmd); |
433330a8 | 2367 | ovs_mutex_unlock(&dp->port_mutex); |
65f13b50 AW |
2368 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
2369 | } | |
8a4e3a85 | 2370 | |
758c456d | 2371 | return 0; |
72865317 BP |
2372 | } |
2373 | ||
1a0c894a BP |
2374 | static void |
2375 | dpif_netdev_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) | |
2376 | { | |
2377 | size_t i; | |
2378 | ||
2379 | for (i = 0; i < n_ops; i++) { | |
2380 | struct dpif_op *op = ops[i]; | |
2381 | ||
2382 | switch (op->type) { | |
2383 | case DPIF_OP_FLOW_PUT: | |
2384 | op->error = dpif_netdev_flow_put(dpif, &op->u.flow_put); | |
2385 | break; | |
2386 | ||
2387 | case DPIF_OP_FLOW_DEL: | |
2388 | op->error = dpif_netdev_flow_del(dpif, &op->u.flow_del); | |
2389 | break; | |
2390 | ||
2391 | case DPIF_OP_EXECUTE: | |
2392 | op->error = dpif_netdev_execute(dpif, &op->u.execute); | |
2393 | break; | |
6fe09f8c JS |
2394 | |
2395 | case DPIF_OP_FLOW_GET: | |
2396 | op->error = dpif_netdev_flow_get(dpif, &op->u.flow_get); | |
2397 | break; | |
1a0c894a BP |
2398 | } |
2399 | } | |
2400 | } | |
2401 | ||
f2eee189 AW |
2402 | /* Returns true if the configuration for rx queues or cpu mask |
2403 | * is changed. */ | |
2404 | static bool | |
2405 | pmd_config_changed(const struct dp_netdev *dp, size_t rxqs, const char *cmask) | |
2406 | { | |
2407 | if (dp->n_dpdk_rxqs != rxqs) { | |
2408 | return true; | |
2409 | } else { | |
2410 | if (dp->pmd_cmask != NULL && cmask != NULL) { | |
2411 | return strcmp(dp->pmd_cmask, cmask); | |
2412 | } else { | |
2413 | return (dp->pmd_cmask != NULL || cmask != NULL); | |
2414 | } | |
2415 | } | |
2416 | } | |
2417 | ||
2418 | /* Resets pmd threads if the configuration for 'rxq's or cpu mask changes. */ | |
2419 | static int | |
2420 | dpif_netdev_pmd_set(struct dpif *dpif, unsigned int n_rxqs, const char *cmask) | |
2421 | { | |
2422 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2423 | ||
2424 | if (pmd_config_changed(dp, n_rxqs, cmask)) { | |
2425 | struct dp_netdev_port *port; | |
2426 | ||
2427 | dp_netdev_destroy_all_pmds(dp); | |
2428 | ||
2429 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
2430 | if (netdev_is_pmd(port->netdev)) { | |
2431 | int i, err; | |
2432 | ||
2433 | /* Closes the existing 'rxq's. */ | |
2434 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2435 | netdev_rxq_close(port->rxq[i]); | |
2436 | port->rxq[i] = NULL; | |
2437 | } | |
2438 | ||
2439 | /* Sets the new rx queue config. */ | |
3bcc10c0 DDP |
2440 | err = netdev_set_multiq(port->netdev, |
2441 | ovs_numa_get_n_cores() + 1, | |
f2eee189 | 2442 | n_rxqs); |
7251515e | 2443 | if (err && (err != EOPNOTSUPP)) { |
f2eee189 AW |
2444 | VLOG_ERR("Failed to set dpdk interface %s rx_queue to:" |
2445 | " %u", netdev_get_name(port->netdev), | |
2446 | n_rxqs); | |
2447 | return err; | |
2448 | } | |
2449 | ||
2450 | /* If the set_multiq() above succeeds, reopens the 'rxq's. */ | |
2451 | port->rxq = xrealloc(port->rxq, sizeof *port->rxq | |
2452 | * netdev_n_rxq(port->netdev)); | |
2453 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2454 | netdev_rxq_open(port->netdev, &port->rxq[i], i); | |
2455 | } | |
2456 | } | |
2457 | } | |
2458 | dp->n_dpdk_rxqs = n_rxqs; | |
2459 | ||
2460 | /* Reconfigures the cpu mask. */ | |
2461 | ovs_numa_set_cpu_mask(cmask); | |
2462 | free(dp->pmd_cmask); | |
2463 | dp->pmd_cmask = cmask ? xstrdup(cmask) : NULL; | |
2464 | ||
2465 | /* Restores the non-pmd. */ | |
2466 | dp_netdev_set_nonpmd(dp); | |
2467 | /* Restores all pmd threads. */ | |
2468 | dp_netdev_reset_pmd_threads(dp); | |
2469 | } | |
2470 | ||
2471 | return 0; | |
2472 | } | |
2473 | ||
5bf93d67 EJ |
2474 | static int |
2475 | dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED, | |
2476 | uint32_t queue_id, uint32_t *priority) | |
2477 | { | |
2478 | *priority = queue_id; | |
2479 | return 0; | |
2480 | } | |
2481 | ||
72865317 | 2482 | \f |
9ff55ae2 DDP |
2483 | /* Creates and returns a new 'struct dp_netdev_actions', whose actions are |
2484 | * a copy of the 'ofpacts_len' bytes of 'ofpacts'. */ | |
a84cb64a BP |
2485 | struct dp_netdev_actions * |
2486 | dp_netdev_actions_create(const struct nlattr *actions, size_t size) | |
2487 | { | |
2488 | struct dp_netdev_actions *netdev_actions; | |
2489 | ||
9ff55ae2 DDP |
2490 | netdev_actions = xmalloc(sizeof *netdev_actions + size); |
2491 | memcpy(netdev_actions->actions, actions, size); | |
a84cb64a BP |
2492 | netdev_actions->size = size; |
2493 | ||
2494 | return netdev_actions; | |
2495 | } | |
2496 | ||
a84cb64a | 2497 | struct dp_netdev_actions * |
61e7deb1 | 2498 | dp_netdev_flow_get_actions(const struct dp_netdev_flow *flow) |
a84cb64a | 2499 | { |
61e7deb1 | 2500 | return ovsrcu_get(struct dp_netdev_actions *, &flow->actions); |
a84cb64a BP |
2501 | } |
2502 | ||
61e7deb1 BP |
2503 | static void |
2504 | dp_netdev_actions_free(struct dp_netdev_actions *actions) | |
a84cb64a | 2505 | { |
61e7deb1 | 2506 | free(actions); |
a84cb64a BP |
2507 | } |
2508 | \f | |
55e3ca97 DDP |
2509 | static inline unsigned long long |
2510 | cycles_counter(void) | |
2511 | { | |
2512 | #ifdef DPDK_NETDEV | |
2513 | return rte_get_tsc_cycles(); | |
2514 | #else | |
2515 | return 0; | |
2516 | #endif | |
2517 | } | |
2518 | ||
2519 | /* Fake mutex to make sure that the calls to cycles_count_* are balanced */ | |
2520 | extern struct ovs_mutex cycles_counter_fake_mutex; | |
2521 | ||
2522 | /* Start counting cycles. Must be followed by 'cycles_count_end()' */ | |
2523 | static inline void | |
2524 | cycles_count_start(struct dp_netdev_pmd_thread *pmd) | |
2525 | OVS_ACQUIRES(&cycles_counter_fake_mutex) | |
2526 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2527 | { | |
2528 | pmd->last_cycles = cycles_counter(); | |
2529 | } | |
2530 | ||
2531 | /* Stop counting cycles and add them to the counter 'type' */ | |
2532 | static inline void | |
2533 | cycles_count_end(struct dp_netdev_pmd_thread *pmd, | |
2534 | enum pmd_cycles_counter_type type) | |
2535 | OVS_RELEASES(&cycles_counter_fake_mutex) | |
2536 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2537 | { | |
2538 | unsigned long long interval = cycles_counter() - pmd->last_cycles; | |
2539 | ||
2540 | non_atomic_ullong_add(&pmd->cycles.n[type], interval); | |
2541 | } | |
e4cfed38 | 2542 | |
5794e276 | 2543 | static void |
65f13b50 | 2544 | dp_netdev_process_rxq_port(struct dp_netdev_pmd_thread *pmd, |
9bbf1c3d DDP |
2545 | struct dp_netdev_port *port, |
2546 | struct netdev_rxq *rxq) | |
e4cfed38 | 2547 | { |
cd159f1a | 2548 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 | 2549 | int error, cnt; |
e4cfed38 | 2550 | |
55e3ca97 | 2551 | cycles_count_start(pmd); |
8cbf4f47 | 2552 | error = netdev_rxq_recv(rxq, packets, &cnt); |
55e3ca97 | 2553 | cycles_count_end(pmd, PMD_CYCLES_POLLING); |
e4cfed38 | 2554 | if (!error) { |
41ccaa24 | 2555 | int i; |
3c33f0ff JR |
2556 | |
2557 | *recirc_depth_get() = 0; | |
41ccaa24 PS |
2558 | |
2559 | /* XXX: initialize md in netdev implementation. */ | |
2560 | for (i = 0; i < cnt; i++) { | |
35303d71 | 2561 | pkt_metadata_init(&packets[i]->md, port->port_no); |
41ccaa24 | 2562 | } |
55e3ca97 | 2563 | cycles_count_start(pmd); |
41ccaa24 | 2564 | dp_netdev_input(pmd, packets, cnt); |
55e3ca97 | 2565 | cycles_count_end(pmd, PMD_CYCLES_PROCESSING); |
e4cfed38 | 2566 | } else if (error != EAGAIN && error != EOPNOTSUPP) { |
3c33f0ff | 2567 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
e4cfed38 PS |
2568 | |
2569 | VLOG_ERR_RL(&rl, "error receiving data from %s: %s", | |
3c33f0ff | 2570 | netdev_get_name(port->netdev), ovs_strerror(error)); |
e4cfed38 PS |
2571 | } |
2572 | } | |
2573 | ||
a36de779 PS |
2574 | /* Return true if needs to revalidate datapath flows. */ |
2575 | static bool | |
e4cfed38 PS |
2576 | dpif_netdev_run(struct dpif *dpif) |
2577 | { | |
2578 | struct dp_netdev_port *port; | |
2579 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
b19befae AW |
2580 | struct dp_netdev_pmd_thread *non_pmd = dp_netdev_get_pmd(dp, |
2581 | NON_PMD_CORE_ID); | |
a36de779 | 2582 | uint64_t new_tnl_seq; |
e4cfed38 | 2583 | |
65f13b50 | 2584 | ovs_mutex_lock(&dp->non_pmd_mutex); |
a532e683 | 2585 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2586 | if (!netdev_is_pmd(port->netdev)) { |
2587 | int i; | |
2588 | ||
2589 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
65f13b50 | 2590 | dp_netdev_process_rxq_port(non_pmd, port, port->rxq[i]); |
55c955bd | 2591 | } |
e4cfed38 PS |
2592 | } |
2593 | } | |
65f13b50 | 2594 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
1c1e46ed AW |
2595 | dp_netdev_pmd_unref(non_pmd); |
2596 | ||
a36de779 PS |
2597 | tnl_arp_cache_run(); |
2598 | new_tnl_seq = seq_read(tnl_conf_seq); | |
2599 | ||
2600 | if (dp->last_tnl_conf_seq != new_tnl_seq) { | |
2601 | dp->last_tnl_conf_seq = new_tnl_seq; | |
2602 | return true; | |
2603 | } | |
2604 | return false; | |
e4cfed38 PS |
2605 | } |
2606 | ||
2607 | static void | |
2608 | dpif_netdev_wait(struct dpif *dpif) | |
2609 | { | |
2610 | struct dp_netdev_port *port; | |
2611 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2612 | ||
59e6d833 | 2613 | ovs_mutex_lock(&dp_netdev_mutex); |
a532e683 | 2614 | CMAP_FOR_EACH (port, node, &dp->ports) { |
55c955bd PS |
2615 | if (!netdev_is_pmd(port->netdev)) { |
2616 | int i; | |
2617 | ||
2618 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
2619 | netdev_rxq_wait(port->rxq[i]); | |
2620 | } | |
e4cfed38 PS |
2621 | } |
2622 | } | |
59e6d833 | 2623 | ovs_mutex_unlock(&dp_netdev_mutex); |
a36de779 | 2624 | seq_wait(tnl_conf_seq, dp->last_tnl_conf_seq); |
e4cfed38 PS |
2625 | } |
2626 | ||
f7791740 | 2627 | struct rxq_poll { |
e4cfed38 | 2628 | struct dp_netdev_port *port; |
55c955bd | 2629 | struct netdev_rxq *rx; |
e4cfed38 PS |
2630 | }; |
2631 | ||
2632 | static int | |
65f13b50 | 2633 | pmd_load_queues(struct dp_netdev_pmd_thread *pmd, |
f7791740 | 2634 | struct rxq_poll **ppoll_list, int poll_cnt) |
e4cfed38 | 2635 | { |
f7791740 | 2636 | struct rxq_poll *poll_list = *ppoll_list; |
e4cfed38 | 2637 | struct dp_netdev_port *port; |
65f13b50 | 2638 | int n_pmds_on_numa, index, i; |
e4cfed38 PS |
2639 | |
2640 | /* Simple scheduler for netdev rx polling. */ | |
e4cfed38 | 2641 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2642 | port_unref(poll_list[i].port); |
e4cfed38 PS |
2643 | } |
2644 | ||
2645 | poll_cnt = 0; | |
65f13b50 | 2646 | n_pmds_on_numa = get_n_pmd_threads_on_numa(pmd->dp, pmd->numa_id); |
e4cfed38 PS |
2647 | index = 0; |
2648 | ||
65f13b50 | 2649 | CMAP_FOR_EACH (port, node, &pmd->dp->ports) { |
a1fdee13 AW |
2650 | /* Calls port_try_ref() to prevent the main thread |
2651 | * from deleting the port. */ | |
2652 | if (port_try_ref(port)) { | |
65f13b50 AW |
2653 | if (netdev_is_pmd(port->netdev) |
2654 | && netdev_get_numa_id(port->netdev) == pmd->numa_id) { | |
a1fdee13 AW |
2655 | int i; |
2656 | ||
2657 | for (i = 0; i < netdev_n_rxq(port->netdev); i++) { | |
65f13b50 | 2658 | if ((index % n_pmds_on_numa) == pmd->index) { |
a1fdee13 AW |
2659 | poll_list = xrealloc(poll_list, |
2660 | sizeof *poll_list * (poll_cnt + 1)); | |
2661 | ||
2662 | port_ref(port); | |
2663 | poll_list[poll_cnt].port = port; | |
2664 | poll_list[poll_cnt].rx = port->rxq[i]; | |
2665 | poll_cnt++; | |
2666 | } | |
2667 | index++; | |
55c955bd | 2668 | } |
e4cfed38 | 2669 | } |
a1fdee13 AW |
2670 | /* Unrefs the port_try_ref(). */ |
2671 | port_unref(port); | |
e4cfed38 PS |
2672 | } |
2673 | } | |
2674 | ||
e4cfed38 PS |
2675 | *ppoll_list = poll_list; |
2676 | return poll_cnt; | |
2677 | } | |
2678 | ||
6c3eee82 | 2679 | static void * |
e4cfed38 | 2680 | pmd_thread_main(void *f_) |
6c3eee82 | 2681 | { |
65f13b50 | 2682 | struct dp_netdev_pmd_thread *pmd = f_; |
e4cfed38 | 2683 | unsigned int lc = 0; |
f7791740 | 2684 | struct rxq_poll *poll_list; |
84067a4c | 2685 | unsigned int port_seq = PMD_INITIAL_SEQ; |
e4cfed38 PS |
2686 | int poll_cnt; |
2687 | int i; | |
6c3eee82 | 2688 | |
e4cfed38 PS |
2689 | poll_cnt = 0; |
2690 | poll_list = NULL; | |
2691 | ||
65f13b50 AW |
2692 | /* Stores the pmd thread's 'pmd' to 'per_pmd_key'. */ |
2693 | ovsthread_setspecific(pmd->dp->per_pmd_key, pmd); | |
2694 | pmd_thread_setaffinity_cpu(pmd->core_id); | |
e4cfed38 | 2695 | reload: |
65f13b50 AW |
2696 | emc_cache_init(&pmd->flow_cache); |
2697 | poll_cnt = pmd_load_queues(pmd, &poll_list, poll_cnt); | |
6c3eee82 | 2698 | |
7dd671f0 MK |
2699 | /* List port/core affinity */ |
2700 | for (i = 0; i < poll_cnt; i++) { | |
2701 | VLOG_INFO("Core %d processing port \'%s\'\n", pmd->core_id, netdev_get_name(poll_list[i].port->netdev)); | |
2702 | } | |
2703 | ||
accf8626 AW |
2704 | /* Signal here to make sure the pmd finishes |
2705 | * reloading the updated configuration. */ | |
2706 | dp_netdev_pmd_reload_done(pmd); | |
2707 | ||
e4cfed38 | 2708 | for (;;) { |
6c3eee82 BP |
2709 | int i; |
2710 | ||
e4cfed38 | 2711 | for (i = 0; i < poll_cnt; i++) { |
65f13b50 | 2712 | dp_netdev_process_rxq_port(pmd, poll_list[i].port, poll_list[i].rx); |
e4cfed38 PS |
2713 | } |
2714 | ||
2715 | if (lc++ > 1024) { | |
84067a4c | 2716 | unsigned int seq; |
6c3eee82 | 2717 | |
e4cfed38 | 2718 | lc = 0; |
84067a4c | 2719 | |
67ad54cb | 2720 | emc_cache_slow_sweep(&pmd->flow_cache); |
84067a4c JR |
2721 | ovsrcu_quiesce(); |
2722 | ||
65f13b50 | 2723 | atomic_read_relaxed(&pmd->change_seq, &seq); |
84067a4c JR |
2724 | if (seq != port_seq) { |
2725 | port_seq = seq; | |
6c3eee82 BP |
2726 | break; |
2727 | } | |
2728 | } | |
e4cfed38 | 2729 | } |
6c3eee82 | 2730 | |
65f13b50 | 2731 | emc_cache_uninit(&pmd->flow_cache); |
9bbf1c3d | 2732 | |
65f13b50 | 2733 | if (!latch_is_set(&pmd->exit_latch)){ |
e4cfed38 PS |
2734 | goto reload; |
2735 | } | |
6c3eee82 | 2736 | |
e4cfed38 PS |
2737 | for (i = 0; i < poll_cnt; i++) { |
2738 | port_unref(poll_list[i].port); | |
6c3eee82 | 2739 | } |
6c3eee82 | 2740 | |
accf8626 AW |
2741 | dp_netdev_pmd_reload_done(pmd); |
2742 | ||
e4cfed38 | 2743 | free(poll_list); |
6c3eee82 BP |
2744 | return NULL; |
2745 | } | |
2746 | ||
6b31e073 RW |
2747 | static void |
2748 | dp_netdev_disable_upcall(struct dp_netdev *dp) | |
2749 | OVS_ACQUIRES(dp->upcall_rwlock) | |
2750 | { | |
2751 | fat_rwlock_wrlock(&dp->upcall_rwlock); | |
2752 | } | |
2753 | ||
2754 | static void | |
2755 | dpif_netdev_disable_upcall(struct dpif *dpif) | |
2756 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2757 | { | |
2758 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2759 | dp_netdev_disable_upcall(dp); | |
2760 | } | |
2761 | ||
2762 | static void | |
2763 | dp_netdev_enable_upcall(struct dp_netdev *dp) | |
2764 | OVS_RELEASES(dp->upcall_rwlock) | |
2765 | { | |
2766 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
2767 | } | |
2768 | ||
2769 | static void | |
2770 | dpif_netdev_enable_upcall(struct dpif *dpif) | |
2771 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
2772 | { | |
2773 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2774 | dp_netdev_enable_upcall(dp); | |
2775 | } | |
2776 | ||
accf8626 AW |
2777 | void |
2778 | dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd) | |
2779 | { | |
2780 | ovs_mutex_lock(&pmd->cond_mutex); | |
2781 | xpthread_cond_signal(&pmd->cond); | |
2782 | ovs_mutex_unlock(&pmd->cond_mutex); | |
2783 | } | |
2784 | ||
1c1e46ed AW |
2785 | /* Finds and refs the dp_netdev_pmd_thread on core 'core_id'. Returns |
2786 | * the pointer if succeeds, otherwise, NULL. | |
2787 | * | |
2788 | * Caller must unrefs the returned reference. */ | |
65f13b50 | 2789 | static struct dp_netdev_pmd_thread * |
bd5131ba | 2790 | dp_netdev_get_pmd(struct dp_netdev *dp, unsigned core_id) |
65f13b50 AW |
2791 | { |
2792 | struct dp_netdev_pmd_thread *pmd; | |
55847abe | 2793 | const struct cmap_node *pnode; |
65f13b50 | 2794 | |
b19befae | 2795 | pnode = cmap_find(&dp->poll_threads, hash_int(core_id, 0)); |
1c1e46ed AW |
2796 | if (!pnode) { |
2797 | return NULL; | |
2798 | } | |
65f13b50 AW |
2799 | pmd = CONTAINER_OF(pnode, struct dp_netdev_pmd_thread, node); |
2800 | ||
1c1e46ed | 2801 | return dp_netdev_pmd_try_ref(pmd) ? pmd : NULL; |
65f13b50 AW |
2802 | } |
2803 | ||
f2eee189 AW |
2804 | /* Sets the 'struct dp_netdev_pmd_thread' for non-pmd threads. */ |
2805 | static void | |
2806 | dp_netdev_set_nonpmd(struct dp_netdev *dp) | |
2807 | { | |
2808 | struct dp_netdev_pmd_thread *non_pmd; | |
2809 | ||
2810 | non_pmd = xzalloc(sizeof *non_pmd); | |
2811 | dp_netdev_configure_pmd(non_pmd, dp, 0, NON_PMD_CORE_ID, | |
2812 | OVS_NUMA_UNSPEC); | |
2813 | } | |
2814 | ||
1c1e46ed AW |
2815 | /* Caller must have valid pointer to 'pmd'. */ |
2816 | static bool | |
2817 | dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd) | |
2818 | { | |
2819 | return ovs_refcount_try_ref_rcu(&pmd->ref_cnt); | |
2820 | } | |
2821 | ||
2822 | static void | |
2823 | dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd) | |
2824 | { | |
2825 | if (pmd && ovs_refcount_unref(&pmd->ref_cnt) == 1) { | |
2826 | ovsrcu_postpone(dp_netdev_destroy_pmd, pmd); | |
2827 | } | |
2828 | } | |
2829 | ||
2830 | /* Given cmap position 'pos', tries to ref the next node. If try_ref() | |
2831 | * fails, keeps checking for next node until reaching the end of cmap. | |
2832 | * | |
2833 | * Caller must unrefs the returned reference. */ | |
2834 | static struct dp_netdev_pmd_thread * | |
2835 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos) | |
2836 | { | |
2837 | struct dp_netdev_pmd_thread *next; | |
2838 | ||
2839 | do { | |
2840 | struct cmap_node *node; | |
2841 | ||
2842 | node = cmap_next_position(&dp->poll_threads, pos); | |
2843 | next = node ? CONTAINER_OF(node, struct dp_netdev_pmd_thread, node) | |
2844 | : NULL; | |
2845 | } while (next && !dp_netdev_pmd_try_ref(next)); | |
2846 | ||
2847 | return next; | |
2848 | } | |
2849 | ||
3bcc10c0 DDP |
2850 | static int |
2851 | core_id_to_qid(unsigned core_id) | |
2852 | { | |
2853 | if (core_id != NON_PMD_CORE_ID) { | |
2854 | return core_id; | |
2855 | } else { | |
2856 | return ovs_numa_get_n_cores(); | |
2857 | } | |
2858 | } | |
2859 | ||
65f13b50 | 2860 | /* Configures the 'pmd' based on the input argument. */ |
6c3eee82 | 2861 | static void |
65f13b50 | 2862 | dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp, |
bd5131ba | 2863 | int index, unsigned core_id, int numa_id) |
65f13b50 AW |
2864 | { |
2865 | pmd->dp = dp; | |
2866 | pmd->index = index; | |
2867 | pmd->core_id = core_id; | |
3bcc10c0 | 2868 | pmd->tx_qid = core_id_to_qid(core_id); |
65f13b50 | 2869 | pmd->numa_id = numa_id; |
1c1e46ed AW |
2870 | |
2871 | ovs_refcount_init(&pmd->ref_cnt); | |
65f13b50 AW |
2872 | latch_init(&pmd->exit_latch); |
2873 | atomic_init(&pmd->change_seq, PMD_INITIAL_SEQ); | |
accf8626 AW |
2874 | xpthread_cond_init(&pmd->cond, NULL); |
2875 | ovs_mutex_init(&pmd->cond_mutex); | |
1c1e46ed AW |
2876 | ovs_mutex_init(&pmd->flow_mutex); |
2877 | dpcls_init(&pmd->cls); | |
2878 | cmap_init(&pmd->flow_table); | |
65f13b50 AW |
2879 | /* init the 'flow_cache' since there is no |
2880 | * actual thread created for NON_PMD_CORE_ID. */ | |
2881 | if (core_id == NON_PMD_CORE_ID) { | |
2882 | emc_cache_init(&pmd->flow_cache); | |
2883 | } | |
2884 | cmap_insert(&dp->poll_threads, CONST_CAST(struct cmap_node *, &pmd->node), | |
2885 | hash_int(core_id, 0)); | |
2886 | } | |
2887 | ||
1c1e46ed AW |
2888 | static void |
2889 | dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd) | |
2890 | { | |
2891 | dp_netdev_pmd_flow_flush(pmd); | |
2892 | dpcls_destroy(&pmd->cls); | |
2893 | cmap_destroy(&pmd->flow_table); | |
2894 | ovs_mutex_destroy(&pmd->flow_mutex); | |
2895 | latch_destroy(&pmd->exit_latch); | |
2896 | xpthread_cond_destroy(&pmd->cond); | |
2897 | ovs_mutex_destroy(&pmd->cond_mutex); | |
2898 | free(pmd); | |
2899 | } | |
2900 | ||
2901 | /* Stops the pmd thread, removes it from the 'dp->poll_threads', | |
2902 | * and unrefs the struct. */ | |
65f13b50 AW |
2903 | static void |
2904 | dp_netdev_del_pmd(struct dp_netdev_pmd_thread *pmd) | |
6c3eee82 | 2905 | { |
65f13b50 | 2906 | /* Uninit the 'flow_cache' since there is |
1c1e46ed | 2907 | * no actual thread uninit it for NON_PMD_CORE_ID. */ |
65f13b50 AW |
2908 | if (pmd->core_id == NON_PMD_CORE_ID) { |
2909 | emc_cache_uninit(&pmd->flow_cache); | |
2910 | } else { | |
2911 | latch_set(&pmd->exit_latch); | |
2912 | dp_netdev_reload_pmd__(pmd); | |
2913 | ovs_numa_unpin_core(pmd->core_id); | |
2914 | xpthread_join(pmd->thread, NULL); | |
2915 | } | |
2916 | cmap_remove(&pmd->dp->poll_threads, &pmd->node, hash_int(pmd->core_id, 0)); | |
1c1e46ed | 2917 | dp_netdev_pmd_unref(pmd); |
65f13b50 | 2918 | } |
6c3eee82 | 2919 | |
65f13b50 AW |
2920 | /* Destroys all pmd threads. */ |
2921 | static void | |
2922 | dp_netdev_destroy_all_pmds(struct dp_netdev *dp) | |
2923 | { | |
2924 | struct dp_netdev_pmd_thread *pmd; | |
2925 | ||
2926 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2927 | dp_netdev_del_pmd(pmd); | |
6c3eee82 | 2928 | } |
65f13b50 | 2929 | } |
6c3eee82 | 2930 | |
65f13b50 AW |
2931 | /* Deletes all pmd threads on numa node 'numa_id'. */ |
2932 | static void | |
2933 | dp_netdev_del_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
2934 | { | |
2935 | struct dp_netdev_pmd_thread *pmd; | |
6c3eee82 | 2936 | |
65f13b50 AW |
2937 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { |
2938 | if (pmd->numa_id == numa_id) { | |
2939 | dp_netdev_del_pmd(pmd); | |
2940 | } | |
6c3eee82 | 2941 | } |
65f13b50 | 2942 | } |
6c3eee82 | 2943 | |
65f13b50 AW |
2944 | /* Checks the numa node id of 'netdev' and starts pmd threads for |
2945 | * the numa node. */ | |
2946 | static void | |
2947 | dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id) | |
2948 | { | |
2949 | int n_pmds; | |
e4cfed38 | 2950 | |
65f13b50 AW |
2951 | if (!ovs_numa_numa_id_is_valid(numa_id)) { |
2952 | VLOG_ERR("Cannot create pmd threads due to numa id (%d)" | |
2953 | "invalid", numa_id); | |
2954 | return ; | |
2955 | } | |
2956 | ||
2957 | n_pmds = get_n_pmd_threads_on_numa(dp, numa_id); | |
2958 | ||
2959 | /* If there are already pmd threads created for the numa node | |
2960 | * in which 'netdev' is on, do nothing. Else, creates the | |
2961 | * pmd threads for the numa node. */ | |
2962 | if (!n_pmds) { | |
2963 | int can_have, n_unpinned, i; | |
2964 | ||
2965 | n_unpinned = ovs_numa_get_n_unpinned_cores_on_numa(numa_id); | |
2966 | if (!n_unpinned) { | |
2967 | VLOG_ERR("Cannot create pmd threads due to out of unpinned " | |
2968 | "cores on numa node"); | |
2969 | return; | |
2970 | } | |
6c3eee82 | 2971 | |
f2eee189 AW |
2972 | /* If cpu mask is specified, uses all unpinned cores, otherwise |
2973 | * tries creating NR_PMD_THREADS pmd threads. */ | |
2974 | can_have = dp->pmd_cmask ? n_unpinned : MIN(n_unpinned, NR_PMD_THREADS); | |
65f13b50 AW |
2975 | for (i = 0; i < can_have; i++) { |
2976 | struct dp_netdev_pmd_thread *pmd = xzalloc(sizeof *pmd); | |
bd5131ba | 2977 | unsigned core_id = ovs_numa_get_unpinned_core_on_numa(numa_id); |
e4cfed38 | 2978 | |
65f13b50 AW |
2979 | dp_netdev_configure_pmd(pmd, dp, i, core_id, numa_id); |
2980 | /* Each thread will distribute all devices rx-queues among | |
2981 | * themselves. */ | |
2982 | pmd->thread = ovs_thread_create("pmd", pmd_thread_main, pmd); | |
2983 | } | |
2984 | VLOG_INFO("Created %d pmd threads on numa node %d", can_have, numa_id); | |
6c3eee82 BP |
2985 | } |
2986 | } | |
e4cfed38 | 2987 | |
6c3eee82 | 2988 | \f |
f2eee189 AW |
2989 | /* Called after pmd threads config change. Restarts pmd threads with |
2990 | * new configuration. */ | |
2991 | static void | |
2992 | dp_netdev_reset_pmd_threads(struct dp_netdev *dp) | |
2993 | { | |
2994 | struct dp_netdev_port *port; | |
2995 | ||
2996 | CMAP_FOR_EACH (port, node, &dp->ports) { | |
2997 | if (netdev_is_pmd(port->netdev)) { | |
2998 | int numa_id = netdev_get_numa_id(port->netdev); | |
2999 | ||
3000 | dp_netdev_set_pmds_on_numa(dp, numa_id); | |
3001 | } | |
3002 | } | |
3003 | } | |
3004 | ||
b5cbbcf6 AZ |
3005 | static char * |
3006 | dpif_netdev_get_datapath_version(void) | |
3007 | { | |
3008 | return xstrdup("<built-in>"); | |
3009 | } | |
3010 | ||
72865317 | 3011 | static void |
1c1e46ed | 3012 | dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow, int cnt, int size, |
11bfdadd | 3013 | uint16_t tcp_flags, long long now) |
72865317 | 3014 | { |
eb94da30 | 3015 | uint16_t flags; |
72865317 | 3016 | |
eb94da30 DDP |
3017 | atomic_store_relaxed(&netdev_flow->stats.used, now); |
3018 | non_atomic_ullong_add(&netdev_flow->stats.packet_count, cnt); | |
3019 | non_atomic_ullong_add(&netdev_flow->stats.byte_count, size); | |
3020 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
3021 | flags |= tcp_flags; | |
3022 | atomic_store_relaxed(&netdev_flow->stats.tcp_flags, flags); | |
51852a57 BP |
3023 | } |
3024 | ||
3025 | static void | |
1c1e46ed AW |
3026 | dp_netdev_count_packet(struct dp_netdev_pmd_thread *pmd, |
3027 | enum dp_stat_type type, int cnt) | |
51852a57 | 3028 | { |
eb94da30 | 3029 | non_atomic_ullong_add(&pmd->stats.n[type], cnt); |
51852a57 BP |
3030 | } |
3031 | ||
623540e4 | 3032 | static int |
e14deea0 | 3033 | dp_netdev_upcall(struct dp_netdev_pmd_thread *pmd, struct dp_packet *packet_, |
7af12bd7 | 3034 | struct flow *flow, struct flow_wildcards *wc, ovs_u128 *ufid, |
623540e4 EJ |
3035 | enum dpif_upcall_type type, const struct nlattr *userdata, |
3036 | struct ofpbuf *actions, struct ofpbuf *put_actions) | |
3037 | { | |
1c1e46ed | 3038 | struct dp_netdev *dp = pmd->dp; |
623540e4 | 3039 | |
623540e4 EJ |
3040 | if (OVS_UNLIKELY(!dp->upcall_cb)) { |
3041 | return ENODEV; | |
3042 | } | |
3043 | ||
3044 | if (OVS_UNLIKELY(!VLOG_DROP_DBG(&upcall_rl))) { | |
3045 | struct ds ds = DS_EMPTY_INITIALIZER; | |
623540e4 | 3046 | char *packet_str; |
cf62fa4c | 3047 | struct ofpbuf key; |
5262eea1 JG |
3048 | struct odp_flow_key_parms odp_parms = { |
3049 | .flow = flow, | |
3050 | .mask = &wc->masks, | |
3051 | .odp_in_port = flow->in_port.odp_port, | |
2494ccd7 | 3052 | .support = dp_netdev_support, |
5262eea1 | 3053 | }; |
623540e4 EJ |
3054 | |
3055 | ofpbuf_init(&key, 0); | |
5262eea1 | 3056 | odp_flow_key_from_flow(&odp_parms, &key); |
cf62fa4c PS |
3057 | packet_str = ofp_packet_to_string(dp_packet_data(packet_), |
3058 | dp_packet_size(packet_)); | |
623540e4 | 3059 | |
6fd6ed71 | 3060 | odp_flow_key_format(key.data, key.size, &ds); |
623540e4 EJ |
3061 | |
3062 | VLOG_DBG("%s: %s upcall:\n%s\n%s", dp->name, | |
3063 | dpif_upcall_type_to_string(type), ds_cstr(&ds), packet_str); | |
3064 | ||
3065 | ofpbuf_uninit(&key); | |
3066 | free(packet_str); | |
6fd6ed71 | 3067 | |
623540e4 EJ |
3068 | ds_destroy(&ds); |
3069 | } | |
3070 | ||
cf62fa4c | 3071 | return dp->upcall_cb(packet_, flow, ufid, pmd->core_id, type, userdata, |
1c1e46ed | 3072 | actions, wc, put_actions, dp->upcall_aux); |
623540e4 EJ |
3073 | } |
3074 | ||
9bbf1c3d | 3075 | static inline uint32_t |
048963aa DDP |
3076 | dpif_netdev_packet_get_rss_hash(struct dp_packet *packet, |
3077 | const struct miniflow *mf) | |
9bbf1c3d | 3078 | { |
048963aa | 3079 | uint32_t hash, recirc_depth; |
9bbf1c3d | 3080 | |
2bc1bbd2 | 3081 | hash = dp_packet_get_rss_hash(packet); |
9bbf1c3d DDP |
3082 | if (OVS_UNLIKELY(!hash)) { |
3083 | hash = miniflow_hash_5tuple(mf, 0); | |
2bc1bbd2 | 3084 | dp_packet_set_rss_hash(packet, hash); |
9bbf1c3d | 3085 | } |
048963aa DDP |
3086 | |
3087 | /* The RSS hash must account for the recirculation depth to avoid | |
3088 | * collisions in the exact match cache */ | |
3089 | recirc_depth = *recirc_depth_get_unsafe(); | |
3090 | if (OVS_UNLIKELY(recirc_depth)) { | |
3091 | hash = hash_finish(hash, recirc_depth); | |
3092 | dp_packet_set_rss_hash(packet, hash); | |
3093 | } | |
9bbf1c3d DDP |
3094 | return hash; |
3095 | } | |
3096 | ||
567bbb2e | 3097 | struct packet_batch { |
8cbf4f47 DDP |
3098 | unsigned int packet_count; |
3099 | unsigned int byte_count; | |
3100 | uint16_t tcp_flags; | |
3101 | ||
3102 | struct dp_netdev_flow *flow; | |
3103 | ||
cd159f1a | 3104 | struct dp_packet *packets[NETDEV_MAX_BURST]; |
8cbf4f47 DDP |
3105 | }; |
3106 | ||
3107 | static inline void | |
e14deea0 | 3108 | packet_batch_update(struct packet_batch *batch, struct dp_packet *packet, |
9bbf1c3d | 3109 | const struct miniflow *mf) |
8cbf4f47 DDP |
3110 | { |
3111 | batch->tcp_flags |= miniflow_get_tcp_flags(mf); | |
3112 | batch->packets[batch->packet_count++] = packet; | |
cf62fa4c | 3113 | batch->byte_count += dp_packet_size(packet); |
8cbf4f47 DDP |
3114 | } |
3115 | ||
3116 | static inline void | |
41ccaa24 | 3117 | packet_batch_init(struct packet_batch *batch, struct dp_netdev_flow *flow) |
8cbf4f47 | 3118 | { |
11e5cf1f | 3119 | flow->batch = batch; |
8cbf4f47 | 3120 | |
11e5cf1f | 3121 | batch->flow = flow; |
8cbf4f47 DDP |
3122 | batch->packet_count = 0; |
3123 | batch->byte_count = 0; | |
3124 | batch->tcp_flags = 0; | |
8cbf4f47 DDP |
3125 | } |
3126 | ||
3127 | static inline void | |
65f13b50 | 3128 | packet_batch_execute(struct packet_batch *batch, |
abcf3ef4 | 3129 | struct dp_netdev_pmd_thread *pmd, |
11bfdadd | 3130 | long long now) |
8cbf4f47 DDP |
3131 | { |
3132 | struct dp_netdev_actions *actions; | |
3133 | struct dp_netdev_flow *flow = batch->flow; | |
3134 | ||
11e5cf1f | 3135 | dp_netdev_flow_used(flow, batch->packet_count, batch->byte_count, |
11bfdadd | 3136 | batch->tcp_flags, now); |
8cbf4f47 DDP |
3137 | |
3138 | actions = dp_netdev_flow_get_actions(flow); | |
3139 | ||
65f13b50 | 3140 | dp_netdev_execute_actions(pmd, batch->packets, batch->packet_count, true, |
41ccaa24 | 3141 | actions->actions, actions->size); |
8cbf4f47 DDP |
3142 | } |
3143 | ||
8aaa125d | 3144 | static inline void |
e14deea0 | 3145 | dp_netdev_queue_batches(struct dp_packet *pkt, |
9bbf1c3d | 3146 | struct dp_netdev_flow *flow, const struct miniflow *mf, |
8aaa125d | 3147 | struct packet_batch *batches, size_t *n_batches) |
9bbf1c3d | 3148 | { |
8aaa125d | 3149 | struct packet_batch *batch = flow->batch; |
11e5cf1f DDP |
3150 | |
3151 | if (OVS_LIKELY(batch)) { | |
3152 | packet_batch_update(batch, pkt, mf); | |
8aaa125d | 3153 | return; |
9bbf1c3d DDP |
3154 | } |
3155 | ||
3156 | batch = &batches[(*n_batches)++]; | |
41ccaa24 | 3157 | packet_batch_init(batch, flow); |
9bbf1c3d | 3158 | packet_batch_update(batch, pkt, mf); |
9bbf1c3d DDP |
3159 | } |
3160 | ||
3161 | static inline void | |
e14deea0 | 3162 | dp_packet_swap(struct dp_packet **a, struct dp_packet **b) |
9bbf1c3d | 3163 | { |
e14deea0 | 3164 | struct dp_packet *tmp = *a; |
9bbf1c3d DDP |
3165 | *a = *b; |
3166 | *b = tmp; | |
3167 | } | |
3168 | ||
3169 | /* Try to process all ('cnt') the 'packets' using only the exact match cache | |
8aaa125d DDP |
3170 | * 'flow_cache'. If a flow is not found for a packet 'packets[i]', the |
3171 | * miniflow is copied into 'keys' and the packet pointer is moved at the | |
3172 | * beginning of the 'packets' array. | |
9bbf1c3d DDP |
3173 | * |
3174 | * The function returns the number of packets that needs to be processed in the | |
3175 | * 'packets' array (they have been moved to the beginning of the vector). | |
3176 | */ | |
3177 | static inline size_t | |
e14deea0 | 3178 | emc_processing(struct dp_netdev_pmd_thread *pmd, struct dp_packet **packets, |
8aaa125d DDP |
3179 | size_t cnt, struct netdev_flow_key *keys, |
3180 | struct packet_batch batches[], size_t *n_batches) | |
72865317 | 3181 | { |
65f13b50 | 3182 | struct emc_cache *flow_cache = &pmd->flow_cache; |
8aaa125d DDP |
3183 | struct netdev_flow_key key; |
3184 | size_t i, notfound_cnt = 0; | |
8cbf4f47 | 3185 | |
84d6d5eb | 3186 | for (i = 0; i < cnt; i++) { |
9bbf1c3d | 3187 | struct dp_netdev_flow *flow; |
9bbf1c3d | 3188 | |
cf62fa4c | 3189 | if (OVS_UNLIKELY(dp_packet_size(packets[i]) < ETH_HEADER_LEN)) { |
e14deea0 | 3190 | dp_packet_delete(packets[i]); |
84d6d5eb EJ |
3191 | continue; |
3192 | } | |
8cbf4f47 | 3193 | |
72a5e2b8 DDP |
3194 | if (i != cnt - 1) { |
3195 | /* Prefetch next packet data */ | |
3196 | OVS_PREFETCH(dp_packet_data(packets[i+1])); | |
3197 | } | |
3198 | ||
cf62fa4c | 3199 | miniflow_extract(packets[i], &key.mf); |
0de8783a | 3200 | key.len = 0; /* Not computed yet. */ |
048963aa | 3201 | key.hash = dpif_netdev_packet_get_rss_hash(packets[i], &key.mf); |
9bbf1c3d | 3202 | |
0de8783a | 3203 | flow = emc_lookup(flow_cache, &key); |
8aaa125d DDP |
3204 | if (OVS_LIKELY(flow)) { |
3205 | dp_netdev_queue_batches(packets[i], flow, &key.mf, batches, | |
3206 | n_batches); | |
3207 | } else { | |
9bbf1c3d | 3208 | if (i != notfound_cnt) { |
e14deea0 | 3209 | dp_packet_swap(&packets[i], &packets[notfound_cnt]); |
9bbf1c3d DDP |
3210 | } |
3211 | ||
3212 | keys[notfound_cnt++] = key; | |
3213 | } | |
3214 | } | |
3215 | ||
8aaa125d | 3216 | dp_netdev_count_packet(pmd, DP_STAT_EXACT_HIT, cnt - notfound_cnt); |
4f150744 | 3217 | |
9bbf1c3d DDP |
3218 | return notfound_cnt; |
3219 | } | |
3220 | ||
3221 | static inline void | |
65f13b50 | 3222 | fast_path_processing(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3223 | struct dp_packet **packets, size_t cnt, |
8aaa125d DDP |
3224 | struct netdev_flow_key *keys, |
3225 | struct packet_batch batches[], size_t *n_batches) | |
9bbf1c3d | 3226 | { |
1a0d5831 | 3227 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3228 | const size_t PKT_ARRAY_SIZE = cnt; |
3229 | #else | |
1a0d5831 | 3230 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3231 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d | 3232 | #endif |
0de8783a | 3233 | struct dpcls_rule *rules[PKT_ARRAY_SIZE]; |
65f13b50 AW |
3234 | struct dp_netdev *dp = pmd->dp; |
3235 | struct emc_cache *flow_cache = &pmd->flow_cache; | |
8aaa125d | 3236 | int miss_cnt = 0, lost_cnt = 0; |
9bbf1c3d | 3237 | bool any_miss; |
8aaa125d | 3238 | size_t i; |
9bbf1c3d DDP |
3239 | |
3240 | for (i = 0; i < cnt; i++) { | |
0de8783a | 3241 | /* Key length is needed in all the cases, hash computed on demand. */ |
361d808d | 3242 | keys[i].len = netdev_flow_key_size(miniflow_n_values(&keys[i].mf)); |
9bbf1c3d | 3243 | } |
1c1e46ed | 3244 | any_miss = !dpcls_lookup(&pmd->cls, keys, rules, cnt); |
623540e4 EJ |
3245 | if (OVS_UNLIKELY(any_miss) && !fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { |
3246 | uint64_t actions_stub[512 / 8], slow_stub[512 / 8]; | |
3247 | struct ofpbuf actions, put_actions; | |
7af12bd7 | 3248 | ovs_u128 ufid; |
623540e4 EJ |
3249 | |
3250 | ofpbuf_use_stub(&actions, actions_stub, sizeof actions_stub); | |
3251 | ofpbuf_use_stub(&put_actions, slow_stub, sizeof slow_stub); | |
3252 | ||
3253 | for (i = 0; i < cnt; i++) { | |
0de8783a | 3254 | struct dp_netdev_flow *netdev_flow; |
623540e4 | 3255 | struct ofpbuf *add_actions; |
0de8783a | 3256 | struct match match; |
623540e4 EJ |
3257 | int error; |
3258 | ||
0de8783a | 3259 | if (OVS_LIKELY(rules[i])) { |
623540e4 EJ |
3260 | continue; |
3261 | } | |
3262 | ||
3263 | /* It's possible that an earlier slow path execution installed | |
0de8783a | 3264 | * a rule covering this flow. In this case, it's a lot cheaper |
623540e4 | 3265 | * to catch it here than execute a miss. */ |
1c1e46ed | 3266 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); |
623540e4 | 3267 | if (netdev_flow) { |
0de8783a | 3268 | rules[i] = &netdev_flow->cr; |
623540e4 EJ |
3269 | continue; |
3270 | } | |
3271 | ||
60fc3b7b DDP |
3272 | miss_cnt++; |
3273 | ||
0de8783a | 3274 | miniflow_expand(&keys[i].mf, &match.flow); |
623540e4 EJ |
3275 | |
3276 | ofpbuf_clear(&actions); | |
3277 | ofpbuf_clear(&put_actions); | |
3278 | ||
7af12bd7 | 3279 | dpif_flow_hash(dp->dpif, &match.flow, sizeof match.flow, &ufid); |
1c1e46ed | 3280 | error = dp_netdev_upcall(pmd, packets[i], &match.flow, &match.wc, |
7af12bd7 | 3281 | &ufid, DPIF_UC_MISS, NULL, &actions, |
0de8783a | 3282 | &put_actions); |
623540e4 | 3283 | if (OVS_UNLIKELY(error && error != ENOSPC)) { |
7ad20cbd | 3284 | dp_packet_delete(packets[i]); |
60fc3b7b | 3285 | lost_cnt++; |
623540e4 EJ |
3286 | continue; |
3287 | } | |
3288 | ||
3289 | /* We can't allow the packet batching in the next loop to execute | |
3290 | * the actions. Otherwise, if there are any slow path actions, | |
3291 | * we'll send the packet up twice. */ | |
41ccaa24 | 3292 | dp_netdev_execute_actions(pmd, &packets[i], 1, true, |
6fd6ed71 | 3293 | actions.data, actions.size); |
623540e4 | 3294 | |
6fd6ed71 | 3295 | add_actions = put_actions.size ? &put_actions : &actions; |
0de8783a JR |
3296 | if (OVS_LIKELY(error != ENOSPC)) { |
3297 | /* XXX: There's a race window where a flow covering this packet | |
3298 | * could have already been installed since we last did the flow | |
3299 | * lookup before upcall. This could be solved by moving the | |
3300 | * mutex lock outside the loop, but that's an awful long time | |
3301 | * to be locking everyone out of making flow installs. If we | |
3302 | * move to a per-core classifier, it would be reasonable. */ | |
1c1e46ed AW |
3303 | ovs_mutex_lock(&pmd->flow_mutex); |
3304 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i]); | |
0de8783a | 3305 | if (OVS_LIKELY(!netdev_flow)) { |
1c1e46ed | 3306 | netdev_flow = dp_netdev_flow_add(pmd, &match, &ufid, |
6fd6ed71 PS |
3307 | add_actions->data, |
3308 | add_actions->size); | |
0de8783a | 3309 | } |
1c1e46ed | 3310 | ovs_mutex_unlock(&pmd->flow_mutex); |
0de8783a | 3311 | |
0de8783a | 3312 | emc_insert(flow_cache, &keys[i], netdev_flow); |
623540e4 | 3313 | } |
623540e4 EJ |
3314 | } |
3315 | ||
3316 | ofpbuf_uninit(&actions); | |
3317 | ofpbuf_uninit(&put_actions); | |
3318 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
60fc3b7b | 3319 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); |
ac8c2081 | 3320 | } else if (OVS_UNLIKELY(any_miss)) { |
ac8c2081 | 3321 | for (i = 0; i < cnt; i++) { |
0de8783a | 3322 | if (OVS_UNLIKELY(!rules[i])) { |
e14deea0 | 3323 | dp_packet_delete(packets[i]); |
8aaa125d DDP |
3324 | lost_cnt++; |
3325 | miss_cnt++; | |
ac8c2081 DDP |
3326 | } |
3327 | } | |
623540e4 | 3328 | } |
84d6d5eb | 3329 | |
8cbf4f47 | 3330 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3331 | struct dp_packet *packet = packets[i]; |
84d6d5eb | 3332 | struct dp_netdev_flow *flow; |
8cbf4f47 | 3333 | |
0de8783a | 3334 | if (OVS_UNLIKELY(!rules[i])) { |
84d6d5eb EJ |
3335 | continue; |
3336 | } | |
3337 | ||
84d6d5eb | 3338 | flow = dp_netdev_flow_cast(rules[i]); |
0de8783a | 3339 | |
0de8783a | 3340 | emc_insert(flow_cache, &keys[i], flow); |
8aaa125d | 3341 | dp_netdev_queue_batches(packet, flow, &keys[i].mf, batches, n_batches); |
8cbf4f47 DDP |
3342 | } |
3343 | ||
8aaa125d DDP |
3344 | dp_netdev_count_packet(pmd, DP_STAT_MASKED_HIT, cnt - miss_cnt); |
3345 | dp_netdev_count_packet(pmd, DP_STAT_MISS, miss_cnt); | |
3346 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); | |
72865317 BP |
3347 | } |
3348 | ||
adcf00ba | 3349 | static void |
65f13b50 | 3350 | dp_netdev_input(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3351 | struct dp_packet **packets, int cnt) |
9bbf1c3d | 3352 | { |
1a0d5831 | 3353 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
3354 | const size_t PKT_ARRAY_SIZE = cnt; |
3355 | #else | |
1a0d5831 | 3356 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 3357 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d DDP |
3358 | #endif |
3359 | struct netdev_flow_key keys[PKT_ARRAY_SIZE]; | |
8aaa125d | 3360 | struct packet_batch batches[PKT_ARRAY_SIZE]; |
11bfdadd | 3361 | long long now = time_msec(); |
8aaa125d | 3362 | size_t newcnt, n_batches, i; |
9bbf1c3d | 3363 | |
8aaa125d DDP |
3364 | n_batches = 0; |
3365 | newcnt = emc_processing(pmd, packets, cnt, keys, batches, &n_batches); | |
9bbf1c3d | 3366 | if (OVS_UNLIKELY(newcnt)) { |
8aaa125d DDP |
3367 | fast_path_processing(pmd, packets, newcnt, keys, batches, &n_batches); |
3368 | } | |
3369 | ||
603f2ce0 EJ |
3370 | for (i = 0; i < n_batches; i++) { |
3371 | batches[i].flow->batch = NULL; | |
3372 | } | |
3373 | ||
8aaa125d DDP |
3374 | for (i = 0; i < n_batches; i++) { |
3375 | packet_batch_execute(&batches[i], pmd, now); | |
9bbf1c3d DDP |
3376 | } |
3377 | } | |
3378 | ||
9080a111 | 3379 | struct dp_netdev_execute_aux { |
65f13b50 | 3380 | struct dp_netdev_pmd_thread *pmd; |
9080a111 JR |
3381 | }; |
3382 | ||
6b31e073 | 3383 | static void |
623540e4 EJ |
3384 | dpif_netdev_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, |
3385 | void *aux) | |
6b31e073 RW |
3386 | { |
3387 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
623540e4 | 3388 | dp->upcall_aux = aux; |
6b31e073 RW |
3389 | dp->upcall_cb = cb; |
3390 | } | |
3391 | ||
ac8c2081 | 3392 | static void |
3bcc10c0 | 3393 | dp_netdev_drop_packets(struct dp_packet **packets, int cnt, bool may_steal) |
ac8c2081 | 3394 | { |
ac8c2081 | 3395 | if (may_steal) { |
a36de779 PS |
3396 | int i; |
3397 | ||
ac8c2081 | 3398 | for (i = 0; i < cnt; i++) { |
e14deea0 | 3399 | dp_packet_delete(packets[i]); |
ac8c2081 DDP |
3400 | } |
3401 | } | |
3402 | } | |
3403 | ||
a36de779 PS |
3404 | static int |
3405 | push_tnl_action(const struct dp_netdev *dp, | |
3406 | const struct nlattr *attr, | |
e14deea0 | 3407 | struct dp_packet **packets, int cnt) |
a36de779 PS |
3408 | { |
3409 | struct dp_netdev_port *tun_port; | |
3410 | const struct ovs_action_push_tnl *data; | |
3411 | ||
3412 | data = nl_attr_get(attr); | |
3413 | ||
3414 | tun_port = dp_netdev_lookup_port(dp, u32_to_odp(data->tnl_port)); | |
3415 | if (!tun_port) { | |
3416 | return -EINVAL; | |
3417 | } | |
3418 | netdev_push_header(tun_port->netdev, packets, cnt, data); | |
3419 | ||
3420 | return 0; | |
3421 | } | |
3422 | ||
3423 | static void | |
28e2fa02 DDP |
3424 | dp_netdev_clone_pkt_batch(struct dp_packet **dst_pkts, |
3425 | struct dp_packet **src_pkts, int cnt) | |
a36de779 PS |
3426 | { |
3427 | int i; | |
3428 | ||
3429 | for (i = 0; i < cnt; i++) { | |
28e2fa02 | 3430 | dst_pkts[i] = dp_packet_clone(src_pkts[i]); |
a36de779 PS |
3431 | } |
3432 | } | |
3433 | ||
9080a111 | 3434 | static void |
e14deea0 | 3435 | dp_execute_cb(void *aux_, struct dp_packet **packets, int cnt, |
09f9da0b | 3436 | const struct nlattr *a, bool may_steal) |
8a4e3a85 | 3437 | OVS_NO_THREAD_SAFETY_ANALYSIS |
9080a111 JR |
3438 | { |
3439 | struct dp_netdev_execute_aux *aux = aux_; | |
623540e4 | 3440 | uint32_t *depth = recirc_depth_get(); |
28e2fa02 DDP |
3441 | struct dp_netdev_pmd_thread *pmd = aux->pmd; |
3442 | struct dp_netdev *dp = pmd->dp; | |
09f9da0b | 3443 | int type = nl_attr_type(a); |
8a4e3a85 | 3444 | struct dp_netdev_port *p; |
8cbf4f47 | 3445 | int i; |
9080a111 | 3446 | |
09f9da0b JR |
3447 | switch ((enum ovs_action_attr)type) { |
3448 | case OVS_ACTION_ATTR_OUTPUT: | |
623540e4 | 3449 | p = dp_netdev_lookup_port(dp, u32_to_odp(nl_attr_get_u32(a))); |
26a5075b | 3450 | if (OVS_LIKELY(p)) { |
3bcc10c0 | 3451 | netdev_send(p->netdev, pmd->tx_qid, packets, cnt, may_steal); |
ac8c2081 | 3452 | return; |
8a4e3a85 | 3453 | } |
09f9da0b JR |
3454 | break; |
3455 | ||
a36de779 PS |
3456 | case OVS_ACTION_ATTR_TUNNEL_PUSH: |
3457 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3458 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3459 | int err; |
3460 | ||
3461 | if (!may_steal) { | |
3462 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3463 | packets = tnl_pkt; | |
3464 | } | |
3465 | ||
3466 | err = push_tnl_action(dp, a, packets, cnt); | |
3467 | if (!err) { | |
3468 | (*depth)++; | |
3469 | dp_netdev_input(pmd, packets, cnt); | |
3470 | (*depth)--; | |
3471 | } else { | |
3472 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3473 | } | |
3474 | return; | |
3475 | } | |
3476 | break; | |
3477 | ||
3478 | case OVS_ACTION_ATTR_TUNNEL_POP: | |
3479 | if (*depth < MAX_RECIRC_DEPTH) { | |
3480 | odp_port_t portno = u32_to_odp(nl_attr_get_u32(a)); | |
3481 | ||
3482 | p = dp_netdev_lookup_port(dp, portno); | |
3483 | if (p) { | |
cd159f1a | 3484 | struct dp_packet *tnl_pkt[NETDEV_MAX_BURST]; |
a36de779 PS |
3485 | int err; |
3486 | ||
3487 | if (!may_steal) { | |
3488 | dp_netdev_clone_pkt_batch(tnl_pkt, packets, cnt); | |
3489 | packets = tnl_pkt; | |
3490 | } | |
3491 | ||
3492 | err = netdev_pop_header(p->netdev, packets, cnt); | |
3493 | if (!err) { | |
3494 | ||
3495 | for (i = 0; i < cnt; i++) { | |
3496 | packets[i]->md.in_port.odp_port = portno; | |
3497 | } | |
3498 | ||
3499 | (*depth)++; | |
3500 | dp_netdev_input(pmd, packets, cnt); | |
3501 | (*depth)--; | |
3502 | } else { | |
3503 | dp_netdev_drop_packets(tnl_pkt, cnt, !may_steal); | |
3504 | } | |
3505 | return; | |
3506 | } | |
3507 | } | |
3508 | break; | |
3509 | ||
623540e4 EJ |
3510 | case OVS_ACTION_ATTR_USERSPACE: |
3511 | if (!fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { | |
3512 | const struct nlattr *userdata; | |
3513 | struct ofpbuf actions; | |
3514 | struct flow flow; | |
7af12bd7 | 3515 | ovs_u128 ufid; |
4fc65926 | 3516 | |
623540e4 EJ |
3517 | userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); |
3518 | ofpbuf_init(&actions, 0); | |
8cbf4f47 | 3519 | |
623540e4 EJ |
3520 | for (i = 0; i < cnt; i++) { |
3521 | int error; | |
3522 | ||
3523 | ofpbuf_clear(&actions); | |
3524 | ||
cf62fa4c | 3525 | flow_extract(packets[i], &flow); |
7af12bd7 | 3526 | dpif_flow_hash(dp->dpif, &flow, sizeof flow, &ufid); |
1c1e46ed | 3527 | error = dp_netdev_upcall(pmd, packets[i], &flow, NULL, &ufid, |
7af12bd7 | 3528 | DPIF_UC_ACTION, userdata,&actions, |
623540e4 EJ |
3529 | NULL); |
3530 | if (!error || error == ENOSPC) { | |
ac8c2081 | 3531 | dp_netdev_execute_actions(pmd, &packets[i], 1, may_steal, |
6fd6ed71 | 3532 | actions.data, actions.size); |
ac8c2081 | 3533 | } else if (may_steal) { |
e14deea0 | 3534 | dp_packet_delete(packets[i]); |
623540e4 | 3535 | } |
db73f716 | 3536 | } |
623540e4 EJ |
3537 | ofpbuf_uninit(&actions); |
3538 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
6b31e073 | 3539 | |
ac8c2081 DDP |
3540 | return; |
3541 | } | |
09f9da0b | 3542 | break; |
572f732a | 3543 | |
adcf00ba AZ |
3544 | case OVS_ACTION_ATTR_RECIRC: |
3545 | if (*depth < MAX_RECIRC_DEPTH) { | |
cd159f1a | 3546 | struct dp_packet *recirc_pkts[NETDEV_MAX_BURST]; |
572f732a | 3547 | |
28e2fa02 DDP |
3548 | if (!may_steal) { |
3549 | dp_netdev_clone_pkt_batch(recirc_pkts, packets, cnt); | |
3550 | packets = recirc_pkts; | |
3551 | } | |
8cbf4f47 | 3552 | |
28e2fa02 DDP |
3553 | for (i = 0; i < cnt; i++) { |
3554 | packets[i]->md.recirc_id = nl_attr_get_u32(a); | |
8cbf4f47 | 3555 | } |
28e2fa02 DDP |
3556 | |
3557 | (*depth)++; | |
3558 | dp_netdev_input(pmd, packets, cnt); | |
adcf00ba AZ |
3559 | (*depth)--; |
3560 | ||
ac8c2081 | 3561 | return; |
adcf00ba | 3562 | } |
ac8c2081 DDP |
3563 | |
3564 | VLOG_WARN("Packet dropped. Max recirculation depth exceeded."); | |
572f732a | 3565 | break; |
572f732a | 3566 | |
09f9da0b JR |
3567 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3568 | case OVS_ACTION_ATTR_POP_VLAN: | |
3569 | case OVS_ACTION_ATTR_PUSH_MPLS: | |
3570 | case OVS_ACTION_ATTR_POP_MPLS: | |
3571 | case OVS_ACTION_ATTR_SET: | |
6d670e7f | 3572 | case OVS_ACTION_ATTR_SET_MASKED: |
09f9da0b | 3573 | case OVS_ACTION_ATTR_SAMPLE: |
53e1d6f1 | 3574 | case OVS_ACTION_ATTR_HASH: |
09f9da0b JR |
3575 | case OVS_ACTION_ATTR_UNSPEC: |
3576 | case __OVS_ACTION_ATTR_MAX: | |
3577 | OVS_NOT_REACHED(); | |
da546e07 | 3578 | } |
ac8c2081 DDP |
3579 | |
3580 | dp_netdev_drop_packets(packets, cnt, may_steal); | |
98403001 BP |
3581 | } |
3582 | ||
4edb9ae9 | 3583 | static void |
65f13b50 | 3584 | dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
e14deea0 | 3585 | struct dp_packet **packets, int cnt, |
41ccaa24 | 3586 | bool may_steal, |
9080a111 | 3587 | const struct nlattr *actions, size_t actions_len) |
72865317 | 3588 | { |
41ccaa24 | 3589 | struct dp_netdev_execute_aux aux = { pmd }; |
9080a111 | 3590 | |
41ccaa24 | 3591 | odp_execute_actions(&aux, packets, cnt, may_steal, actions, |
8cbf4f47 | 3592 | actions_len, dp_execute_cb); |
72865317 BP |
3593 | } |
3594 | ||
3595 | const struct dpif_class dpif_netdev_class = { | |
72865317 | 3596 | "netdev", |
6553d06b | 3597 | dpif_netdev_init, |
2197d7ab | 3598 | dpif_netdev_enumerate, |
0aeaabc8 | 3599 | dpif_netdev_port_open_type, |
72865317 BP |
3600 | dpif_netdev_open, |
3601 | dpif_netdev_close, | |
7dab847a | 3602 | dpif_netdev_destroy, |
e4cfed38 PS |
3603 | dpif_netdev_run, |
3604 | dpif_netdev_wait, | |
72865317 | 3605 | dpif_netdev_get_stats, |
72865317 BP |
3606 | dpif_netdev_port_add, |
3607 | dpif_netdev_port_del, | |
3608 | dpif_netdev_port_query_by_number, | |
3609 | dpif_netdev_port_query_by_name, | |
98403001 | 3610 | NULL, /* port_get_pid */ |
b0ec0f27 BP |
3611 | dpif_netdev_port_dump_start, |
3612 | dpif_netdev_port_dump_next, | |
3613 | dpif_netdev_port_dump_done, | |
72865317 BP |
3614 | dpif_netdev_port_poll, |
3615 | dpif_netdev_port_poll_wait, | |
72865317 | 3616 | dpif_netdev_flow_flush, |
ac64794a BP |
3617 | dpif_netdev_flow_dump_create, |
3618 | dpif_netdev_flow_dump_destroy, | |
3619 | dpif_netdev_flow_dump_thread_create, | |
3620 | dpif_netdev_flow_dump_thread_destroy, | |
704a1e09 | 3621 | dpif_netdev_flow_dump_next, |
1a0c894a | 3622 | dpif_netdev_operate, |
6b31e073 RW |
3623 | NULL, /* recv_set */ |
3624 | NULL, /* handlers_set */ | |
f2eee189 | 3625 | dpif_netdev_pmd_set, |
5bf93d67 | 3626 | dpif_netdev_queue_to_priority, |
6b31e073 RW |
3627 | NULL, /* recv */ |
3628 | NULL, /* recv_wait */ | |
3629 | NULL, /* recv_purge */ | |
3630 | dpif_netdev_register_upcall_cb, | |
3631 | dpif_netdev_enable_upcall, | |
3632 | dpif_netdev_disable_upcall, | |
b5cbbcf6 | 3633 | dpif_netdev_get_datapath_version, |
72865317 | 3634 | }; |
614c4892 | 3635 | |
74cc3969 BP |
3636 | static void |
3637 | dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3638 | const char *argv[], void *aux OVS_UNUSED) | |
3639 | { | |
59e6d833 BP |
3640 | struct dp_netdev_port *old_port; |
3641 | struct dp_netdev_port *new_port; | |
74cc3969 | 3642 | struct dp_netdev *dp; |
ff073a71 | 3643 | odp_port_t port_no; |
74cc3969 | 3644 | |
8a4e3a85 | 3645 | ovs_mutex_lock(&dp_netdev_mutex); |
74cc3969 BP |
3646 | dp = shash_find_data(&dp_netdevs, argv[1]); |
3647 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
8a4e3a85 | 3648 | ovs_mutex_unlock(&dp_netdev_mutex); |
74cc3969 BP |
3649 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); |
3650 | return; | |
3651 | } | |
8a4e3a85 BP |
3652 | ovs_refcount_ref(&dp->ref_cnt); |
3653 | ovs_mutex_unlock(&dp_netdev_mutex); | |
74cc3969 | 3654 | |
59e6d833 BP |
3655 | ovs_mutex_lock(&dp->port_mutex); |
3656 | if (get_port_by_name(dp, argv[2], &old_port)) { | |
74cc3969 | 3657 | unixctl_command_reply_error(conn, "unknown port"); |
8a4e3a85 | 3658 | goto exit; |
74cc3969 BP |
3659 | } |
3660 | ||
ff073a71 BP |
3661 | port_no = u32_to_odp(atoi(argv[3])); |
3662 | if (!port_no || port_no == ODPP_NONE) { | |
74cc3969 | 3663 | unixctl_command_reply_error(conn, "bad port number"); |
8a4e3a85 | 3664 | goto exit; |
74cc3969 | 3665 | } |
ff073a71 | 3666 | if (dp_netdev_lookup_port(dp, port_no)) { |
74cc3969 | 3667 | unixctl_command_reply_error(conn, "port number already in use"); |
8a4e3a85 | 3668 | goto exit; |
74cc3969 | 3669 | } |
59e6d833 BP |
3670 | |
3671 | /* Remove old port. */ | |
35303d71 | 3672 | cmap_remove(&dp->ports, &old_port->node, hash_port_no(old_port->port_no)); |
59e6d833 BP |
3673 | ovsrcu_postpone(free, old_port); |
3674 | ||
3675 | /* Insert new port (cmap semantics mean we cannot re-insert 'old_port'). */ | |
3676 | new_port = xmemdup(old_port, sizeof *old_port); | |
35303d71 | 3677 | new_port->port_no = port_no; |
59e6d833 BP |
3678 | cmap_insert(&dp->ports, &new_port->node, hash_port_no(port_no)); |
3679 | ||
d33ed218 | 3680 | seq_change(dp->port_seq); |
74cc3969 | 3681 | unixctl_command_reply(conn, NULL); |
8a4e3a85 BP |
3682 | |
3683 | exit: | |
59e6d833 | 3684 | ovs_mutex_unlock(&dp->port_mutex); |
8a4e3a85 | 3685 | dp_netdev_unref(dp); |
74cc3969 BP |
3686 | } |
3687 | ||
c40b890f BP |
3688 | static void |
3689 | dpif_dummy_delete_port(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3690 | const char *argv[], void *aux OVS_UNUSED) | |
3691 | { | |
3692 | struct dp_netdev_port *port; | |
3693 | struct dp_netdev *dp; | |
3694 | ||
3695 | ovs_mutex_lock(&dp_netdev_mutex); | |
3696 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
3697 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
3698 | ovs_mutex_unlock(&dp_netdev_mutex); | |
3699 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); | |
3700 | return; | |
3701 | } | |
3702 | ovs_refcount_ref(&dp->ref_cnt); | |
3703 | ovs_mutex_unlock(&dp_netdev_mutex); | |
3704 | ||
3705 | ovs_mutex_lock(&dp->port_mutex); | |
3706 | if (get_port_by_name(dp, argv[2], &port)) { | |
3707 | unixctl_command_reply_error(conn, "unknown port"); | |
35303d71 | 3708 | } else if (port->port_no == ODPP_LOCAL) { |
c40b890f BP |
3709 | unixctl_command_reply_error(conn, "can't delete local port"); |
3710 | } else { | |
3711 | do_del_port(dp, port); | |
3712 | unixctl_command_reply(conn, NULL); | |
3713 | } | |
3714 | ovs_mutex_unlock(&dp->port_mutex); | |
3715 | ||
3716 | dp_netdev_unref(dp); | |
3717 | } | |
3718 | ||
0cbfe35d BP |
3719 | static void |
3720 | dpif_dummy_register__(const char *type) | |
3721 | { | |
3722 | struct dpif_class *class; | |
3723 | ||
3724 | class = xmalloc(sizeof *class); | |
3725 | *class = dpif_netdev_class; | |
3726 | class->type = xstrdup(type); | |
3727 | dp_register_provider(class); | |
3728 | } | |
3729 | ||
8420c7ad BP |
3730 | static void |
3731 | dpif_dummy_override(const char *type) | |
3732 | { | |
3733 | if (!dp_unregister_provider(type)) { | |
3734 | dpif_dummy_register__(type); | |
3735 | } | |
3736 | } | |
3737 | ||
614c4892 | 3738 | void |
8420c7ad | 3739 | dpif_dummy_register(enum dummy_level level) |
614c4892 | 3740 | { |
8420c7ad | 3741 | if (level == DUMMY_OVERRIDE_ALL) { |
0cbfe35d BP |
3742 | struct sset types; |
3743 | const char *type; | |
3744 | ||
3745 | sset_init(&types); | |
3746 | dp_enumerate_types(&types); | |
3747 | SSET_FOR_EACH (type, &types) { | |
8420c7ad | 3748 | dpif_dummy_override(type); |
0cbfe35d BP |
3749 | } |
3750 | sset_destroy(&types); | |
8420c7ad BP |
3751 | } else if (level == DUMMY_OVERRIDE_SYSTEM) { |
3752 | dpif_dummy_override("system"); | |
614c4892 | 3753 | } |
0cbfe35d BP |
3754 | |
3755 | dpif_dummy_register__("dummy"); | |
74cc3969 BP |
3756 | |
3757 | unixctl_command_register("dpif-dummy/change-port-number", | |
74467d5c | 3758 | "dp port new-number", |
74cc3969 | 3759 | 3, 3, dpif_dummy_change_port_number, NULL); |
74467d5c | 3760 | unixctl_command_register("dpif-dummy/delete-port", "dp port", |
c40b890f | 3761 | 2, 2, dpif_dummy_delete_port, NULL); |
614c4892 | 3762 | } |
0de8783a JR |
3763 | \f |
3764 | /* Datapath Classifier. */ | |
3765 | ||
3766 | /* A set of rules that all have the same fields wildcarded. */ | |
3767 | struct dpcls_subtable { | |
3768 | /* The fields are only used by writers. */ | |
3769 | struct cmap_node cmap_node OVS_GUARDED; /* Within dpcls 'subtables_map'. */ | |
3770 | ||
3771 | /* These fields are accessed by readers. */ | |
3772 | struct cmap rules; /* Contains "struct dpcls_rule"s. */ | |
3773 | struct netdev_flow_key mask; /* Wildcards for fields (const). */ | |
3774 | /* 'mask' must be the last field, additional space is allocated here. */ | |
3775 | }; | |
3776 | ||
3777 | /* Initializes 'cls' as a classifier that initially contains no classification | |
3778 | * rules. */ | |
3779 | static void | |
3780 | dpcls_init(struct dpcls *cls) | |
3781 | { | |
3782 | cmap_init(&cls->subtables_map); | |
3783 | pvector_init(&cls->subtables); | |
3784 | } | |
3785 | ||
3786 | static void | |
3787 | dpcls_destroy_subtable(struct dpcls *cls, struct dpcls_subtable *subtable) | |
3788 | { | |
3789 | pvector_remove(&cls->subtables, subtable); | |
3790 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, | |
3791 | subtable->mask.hash); | |
3792 | cmap_destroy(&subtable->rules); | |
3793 | ovsrcu_postpone(free, subtable); | |
3794 | } | |
3795 | ||
3796 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
3797 | * caller's responsibility. | |
3798 | * May only be called after all the readers have been terminated. */ | |
3799 | static void | |
3800 | dpcls_destroy(struct dpcls *cls) | |
3801 | { | |
3802 | if (cls) { | |
3803 | struct dpcls_subtable *subtable; | |
3804 | ||
3805 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
361d808d | 3806 | ovs_assert(cmap_count(&subtable->rules) == 0); |
0de8783a JR |
3807 | dpcls_destroy_subtable(cls, subtable); |
3808 | } | |
3809 | cmap_destroy(&cls->subtables_map); | |
3810 | pvector_destroy(&cls->subtables); | |
3811 | } | |
3812 | } | |
3813 | ||
3814 | static struct dpcls_subtable * | |
3815 | dpcls_create_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
3816 | { | |
3817 | struct dpcls_subtable *subtable; | |
3818 | ||
3819 | /* Need to add one. */ | |
caeb4906 JR |
3820 | subtable = xmalloc(sizeof *subtable |
3821 | - sizeof subtable->mask.mf + mask->len); | |
0de8783a JR |
3822 | cmap_init(&subtable->rules); |
3823 | netdev_flow_key_clone(&subtable->mask, mask); | |
3824 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, mask->hash); | |
3825 | pvector_insert(&cls->subtables, subtable, 0); | |
802f84ff | 3826 | pvector_publish(&cls->subtables); |
0de8783a JR |
3827 | |
3828 | return subtable; | |
3829 | } | |
3830 | ||
3831 | static inline struct dpcls_subtable * | |
3832 | dpcls_find_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
3833 | { | |
3834 | struct dpcls_subtable *subtable; | |
3835 | ||
3836 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, mask->hash, | |
3837 | &cls->subtables_map) { | |
3838 | if (netdev_flow_key_equal(&subtable->mask, mask)) { | |
3839 | return subtable; | |
3840 | } | |
3841 | } | |
3842 | return dpcls_create_subtable(cls, mask); | |
3843 | } | |
3844 | ||
3845 | /* Insert 'rule' into 'cls'. */ | |
3846 | static void | |
3847 | dpcls_insert(struct dpcls *cls, struct dpcls_rule *rule, | |
3848 | const struct netdev_flow_key *mask) | |
3849 | { | |
3850 | struct dpcls_subtable *subtable = dpcls_find_subtable(cls, mask); | |
3851 | ||
3852 | rule->mask = &subtable->mask; | |
3853 | cmap_insert(&subtable->rules, &rule->cmap_node, rule->flow.hash); | |
3854 | } | |
3855 | ||
3856 | /* Removes 'rule' from 'cls', also destructing the 'rule'. */ | |
3857 | static void | |
3858 | dpcls_remove(struct dpcls *cls, struct dpcls_rule *rule) | |
3859 | { | |
3860 | struct dpcls_subtable *subtable; | |
3861 | ||
3862 | ovs_assert(rule->mask); | |
3863 | ||
3864 | INIT_CONTAINER(subtable, rule->mask, mask); | |
3865 | ||
3866 | if (cmap_remove(&subtable->rules, &rule->cmap_node, rule->flow.hash) | |
3867 | == 0) { | |
3868 | dpcls_destroy_subtable(cls, subtable); | |
802f84ff | 3869 | pvector_publish(&cls->subtables); |
0de8783a JR |
3870 | } |
3871 | } | |
3872 | ||
361d808d JR |
3873 | /* Returns true if 'target' satisfies 'key' in 'mask', that is, if each 1-bit |
3874 | * in 'mask' the values in 'key' and 'target' are the same. */ | |
0de8783a JR |
3875 | static inline bool |
3876 | dpcls_rule_matches_key(const struct dpcls_rule *rule, | |
3877 | const struct netdev_flow_key *target) | |
3878 | { | |
09b0fa9c JR |
3879 | const uint64_t *keyp = miniflow_get_values(&rule->flow.mf); |
3880 | const uint64_t *maskp = miniflow_get_values(&rule->mask->mf); | |
d70e8c28 | 3881 | uint64_t target_u64; |
0de8783a | 3882 | |
361d808d JR |
3883 | NETDEV_FLOW_KEY_FOR_EACH_IN_TNL_MAP(target_u64, target, rule->flow.mf) { |
3884 | if (OVS_UNLIKELY((target_u64 & *maskp++) != *keyp++)) { | |
3885 | return false; | |
3886 | } | |
3887 | } | |
3888 | NETDEV_FLOW_KEY_FOR_EACH_IN_PKT_MAP(target_u64, target, rule->flow.mf) { | |
d70e8c28 | 3889 | if (OVS_UNLIKELY((target_u64 & *maskp++) != *keyp++)) { |
0de8783a JR |
3890 | return false; |
3891 | } | |
3892 | } | |
3893 | return true; | |
3894 | } | |
3895 | ||
3896 | /* For each miniflow in 'flows' performs a classifier lookup writing the result | |
3897 | * into the corresponding slot in 'rules'. If a particular entry in 'flows' is | |
3898 | * NULL it is skipped. | |
3899 | * | |
3900 | * This function is optimized for use in the userspace datapath and therefore | |
3901 | * does not implement a lot of features available in the standard | |
3902 | * classifier_lookup() function. Specifically, it does not implement | |
3903 | * priorities, instead returning any rule which matches the flow. | |
3904 | * | |
3905 | * Returns true if all flows found a corresponding rule. */ | |
3906 | static bool | |
3907 | dpcls_lookup(const struct dpcls *cls, const struct netdev_flow_key keys[], | |
3908 | struct dpcls_rule **rules, const size_t cnt) | |
3909 | { | |
3910 | /* The batch size 16 was experimentally found faster than 8 or 32. */ | |
3911 | typedef uint16_t map_type; | |
3912 | #define MAP_BITS (sizeof(map_type) * CHAR_BIT) | |
3913 | ||
3914 | #if !defined(__CHECKER__) && !defined(_WIN32) | |
3915 | const int N_MAPS = DIV_ROUND_UP(cnt, MAP_BITS); | |
3916 | #else | |
cd159f1a | 3917 | enum { N_MAPS = DIV_ROUND_UP(NETDEV_MAX_BURST, MAP_BITS) }; |
0de8783a JR |
3918 | #endif |
3919 | map_type maps[N_MAPS]; | |
3920 | struct dpcls_subtable *subtable; | |
3921 | ||
3922 | memset(maps, 0xff, sizeof maps); | |
3923 | if (cnt % MAP_BITS) { | |
3924 | maps[N_MAPS - 1] >>= MAP_BITS - cnt % MAP_BITS; /* Clear extra bits. */ | |
3925 | } | |
3926 | memset(rules, 0, cnt * sizeof *rules); | |
3927 | ||
3928 | PVECTOR_FOR_EACH (subtable, &cls->subtables) { | |
3929 | const struct netdev_flow_key *mkeys = keys; | |
3930 | struct dpcls_rule **mrules = rules; | |
3931 | map_type remains = 0; | |
3932 | int m; | |
3933 | ||
3934 | BUILD_ASSERT_DECL(sizeof remains == sizeof *maps); | |
3935 | ||
3936 | for (m = 0; m < N_MAPS; m++, mkeys += MAP_BITS, mrules += MAP_BITS) { | |
3937 | uint32_t hashes[MAP_BITS]; | |
3938 | const struct cmap_node *nodes[MAP_BITS]; | |
3939 | unsigned long map = maps[m]; | |
3940 | int i; | |
3941 | ||
3942 | if (!map) { | |
3943 | continue; /* Skip empty maps. */ | |
3944 | } | |
3945 | ||
3946 | /* Compute hashes for the remaining keys. */ | |
3ee6026a | 3947 | ULLONG_FOR_EACH_1(i, map) { |
0de8783a JR |
3948 | hashes[i] = netdev_flow_key_hash_in_mask(&mkeys[i], |
3949 | &subtable->mask); | |
3950 | } | |
3951 | /* Lookup. */ | |
3952 | map = cmap_find_batch(&subtable->rules, map, hashes, nodes); | |
3953 | /* Check results. */ | |
3ee6026a | 3954 | ULLONG_FOR_EACH_1(i, map) { |
0de8783a JR |
3955 | struct dpcls_rule *rule; |
3956 | ||
3957 | CMAP_NODE_FOR_EACH (rule, cmap_node, nodes[i]) { | |
3958 | if (OVS_LIKELY(dpcls_rule_matches_key(rule, &mkeys[i]))) { | |
3959 | mrules[i] = rule; | |
3960 | goto next; | |
3961 | } | |
3962 | } | |
3ee6026a | 3963 | ULLONG_SET0(map, i); /* Did not match. */ |
0de8783a JR |
3964 | next: |
3965 | ; /* Keep Sparse happy. */ | |
3966 | } | |
3967 | maps[m] &= ~map; /* Clear the found rules. */ | |
3968 | remains |= maps[m]; | |
3969 | } | |
3970 | if (!remains) { | |
3971 | return true; /* All found. */ | |
3972 | } | |
3973 | } | |
3974 | return false; /* Some misses. */ | |
3975 | } |