]>
Commit | Line | Data |
---|---|---|
72865317 | 1 | /* |
f582b6df | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2016, 2017 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> | |
7f3adc00 | 24 | #include <net/if.h> |
7daedce4 | 25 | #include <netinet/in.h> |
cdee00fd | 26 | #include <stdint.h> |
72865317 BP |
27 | #include <stdlib.h> |
28 | #include <string.h> | |
29 | #include <sys/ioctl.h> | |
7daedce4 | 30 | #include <sys/socket.h> |
72865317 | 31 | #include <sys/stat.h> |
72865317 BP |
32 | #include <unistd.h> |
33 | ||
01961bbd DDP |
34 | #ifdef DPDK_NETDEV |
35 | #include <rte_cycles.h> | |
36 | #endif | |
37 | ||
9f861c91 | 38 | #include "bitmap.h" |
59e6d833 | 39 | #include "cmap.h" |
5cf3edb3 | 40 | #include "conntrack.h" |
7daedce4 | 41 | #include "coverage.h" |
4d4e68ed | 42 | #include "ct-dpif.h" |
72865317 | 43 | #include "csum.h" |
e14deea0 | 44 | #include "dp-packet.h" |
614c4892 | 45 | #include "dpif.h" |
72865317 | 46 | #include "dpif-provider.h" |
614c4892 | 47 | #include "dummy.h" |
afae68b1 | 48 | #include "fat-rwlock.h" |
72865317 | 49 | #include "flow.h" |
762d146a | 50 | #include "hmapx.h" |
140dd699 | 51 | #include "id-pool.h" |
6c3eee82 | 52 | #include "latch.h" |
72865317 | 53 | #include "netdev.h" |
de281153 | 54 | #include "netdev-vport.h" |
cdee00fd | 55 | #include "netlink.h" |
f094af7b | 56 | #include "odp-execute.h" |
72865317 | 57 | #include "odp-util.h" |
25d436fb BW |
58 | #include "openvswitch/dynamic-string.h" |
59 | #include "openvswitch/list.h" | |
60 | #include "openvswitch/match.h" | |
61 | #include "openvswitch/ofp-print.h" | |
3eb67853 | 62 | #include "openvswitch/ofp-util.h" |
64c96779 | 63 | #include "openvswitch/ofpbuf.h" |
3eb67853 | 64 | #include "openvswitch/shash.h" |
25d436fb | 65 | #include "openvswitch/vlog.h" |
5a034064 | 66 | #include "ovs-numa.h" |
61e7deb1 | 67 | #include "ovs-rcu.h" |
72865317 BP |
68 | #include "packets.h" |
69 | #include "poll-loop.h" | |
0de8783a | 70 | #include "pvector.h" |
26c6b6cd | 71 | #include "random.h" |
d33ed218 | 72 | #include "seq.h" |
3eb67853 | 73 | #include "smap.h" |
0cbfe35d | 74 | #include "sset.h" |
72865317 | 75 | #include "timeval.h" |
53902038 | 76 | #include "tnl-neigh-cache.h" |
7f9b8504 | 77 | #include "tnl-ports.h" |
74cc3969 | 78 | #include "unixctl.h" |
72865317 | 79 | #include "util.h" |
7daedce4 | 80 | |
d98e6007 | 81 | VLOG_DEFINE_THIS_MODULE(dpif_netdev); |
72865317 | 82 | |
8bb113da | 83 | #define FLOW_DUMP_MAX_BATCH 50 |
adcf00ba AZ |
84 | /* Use per thread recirc_depth to prevent recirculation loop. */ |
85 | #define MAX_RECIRC_DEPTH 5 | |
86 | DEFINE_STATIC_PER_THREAD_DATA(uint32_t, recirc_depth, 0) | |
e4cfed38 | 87 | |
72865317 | 88 | /* Configuration parameters. */ |
72865317 | 89 | enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ |
4b27db64 JR |
90 | enum { MAX_METERS = 65536 }; /* Maximum number of meters. */ |
91 | enum { MAX_BANDS = 8 }; /* Maximum number of bands / meter. */ | |
92 | enum { N_METER_LOCKS = 64 }; /* Maximum number of meters. */ | |
72865317 | 93 | |
8a4e3a85 BP |
94 | /* Protects against changes to 'dp_netdevs'. */ |
95 | static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER; | |
96 | ||
97 | /* Contains all 'struct dp_netdev's. */ | |
98 | static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex) | |
99 | = SHASH_INITIALIZER(&dp_netdevs); | |
100 | ||
623540e4 | 101 | static struct vlog_rate_limit upcall_rl = VLOG_RATE_LIMIT_INIT(600, 600); |
6b31e073 | 102 | |
5cf3edb3 | 103 | #define DP_NETDEV_CS_SUPPORTED_MASK (CS_NEW | CS_ESTABLISHED | CS_RELATED \ |
4cddb1f0 DB |
104 | | CS_INVALID | CS_REPLY_DIR | CS_TRACKED \ |
105 | | CS_SRC_NAT | CS_DST_NAT) | |
5cf3edb3 DDP |
106 | #define DP_NETDEV_CS_UNSUPPORTED_MASK (~(uint32_t)DP_NETDEV_CS_SUPPORTED_MASK) |
107 | ||
2494ccd7 | 108 | static struct odp_support dp_netdev_support = { |
f0fb825a | 109 | .max_vlan_headers = SIZE_MAX, |
2494ccd7 JS |
110 | .max_mpls_depth = SIZE_MAX, |
111 | .recirc = true, | |
5cf3edb3 DDP |
112 | .ct_state = true, |
113 | .ct_zone = true, | |
114 | .ct_mark = true, | |
115 | .ct_label = true, | |
2575df07 JP |
116 | .ct_state_nat = true, |
117 | .ct_orig_tuple = true, | |
118 | .ct_orig_tuple6 = true, | |
2494ccd7 JS |
119 | }; |
120 | ||
79df317f | 121 | /* Stores a miniflow with inline values */ |
9bbf1c3d | 122 | |
9bbf1c3d | 123 | struct netdev_flow_key { |
caeb4906 JR |
124 | uint32_t hash; /* Hash function differs for different users. */ |
125 | uint32_t len; /* Length of the following miniflow (incl. map). */ | |
0de8783a | 126 | struct miniflow mf; |
8fd47924 | 127 | uint64_t buf[FLOW_MAX_PACKET_U64S]; |
9bbf1c3d DDP |
128 | }; |
129 | ||
130 | /* Exact match cache for frequently used flows | |
131 | * | |
132 | * The cache uses a 32-bit hash of the packet (which can be the RSS hash) to | |
133 | * search its entries for a miniflow that matches exactly the miniflow of the | |
0de8783a | 134 | * packet. It stores the 'dpcls_rule' (rule) that matches the miniflow. |
9bbf1c3d DDP |
135 | * |
136 | * A cache entry holds a reference to its 'dp_netdev_flow'. | |
137 | * | |
138 | * A miniflow with a given hash can be in one of EM_FLOW_HASH_SEGS different | |
139 | * entries. The 32-bit hash is split into EM_FLOW_HASH_SEGS values (each of | |
140 | * them is EM_FLOW_HASH_SHIFT bits wide and the remainder is thrown away). Each | |
141 | * value is the index of a cache entry where the miniflow could be. | |
142 | * | |
143 | * | |
144 | * Thread-safety | |
145 | * ============= | |
146 | * | |
147 | * Each pmd_thread has its own private exact match cache. | |
148 | * If dp_netdev_input is not called from a pmd thread, a mutex is used. | |
149 | */ | |
150 | ||
fc82e877 | 151 | #define EM_FLOW_HASH_SHIFT 13 |
9bbf1c3d DDP |
152 | #define EM_FLOW_HASH_ENTRIES (1u << EM_FLOW_HASH_SHIFT) |
153 | #define EM_FLOW_HASH_MASK (EM_FLOW_HASH_ENTRIES - 1) | |
154 | #define EM_FLOW_HASH_SEGS 2 | |
155 | ||
4c30b246 CL |
156 | /* Default EMC insert probability is 1 / DEFAULT_EM_FLOW_INSERT_INV_PROB */ |
157 | #define DEFAULT_EM_FLOW_INSERT_INV_PROB 100 | |
158 | #define DEFAULT_EM_FLOW_INSERT_MIN (UINT32_MAX / \ | |
159 | DEFAULT_EM_FLOW_INSERT_INV_PROB) | |
160 | ||
9bbf1c3d | 161 | struct emc_entry { |
9bbf1c3d | 162 | struct dp_netdev_flow *flow; |
0de8783a | 163 | struct netdev_flow_key key; /* key.hash used for emc hash value. */ |
9bbf1c3d DDP |
164 | }; |
165 | ||
166 | struct emc_cache { | |
167 | struct emc_entry entries[EM_FLOW_HASH_ENTRIES]; | |
67ad54cb | 168 | int sweep_idx; /* For emc_cache_slow_sweep(). */ |
9bbf1c3d DDP |
169 | }; |
170 | ||
171 | /* Iterate in the exact match cache through every entry that might contain a | |
172 | * miniflow with hash 'HASH'. */ | |
173 | #define EMC_FOR_EACH_POS_WITH_HASH(EMC, CURRENT_ENTRY, HASH) \ | |
174 | for (uint32_t i__ = 0, srch_hash__ = (HASH); \ | |
175 | (CURRENT_ENTRY) = &(EMC)->entries[srch_hash__ & EM_FLOW_HASH_MASK], \ | |
176 | i__ < EM_FLOW_HASH_SEGS; \ | |
177 | i__++, srch_hash__ >>= EM_FLOW_HASH_SHIFT) | |
0de8783a JR |
178 | \f |
179 | /* Simple non-wildcarding single-priority classifier. */ | |
180 | ||
3453b4d6 JS |
181 | /* Time in ms between successive optimizations of the dpcls subtable vector */ |
182 | #define DPCLS_OPTIMIZATION_INTERVAL 1000 | |
183 | ||
0de8783a | 184 | struct dpcls { |
3453b4d6 JS |
185 | struct cmap_node node; /* Within dp_netdev_pmd_thread.classifiers */ |
186 | odp_port_t in_port; | |
0de8783a | 187 | struct cmap subtables_map; |
da9cfca6 | 188 | struct pvector subtables; |
0de8783a | 189 | }; |
9bbf1c3d | 190 | |
0de8783a JR |
191 | /* A rule to be inserted to the classifier. */ |
192 | struct dpcls_rule { | |
193 | struct cmap_node cmap_node; /* Within struct dpcls_subtable 'rules'. */ | |
194 | struct netdev_flow_key *mask; /* Subtable's mask. */ | |
195 | struct netdev_flow_key flow; /* Matching key. */ | |
196 | /* 'flow' must be the last field, additional space is allocated here. */ | |
197 | }; | |
198 | ||
199 | static void dpcls_init(struct dpcls *); | |
200 | static void dpcls_destroy(struct dpcls *); | |
3453b4d6 | 201 | static void dpcls_sort_subtable_vector(struct dpcls *); |
0de8783a JR |
202 | static void dpcls_insert(struct dpcls *, struct dpcls_rule *, |
203 | const struct netdev_flow_key *mask); | |
204 | static void dpcls_remove(struct dpcls *, struct dpcls_rule *); | |
3453b4d6 | 205 | static bool dpcls_lookup(struct dpcls *cls, |
0de8783a | 206 | const struct netdev_flow_key keys[], |
3453b4d6 JS |
207 | struct dpcls_rule **rules, size_t cnt, |
208 | int *num_lookups_p); | |
0de8783a | 209 | \f |
4b27db64 JR |
210 | /* Set of supported meter flags */ |
211 | #define DP_SUPPORTED_METER_FLAGS_MASK \ | |
212 | (OFPMF13_STATS | OFPMF13_PKTPS | OFPMF13_KBPS | OFPMF13_BURST) | |
213 | ||
214 | /* Set of supported meter band types */ | |
215 | #define DP_SUPPORTED_METER_BAND_TYPES \ | |
216 | ( 1 << OFPMBT13_DROP ) | |
217 | ||
218 | struct dp_meter_band { | |
219 | struct ofputil_meter_band up; /* type, prec_level, pad, rate, burst_size */ | |
220 | uint32_t bucket; /* In 1/1000 packets (for PKTPS), or in bits (for KBPS) */ | |
221 | uint64_t packet_count; | |
222 | uint64_t byte_count; | |
223 | }; | |
224 | ||
225 | struct dp_meter { | |
226 | uint16_t flags; | |
227 | uint16_t n_bands; | |
228 | uint32_t max_delta_t; | |
229 | uint64_t used; | |
230 | uint64_t packet_count; | |
231 | uint64_t byte_count; | |
232 | struct dp_meter_band bands[]; | |
233 | }; | |
234 | ||
8a4e3a85 BP |
235 | /* Datapath based on the network device interface from netdev.h. |
236 | * | |
237 | * | |
238 | * Thread-safety | |
239 | * ============= | |
240 | * | |
241 | * Some members, marked 'const', are immutable. Accessing other members | |
242 | * requires synchronization, as noted in more detail below. | |
243 | * | |
244 | * Acquisition order is, from outermost to innermost: | |
245 | * | |
246 | * dp_netdev_mutex (global) | |
59e6d833 | 247 | * port_mutex |
d0cca6c3 | 248 | * non_pmd_mutex |
8a4e3a85 | 249 | */ |
72865317 | 250 | struct dp_netdev { |
8a4e3a85 BP |
251 | const struct dpif_class *const class; |
252 | const char *const name; | |
6b31e073 | 253 | struct dpif *dpif; |
6a8267c5 BP |
254 | struct ovs_refcount ref_cnt; |
255 | atomic_flag destroyed; | |
72865317 | 256 | |
8a4e3a85 BP |
257 | /* Ports. |
258 | * | |
e9985d6a DDP |
259 | * Any lookup into 'ports' or any access to the dp_netdev_ports found |
260 | * through 'ports' requires taking 'port_mutex'. */ | |
59e6d833 | 261 | struct ovs_mutex port_mutex; |
e9985d6a | 262 | struct hmap ports; |
d33ed218 | 263 | struct seq *port_seq; /* Incremented whenever a port changes. */ |
6c3eee82 | 264 | |
4b27db64 JR |
265 | /* Meters. */ |
266 | struct ovs_mutex meter_locks[N_METER_LOCKS]; | |
267 | struct dp_meter *meters[MAX_METERS]; /* Meter bands. */ | |
4b27db64 | 268 | |
65dcf3da BB |
269 | /* Probability of EMC insertions is a factor of 'emc_insert_min'.*/ |
270 | OVS_ALIGNED_VAR(CACHE_LINE_SIZE) atomic_uint32_t emc_insert_min; | |
271 | ||
6b31e073 RW |
272 | /* Protects access to ofproto-dpif-upcall interface during revalidator |
273 | * thread synchronization. */ | |
274 | struct fat_rwlock upcall_rwlock; | |
623540e4 EJ |
275 | upcall_callback *upcall_cb; /* Callback function for executing upcalls. */ |
276 | void *upcall_aux; | |
6b31e073 | 277 | |
e4e74c3a AW |
278 | /* Callback function for notifying the purging of dp flows (during |
279 | * reseting pmd deletion). */ | |
280 | dp_purge_callback *dp_purge_cb; | |
281 | void *dp_purge_aux; | |
282 | ||
65f13b50 AW |
283 | /* Stores all 'struct dp_netdev_pmd_thread's. */ |
284 | struct cmap poll_threads; | |
140dd699 IM |
285 | /* id pool for per thread static_tx_qid. */ |
286 | struct id_pool *tx_qid_pool; | |
287 | struct ovs_mutex tx_qid_pool_mutex; | |
65f13b50 AW |
288 | |
289 | /* Protects the access of the 'struct dp_netdev_pmd_thread' | |
290 | * instance for non-pmd thread. */ | |
291 | struct ovs_mutex non_pmd_mutex; | |
292 | ||
293 | /* Each pmd thread will store its pointer to | |
294 | * 'struct dp_netdev_pmd_thread' in 'per_pmd_key'. */ | |
295 | ovsthread_key_t per_pmd_key; | |
f2eee189 | 296 | |
a6a426d6 IM |
297 | struct seq *reconfigure_seq; |
298 | uint64_t last_reconfigure_seq; | |
299 | ||
a14b8947 | 300 | /* Cpu mask for pin of pmd threads. */ |
f2eee189 | 301 | char *pmd_cmask; |
6e3c6fa4 | 302 | |
a36de779 | 303 | uint64_t last_tnl_conf_seq; |
5cf3edb3 DDP |
304 | |
305 | struct conntrack conntrack; | |
72865317 BP |
306 | }; |
307 | ||
4b27db64 JR |
308 | static void meter_lock(const struct dp_netdev *dp, uint32_t meter_id) |
309 | OVS_ACQUIRES(dp->meter_locks[meter_id % N_METER_LOCKS]) | |
310 | { | |
311 | ovs_mutex_lock(&dp->meter_locks[meter_id % N_METER_LOCKS]); | |
312 | } | |
313 | ||
314 | static void meter_unlock(const struct dp_netdev *dp, uint32_t meter_id) | |
315 | OVS_RELEASES(dp->meter_locks[meter_id % N_METER_LOCKS]) | |
316 | { | |
317 | ovs_mutex_unlock(&dp->meter_locks[meter_id % N_METER_LOCKS]); | |
318 | } | |
319 | ||
320 | ||
8a4e3a85 | 321 | static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp, |
e9985d6a DDP |
322 | odp_port_t) |
323 | OVS_REQUIRES(dp->port_mutex); | |
ff073a71 | 324 | |
51852a57 | 325 | enum dp_stat_type { |
abcf3ef4 DDP |
326 | DP_STAT_EXACT_HIT, /* Packets that had an exact match (emc). */ |
327 | DP_STAT_MASKED_HIT, /* Packets that matched in the flow table. */ | |
51852a57 BP |
328 | DP_STAT_MISS, /* Packets that did not match. */ |
329 | DP_STAT_LOST, /* Packets not passed up to the client. */ | |
3453b4d6 JS |
330 | DP_STAT_LOOKUP_HIT, /* Number of subtable lookups for flow table |
331 | hits */ | |
51852a57 BP |
332 | DP_N_STATS |
333 | }; | |
334 | ||
55e3ca97 | 335 | enum pmd_cycles_counter_type { |
a2ac666d CL |
336 | PMD_CYCLES_IDLE, /* Cycles spent idle or unsuccessful polling */ |
337 | PMD_CYCLES_PROCESSING, /* Cycles spent successfully polling and | |
338 | * processing polled packets */ | |
55e3ca97 DDP |
339 | PMD_N_CYCLES |
340 | }; | |
341 | ||
324c8374 IM |
342 | #define XPS_TIMEOUT_MS 500LL |
343 | ||
3eb67853 IM |
344 | /* Contained by struct dp_netdev_port's 'rxqs' member. */ |
345 | struct dp_netdev_rxq { | |
947dc567 DDP |
346 | struct dp_netdev_port *port; |
347 | struct netdev_rxq *rx; | |
348 | unsigned core_id; /* Core to which this queue should be | |
349 | pinned. OVS_CORE_UNSPEC if the | |
350 | queue doesn't need to be pinned to a | |
351 | particular core. */ | |
47a45d86 | 352 | struct dp_netdev_pmd_thread *pmd; /* pmd thread that polls this queue. */ |
3eb67853 IM |
353 | }; |
354 | ||
72865317 BP |
355 | /* A port in a netdev-based datapath. */ |
356 | struct dp_netdev_port { | |
35303d71 | 357 | odp_port_t port_no; |
72865317 | 358 | struct netdev *netdev; |
e9985d6a | 359 | struct hmap_node node; /* Node in dp_netdev's 'ports'. */ |
4b609110 | 360 | struct netdev_saved_flags *sf; |
3eb67853 | 361 | struct dp_netdev_rxq *rxqs; |
85a4f238 | 362 | unsigned n_rxq; /* Number of elements in 'rxqs' */ |
324c8374 | 363 | bool dynamic_txqs; /* If true XPS will be used. */ |
47a45d86 | 364 | unsigned *txq_used; /* Number of threads that use each tx queue. */ |
324c8374 | 365 | struct ovs_mutex txq_used_mutex; |
0cbfe35d | 366 | char *type; /* Port type as requested by user. */ |
3eb67853 | 367 | char *rxq_affinity_list; /* Requested affinity of rx queues. */ |
e32971b8 | 368 | bool need_reconfigure; /* True if we should reconfigure netdev. */ |
72865317 BP |
369 | }; |
370 | ||
1c1e46ed AW |
371 | /* Contained by struct dp_netdev_flow's 'stats' member. */ |
372 | struct dp_netdev_flow_stats { | |
eb94da30 DDP |
373 | atomic_llong used; /* Last used time, in monotonic msecs. */ |
374 | atomic_ullong packet_count; /* Number of packets matched. */ | |
375 | atomic_ullong byte_count; /* Number of bytes matched. */ | |
376 | atomic_uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */ | |
1c1e46ed AW |
377 | }; |
378 | ||
379 | /* A flow in 'dp_netdev_pmd_thread's 'flow_table'. | |
8a4e3a85 BP |
380 | * |
381 | * | |
382 | * Thread-safety | |
383 | * ============= | |
384 | * | |
385 | * Except near the beginning or ending of its lifespan, rule 'rule' belongs to | |
1c1e46ed | 386 | * its pmd thread's classifier. The text below calls this classifier 'cls'. |
8a4e3a85 BP |
387 | * |
388 | * Motivation | |
389 | * ---------- | |
390 | * | |
391 | * The thread safety rules described here for "struct dp_netdev_flow" are | |
392 | * motivated by two goals: | |
393 | * | |
394 | * - Prevent threads that read members of "struct dp_netdev_flow" from | |
395 | * reading bad data due to changes by some thread concurrently modifying | |
396 | * those members. | |
397 | * | |
398 | * - Prevent two threads making changes to members of a given "struct | |
399 | * dp_netdev_flow" from interfering with each other. | |
400 | * | |
401 | * | |
402 | * Rules | |
403 | * ----- | |
404 | * | |
ed79f89a DDP |
405 | * A flow 'flow' may be accessed without a risk of being freed during an RCU |
406 | * grace period. Code that needs to hold onto a flow for a while | |
407 | * should try incrementing 'flow->ref_cnt' with dp_netdev_flow_ref(). | |
8a4e3a85 BP |
408 | * |
409 | * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the | |
ed79f89a DDP |
410 | * flow from being deleted from 'cls' and it doesn't protect members of 'flow' |
411 | * from modification. | |
8a4e3a85 BP |
412 | * |
413 | * Some members, marked 'const', are immutable. Accessing other members | |
414 | * requires synchronization, as noted in more detail below. | |
415 | */ | |
72865317 | 416 | struct dp_netdev_flow { |
11e5cf1f | 417 | const struct flow flow; /* Unmasked flow that created this entry. */ |
8a4e3a85 | 418 | /* Hash table index by unmasked flow. */ |
1c1e46ed AW |
419 | const struct cmap_node node; /* In owning dp_netdev_pmd_thread's */ |
420 | /* 'flow_table'. */ | |
70e5ed6f | 421 | const ovs_u128 ufid; /* Unique flow identifier. */ |
bd5131ba | 422 | const unsigned pmd_id; /* The 'core_id' of pmd thread owning this */ |
1c1e46ed | 423 | /* flow. */ |
72865317 | 424 | |
ed79f89a DDP |
425 | /* Number of references. |
426 | * The classifier owns one reference. | |
427 | * Any thread trying to keep a rule from being freed should hold its own | |
428 | * reference. */ | |
429 | struct ovs_refcount ref_cnt; | |
430 | ||
11e5cf1f DDP |
431 | bool dead; |
432 | ||
1c1e46ed AW |
433 | /* Statistics. */ |
434 | struct dp_netdev_flow_stats stats; | |
8a4e3a85 | 435 | |
45c626a3 | 436 | /* Actions. */ |
61e7deb1 | 437 | OVSRCU_TYPE(struct dp_netdev_actions *) actions; |
0de8783a | 438 | |
11e5cf1f DDP |
439 | /* While processing a group of input packets, the datapath uses the next |
440 | * member to store a pointer to the output batch for the flow. It is | |
441 | * reset after the batch has been sent out (See dp_netdev_queue_batches(), | |
f7ce4811 PS |
442 | * packet_batch_per_flow_init() and packet_batch_per_flow_execute()). */ |
443 | struct packet_batch_per_flow *batch; | |
11e5cf1f | 444 | |
0de8783a JR |
445 | /* Packet classification. */ |
446 | struct dpcls_rule cr; /* In owning dp_netdev's 'cls'. */ | |
447 | /* 'cr' must be the last member. */ | |
72865317 BP |
448 | }; |
449 | ||
ed79f89a | 450 | static void dp_netdev_flow_unref(struct dp_netdev_flow *); |
9bbf1c3d | 451 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *); |
70e5ed6f | 452 | static int dpif_netdev_flow_from_nlattrs(const struct nlattr *, uint32_t, |
f0fb825a | 453 | struct flow *, bool); |
8a4e3a85 | 454 | |
a84cb64a BP |
455 | /* A set of datapath actions within a "struct dp_netdev_flow". |
456 | * | |
457 | * | |
458 | * Thread-safety | |
459 | * ============= | |
460 | * | |
45c626a3 | 461 | * A struct dp_netdev_actions 'actions' is protected with RCU. */ |
a84cb64a | 462 | struct dp_netdev_actions { |
a84cb64a BP |
463 | /* These members are immutable: they do not change during the struct's |
464 | * lifetime. */ | |
a84cb64a | 465 | unsigned int size; /* Size of 'actions', in bytes. */ |
9ff55ae2 | 466 | struct nlattr actions[]; /* Sequence of OVS_ACTION_ATTR_* attributes. */ |
a84cb64a BP |
467 | }; |
468 | ||
469 | struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *, | |
470 | size_t); | |
61e7deb1 BP |
471 | struct dp_netdev_actions *dp_netdev_flow_get_actions( |
472 | const struct dp_netdev_flow *); | |
473 | static void dp_netdev_actions_free(struct dp_netdev_actions *); | |
a84cb64a | 474 | |
1c1e46ed AW |
475 | /* Contained by struct dp_netdev_pmd_thread's 'stats' member. */ |
476 | struct dp_netdev_pmd_stats { | |
477 | /* Indexed by DP_STAT_*. */ | |
eb94da30 | 478 | atomic_ullong n[DP_N_STATS]; |
1c1e46ed AW |
479 | }; |
480 | ||
55e3ca97 DDP |
481 | /* Contained by struct dp_netdev_pmd_thread's 'cycle' member. */ |
482 | struct dp_netdev_pmd_cycles { | |
483 | /* Indexed by PMD_CYCLES_*. */ | |
484 | atomic_ullong n[PMD_N_CYCLES]; | |
485 | }; | |
486 | ||
947dc567 DDP |
487 | struct polled_queue { |
488 | struct netdev_rxq *rx; | |
489 | odp_port_t port_no; | |
490 | }; | |
491 | ||
ae7ad0a1 IM |
492 | /* Contained by struct dp_netdev_pmd_thread's 'poll_list' member. */ |
493 | struct rxq_poll { | |
947dc567 DDP |
494 | struct dp_netdev_rxq *rxq; |
495 | struct hmap_node node; | |
ae7ad0a1 IM |
496 | }; |
497 | ||
57eebbb4 DDP |
498 | /* Contained by struct dp_netdev_pmd_thread's 'send_port_cache', |
499 | * 'tnl_port_cache' or 'tx_ports'. */ | |
d0cca6c3 | 500 | struct tx_port { |
324c8374 IM |
501 | struct dp_netdev_port *port; |
502 | int qid; | |
503 | long long last_used; | |
d0cca6c3 DDP |
504 | struct hmap_node node; |
505 | }; | |
506 | ||
e4cfed38 PS |
507 | /* PMD: Poll modes drivers. PMD accesses devices via polling to eliminate |
508 | * the performance overhead of interrupt processing. Therefore netdev can | |
509 | * not implement rx-wait for these devices. dpif-netdev needs to poll | |
510 | * these device to check for recv buffer. pmd-thread does polling for | |
1c1e46ed | 511 | * devices assigned to itself. |
e4cfed38 PS |
512 | * |
513 | * DPDK used PMD for accessing NIC. | |
514 | * | |
65f13b50 AW |
515 | * Note, instance with cpu core id NON_PMD_CORE_ID will be reserved for |
516 | * I/O of all non-pmd threads. There will be no actual thread created | |
517 | * for the instance. | |
1c1e46ed | 518 | * |
1859876c BB |
519 | * Each struct has its own flow cache and classifier per managed ingress port. |
520 | * For packets received on ingress port, a look up is done on corresponding PMD | |
521 | * thread's flow cache and in case of a miss, lookup is performed in the | |
522 | * corresponding classifier of port. Packets are executed with the found | |
523 | * actions in either case. | |
1c1e46ed | 524 | * */ |
65f13b50 | 525 | struct dp_netdev_pmd_thread { |
6c3eee82 | 526 | struct dp_netdev *dp; |
1c1e46ed | 527 | struct ovs_refcount ref_cnt; /* Every reference must be refcount'ed. */ |
65f13b50 | 528 | struct cmap_node node; /* In 'dp->poll_threads'. */ |
accf8626 AW |
529 | |
530 | pthread_cond_t cond; /* For synchronizing pmd thread reload. */ | |
531 | struct ovs_mutex cond_mutex; /* Mutex for condition variable. */ | |
532 | ||
65f13b50 AW |
533 | /* Per thread exact-match cache. Note, the instance for cpu core |
534 | * NON_PMD_CORE_ID can be accessed by multiple threads, and thusly | |
d0cca6c3 DDP |
535 | * need to be protected by 'non_pmd_mutex'. Every other instance |
536 | * will only be accessed by its own pmd thread. */ | |
9bbf1c3d | 537 | struct emc_cache flow_cache; |
1c1e46ed | 538 | |
3453b4d6 | 539 | /* Flow-Table and classifiers |
1c1e46ed AW |
540 | * |
541 | * Writers of 'flow_table' must take the 'flow_mutex'. Corresponding | |
3453b4d6 JS |
542 | * changes to 'classifiers' must be made while still holding the |
543 | * 'flow_mutex'. | |
1c1e46ed AW |
544 | */ |
545 | struct ovs_mutex flow_mutex; | |
1c1e46ed AW |
546 | struct cmap flow_table OVS_GUARDED; /* Flow table. */ |
547 | ||
3453b4d6 JS |
548 | /* One classifier per in_port polled by the pmd */ |
549 | struct cmap classifiers; | |
550 | /* Periodically sort subtable vectors according to hit frequencies */ | |
551 | long long int next_optimization; | |
552 | ||
1c1e46ed AW |
553 | /* Statistics. */ |
554 | struct dp_netdev_pmd_stats stats; | |
555 | ||
55e3ca97 DDP |
556 | /* Cycles counters */ |
557 | struct dp_netdev_pmd_cycles cycles; | |
558 | ||
559 | /* Used to count cicles. See 'cycles_counter_end()' */ | |
560 | unsigned long long last_cycles; | |
561 | ||
65f13b50 | 562 | struct latch exit_latch; /* For terminating the pmd thread. */ |
2788a1b1 DDP |
563 | struct seq *reload_seq; |
564 | uint64_t last_reload_seq; | |
14e3e12a | 565 | atomic_bool reload; /* Do we need to reload ports? */ |
6c3eee82 | 566 | pthread_t thread; |
bd5131ba | 567 | unsigned core_id; /* CPU core id of this pmd thread. */ |
65f13b50 | 568 | int numa_id; /* numa node id of this pmd thread. */ |
3eb67853 | 569 | bool isolated; |
81acebda | 570 | |
324c8374 IM |
571 | /* Queue id used by this pmd thread to send packets on all netdevs if |
572 | * XPS disabled for this netdev. All static_tx_qid's are unique and less | |
e32971b8 | 573 | * than 'cmap_count(dp->poll_threads)'. */ |
140dd699 | 574 | uint32_t static_tx_qid; |
6553d06b | 575 | |
d0cca6c3 | 576 | struct ovs_mutex port_mutex; /* Mutex for 'poll_list' and 'tx_ports'. */ |
ae7ad0a1 | 577 | /* List of rx queues to poll. */ |
947dc567 | 578 | struct hmap poll_list OVS_GUARDED; |
d0cca6c3 DDP |
579 | /* Map of 'tx_port's used for transmission. Written by the main thread, |
580 | * read by the pmd thread. */ | |
581 | struct hmap tx_ports OVS_GUARDED; | |
582 | ||
57eebbb4 DDP |
583 | /* These are thread-local copies of 'tx_ports'. One contains only tunnel |
584 | * ports (that support push_tunnel/pop_tunnel), the other contains ports | |
585 | * with at least one txq (that support send). A port can be in both. | |
586 | * | |
587 | * There are two separate maps to make sure that we don't try to execute | |
588 | * OUTPUT on a device which has 0 txqs or PUSH/POP on a non-tunnel device. | |
589 | * | |
590 | * The instances for cpu core NON_PMD_CORE_ID can be accessed by multiple | |
591 | * threads, and thusly need to be protected by 'non_pmd_mutex'. Every | |
592 | * other instance will only be accessed by its own pmd thread. */ | |
593 | struct hmap tnl_port_cache; | |
594 | struct hmap send_port_cache; | |
ae7ad0a1 | 595 | |
6553d06b DDP |
596 | /* Only a pmd thread can write on its own 'cycles' and 'stats'. |
597 | * The main thread keeps 'stats_zero' and 'cycles_zero' as base | |
598 | * values and subtracts them from 'stats' and 'cycles' before | |
599 | * reporting to the user */ | |
600 | unsigned long long stats_zero[DP_N_STATS]; | |
601 | uint64_t cycles_zero[PMD_N_CYCLES]; | |
e32971b8 DDP |
602 | |
603 | /* Set to true if the pmd thread needs to be reloaded. */ | |
604 | bool need_reload; | |
6c3eee82 BP |
605 | }; |
606 | ||
72865317 BP |
607 | /* Interface to netdev-based datapath. */ |
608 | struct dpif_netdev { | |
609 | struct dpif dpif; | |
610 | struct dp_netdev *dp; | |
d33ed218 | 611 | uint64_t last_port_seq; |
72865317 BP |
612 | }; |
613 | ||
8a4e3a85 | 614 | static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no, |
e9985d6a DDP |
615 | struct dp_netdev_port **portp) |
616 | OVS_REQUIRES(dp->port_mutex); | |
8a4e3a85 | 617 | static int get_port_by_name(struct dp_netdev *dp, const char *devname, |
e9985d6a DDP |
618 | struct dp_netdev_port **portp) |
619 | OVS_REQUIRES(dp->port_mutex); | |
8a4e3a85 BP |
620 | static void dp_netdev_free(struct dp_netdev *) |
621 | OVS_REQUIRES(dp_netdev_mutex); | |
8a4e3a85 BP |
622 | static int do_add_port(struct dp_netdev *dp, const char *devname, |
623 | const char *type, odp_port_t port_no) | |
59e6d833 | 624 | OVS_REQUIRES(dp->port_mutex); |
c40b890f | 625 | static void do_del_port(struct dp_netdev *dp, struct dp_netdev_port *) |
59e6d833 | 626 | OVS_REQUIRES(dp->port_mutex); |
614c4892 BP |
627 | static int dpif_netdev_open(const struct dpif_class *, const char *name, |
628 | bool create, struct dpif **); | |
65f13b50 | 629 | static void dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
1895cc8d | 630 | struct dp_packet_batch *, |
66e4ad8a | 631 | bool may_steal, const struct flow *flow, |
4edb9ae9 | 632 | const struct nlattr *actions, |
324c8374 IM |
633 | size_t actions_len, |
634 | long long now); | |
65f13b50 | 635 | static void dp_netdev_input(struct dp_netdev_pmd_thread *, |
1895cc8d | 636 | struct dp_packet_batch *, odp_port_t port_no); |
a90ed026 | 637 | static void dp_netdev_recirculate(struct dp_netdev_pmd_thread *, |
1895cc8d | 638 | struct dp_packet_batch *); |
41ccaa24 | 639 | |
6b31e073 | 640 | static void dp_netdev_disable_upcall(struct dp_netdev *); |
ae7ad0a1 | 641 | static void dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd); |
65f13b50 | 642 | static void dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, |
00873463 DDP |
643 | struct dp_netdev *dp, unsigned core_id, |
644 | int numa_id); | |
1c1e46ed | 645 | static void dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd); |
e9985d6a DDP |
646 | static void dp_netdev_set_nonpmd(struct dp_netdev *dp) |
647 | OVS_REQUIRES(dp->port_mutex); | |
648 | ||
e32971b8 | 649 | static void *pmd_thread_main(void *); |
b19befae | 650 | static struct dp_netdev_pmd_thread *dp_netdev_get_pmd(struct dp_netdev *dp, |
bd5131ba | 651 | unsigned core_id); |
1c1e46ed AW |
652 | static struct dp_netdev_pmd_thread * |
653 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos); | |
140dd699 IM |
654 | static void dp_netdev_del_pmd(struct dp_netdev *dp, |
655 | struct dp_netdev_pmd_thread *pmd); | |
e32971b8 | 656 | static void dp_netdev_destroy_all_pmds(struct dp_netdev *dp, bool non_pmd); |
d0cca6c3 | 657 | static void dp_netdev_pmd_clear_ports(struct dp_netdev_pmd_thread *pmd); |
d0cca6c3 | 658 | static void dp_netdev_add_port_tx_to_pmd(struct dp_netdev_pmd_thread *pmd, |
e32971b8 DDP |
659 | struct dp_netdev_port *port) |
660 | OVS_REQUIRES(pmd->port_mutex); | |
661 | static void dp_netdev_del_port_tx_from_pmd(struct dp_netdev_pmd_thread *pmd, | |
662 | struct tx_port *tx) | |
663 | OVS_REQUIRES(pmd->port_mutex); | |
d0cca6c3 | 664 | static void dp_netdev_add_rxq_to_pmd(struct dp_netdev_pmd_thread *pmd, |
947dc567 DDP |
665 | struct dp_netdev_rxq *rxq) |
666 | OVS_REQUIRES(pmd->port_mutex); | |
e32971b8 DDP |
667 | static void dp_netdev_del_rxq_from_pmd(struct dp_netdev_pmd_thread *pmd, |
668 | struct rxq_poll *poll) | |
669 | OVS_REQUIRES(pmd->port_mutex); | |
670 | static void reconfigure_datapath(struct dp_netdev *dp) | |
3eb67853 | 671 | OVS_REQUIRES(dp->port_mutex); |
1c1e46ed AW |
672 | static bool dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd); |
673 | static void dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd); | |
674 | static void dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd); | |
d0cca6c3 DDP |
675 | static void pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd) |
676 | OVS_REQUIRES(pmd->port_mutex); | |
3453b4d6 JS |
677 | static inline void |
678 | dp_netdev_pmd_try_optimize(struct dp_netdev_pmd_thread *pmd); | |
72865317 | 679 | |
324c8374 IM |
680 | static void |
681 | dpif_netdev_xps_revalidate_pmd(const struct dp_netdev_pmd_thread *pmd, | |
682 | long long now, bool purge); | |
683 | static int dpif_netdev_xps_get_tx_qid(const struct dp_netdev_pmd_thread *pmd, | |
684 | struct tx_port *tx, long long now); | |
685 | ||
67ad54cb | 686 | static inline bool emc_entry_alive(struct emc_entry *ce); |
9bbf1c3d DDP |
687 | static void emc_clear_entry(struct emc_entry *ce); |
688 | ||
689 | static void | |
690 | emc_cache_init(struct emc_cache *flow_cache) | |
691 | { | |
692 | int i; | |
693 | ||
67ad54cb | 694 | flow_cache->sweep_idx = 0; |
9bbf1c3d DDP |
695 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { |
696 | flow_cache->entries[i].flow = NULL; | |
0de8783a | 697 | flow_cache->entries[i].key.hash = 0; |
09b0fa9c | 698 | flow_cache->entries[i].key.len = sizeof(struct miniflow); |
5fcff47b | 699 | flowmap_init(&flow_cache->entries[i].key.mf.map); |
9bbf1c3d DDP |
700 | } |
701 | } | |
702 | ||
703 | static void | |
704 | emc_cache_uninit(struct emc_cache *flow_cache) | |
705 | { | |
706 | int i; | |
707 | ||
708 | for (i = 0; i < ARRAY_SIZE(flow_cache->entries); i++) { | |
709 | emc_clear_entry(&flow_cache->entries[i]); | |
710 | } | |
711 | } | |
712 | ||
67ad54cb AW |
713 | /* Check and clear dead flow references slowly (one entry at each |
714 | * invocation). */ | |
715 | static void | |
716 | emc_cache_slow_sweep(struct emc_cache *flow_cache) | |
717 | { | |
718 | struct emc_entry *entry = &flow_cache->entries[flow_cache->sweep_idx]; | |
719 | ||
720 | if (!emc_entry_alive(entry)) { | |
721 | emc_clear_entry(entry); | |
722 | } | |
723 | flow_cache->sweep_idx = (flow_cache->sweep_idx + 1) & EM_FLOW_HASH_MASK; | |
724 | } | |
725 | ||
c4ea7529 BP |
726 | /* Returns true if 'dpif' is a netdev or dummy dpif, false otherwise. */ |
727 | bool | |
728 | dpif_is_netdev(const struct dpif *dpif) | |
729 | { | |
730 | return dpif->dpif_class->open == dpif_netdev_open; | |
731 | } | |
732 | ||
72865317 BP |
733 | static struct dpif_netdev * |
734 | dpif_netdev_cast(const struct dpif *dpif) | |
735 | { | |
c4ea7529 | 736 | ovs_assert(dpif_is_netdev(dpif)); |
72865317 BP |
737 | return CONTAINER_OF(dpif, struct dpif_netdev, dpif); |
738 | } | |
739 | ||
740 | static struct dp_netdev * | |
741 | get_dp_netdev(const struct dpif *dpif) | |
742 | { | |
743 | return dpif_netdev_cast(dpif)->dp; | |
744 | } | |
6553d06b DDP |
745 | \f |
746 | enum pmd_info_type { | |
ce179f11 IM |
747 | PMD_INFO_SHOW_STATS, /* Show how cpu cycles are spent. */ |
748 | PMD_INFO_CLEAR_STATS, /* Set the cycles count to 0. */ | |
749 | PMD_INFO_SHOW_RXQ /* Show poll-lists of pmd threads. */ | |
6553d06b DDP |
750 | }; |
751 | ||
752 | static void | |
753 | pmd_info_show_stats(struct ds *reply, | |
754 | struct dp_netdev_pmd_thread *pmd, | |
755 | unsigned long long stats[DP_N_STATS], | |
756 | uint64_t cycles[PMD_N_CYCLES]) | |
757 | { | |
758 | unsigned long long total_packets = 0; | |
759 | uint64_t total_cycles = 0; | |
760 | int i; | |
761 | ||
762 | /* These loops subtracts reference values ('*_zero') from the counters. | |
763 | * Since loads and stores are relaxed, it might be possible for a '*_zero' | |
764 | * value to be more recent than the current value we're reading from the | |
765 | * counter. This is not a big problem, since these numbers are not | |
766 | * supposed to be too accurate, but we should at least make sure that | |
767 | * the result is not negative. */ | |
768 | for (i = 0; i < DP_N_STATS; i++) { | |
769 | if (stats[i] > pmd->stats_zero[i]) { | |
770 | stats[i] -= pmd->stats_zero[i]; | |
771 | } else { | |
772 | stats[i] = 0; | |
773 | } | |
774 | ||
775 | if (i != DP_STAT_LOST) { | |
776 | /* Lost packets are already included in DP_STAT_MISS */ | |
777 | total_packets += stats[i]; | |
778 | } | |
779 | } | |
780 | ||
781 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
782 | if (cycles[i] > pmd->cycles_zero[i]) { | |
783 | cycles[i] -= pmd->cycles_zero[i]; | |
784 | } else { | |
785 | cycles[i] = 0; | |
786 | } | |
787 | ||
788 | total_cycles += cycles[i]; | |
789 | } | |
790 | ||
791 | ds_put_cstr(reply, (pmd->core_id == NON_PMD_CORE_ID) | |
792 | ? "main thread" : "pmd thread"); | |
793 | ||
794 | if (pmd->numa_id != OVS_NUMA_UNSPEC) { | |
795 | ds_put_format(reply, " numa_id %d", pmd->numa_id); | |
796 | } | |
d5c199ea | 797 | if (pmd->core_id != OVS_CORE_UNSPEC && pmd->core_id != NON_PMD_CORE_ID) { |
bd5131ba | 798 | ds_put_format(reply, " core_id %u", pmd->core_id); |
6553d06b DDP |
799 | } |
800 | ds_put_cstr(reply, ":\n"); | |
801 | ||
802 | ds_put_format(reply, | |
803 | "\temc hits:%llu\n\tmegaflow hits:%llu\n" | |
3453b4d6 | 804 | "\tavg. subtable lookups per hit:%.2f\n" |
6553d06b DDP |
805 | "\tmiss:%llu\n\tlost:%llu\n", |
806 | stats[DP_STAT_EXACT_HIT], stats[DP_STAT_MASKED_HIT], | |
3453b4d6 JS |
807 | stats[DP_STAT_MASKED_HIT] > 0 |
808 | ? (1.0*stats[DP_STAT_LOOKUP_HIT])/stats[DP_STAT_MASKED_HIT] | |
809 | : 0, | |
6553d06b DDP |
810 | stats[DP_STAT_MISS], stats[DP_STAT_LOST]); |
811 | ||
812 | if (total_cycles == 0) { | |
813 | return; | |
814 | } | |
815 | ||
816 | ds_put_format(reply, | |
a2ac666d | 817 | "\tidle cycles:%"PRIu64" (%.02f%%)\n" |
6553d06b | 818 | "\tprocessing cycles:%"PRIu64" (%.02f%%)\n", |
a2ac666d CL |
819 | cycles[PMD_CYCLES_IDLE], |
820 | cycles[PMD_CYCLES_IDLE] / (double)total_cycles * 100, | |
6553d06b DDP |
821 | cycles[PMD_CYCLES_PROCESSING], |
822 | cycles[PMD_CYCLES_PROCESSING] / (double)total_cycles * 100); | |
823 | ||
824 | if (total_packets == 0) { | |
825 | return; | |
826 | } | |
827 | ||
828 | ds_put_format(reply, | |
829 | "\tavg cycles per packet: %.02f (%"PRIu64"/%llu)\n", | |
830 | total_cycles / (double)total_packets, | |
831 | total_cycles, total_packets); | |
832 | ||
833 | ds_put_format(reply, | |
834 | "\tavg processing cycles per packet: " | |
835 | "%.02f (%"PRIu64"/%llu)\n", | |
836 | cycles[PMD_CYCLES_PROCESSING] / (double)total_packets, | |
837 | cycles[PMD_CYCLES_PROCESSING], total_packets); | |
838 | } | |
839 | ||
840 | static void | |
841 | pmd_info_clear_stats(struct ds *reply OVS_UNUSED, | |
842 | struct dp_netdev_pmd_thread *pmd, | |
843 | unsigned long long stats[DP_N_STATS], | |
844 | uint64_t cycles[PMD_N_CYCLES]) | |
845 | { | |
846 | int i; | |
847 | ||
848 | /* We cannot write 'stats' and 'cycles' (because they're written by other | |
849 | * threads) and we shouldn't change 'stats' (because they're used to count | |
850 | * datapath stats, which must not be cleared here). Instead, we save the | |
851 | * current values and subtract them from the values to be displayed in the | |
852 | * future */ | |
853 | for (i = 0; i < DP_N_STATS; i++) { | |
854 | pmd->stats_zero[i] = stats[i]; | |
855 | } | |
856 | for (i = 0; i < PMD_N_CYCLES; i++) { | |
857 | pmd->cycles_zero[i] = cycles[i]; | |
858 | } | |
859 | } | |
860 | ||
947dc567 DDP |
861 | static int |
862 | compare_poll_list(const void *a_, const void *b_) | |
863 | { | |
864 | const struct rxq_poll *a = a_; | |
865 | const struct rxq_poll *b = b_; | |
866 | ||
867 | const char *namea = netdev_rxq_get_name(a->rxq->rx); | |
868 | const char *nameb = netdev_rxq_get_name(b->rxq->rx); | |
869 | ||
870 | int cmp = strcmp(namea, nameb); | |
871 | if (!cmp) { | |
872 | return netdev_rxq_get_queue_id(a->rxq->rx) | |
873 | - netdev_rxq_get_queue_id(b->rxq->rx); | |
874 | } else { | |
875 | return cmp; | |
876 | } | |
877 | } | |
878 | ||
879 | static void | |
880 | sorted_poll_list(struct dp_netdev_pmd_thread *pmd, struct rxq_poll **list, | |
881 | size_t *n) | |
882 | { | |
883 | struct rxq_poll *ret, *poll; | |
884 | size_t i; | |
885 | ||
886 | *n = hmap_count(&pmd->poll_list); | |
887 | if (!*n) { | |
888 | ret = NULL; | |
889 | } else { | |
890 | ret = xcalloc(*n, sizeof *ret); | |
891 | i = 0; | |
892 | HMAP_FOR_EACH (poll, node, &pmd->poll_list) { | |
893 | ret[i] = *poll; | |
894 | i++; | |
895 | } | |
896 | ovs_assert(i == *n); | |
1cc1b5f6 | 897 | qsort(ret, *n, sizeof *ret, compare_poll_list); |
947dc567 DDP |
898 | } |
899 | ||
947dc567 DDP |
900 | *list = ret; |
901 | } | |
902 | ||
ce179f11 IM |
903 | static void |
904 | pmd_info_show_rxq(struct ds *reply, struct dp_netdev_pmd_thread *pmd) | |
905 | { | |
906 | if (pmd->core_id != NON_PMD_CORE_ID) { | |
ce179f11 | 907 | const char *prev_name = NULL; |
947dc567 DDP |
908 | struct rxq_poll *list; |
909 | size_t i, n; | |
ce179f11 | 910 | |
3eb67853 IM |
911 | ds_put_format(reply, |
912 | "pmd thread numa_id %d core_id %u:\n\tisolated : %s\n", | |
913 | pmd->numa_id, pmd->core_id, (pmd->isolated) | |
914 | ? "true" : "false"); | |
ce179f11 | 915 | |
d0cca6c3 | 916 | ovs_mutex_lock(&pmd->port_mutex); |
947dc567 DDP |
917 | sorted_poll_list(pmd, &list, &n); |
918 | for (i = 0; i < n; i++) { | |
919 | const char *name = netdev_rxq_get_name(list[i].rxq->rx); | |
ce179f11 IM |
920 | |
921 | if (!prev_name || strcmp(name, prev_name)) { | |
922 | if (prev_name) { | |
923 | ds_put_cstr(reply, "\n"); | |
924 | } | |
947dc567 | 925 | ds_put_format(reply, "\tport: %s\tqueue-id:", name); |
ce179f11 | 926 | } |
947dc567 DDP |
927 | ds_put_format(reply, " %d", |
928 | netdev_rxq_get_queue_id(list[i].rxq->rx)); | |
ce179f11 IM |
929 | prev_name = name; |
930 | } | |
d0cca6c3 | 931 | ovs_mutex_unlock(&pmd->port_mutex); |
ce179f11 | 932 | ds_put_cstr(reply, "\n"); |
947dc567 | 933 | free(list); |
ce179f11 IM |
934 | } |
935 | } | |
936 | ||
34d8e04b EC |
937 | static int |
938 | compare_poll_thread_list(const void *a_, const void *b_) | |
939 | { | |
940 | const struct dp_netdev_pmd_thread *a, *b; | |
941 | ||
942 | a = *(struct dp_netdev_pmd_thread **)a_; | |
943 | b = *(struct dp_netdev_pmd_thread **)b_; | |
944 | ||
945 | if (a->core_id < b->core_id) { | |
946 | return -1; | |
947 | } | |
948 | if (a->core_id > b->core_id) { | |
949 | return 1; | |
950 | } | |
951 | return 0; | |
952 | } | |
953 | ||
954 | /* Create a sorted list of pmd's from the dp->poll_threads cmap. We can use | |
955 | * this list, as long as we do not go to quiescent state. */ | |
956 | static void | |
957 | sorted_poll_thread_list(struct dp_netdev *dp, | |
958 | struct dp_netdev_pmd_thread ***list, | |
959 | size_t *n) | |
960 | { | |
961 | struct dp_netdev_pmd_thread *pmd; | |
962 | struct dp_netdev_pmd_thread **pmd_list; | |
963 | size_t k = 0, n_pmds; | |
964 | ||
965 | n_pmds = cmap_count(&dp->poll_threads); | |
966 | pmd_list = xcalloc(n_pmds, sizeof *pmd_list); | |
967 | ||
968 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
969 | if (k >= n_pmds) { | |
970 | break; | |
971 | } | |
972 | pmd_list[k++] = pmd; | |
973 | } | |
974 | ||
975 | qsort(pmd_list, k, sizeof *pmd_list, compare_poll_thread_list); | |
976 | ||
977 | *list = pmd_list; | |
978 | *n = k; | |
979 | } | |
980 | ||
6553d06b DDP |
981 | static void |
982 | dpif_netdev_pmd_info(struct unixctl_conn *conn, int argc, const char *argv[], | |
983 | void *aux) | |
984 | { | |
985 | struct ds reply = DS_EMPTY_INITIALIZER; | |
34d8e04b | 986 | struct dp_netdev_pmd_thread **pmd_list; |
6553d06b | 987 | struct dp_netdev *dp = NULL; |
34d8e04b | 988 | size_t n; |
6553d06b DDP |
989 | enum pmd_info_type type = *(enum pmd_info_type *) aux; |
990 | ||
991 | ovs_mutex_lock(&dp_netdev_mutex); | |
992 | ||
993 | if (argc == 2) { | |
994 | dp = shash_find_data(&dp_netdevs, argv[1]); | |
995 | } else if (shash_count(&dp_netdevs) == 1) { | |
996 | /* There's only one datapath */ | |
997 | dp = shash_first(&dp_netdevs)->data; | |
998 | } | |
999 | ||
1000 | if (!dp) { | |
1001 | ovs_mutex_unlock(&dp_netdev_mutex); | |
1002 | unixctl_command_reply_error(conn, | |
1003 | "please specify an existing datapath"); | |
1004 | return; | |
1005 | } | |
1006 | ||
34d8e04b EC |
1007 | sorted_poll_thread_list(dp, &pmd_list, &n); |
1008 | for (size_t i = 0; i < n; i++) { | |
1009 | struct dp_netdev_pmd_thread *pmd = pmd_list[i]; | |
1010 | if (!pmd) { | |
1011 | break; | |
1012 | } | |
1013 | ||
ce179f11 IM |
1014 | if (type == PMD_INFO_SHOW_RXQ) { |
1015 | pmd_info_show_rxq(&reply, pmd); | |
1016 | } else { | |
1017 | unsigned long long stats[DP_N_STATS]; | |
1018 | uint64_t cycles[PMD_N_CYCLES]; | |
1019 | int i; | |
6553d06b | 1020 | |
ce179f11 IM |
1021 | /* Read current stats and cycle counters */ |
1022 | for (i = 0; i < ARRAY_SIZE(stats); i++) { | |
1023 | atomic_read_relaxed(&pmd->stats.n[i], &stats[i]); | |
1024 | } | |
1025 | for (i = 0; i < ARRAY_SIZE(cycles); i++) { | |
1026 | atomic_read_relaxed(&pmd->cycles.n[i], &cycles[i]); | |
1027 | } | |
6553d06b | 1028 | |
ce179f11 IM |
1029 | if (type == PMD_INFO_CLEAR_STATS) { |
1030 | pmd_info_clear_stats(&reply, pmd, stats, cycles); | |
1031 | } else if (type == PMD_INFO_SHOW_STATS) { | |
1032 | pmd_info_show_stats(&reply, pmd, stats, cycles); | |
1033 | } | |
6553d06b DDP |
1034 | } |
1035 | } | |
34d8e04b | 1036 | free(pmd_list); |
6553d06b DDP |
1037 | |
1038 | ovs_mutex_unlock(&dp_netdev_mutex); | |
1039 | ||
1040 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
1041 | ds_destroy(&reply); | |
1042 | } | |
1043 | \f | |
1044 | static int | |
1045 | dpif_netdev_init(void) | |
1046 | { | |
1047 | static enum pmd_info_type show_aux = PMD_INFO_SHOW_STATS, | |
ce179f11 IM |
1048 | clear_aux = PMD_INFO_CLEAR_STATS, |
1049 | poll_aux = PMD_INFO_SHOW_RXQ; | |
6553d06b DDP |
1050 | |
1051 | unixctl_command_register("dpif-netdev/pmd-stats-show", "[dp]", | |
1052 | 0, 1, dpif_netdev_pmd_info, | |
1053 | (void *)&show_aux); | |
1054 | unixctl_command_register("dpif-netdev/pmd-stats-clear", "[dp]", | |
1055 | 0, 1, dpif_netdev_pmd_info, | |
1056 | (void *)&clear_aux); | |
ce179f11 IM |
1057 | unixctl_command_register("dpif-netdev/pmd-rxq-show", "[dp]", |
1058 | 0, 1, dpif_netdev_pmd_info, | |
1059 | (void *)&poll_aux); | |
6553d06b DDP |
1060 | return 0; |
1061 | } | |
72865317 | 1062 | |
2197d7ab | 1063 | static int |
2240af25 DDP |
1064 | dpif_netdev_enumerate(struct sset *all_dps, |
1065 | const struct dpif_class *dpif_class) | |
2197d7ab GL |
1066 | { |
1067 | struct shash_node *node; | |
1068 | ||
97be1538 | 1069 | ovs_mutex_lock(&dp_netdev_mutex); |
2197d7ab | 1070 | SHASH_FOR_EACH(node, &dp_netdevs) { |
2240af25 DDP |
1071 | struct dp_netdev *dp = node->data; |
1072 | if (dpif_class != dp->class) { | |
1073 | /* 'dp_netdevs' contains both "netdev" and "dummy" dpifs. | |
1074 | * If the class doesn't match, skip this dpif. */ | |
1075 | continue; | |
1076 | } | |
2197d7ab GL |
1077 | sset_add(all_dps, node->name); |
1078 | } | |
97be1538 | 1079 | ovs_mutex_unlock(&dp_netdev_mutex); |
5279f8fd | 1080 | |
2197d7ab GL |
1081 | return 0; |
1082 | } | |
1083 | ||
add90f6f EJ |
1084 | static bool |
1085 | dpif_netdev_class_is_dummy(const struct dpif_class *class) | |
1086 | { | |
1087 | return class != &dpif_netdev_class; | |
1088 | } | |
1089 | ||
0aeaabc8 JP |
1090 | static const char * |
1091 | dpif_netdev_port_open_type(const struct dpif_class *class, const char *type) | |
1092 | { | |
1093 | return strcmp(type, "internal") ? type | |
e98d0cb3 | 1094 | : dpif_netdev_class_is_dummy(class) ? "dummy-internal" |
0aeaabc8 JP |
1095 | : "tap"; |
1096 | } | |
1097 | ||
72865317 BP |
1098 | static struct dpif * |
1099 | create_dpif_netdev(struct dp_netdev *dp) | |
1100 | { | |
462278db | 1101 | uint16_t netflow_id = hash_string(dp->name, 0); |
72865317 | 1102 | struct dpif_netdev *dpif; |
72865317 | 1103 | |
6a8267c5 | 1104 | ovs_refcount_ref(&dp->ref_cnt); |
72865317 | 1105 | |
72865317 | 1106 | dpif = xmalloc(sizeof *dpif); |
614c4892 | 1107 | dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id); |
72865317 | 1108 | dpif->dp = dp; |
d33ed218 | 1109 | dpif->last_port_seq = seq_read(dp->port_seq); |
72865317 BP |
1110 | |
1111 | return &dpif->dpif; | |
1112 | } | |
1113 | ||
4e022ec0 AW |
1114 | /* Choose an unused, non-zero port number and return it on success. |
1115 | * Return ODPP_NONE on failure. */ | |
1116 | static odp_port_t | |
e44768b7 | 1117 | choose_port(struct dp_netdev *dp, const char *name) |
59e6d833 | 1118 | OVS_REQUIRES(dp->port_mutex) |
e44768b7 | 1119 | { |
4e022ec0 | 1120 | uint32_t port_no; |
e44768b7 JP |
1121 | |
1122 | if (dp->class != &dpif_netdev_class) { | |
1123 | const char *p; | |
1124 | int start_no = 0; | |
1125 | ||
1126 | /* If the port name begins with "br", start the number search at | |
1127 | * 100 to make writing tests easier. */ | |
1128 | if (!strncmp(name, "br", 2)) { | |
1129 | start_no = 100; | |
1130 | } | |
1131 | ||
1132 | /* If the port name contains a number, try to assign that port number. | |
1133 | * This can make writing unit tests easier because port numbers are | |
1134 | * predictable. */ | |
1135 | for (p = name; *p != '\0'; p++) { | |
1136 | if (isdigit((unsigned char) *p)) { | |
1137 | port_no = start_no + strtol(p, NULL, 10); | |
ff073a71 BP |
1138 | if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE) |
1139 | && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 1140 | return u32_to_odp(port_no); |
e44768b7 JP |
1141 | } |
1142 | break; | |
1143 | } | |
1144 | } | |
1145 | } | |
1146 | ||
ff073a71 BP |
1147 | for (port_no = 1; port_no <= UINT16_MAX; port_no++) { |
1148 | if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) { | |
4e022ec0 | 1149 | return u32_to_odp(port_no); |
e44768b7 JP |
1150 | } |
1151 | } | |
1152 | ||
4e022ec0 | 1153 | return ODPP_NONE; |
e44768b7 JP |
1154 | } |
1155 | ||
72865317 | 1156 | static int |
614c4892 BP |
1157 | create_dp_netdev(const char *name, const struct dpif_class *class, |
1158 | struct dp_netdev **dpp) | |
8a4e3a85 | 1159 | OVS_REQUIRES(dp_netdev_mutex) |
72865317 BP |
1160 | { |
1161 | struct dp_netdev *dp; | |
1162 | int error; | |
72865317 | 1163 | |
462278db | 1164 | dp = xzalloc(sizeof *dp); |
8a4e3a85 BP |
1165 | shash_add(&dp_netdevs, name, dp); |
1166 | ||
1167 | *CONST_CAST(const struct dpif_class **, &dp->class) = class; | |
1168 | *CONST_CAST(const char **, &dp->name) = xstrdup(name); | |
6a8267c5 | 1169 | ovs_refcount_init(&dp->ref_cnt); |
1a65ba85 | 1170 | atomic_flag_clear(&dp->destroyed); |
8a4e3a85 | 1171 | |
59e6d833 | 1172 | ovs_mutex_init(&dp->port_mutex); |
e9985d6a | 1173 | hmap_init(&dp->ports); |
d33ed218 | 1174 | dp->port_seq = seq_create(); |
6b31e073 RW |
1175 | fat_rwlock_init(&dp->upcall_rwlock); |
1176 | ||
a6a426d6 IM |
1177 | dp->reconfigure_seq = seq_create(); |
1178 | dp->last_reconfigure_seq = seq_read(dp->reconfigure_seq); | |
1179 | ||
4b27db64 JR |
1180 | for (int i = 0; i < N_METER_LOCKS; ++i) { |
1181 | ovs_mutex_init_adaptive(&dp->meter_locks[i]); | |
1182 | } | |
1183 | ||
6b31e073 RW |
1184 | /* Disable upcalls by default. */ |
1185 | dp_netdev_disable_upcall(dp); | |
623540e4 | 1186 | dp->upcall_aux = NULL; |
6b31e073 | 1187 | dp->upcall_cb = NULL; |
e44768b7 | 1188 | |
5cf3edb3 DDP |
1189 | conntrack_init(&dp->conntrack); |
1190 | ||
4c30b246 CL |
1191 | atomic_init(&dp->emc_insert_min, DEFAULT_EM_FLOW_INSERT_MIN); |
1192 | ||
65f13b50 | 1193 | cmap_init(&dp->poll_threads); |
140dd699 IM |
1194 | |
1195 | ovs_mutex_init(&dp->tx_qid_pool_mutex); | |
1196 | /* We need 1 Tx queue for each possible core + 1 for non-PMD threads. */ | |
1197 | dp->tx_qid_pool = id_pool_create(0, ovs_numa_get_n_cores() + 1); | |
1198 | ||
65f13b50 AW |
1199 | ovs_mutex_init_recursive(&dp->non_pmd_mutex); |
1200 | ovsthread_key_create(&dp->per_pmd_key, NULL); | |
1201 | ||
e9985d6a | 1202 | ovs_mutex_lock(&dp->port_mutex); |
140dd699 IM |
1203 | /* non-PMD will be created before all other threads and will |
1204 | * allocate static_tx_qid = 0. */ | |
f2eee189 | 1205 | dp_netdev_set_nonpmd(dp); |
65f13b50 | 1206 | |
a3e8437a TLSC |
1207 | error = do_add_port(dp, name, dpif_netdev_port_open_type(dp->class, |
1208 | "internal"), | |
1209 | ODPP_LOCAL); | |
59e6d833 | 1210 | ovs_mutex_unlock(&dp->port_mutex); |
72865317 BP |
1211 | if (error) { |
1212 | dp_netdev_free(dp); | |
462278db | 1213 | return error; |
72865317 BP |
1214 | } |
1215 | ||
a36de779 | 1216 | dp->last_tnl_conf_seq = seq_read(tnl_conf_seq); |
462278db | 1217 | *dpp = dp; |
72865317 BP |
1218 | return 0; |
1219 | } | |
1220 | ||
a6a426d6 IM |
1221 | static void |
1222 | dp_netdev_request_reconfigure(struct dp_netdev *dp) | |
1223 | { | |
1224 | seq_change(dp->reconfigure_seq); | |
1225 | } | |
1226 | ||
1227 | static bool | |
1228 | dp_netdev_is_reconf_required(struct dp_netdev *dp) | |
1229 | { | |
1230 | return seq_read(dp->reconfigure_seq) != dp->last_reconfigure_seq; | |
1231 | } | |
1232 | ||
72865317 | 1233 | static int |
614c4892 | 1234 | dpif_netdev_open(const struct dpif_class *class, const char *name, |
4a387741 | 1235 | bool create, struct dpif **dpifp) |
72865317 | 1236 | { |
462278db | 1237 | struct dp_netdev *dp; |
5279f8fd | 1238 | int error; |
462278db | 1239 | |
97be1538 | 1240 | ovs_mutex_lock(&dp_netdev_mutex); |
462278db BP |
1241 | dp = shash_find_data(&dp_netdevs, name); |
1242 | if (!dp) { | |
5279f8fd | 1243 | error = create ? create_dp_netdev(name, class, &dp) : ENODEV; |
72865317 | 1244 | } else { |
5279f8fd BP |
1245 | error = (dp->class != class ? EINVAL |
1246 | : create ? EEXIST | |
1247 | : 0); | |
1248 | } | |
1249 | if (!error) { | |
1250 | *dpifp = create_dpif_netdev(dp); | |
6b31e073 | 1251 | dp->dpif = *dpifp; |
72865317 | 1252 | } |
97be1538 | 1253 | ovs_mutex_unlock(&dp_netdev_mutex); |
462278db | 1254 | |
5279f8fd | 1255 | return error; |
72865317 BP |
1256 | } |
1257 | ||
88ace79b DDP |
1258 | static void |
1259 | dp_netdev_destroy_upcall_lock(struct dp_netdev *dp) | |
1260 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
1261 | { | |
1262 | /* Check that upcalls are disabled, i.e. that the rwlock is taken */ | |
1263 | ovs_assert(fat_rwlock_tryrdlock(&dp->upcall_rwlock)); | |
1264 | ||
1265 | /* Before freeing a lock we should release it */ | |
1266 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
1267 | fat_rwlock_destroy(&dp->upcall_rwlock); | |
1268 | } | |
1269 | ||
4b27db64 JR |
1270 | static void |
1271 | dp_delete_meter(struct dp_netdev *dp, uint32_t meter_id) | |
1272 | OVS_REQUIRES(dp->meter_locks[meter_id % N_METER_LOCKS]) | |
1273 | { | |
1274 | if (dp->meters[meter_id]) { | |
1275 | free(dp->meters[meter_id]); | |
1276 | dp->meters[meter_id] = NULL; | |
1277 | } | |
1278 | } | |
1279 | ||
8a4e3a85 BP |
1280 | /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp' |
1281 | * through the 'dp_netdevs' shash while freeing 'dp'. */ | |
1ba530f4 BP |
1282 | static void |
1283 | dp_netdev_free(struct dp_netdev *dp) | |
8a4e3a85 | 1284 | OVS_REQUIRES(dp_netdev_mutex) |
1ba530f4 | 1285 | { |
e9985d6a | 1286 | struct dp_netdev_port *port, *next; |
4ad28026 | 1287 | |
8a4e3a85 BP |
1288 | shash_find_and_delete(&dp_netdevs, dp->name); |
1289 | ||
59e6d833 | 1290 | ovs_mutex_lock(&dp->port_mutex); |
e9985d6a | 1291 | HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) { |
c40b890f | 1292 | do_del_port(dp, port); |
1ba530f4 | 1293 | } |
59e6d833 | 1294 | ovs_mutex_unlock(&dp->port_mutex); |
4b27db64 | 1295 | |
e32971b8 | 1296 | dp_netdev_destroy_all_pmds(dp, true); |
d916785c | 1297 | cmap_destroy(&dp->poll_threads); |
51852a57 | 1298 | |
140dd699 IM |
1299 | ovs_mutex_destroy(&dp->tx_qid_pool_mutex); |
1300 | id_pool_destroy(dp->tx_qid_pool); | |
1301 | ||
b9584f21 DDP |
1302 | ovs_mutex_destroy(&dp->non_pmd_mutex); |
1303 | ovsthread_key_delete(dp->per_pmd_key); | |
1304 | ||
1305 | conntrack_destroy(&dp->conntrack); | |
1306 | ||
1307 | ||
a6a426d6 IM |
1308 | seq_destroy(dp->reconfigure_seq); |
1309 | ||
d33ed218 | 1310 | seq_destroy(dp->port_seq); |
e9985d6a | 1311 | hmap_destroy(&dp->ports); |
3186ea46 | 1312 | ovs_mutex_destroy(&dp->port_mutex); |
88ace79b DDP |
1313 | |
1314 | /* Upcalls must be disabled at this point */ | |
1315 | dp_netdev_destroy_upcall_lock(dp); | |
9bbf1c3d | 1316 | |
4b27db64 JR |
1317 | int i; |
1318 | ||
1319 | for (i = 0; i < MAX_METERS; ++i) { | |
1320 | meter_lock(dp, i); | |
1321 | dp_delete_meter(dp, i); | |
1322 | meter_unlock(dp, i); | |
1323 | } | |
1324 | for (i = 0; i < N_METER_LOCKS; ++i) { | |
1325 | ovs_mutex_destroy(&dp->meter_locks[i]); | |
1326 | } | |
1327 | ||
f2eee189 | 1328 | free(dp->pmd_cmask); |
8a4e3a85 | 1329 | free(CONST_CAST(char *, dp->name)); |
72865317 BP |
1330 | free(dp); |
1331 | } | |
1332 | ||
8a4e3a85 BP |
1333 | static void |
1334 | dp_netdev_unref(struct dp_netdev *dp) | |
1335 | { | |
1336 | if (dp) { | |
1337 | /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't | |
1338 | * get a new reference to 'dp' through the 'dp_netdevs' shash. */ | |
1339 | ovs_mutex_lock(&dp_netdev_mutex); | |
24f83812 | 1340 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
8a4e3a85 BP |
1341 | dp_netdev_free(dp); |
1342 | } | |
1343 | ovs_mutex_unlock(&dp_netdev_mutex); | |
1344 | } | |
1345 | } | |
1346 | ||
72865317 BP |
1347 | static void |
1348 | dpif_netdev_close(struct dpif *dpif) | |
1349 | { | |
1350 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 1351 | |
8a4e3a85 | 1352 | dp_netdev_unref(dp); |
72865317 BP |
1353 | free(dpif); |
1354 | } | |
1355 | ||
1356 | static int | |
7dab847a | 1357 | dpif_netdev_destroy(struct dpif *dpif) |
72865317 BP |
1358 | { |
1359 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd | 1360 | |
6a8267c5 | 1361 | if (!atomic_flag_test_and_set(&dp->destroyed)) { |
24f83812 | 1362 | if (ovs_refcount_unref_relaxed(&dp->ref_cnt) == 1) { |
6a8267c5 BP |
1363 | /* Can't happen: 'dpif' still owns a reference to 'dp'. */ |
1364 | OVS_NOT_REACHED(); | |
1365 | } | |
1366 | } | |
5279f8fd | 1367 | |
72865317 BP |
1368 | return 0; |
1369 | } | |
1370 | ||
eb94da30 DDP |
1371 | /* Add 'n' to the atomic variable 'var' non-atomically and using relaxed |
1372 | * load/store semantics. While the increment is not atomic, the load and | |
1373 | * store operations are, making it impossible to read inconsistent values. | |
1374 | * | |
1375 | * This is used to update thread local stats counters. */ | |
1376 | static void | |
1377 | non_atomic_ullong_add(atomic_ullong *var, unsigned long long n) | |
1378 | { | |
1379 | unsigned long long tmp; | |
1380 | ||
1381 | atomic_read_relaxed(var, &tmp); | |
1382 | tmp += n; | |
1383 | atomic_store_relaxed(var, tmp); | |
1384 | } | |
1385 | ||
72865317 | 1386 | static int |
a8d9304d | 1387 | dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) |
72865317 BP |
1388 | { |
1389 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed | 1390 | struct dp_netdev_pmd_thread *pmd; |
8a4e3a85 | 1391 | |
1c1e46ed AW |
1392 | stats->n_flows = stats->n_hit = stats->n_missed = stats->n_lost = 0; |
1393 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
eb94da30 | 1394 | unsigned long long n; |
1c1e46ed | 1395 | stats->n_flows += cmap_count(&pmd->flow_table); |
eb94da30 | 1396 | |
abcf3ef4 DDP |
1397 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MASKED_HIT], &n); |
1398 | stats->n_hit += n; | |
1399 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_EXACT_HIT], &n); | |
eb94da30 DDP |
1400 | stats->n_hit += n; |
1401 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_MISS], &n); | |
1402 | stats->n_missed += n; | |
1403 | atomic_read_relaxed(&pmd->stats.n[DP_STAT_LOST], &n); | |
1404 | stats->n_lost += n; | |
51852a57 | 1405 | } |
1ce3fa06 | 1406 | stats->n_masks = UINT32_MAX; |
847108dc | 1407 | stats->n_mask_hit = UINT64_MAX; |
5279f8fd | 1408 | |
72865317 BP |
1409 | return 0; |
1410 | } | |
1411 | ||
e4cfed38 | 1412 | static void |
65f13b50 | 1413 | dp_netdev_reload_pmd__(struct dp_netdev_pmd_thread *pmd) |
e4cfed38 | 1414 | { |
accf8626 | 1415 | if (pmd->core_id == NON_PMD_CORE_ID) { |
d0cca6c3 DDP |
1416 | ovs_mutex_lock(&pmd->dp->non_pmd_mutex); |
1417 | ovs_mutex_lock(&pmd->port_mutex); | |
1418 | pmd_load_cached_ports(pmd); | |
1419 | ovs_mutex_unlock(&pmd->port_mutex); | |
1420 | ovs_mutex_unlock(&pmd->dp->non_pmd_mutex); | |
accf8626 AW |
1421 | return; |
1422 | } | |
1423 | ||
1424 | ovs_mutex_lock(&pmd->cond_mutex); | |
2788a1b1 | 1425 | seq_change(pmd->reload_seq); |
14e3e12a | 1426 | atomic_store_relaxed(&pmd->reload, true); |
accf8626 AW |
1427 | ovs_mutex_cond_wait(&pmd->cond, &pmd->cond_mutex); |
1428 | ovs_mutex_unlock(&pmd->cond_mutex); | |
65f13b50 | 1429 | } |
e4cfed38 | 1430 | |
59e6d833 BP |
1431 | static uint32_t |
1432 | hash_port_no(odp_port_t port_no) | |
1433 | { | |
1434 | return hash_int(odp_to_u32(port_no), 0); | |
1435 | } | |
1436 | ||
72865317 | 1437 | static int |
a3e8437a | 1438 | port_create(const char *devname, const char *type, |
b8d29252 | 1439 | odp_port_t port_no, struct dp_netdev_port **portp) |
72865317 | 1440 | { |
4b609110 | 1441 | struct netdev_saved_flags *sf; |
72865317 | 1442 | struct dp_netdev_port *port; |
2499a8ce | 1443 | enum netdev_flags flags; |
b8d29252 | 1444 | struct netdev *netdev; |
e32971b8 | 1445 | int error; |
72865317 | 1446 | |
b8d29252 | 1447 | *portp = NULL; |
72865317 BP |
1448 | |
1449 | /* Open and validate network device. */ | |
a3e8437a | 1450 | error = netdev_open(devname, type, &netdev); |
72865317 | 1451 | if (error) { |
b8d29252 | 1452 | return error; |
72865317 | 1453 | } |
72865317 BP |
1454 | /* XXX reject non-Ethernet devices */ |
1455 | ||
2499a8ce AC |
1456 | netdev_get_flags(netdev, &flags); |
1457 | if (flags & NETDEV_LOOPBACK) { | |
1458 | VLOG_ERR("%s: cannot add a loopback device", devname); | |
d17f4f08 | 1459 | error = EINVAL; |
b8d29252 | 1460 | goto out; |
2499a8ce AC |
1461 | } |
1462 | ||
e32971b8 DDP |
1463 | error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf); |
1464 | if (error) { | |
1465 | VLOG_ERR("%s: cannot set promisc flag", devname); | |
1466 | goto out; | |
324c8374 IM |
1467 | } |
1468 | ||
e4cfed38 | 1469 | port = xzalloc(sizeof *port); |
35303d71 | 1470 | port->port_no = port_no; |
e4cfed38 PS |
1471 | port->netdev = netdev; |
1472 | port->type = xstrdup(type); | |
4b609110 | 1473 | port->sf = sf; |
e32971b8 DDP |
1474 | port->need_reconfigure = true; |
1475 | ovs_mutex_init(&port->txq_used_mutex); | |
e4cfed38 | 1476 | |
b8d29252 | 1477 | *portp = port; |
72865317 BP |
1478 | |
1479 | return 0; | |
d17f4f08 | 1480 | |
d17f4f08 | 1481 | out: |
b8d29252 | 1482 | netdev_close(netdev); |
d17f4f08 | 1483 | return error; |
72865317 BP |
1484 | } |
1485 | ||
b8d29252 DDP |
1486 | static int |
1487 | do_add_port(struct dp_netdev *dp, const char *devname, const char *type, | |
1488 | odp_port_t port_no) | |
1489 | OVS_REQUIRES(dp->port_mutex) | |
1490 | { | |
1491 | struct dp_netdev_port *port; | |
1492 | int error; | |
1493 | ||
1494 | /* Reject devices already in 'dp'. */ | |
1495 | if (!get_port_by_name(dp, devname, &port)) { | |
1496 | return EEXIST; | |
1497 | } | |
1498 | ||
a3e8437a | 1499 | error = port_create(devname, type, port_no, &port); |
b8d29252 DDP |
1500 | if (error) { |
1501 | return error; | |
1502 | } | |
1503 | ||
e9985d6a | 1504 | hmap_insert(&dp->ports, &port->node, hash_port_no(port_no)); |
b8d29252 DDP |
1505 | seq_change(dp->port_seq); |
1506 | ||
e32971b8 DDP |
1507 | reconfigure_datapath(dp); |
1508 | ||
b8d29252 DDP |
1509 | return 0; |
1510 | } | |
1511 | ||
247527db BP |
1512 | static int |
1513 | dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev, | |
4e022ec0 | 1514 | odp_port_t *port_nop) |
247527db BP |
1515 | { |
1516 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3aa30359 BP |
1517 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1518 | const char *dpif_port; | |
4e022ec0 | 1519 | odp_port_t port_no; |
5279f8fd | 1520 | int error; |
247527db | 1521 | |
59e6d833 | 1522 | ovs_mutex_lock(&dp->port_mutex); |
3aa30359 | 1523 | dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
4e022ec0 | 1524 | if (*port_nop != ODPP_NONE) { |
ff073a71 BP |
1525 | port_no = *port_nop; |
1526 | error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0; | |
232dfa4a | 1527 | } else { |
3aa30359 | 1528 | port_no = choose_port(dp, dpif_port); |
5279f8fd | 1529 | error = port_no == ODPP_NONE ? EFBIG : 0; |
232dfa4a | 1530 | } |
5279f8fd | 1531 | if (!error) { |
247527db | 1532 | *port_nop = port_no; |
5279f8fd | 1533 | error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no); |
247527db | 1534 | } |
59e6d833 | 1535 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1536 | |
1537 | return error; | |
72865317 BP |
1538 | } |
1539 | ||
1540 | static int | |
4e022ec0 | 1541 | dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no) |
72865317 BP |
1542 | { |
1543 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5279f8fd BP |
1544 | int error; |
1545 | ||
59e6d833 | 1546 | ovs_mutex_lock(&dp->port_mutex); |
c40b890f BP |
1547 | if (port_no == ODPP_LOCAL) { |
1548 | error = EINVAL; | |
1549 | } else { | |
1550 | struct dp_netdev_port *port; | |
1551 | ||
1552 | error = get_port_by_number(dp, port_no, &port); | |
1553 | if (!error) { | |
1554 | do_del_port(dp, port); | |
1555 | } | |
1556 | } | |
59e6d833 | 1557 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd BP |
1558 | |
1559 | return error; | |
72865317 BP |
1560 | } |
1561 | ||
1562 | static bool | |
4e022ec0 | 1563 | is_valid_port_number(odp_port_t port_no) |
72865317 | 1564 | { |
ff073a71 BP |
1565 | return port_no != ODPP_NONE; |
1566 | } | |
1567 | ||
1568 | static struct dp_netdev_port * | |
1569 | dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no) | |
e9985d6a | 1570 | OVS_REQUIRES(dp->port_mutex) |
ff073a71 BP |
1571 | { |
1572 | struct dp_netdev_port *port; | |
1573 | ||
e9985d6a | 1574 | HMAP_FOR_EACH_WITH_HASH (port, node, hash_port_no(port_no), &dp->ports) { |
35303d71 | 1575 | if (port->port_no == port_no) { |
ff073a71 BP |
1576 | return port; |
1577 | } | |
1578 | } | |
1579 | return NULL; | |
72865317 BP |
1580 | } |
1581 | ||
1582 | static int | |
1583 | get_port_by_number(struct dp_netdev *dp, | |
4e022ec0 | 1584 | odp_port_t port_no, struct dp_netdev_port **portp) |
e9985d6a | 1585 | OVS_REQUIRES(dp->port_mutex) |
72865317 BP |
1586 | { |
1587 | if (!is_valid_port_number(port_no)) { | |
1588 | *portp = NULL; | |
1589 | return EINVAL; | |
1590 | } else { | |
ff073a71 | 1591 | *portp = dp_netdev_lookup_port(dp, port_no); |
0f6a066f | 1592 | return *portp ? 0 : ENODEV; |
72865317 BP |
1593 | } |
1594 | } | |
1595 | ||
b284085e | 1596 | static void |
62453dad | 1597 | port_destroy(struct dp_netdev_port *port) |
b284085e | 1598 | { |
62453dad DDP |
1599 | if (!port) { |
1600 | return; | |
b284085e | 1601 | } |
b284085e | 1602 | |
62453dad DDP |
1603 | netdev_close(port->netdev); |
1604 | netdev_restore_flags(port->sf); | |
accf8626 | 1605 | |
62453dad | 1606 | for (unsigned i = 0; i < port->n_rxq; i++) { |
947dc567 | 1607 | netdev_rxq_close(port->rxqs[i].rx); |
b284085e | 1608 | } |
324c8374 | 1609 | ovs_mutex_destroy(&port->txq_used_mutex); |
3eb67853 | 1610 | free(port->rxq_affinity_list); |
324c8374 | 1611 | free(port->txq_used); |
3eb67853 | 1612 | free(port->rxqs); |
62453dad DDP |
1613 | free(port->type); |
1614 | free(port); | |
b284085e PS |
1615 | } |
1616 | ||
72865317 BP |
1617 | static int |
1618 | get_port_by_name(struct dp_netdev *dp, | |
1619 | const char *devname, struct dp_netdev_port **portp) | |
59e6d833 | 1620 | OVS_REQUIRES(dp->port_mutex) |
72865317 BP |
1621 | { |
1622 | struct dp_netdev_port *port; | |
1623 | ||
e9985d6a | 1624 | HMAP_FOR_EACH (port, node, &dp->ports) { |
3efb6063 | 1625 | if (!strcmp(netdev_get_name(port->netdev), devname)) { |
72865317 BP |
1626 | *portp = port; |
1627 | return 0; | |
1628 | } | |
1629 | } | |
0f6a066f DDP |
1630 | |
1631 | /* Callers of dpif_netdev_port_query_by_name() expect ENODEV for a non | |
1632 | * existing port. */ | |
1633 | return ENODEV; | |
72865317 BP |
1634 | } |
1635 | ||
b9584f21 | 1636 | /* Returns 'true' if there is a port with pmd netdev. */ |
65f13b50 | 1637 | static bool |
b9584f21 | 1638 | has_pmd_port(struct dp_netdev *dp) |
e9985d6a | 1639 | OVS_REQUIRES(dp->port_mutex) |
65f13b50 AW |
1640 | { |
1641 | struct dp_netdev_port *port; | |
1642 | ||
e9985d6a | 1643 | HMAP_FOR_EACH (port, node, &dp->ports) { |
5dd57e80 | 1644 | if (netdev_is_pmd(port->netdev)) { |
b9584f21 | 1645 | return true; |
65f13b50 AW |
1646 | } |
1647 | } | |
1648 | ||
1649 | return false; | |
1650 | } | |
1651 | ||
c40b890f BP |
1652 | static void |
1653 | do_del_port(struct dp_netdev *dp, struct dp_netdev_port *port) | |
59e6d833 | 1654 | OVS_REQUIRES(dp->port_mutex) |
72865317 | 1655 | { |
e9985d6a | 1656 | hmap_remove(&dp->ports, &port->node); |
d33ed218 | 1657 | seq_change(dp->port_seq); |
d0cca6c3 | 1658 | |
e32971b8 | 1659 | reconfigure_datapath(dp); |
72865317 | 1660 | |
62453dad | 1661 | port_destroy(port); |
72865317 BP |
1662 | } |
1663 | ||
1664 | static void | |
4c738a8d BP |
1665 | answer_port_query(const struct dp_netdev_port *port, |
1666 | struct dpif_port *dpif_port) | |
72865317 | 1667 | { |
3efb6063 | 1668 | dpif_port->name = xstrdup(netdev_get_name(port->netdev)); |
0cbfe35d | 1669 | dpif_port->type = xstrdup(port->type); |
35303d71 | 1670 | dpif_port->port_no = port->port_no; |
72865317 BP |
1671 | } |
1672 | ||
1673 | static int | |
4e022ec0 | 1674 | dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, |
4c738a8d | 1675 | struct dpif_port *dpif_port) |
72865317 BP |
1676 | { |
1677 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1678 | struct dp_netdev_port *port; | |
1679 | int error; | |
1680 | ||
e9985d6a | 1681 | ovs_mutex_lock(&dp->port_mutex); |
72865317 | 1682 | error = get_port_by_number(dp, port_no, &port); |
4afba28d | 1683 | if (!error && dpif_port) { |
4c738a8d | 1684 | answer_port_query(port, dpif_port); |
72865317 | 1685 | } |
e9985d6a | 1686 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1687 | |
72865317 BP |
1688 | return error; |
1689 | } | |
1690 | ||
1691 | static int | |
1692 | dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname, | |
4c738a8d | 1693 | struct dpif_port *dpif_port) |
72865317 BP |
1694 | { |
1695 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1696 | struct dp_netdev_port *port; | |
1697 | int error; | |
1698 | ||
59e6d833 | 1699 | ovs_mutex_lock(&dp->port_mutex); |
72865317 | 1700 | error = get_port_by_name(dp, devname, &port); |
4afba28d | 1701 | if (!error && dpif_port) { |
4c738a8d | 1702 | answer_port_query(port, dpif_port); |
72865317 | 1703 | } |
59e6d833 | 1704 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1705 | |
72865317 BP |
1706 | return error; |
1707 | } | |
1708 | ||
61e7deb1 BP |
1709 | static void |
1710 | dp_netdev_flow_free(struct dp_netdev_flow *flow) | |
1711 | { | |
61e7deb1 | 1712 | dp_netdev_actions_free(dp_netdev_flow_get_actions(flow)); |
61e7deb1 BP |
1713 | free(flow); |
1714 | } | |
1715 | ||
ed79f89a DDP |
1716 | static void dp_netdev_flow_unref(struct dp_netdev_flow *flow) |
1717 | { | |
1718 | if (ovs_refcount_unref_relaxed(&flow->ref_cnt) == 1) { | |
1719 | ovsrcu_postpone(dp_netdev_flow_free, flow); | |
1720 | } | |
1721 | } | |
1722 | ||
70e5ed6f JS |
1723 | static uint32_t |
1724 | dp_netdev_flow_hash(const ovs_u128 *ufid) | |
1725 | { | |
1726 | return ufid->u32[0]; | |
1727 | } | |
1728 | ||
3453b4d6 JS |
1729 | static inline struct dpcls * |
1730 | dp_netdev_pmd_lookup_dpcls(struct dp_netdev_pmd_thread *pmd, | |
1731 | odp_port_t in_port) | |
1732 | { | |
1733 | struct dpcls *cls; | |
1734 | uint32_t hash = hash_port_no(in_port); | |
1735 | CMAP_FOR_EACH_WITH_HASH (cls, node, hash, &pmd->classifiers) { | |
1736 | if (cls->in_port == in_port) { | |
1737 | /* Port classifier exists already */ | |
1738 | return cls; | |
1739 | } | |
1740 | } | |
1741 | return NULL; | |
1742 | } | |
1743 | ||
1744 | static inline struct dpcls * | |
1745 | dp_netdev_pmd_find_dpcls(struct dp_netdev_pmd_thread *pmd, | |
1746 | odp_port_t in_port) | |
1747 | OVS_REQUIRES(pmd->flow_mutex) | |
1748 | { | |
1749 | struct dpcls *cls = dp_netdev_pmd_lookup_dpcls(pmd, in_port); | |
1750 | uint32_t hash = hash_port_no(in_port); | |
1751 | ||
1752 | if (!cls) { | |
1753 | /* Create new classifier for in_port */ | |
1754 | cls = xmalloc(sizeof(*cls)); | |
1755 | dpcls_init(cls); | |
1756 | cls->in_port = in_port; | |
1757 | cmap_insert(&pmd->classifiers, &cls->node, hash); | |
1758 | VLOG_DBG("Creating dpcls %p for in_port %d", cls, in_port); | |
1759 | } | |
1760 | return cls; | |
1761 | } | |
1762 | ||
72865317 | 1763 | static void |
1c1e46ed AW |
1764 | dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, |
1765 | struct dp_netdev_flow *flow) | |
1766 | OVS_REQUIRES(pmd->flow_mutex) | |
72865317 | 1767 | { |
9f361d6b | 1768 | struct cmap_node *node = CONST_CAST(struct cmap_node *, &flow->node); |
3453b4d6 JS |
1769 | struct dpcls *cls; |
1770 | odp_port_t in_port = flow->flow.in_port.odp_port; | |
2c0ea78f | 1771 | |
3453b4d6 JS |
1772 | cls = dp_netdev_pmd_lookup_dpcls(pmd, in_port); |
1773 | ovs_assert(cls != NULL); | |
1774 | dpcls_remove(cls, &flow->cr); | |
1c1e46ed | 1775 | cmap_remove(&pmd->flow_table, node, dp_netdev_flow_hash(&flow->ufid)); |
9bbf1c3d | 1776 | flow->dead = true; |
ed79f89a DDP |
1777 | |
1778 | dp_netdev_flow_unref(flow); | |
72865317 BP |
1779 | } |
1780 | ||
1781 | static void | |
1c1e46ed | 1782 | dp_netdev_pmd_flow_flush(struct dp_netdev_pmd_thread *pmd) |
72865317 | 1783 | { |
78c8df12 | 1784 | struct dp_netdev_flow *netdev_flow; |
72865317 | 1785 | |
1c1e46ed AW |
1786 | ovs_mutex_lock(&pmd->flow_mutex); |
1787 | CMAP_FOR_EACH (netdev_flow, node, &pmd->flow_table) { | |
1788 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); | |
72865317 | 1789 | } |
1c1e46ed | 1790 | ovs_mutex_unlock(&pmd->flow_mutex); |
72865317 BP |
1791 | } |
1792 | ||
1793 | static int | |
1794 | dpif_netdev_flow_flush(struct dpif *dpif) | |
1795 | { | |
1796 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1c1e46ed AW |
1797 | struct dp_netdev_pmd_thread *pmd; |
1798 | ||
1799 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
1800 | dp_netdev_pmd_flow_flush(pmd); | |
1801 | } | |
5279f8fd | 1802 | |
72865317 BP |
1803 | return 0; |
1804 | } | |
1805 | ||
b0ec0f27 | 1806 | struct dp_netdev_port_state { |
e9985d6a | 1807 | struct hmap_position position; |
4c738a8d | 1808 | char *name; |
b0ec0f27 BP |
1809 | }; |
1810 | ||
1811 | static int | |
1812 | dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) | |
1813 | { | |
1814 | *statep = xzalloc(sizeof(struct dp_netdev_port_state)); | |
1815 | return 0; | |
1816 | } | |
1817 | ||
72865317 | 1818 | static int |
b0ec0f27 | 1819 | dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_, |
4c738a8d | 1820 | struct dpif_port *dpif_port) |
72865317 | 1821 | { |
b0ec0f27 | 1822 | struct dp_netdev_port_state *state = state_; |
72865317 | 1823 | struct dp_netdev *dp = get_dp_netdev(dpif); |
e9985d6a | 1824 | struct hmap_node *node; |
ff073a71 | 1825 | int retval; |
72865317 | 1826 | |
e9985d6a DDP |
1827 | ovs_mutex_lock(&dp->port_mutex); |
1828 | node = hmap_at_position(&dp->ports, &state->position); | |
ff073a71 BP |
1829 | if (node) { |
1830 | struct dp_netdev_port *port; | |
5279f8fd | 1831 | |
ff073a71 BP |
1832 | port = CONTAINER_OF(node, struct dp_netdev_port, node); |
1833 | ||
1834 | free(state->name); | |
1835 | state->name = xstrdup(netdev_get_name(port->netdev)); | |
1836 | dpif_port->name = state->name; | |
1837 | dpif_port->type = port->type; | |
35303d71 | 1838 | dpif_port->port_no = port->port_no; |
ff073a71 BP |
1839 | |
1840 | retval = 0; | |
1841 | } else { | |
1842 | retval = EOF; | |
72865317 | 1843 | } |
e9985d6a | 1844 | ovs_mutex_unlock(&dp->port_mutex); |
5279f8fd | 1845 | |
ff073a71 | 1846 | return retval; |
b0ec0f27 BP |
1847 | } |
1848 | ||
1849 | static int | |
4c738a8d | 1850 | dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) |
b0ec0f27 | 1851 | { |
4c738a8d BP |
1852 | struct dp_netdev_port_state *state = state_; |
1853 | free(state->name); | |
b0ec0f27 BP |
1854 | free(state); |
1855 | return 0; | |
72865317 BP |
1856 | } |
1857 | ||
1858 | static int | |
67a4917b | 1859 | dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED) |
72865317 BP |
1860 | { |
1861 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
d33ed218 | 1862 | uint64_t new_port_seq; |
5279f8fd BP |
1863 | int error; |
1864 | ||
d33ed218 BP |
1865 | new_port_seq = seq_read(dpif->dp->port_seq); |
1866 | if (dpif->last_port_seq != new_port_seq) { | |
1867 | dpif->last_port_seq = new_port_seq; | |
5279f8fd | 1868 | error = ENOBUFS; |
72865317 | 1869 | } else { |
5279f8fd | 1870 | error = EAGAIN; |
72865317 | 1871 | } |
5279f8fd BP |
1872 | |
1873 | return error; | |
72865317 BP |
1874 | } |
1875 | ||
1876 | static void | |
1877 | dpif_netdev_port_poll_wait(const struct dpif *dpif_) | |
1878 | { | |
1879 | struct dpif_netdev *dpif = dpif_netdev_cast(dpif_); | |
5279f8fd | 1880 | |
d33ed218 | 1881 | seq_wait(dpif->dp->port_seq, dpif->last_port_seq); |
8a4e3a85 BP |
1882 | } |
1883 | ||
1884 | static struct dp_netdev_flow * | |
0de8783a | 1885 | dp_netdev_flow_cast(const struct dpcls_rule *cr) |
8a4e3a85 BP |
1886 | { |
1887 | return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL; | |
72865317 BP |
1888 | } |
1889 | ||
9bbf1c3d DDP |
1890 | static bool dp_netdev_flow_ref(struct dp_netdev_flow *flow) |
1891 | { | |
1892 | return ovs_refcount_try_ref_rcu(&flow->ref_cnt); | |
1893 | } | |
1894 | ||
79df317f DDP |
1895 | /* netdev_flow_key utilities. |
1896 | * | |
1897 | * netdev_flow_key is basically a miniflow. We use these functions | |
1898 | * (netdev_flow_key_clone, netdev_flow_key_equal, ...) instead of the miniflow | |
1899 | * functions (miniflow_clone_inline, miniflow_equal, ...), because: | |
1900 | * | |
1901 | * - Since we are dealing exclusively with miniflows created by | |
1902 | * miniflow_extract(), if the map is different the miniflow is different. | |
1903 | * Therefore we can be faster by comparing the map and the miniflow in a | |
1904 | * single memcmp(). | |
5fcff47b | 1905 | * - These functions can be inlined by the compiler. */ |
79df317f | 1906 | |
361d808d | 1907 | /* Given the number of bits set in miniflow's maps, returns the size of the |
caeb4906 | 1908 | * 'netdev_flow_key.mf' */ |
361d808d JR |
1909 | static inline size_t |
1910 | netdev_flow_key_size(size_t flow_u64s) | |
79df317f | 1911 | { |
361d808d | 1912 | return sizeof(struct miniflow) + MINIFLOW_VALUES_SIZE(flow_u64s); |
79df317f DDP |
1913 | } |
1914 | ||
79df317f DDP |
1915 | static inline bool |
1916 | netdev_flow_key_equal(const struct netdev_flow_key *a, | |
0de8783a JR |
1917 | const struct netdev_flow_key *b) |
1918 | { | |
caeb4906 JR |
1919 | /* 'b->len' may be not set yet. */ |
1920 | return a->hash == b->hash && !memcmp(&a->mf, &b->mf, a->len); | |
0de8783a JR |
1921 | } |
1922 | ||
1923 | /* Used to compare 'netdev_flow_key' in the exact match cache to a miniflow. | |
d79a39fe | 1924 | * The maps are compared bitwise, so both 'key->mf' and 'mf' must have been |
0de8783a JR |
1925 | * generated by miniflow_extract. */ |
1926 | static inline bool | |
1927 | netdev_flow_key_equal_mf(const struct netdev_flow_key *key, | |
1928 | const struct miniflow *mf) | |
79df317f | 1929 | { |
caeb4906 | 1930 | return !memcmp(&key->mf, mf, key->len); |
79df317f DDP |
1931 | } |
1932 | ||
1933 | static inline void | |
1934 | netdev_flow_key_clone(struct netdev_flow_key *dst, | |
0de8783a JR |
1935 | const struct netdev_flow_key *src) |
1936 | { | |
caeb4906 JR |
1937 | memcpy(dst, src, |
1938 | offsetof(struct netdev_flow_key, mf) + src->len); | |
0de8783a JR |
1939 | } |
1940 | ||
0de8783a JR |
1941 | /* Initialize a netdev_flow_key 'mask' from 'match'. */ |
1942 | static inline void | |
1943 | netdev_flow_mask_init(struct netdev_flow_key *mask, | |
1944 | const struct match *match) | |
1945 | { | |
09b0fa9c | 1946 | uint64_t *dst = miniflow_values(&mask->mf); |
5fcff47b | 1947 | struct flowmap fmap; |
0de8783a | 1948 | uint32_t hash = 0; |
5fcff47b | 1949 | size_t idx; |
0de8783a JR |
1950 | |
1951 | /* Only check masks that make sense for the flow. */ | |
5fcff47b JR |
1952 | flow_wc_map(&match->flow, &fmap); |
1953 | flowmap_init(&mask->mf.map); | |
0de8783a | 1954 | |
5fcff47b JR |
1955 | FLOWMAP_FOR_EACH_INDEX(idx, fmap) { |
1956 | uint64_t mask_u64 = flow_u64_value(&match->wc.masks, idx); | |
0de8783a | 1957 | |
5fcff47b JR |
1958 | if (mask_u64) { |
1959 | flowmap_set(&mask->mf.map, idx, 1); | |
1960 | *dst++ = mask_u64; | |
1961 | hash = hash_add64(hash, mask_u64); | |
0de8783a | 1962 | } |
0de8783a JR |
1963 | } |
1964 | ||
5fcff47b | 1965 | map_t map; |
0de8783a | 1966 | |
5fcff47b JR |
1967 | FLOWMAP_FOR_EACH_MAP (map, mask->mf.map) { |
1968 | hash = hash_add64(hash, map); | |
1969 | } | |
0de8783a | 1970 | |
5fcff47b | 1971 | size_t n = dst - miniflow_get_values(&mask->mf); |
0de8783a | 1972 | |
d70e8c28 | 1973 | mask->hash = hash_finish(hash, n * 8); |
0de8783a JR |
1974 | mask->len = netdev_flow_key_size(n); |
1975 | } | |
1976 | ||
361d808d | 1977 | /* Initializes 'dst' as a copy of 'flow' masked with 'mask'. */ |
0de8783a JR |
1978 | static inline void |
1979 | netdev_flow_key_init_masked(struct netdev_flow_key *dst, | |
1980 | const struct flow *flow, | |
1981 | const struct netdev_flow_key *mask) | |
79df317f | 1982 | { |
09b0fa9c JR |
1983 | uint64_t *dst_u64 = miniflow_values(&dst->mf); |
1984 | const uint64_t *mask_u64 = miniflow_get_values(&mask->mf); | |
0de8783a | 1985 | uint32_t hash = 0; |
d70e8c28 | 1986 | uint64_t value; |
0de8783a JR |
1987 | |
1988 | dst->len = mask->len; | |
361d808d | 1989 | dst->mf = mask->mf; /* Copy maps. */ |
0de8783a | 1990 | |
5fcff47b | 1991 | FLOW_FOR_EACH_IN_MAPS(value, flow, mask->mf.map) { |
d70e8c28 JR |
1992 | *dst_u64 = value & *mask_u64++; |
1993 | hash = hash_add64(hash, *dst_u64++); | |
0de8783a | 1994 | } |
09b0fa9c JR |
1995 | dst->hash = hash_finish(hash, |
1996 | (dst_u64 - miniflow_get_values(&dst->mf)) * 8); | |
0de8783a JR |
1997 | } |
1998 | ||
5fcff47b JR |
1999 | /* Iterate through netdev_flow_key TNL u64 values specified by 'FLOWMAP'. */ |
2000 | #define NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(VALUE, KEY, FLOWMAP) \ | |
2001 | MINIFLOW_FOR_EACH_IN_FLOWMAP(VALUE, &(KEY)->mf, FLOWMAP) | |
0de8783a JR |
2002 | |
2003 | /* Returns a hash value for the bits of 'key' where there are 1-bits in | |
2004 | * 'mask'. */ | |
2005 | static inline uint32_t | |
2006 | netdev_flow_key_hash_in_mask(const struct netdev_flow_key *key, | |
2007 | const struct netdev_flow_key *mask) | |
2008 | { | |
09b0fa9c | 2009 | const uint64_t *p = miniflow_get_values(&mask->mf); |
0de8783a | 2010 | uint32_t hash = 0; |
5fcff47b | 2011 | uint64_t value; |
0de8783a | 2012 | |
5fcff47b JR |
2013 | NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(value, key, mask->mf.map) { |
2014 | hash = hash_add64(hash, value & *p++); | |
0de8783a JR |
2015 | } |
2016 | ||
09b0fa9c | 2017 | return hash_finish(hash, (p - miniflow_get_values(&mask->mf)) * 8); |
79df317f DDP |
2018 | } |
2019 | ||
9bbf1c3d DDP |
2020 | static inline bool |
2021 | emc_entry_alive(struct emc_entry *ce) | |
2022 | { | |
2023 | return ce->flow && !ce->flow->dead; | |
2024 | } | |
2025 | ||
2026 | static void | |
2027 | emc_clear_entry(struct emc_entry *ce) | |
2028 | { | |
2029 | if (ce->flow) { | |
2030 | dp_netdev_flow_unref(ce->flow); | |
2031 | ce->flow = NULL; | |
2032 | } | |
2033 | } | |
2034 | ||
2035 | static inline void | |
2036 | emc_change_entry(struct emc_entry *ce, struct dp_netdev_flow *flow, | |
0de8783a | 2037 | const struct netdev_flow_key *key) |
9bbf1c3d DDP |
2038 | { |
2039 | if (ce->flow != flow) { | |
2040 | if (ce->flow) { | |
2041 | dp_netdev_flow_unref(ce->flow); | |
2042 | } | |
2043 | ||
2044 | if (dp_netdev_flow_ref(flow)) { | |
2045 | ce->flow = flow; | |
2046 | } else { | |
2047 | ce->flow = NULL; | |
2048 | } | |
2049 | } | |
0de8783a JR |
2050 | if (key) { |
2051 | netdev_flow_key_clone(&ce->key, key); | |
9bbf1c3d DDP |
2052 | } |
2053 | } | |
2054 | ||
2055 | static inline void | |
0de8783a | 2056 | emc_insert(struct emc_cache *cache, const struct netdev_flow_key *key, |
9bbf1c3d DDP |
2057 | struct dp_netdev_flow *flow) |
2058 | { | |
2059 | struct emc_entry *to_be_replaced = NULL; | |
2060 | struct emc_entry *current_entry; | |
2061 | ||
0de8783a JR |
2062 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
2063 | if (netdev_flow_key_equal(¤t_entry->key, key)) { | |
9bbf1c3d | 2064 | /* We found the entry with the 'mf' miniflow */ |
0de8783a | 2065 | emc_change_entry(current_entry, flow, NULL); |
9bbf1c3d DDP |
2066 | return; |
2067 | } | |
2068 | ||
2069 | /* Replacement policy: put the flow in an empty (not alive) entry, or | |
2070 | * in the first entry where it can be */ | |
2071 | if (!to_be_replaced | |
2072 | || (emc_entry_alive(to_be_replaced) | |
2073 | && !emc_entry_alive(current_entry)) | |
0de8783a | 2074 | || current_entry->key.hash < to_be_replaced->key.hash) { |
9bbf1c3d DDP |
2075 | to_be_replaced = current_entry; |
2076 | } | |
2077 | } | |
2078 | /* We didn't find the miniflow in the cache. | |
2079 | * The 'to_be_replaced' entry is where the new flow will be stored */ | |
2080 | ||
0de8783a | 2081 | emc_change_entry(to_be_replaced, flow, key); |
9bbf1c3d DDP |
2082 | } |
2083 | ||
4c30b246 CL |
2084 | static inline void |
2085 | emc_probabilistic_insert(struct dp_netdev_pmd_thread *pmd, | |
2086 | const struct netdev_flow_key *key, | |
2087 | struct dp_netdev_flow *flow) | |
2088 | { | |
2089 | /* Insert an entry into the EMC based on probability value 'min'. By | |
2090 | * default the value is UINT32_MAX / 100 which yields an insertion | |
2091 | * probability of 1/100 ie. 1% */ | |
2092 | ||
2093 | uint32_t min; | |
2094 | atomic_read_relaxed(&pmd->dp->emc_insert_min, &min); | |
2095 | ||
656238ee | 2096 | if (min && random_uint32() <= min) { |
4c30b246 CL |
2097 | emc_insert(&pmd->flow_cache, key, flow); |
2098 | } | |
2099 | } | |
2100 | ||
9bbf1c3d | 2101 | static inline struct dp_netdev_flow * |
0de8783a | 2102 | emc_lookup(struct emc_cache *cache, const struct netdev_flow_key *key) |
9bbf1c3d DDP |
2103 | { |
2104 | struct emc_entry *current_entry; | |
2105 | ||
0de8783a JR |
2106 | EMC_FOR_EACH_POS_WITH_HASH(cache, current_entry, key->hash) { |
2107 | if (current_entry->key.hash == key->hash | |
2108 | && emc_entry_alive(current_entry) | |
2109 | && netdev_flow_key_equal_mf(¤t_entry->key, &key->mf)) { | |
9bbf1c3d | 2110 | |
0de8783a | 2111 | /* We found the entry with the 'key->mf' miniflow */ |
9bbf1c3d DDP |
2112 | return current_entry->flow; |
2113 | } | |
2114 | } | |
2115 | ||
2116 | return NULL; | |
2117 | } | |
2118 | ||
72865317 | 2119 | static struct dp_netdev_flow * |
3453b4d6 JS |
2120 | dp_netdev_pmd_lookup_flow(struct dp_netdev_pmd_thread *pmd, |
2121 | const struct netdev_flow_key *key, | |
2122 | int *lookup_num_p) | |
2c0ea78f | 2123 | { |
3453b4d6 | 2124 | struct dpcls *cls; |
0de8783a | 2125 | struct dpcls_rule *rule; |
3453b4d6 JS |
2126 | odp_port_t in_port = u32_to_odp(MINIFLOW_GET_U32(&key->mf, in_port)); |
2127 | struct dp_netdev_flow *netdev_flow = NULL; | |
2c0ea78f | 2128 | |
3453b4d6 JS |
2129 | cls = dp_netdev_pmd_lookup_dpcls(pmd, in_port); |
2130 | if (OVS_LIKELY(cls)) { | |
2131 | dpcls_lookup(cls, key, &rule, 1, lookup_num_p); | |
2132 | netdev_flow = dp_netdev_flow_cast(rule); | |
2133 | } | |
8a4e3a85 | 2134 | return netdev_flow; |
2c0ea78f GS |
2135 | } |
2136 | ||
2137 | static struct dp_netdev_flow * | |
1c1e46ed AW |
2138 | dp_netdev_pmd_find_flow(const struct dp_netdev_pmd_thread *pmd, |
2139 | const ovs_u128 *ufidp, const struct nlattr *key, | |
2140 | size_t key_len) | |
72865317 | 2141 | { |
1763b4b8 | 2142 | struct dp_netdev_flow *netdev_flow; |
70e5ed6f JS |
2143 | struct flow flow; |
2144 | ovs_u128 ufid; | |
2145 | ||
2146 | /* If a UFID is not provided, determine one based on the key. */ | |
2147 | if (!ufidp && key && key_len | |
f0fb825a | 2148 | && !dpif_netdev_flow_from_nlattrs(key, key_len, &flow, false)) { |
1c1e46ed | 2149 | dpif_flow_hash(pmd->dp->dpif, &flow, sizeof flow, &ufid); |
70e5ed6f JS |
2150 | ufidp = &ufid; |
2151 | } | |
72865317 | 2152 | |
70e5ed6f JS |
2153 | if (ufidp) { |
2154 | CMAP_FOR_EACH_WITH_HASH (netdev_flow, node, dp_netdev_flow_hash(ufidp), | |
1c1e46ed | 2155 | &pmd->flow_table) { |
2ff8484b | 2156 | if (ovs_u128_equals(netdev_flow->ufid, *ufidp)) { |
70e5ed6f JS |
2157 | return netdev_flow; |
2158 | } | |
72865317 BP |
2159 | } |
2160 | } | |
8a4e3a85 | 2161 | |
72865317 BP |
2162 | return NULL; |
2163 | } | |
2164 | ||
2165 | static void | |
eb94da30 | 2166 | get_dpif_flow_stats(const struct dp_netdev_flow *netdev_flow_, |
1763b4b8 | 2167 | struct dpif_flow_stats *stats) |
feebdea2 | 2168 | { |
eb94da30 DDP |
2169 | struct dp_netdev_flow *netdev_flow; |
2170 | unsigned long long n; | |
2171 | long long used; | |
2172 | uint16_t flags; | |
2173 | ||
2174 | netdev_flow = CONST_CAST(struct dp_netdev_flow *, netdev_flow_); | |
2175 | ||
2176 | atomic_read_relaxed(&netdev_flow->stats.packet_count, &n); | |
2177 | stats->n_packets = n; | |
2178 | atomic_read_relaxed(&netdev_flow->stats.byte_count, &n); | |
2179 | stats->n_bytes = n; | |
2180 | atomic_read_relaxed(&netdev_flow->stats.used, &used); | |
2181 | stats->used = used; | |
2182 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
2183 | stats->tcp_flags = flags; | |
72865317 BP |
2184 | } |
2185 | ||
7af12bd7 JS |
2186 | /* Converts to the dpif_flow format, using 'key_buf' and 'mask_buf' for |
2187 | * storing the netlink-formatted key/mask. 'key_buf' may be the same as | |
2188 | * 'mask_buf'. Actions will be returned without copying, by relying on RCU to | |
2189 | * protect them. */ | |
6fe09f8c | 2190 | static void |
70e5ed6f | 2191 | dp_netdev_flow_to_dpif_flow(const struct dp_netdev_flow *netdev_flow, |
7af12bd7 | 2192 | struct ofpbuf *key_buf, struct ofpbuf *mask_buf, |
64bb477f | 2193 | struct dpif_flow *flow, bool terse) |
6fe09f8c | 2194 | { |
64bb477f JS |
2195 | if (terse) { |
2196 | memset(flow, 0, sizeof *flow); | |
2197 | } else { | |
2198 | struct flow_wildcards wc; | |
2199 | struct dp_netdev_actions *actions; | |
2200 | size_t offset; | |
5262eea1 JG |
2201 | struct odp_flow_key_parms odp_parms = { |
2202 | .flow = &netdev_flow->flow, | |
2203 | .mask = &wc.masks, | |
2494ccd7 | 2204 | .support = dp_netdev_support, |
5262eea1 | 2205 | }; |
64bb477f JS |
2206 | |
2207 | miniflow_expand(&netdev_flow->cr.mask->mf, &wc.masks); | |
f4b835bb JR |
2208 | /* in_port is exact matched, but we have left it out from the mask for |
2209 | * optimnization reasons. Add in_port back to the mask. */ | |
2210 | wc.masks.in_port.odp_port = ODPP_NONE; | |
64bb477f JS |
2211 | |
2212 | /* Key */ | |
6fd6ed71 | 2213 | offset = key_buf->size; |
64bb477f | 2214 | flow->key = ofpbuf_tail(key_buf); |
5262eea1 | 2215 | odp_flow_key_from_flow(&odp_parms, key_buf); |
6fd6ed71 | 2216 | flow->key_len = key_buf->size - offset; |
64bb477f JS |
2217 | |
2218 | /* Mask */ | |
6fd6ed71 | 2219 | offset = mask_buf->size; |
64bb477f | 2220 | flow->mask = ofpbuf_tail(mask_buf); |
ec1f6f32 | 2221 | odp_parms.key_buf = key_buf; |
5262eea1 | 2222 | odp_flow_key_from_mask(&odp_parms, mask_buf); |
6fd6ed71 | 2223 | flow->mask_len = mask_buf->size - offset; |
64bb477f JS |
2224 | |
2225 | /* Actions */ | |
2226 | actions = dp_netdev_flow_get_actions(netdev_flow); | |
2227 | flow->actions = actions->actions; | |
2228 | flow->actions_len = actions->size; | |
2229 | } | |
6fe09f8c | 2230 | |
70e5ed6f JS |
2231 | flow->ufid = netdev_flow->ufid; |
2232 | flow->ufid_present = true; | |
1c1e46ed | 2233 | flow->pmd_id = netdev_flow->pmd_id; |
6fe09f8c JS |
2234 | get_dpif_flow_stats(netdev_flow, &flow->stats); |
2235 | } | |
2236 | ||
36956a7d | 2237 | static int |
8c301900 JR |
2238 | dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len, |
2239 | const struct nlattr *mask_key, | |
2240 | uint32_t mask_key_len, const struct flow *flow, | |
f0fb825a | 2241 | struct flow_wildcards *wc, bool probe) |
8c301900 | 2242 | { |
ca8d3442 DDP |
2243 | enum odp_key_fitness fitness; |
2244 | ||
8d8ab6c2 | 2245 | fitness = odp_flow_key_to_mask(mask_key, mask_key_len, wc, flow); |
ca8d3442 | 2246 | if (fitness) { |
f0fb825a EG |
2247 | if (!probe) { |
2248 | /* This should not happen: it indicates that | |
2249 | * odp_flow_key_from_mask() and odp_flow_key_to_mask() | |
2250 | * disagree on the acceptable form of a mask. Log the problem | |
2251 | * as an error, with enough details to enable debugging. */ | |
2252 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2253 | ||
2254 | if (!VLOG_DROP_ERR(&rl)) { | |
2255 | struct ds s; | |
2256 | ||
2257 | ds_init(&s); | |
2258 | odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s, | |
2259 | true); | |
2260 | VLOG_ERR("internal error parsing flow mask %s (%s)", | |
2261 | ds_cstr(&s), odp_key_fitness_to_string(fitness)); | |
2262 | ds_destroy(&s); | |
2263 | } | |
8c301900 | 2264 | } |
ca8d3442 DDP |
2265 | |
2266 | return EINVAL; | |
8c301900 JR |
2267 | } |
2268 | ||
2269 | return 0; | |
2270 | } | |
2271 | ||
2272 | static int | |
2273 | dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len, | |
f0fb825a | 2274 | struct flow *flow, bool probe) |
36956a7d | 2275 | { |
8d8ab6c2 | 2276 | if (odp_flow_key_to_flow(key, key_len, flow)) { |
f0fb825a EG |
2277 | if (!probe) { |
2278 | /* This should not happen: it indicates that | |
2279 | * odp_flow_key_from_flow() and odp_flow_key_to_flow() disagree on | |
2280 | * the acceptable form of a flow. Log the problem as an error, | |
2281 | * with enough details to enable debugging. */ | |
2282 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2283 | ||
2284 | if (!VLOG_DROP_ERR(&rl)) { | |
2285 | struct ds s; | |
2286 | ||
2287 | ds_init(&s); | |
2288 | odp_flow_format(key, key_len, NULL, 0, NULL, &s, true); | |
2289 | VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s)); | |
2290 | ds_destroy(&s); | |
2291 | } | |
36956a7d BP |
2292 | } |
2293 | ||
2294 | return EINVAL; | |
2295 | } | |
2296 | ||
5cf3edb3 | 2297 | if (flow->ct_state & DP_NETDEV_CS_UNSUPPORTED_MASK) { |
07659514 JS |
2298 | return EINVAL; |
2299 | } | |
2300 | ||
36956a7d BP |
2301 | return 0; |
2302 | } | |
2303 | ||
72865317 | 2304 | static int |
6fe09f8c | 2305 | dpif_netdev_flow_get(const struct dpif *dpif, const struct dpif_flow_get *get) |
72865317 BP |
2306 | { |
2307 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
1763b4b8 | 2308 | struct dp_netdev_flow *netdev_flow; |
1c1e46ed | 2309 | struct dp_netdev_pmd_thread *pmd; |
c673049c IM |
2310 | struct hmapx to_find = HMAPX_INITIALIZER(&to_find); |
2311 | struct hmapx_node *node; | |
2312 | int error = EINVAL; | |
2313 | ||
2314 | if (get->pmd_id == PMD_ID_NULL) { | |
2315 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2316 | if (dp_netdev_pmd_try_ref(pmd) && !hmapx_add(&to_find, pmd)) { | |
2317 | dp_netdev_pmd_unref(pmd); | |
2318 | } | |
2319 | } | |
2320 | } else { | |
2321 | pmd = dp_netdev_get_pmd(dp, get->pmd_id); | |
2322 | if (!pmd) { | |
2323 | goto out; | |
2324 | } | |
2325 | hmapx_add(&to_find, pmd); | |
1c1e46ed AW |
2326 | } |
2327 | ||
c673049c IM |
2328 | if (!hmapx_count(&to_find)) { |
2329 | goto out; | |
72865317 | 2330 | } |
1c1e46ed | 2331 | |
c673049c IM |
2332 | HMAPX_FOR_EACH (node, &to_find) { |
2333 | pmd = (struct dp_netdev_pmd_thread *) node->data; | |
2334 | netdev_flow = dp_netdev_pmd_find_flow(pmd, get->ufid, get->key, | |
2335 | get->key_len); | |
2336 | if (netdev_flow) { | |
2337 | dp_netdev_flow_to_dpif_flow(netdev_flow, get->buffer, get->buffer, | |
2338 | get->flow, false); | |
2339 | error = 0; | |
2340 | break; | |
2341 | } else { | |
2342 | error = ENOENT; | |
2343 | } | |
2344 | } | |
bc4a05c6 | 2345 | |
c673049c IM |
2346 | HMAPX_FOR_EACH (node, &to_find) { |
2347 | pmd = (struct dp_netdev_pmd_thread *) node->data; | |
2348 | dp_netdev_pmd_unref(pmd); | |
2349 | } | |
2350 | out: | |
2351 | hmapx_destroy(&to_find); | |
5279f8fd | 2352 | return error; |
72865317 BP |
2353 | } |
2354 | ||
0de8783a | 2355 | static struct dp_netdev_flow * |
1c1e46ed AW |
2356 | dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd, |
2357 | struct match *match, const ovs_u128 *ufid, | |
ae2ceebd | 2358 | const struct nlattr *actions, size_t actions_len) |
1c1e46ed | 2359 | OVS_REQUIRES(pmd->flow_mutex) |
72865317 | 2360 | { |
0de8783a JR |
2361 | struct dp_netdev_flow *flow; |
2362 | struct netdev_flow_key mask; | |
3453b4d6 | 2363 | struct dpcls *cls; |
f4b835bb JR |
2364 | |
2365 | /* Make sure in_port is exact matched before we read it. */ | |
2366 | ovs_assert(match->wc.masks.in_port.odp_port == ODPP_NONE); | |
3453b4d6 | 2367 | odp_port_t in_port = match->flow.in_port.odp_port; |
ed79f89a | 2368 | |
f4b835bb JR |
2369 | /* As we select the dpcls based on the port number, each netdev flow |
2370 | * belonging to the same dpcls will have the same odp_port value. | |
2371 | * For performance reasons we wildcard odp_port here in the mask. In the | |
2372 | * typical case dp_hash is also wildcarded, and the resulting 8-byte | |
2373 | * chunk {dp_hash, in_port} will be ignored by netdev_flow_mask_init() and | |
2374 | * will not be part of the subtable mask. | |
2375 | * This will speed up the hash computation during dpcls_lookup() because | |
2376 | * there is one less call to hash_add64() in this case. */ | |
2377 | match->wc.masks.in_port.odp_port = 0; | |
0de8783a | 2378 | netdev_flow_mask_init(&mask, match); |
f4b835bb JR |
2379 | match->wc.masks.in_port.odp_port = ODPP_NONE; |
2380 | ||
0de8783a | 2381 | /* Make sure wc does not have metadata. */ |
5fcff47b JR |
2382 | ovs_assert(!FLOWMAP_HAS_FIELD(&mask.mf.map, metadata) |
2383 | && !FLOWMAP_HAS_FIELD(&mask.mf.map, regs)); | |
679ba04c | 2384 | |
0de8783a | 2385 | /* Do not allocate extra space. */ |
caeb4906 | 2386 | flow = xmalloc(sizeof *flow - sizeof flow->cr.flow.mf + mask.len); |
1c1e46ed | 2387 | memset(&flow->stats, 0, sizeof flow->stats); |
0de8783a | 2388 | flow->dead = false; |
11e5cf1f | 2389 | flow->batch = NULL; |
bd5131ba | 2390 | *CONST_CAST(unsigned *, &flow->pmd_id) = pmd->core_id; |
0de8783a | 2391 | *CONST_CAST(struct flow *, &flow->flow) = match->flow; |
70e5ed6f | 2392 | *CONST_CAST(ovs_u128 *, &flow->ufid) = *ufid; |
0de8783a | 2393 | ovs_refcount_init(&flow->ref_cnt); |
0de8783a | 2394 | ovsrcu_set(&flow->actions, dp_netdev_actions_create(actions, actions_len)); |
2c0ea78f | 2395 | |
0de8783a | 2396 | netdev_flow_key_init_masked(&flow->cr.flow, &match->flow, &mask); |
3453b4d6 | 2397 | |
f4b835bb | 2398 | /* Select dpcls for in_port. Relies on in_port to be exact match. */ |
3453b4d6 JS |
2399 | cls = dp_netdev_pmd_find_dpcls(pmd, in_port); |
2400 | dpcls_insert(cls, &flow->cr, &mask); | |
72865317 | 2401 | |
4c75aaab EJ |
2402 | cmap_insert(&pmd->flow_table, CONST_CAST(struct cmap_node *, &flow->node), |
2403 | dp_netdev_flow_hash(&flow->ufid)); | |
2404 | ||
beb75a40 | 2405 | if (OVS_UNLIKELY(!VLOG_DROP_DBG((&upcall_rl)))) { |
623540e4 | 2406 | struct ds ds = DS_EMPTY_INITIALIZER; |
9044f2c1 JG |
2407 | struct ofpbuf key_buf, mask_buf; |
2408 | struct odp_flow_key_parms odp_parms = { | |
2409 | .flow = &match->flow, | |
2410 | .mask = &match->wc.masks, | |
2411 | .support = dp_netdev_support, | |
2412 | }; | |
2413 | ||
2414 | ofpbuf_init(&key_buf, 0); | |
2415 | ofpbuf_init(&mask_buf, 0); | |
623540e4 | 2416 | |
9044f2c1 JG |
2417 | odp_flow_key_from_flow(&odp_parms, &key_buf); |
2418 | odp_parms.key_buf = &key_buf; | |
2419 | odp_flow_key_from_mask(&odp_parms, &mask_buf); | |
0de8783a | 2420 | |
623540e4 | 2421 | ds_put_cstr(&ds, "flow_add: "); |
70e5ed6f JS |
2422 | odp_format_ufid(ufid, &ds); |
2423 | ds_put_cstr(&ds, " "); | |
9044f2c1 JG |
2424 | odp_flow_format(key_buf.data, key_buf.size, |
2425 | mask_buf.data, mask_buf.size, | |
2426 | NULL, &ds, false); | |
623540e4 | 2427 | ds_put_cstr(&ds, ", actions:"); |
0722f341 | 2428 | format_odp_actions(&ds, actions, actions_len, NULL); |
623540e4 | 2429 | |
beb75a40 | 2430 | VLOG_DBG("%s", ds_cstr(&ds)); |
623540e4 | 2431 | |
9044f2c1 JG |
2432 | ofpbuf_uninit(&key_buf); |
2433 | ofpbuf_uninit(&mask_buf); | |
beb75a40 JS |
2434 | |
2435 | /* Add a printout of the actual match installed. */ | |
2436 | struct match m; | |
2437 | ds_clear(&ds); | |
2438 | ds_put_cstr(&ds, "flow match: "); | |
2439 | miniflow_expand(&flow->cr.flow.mf, &m.flow); | |
2440 | miniflow_expand(&flow->cr.mask->mf, &m.wc.masks); | |
b2f4b622 | 2441 | memset(&m.tun_md, 0, sizeof m.tun_md); |
beb75a40 JS |
2442 | match_format(&m, NULL, &ds, OFP_DEFAULT_PRIORITY); |
2443 | ||
2444 | VLOG_DBG("%s", ds_cstr(&ds)); | |
2445 | ||
623540e4 EJ |
2446 | ds_destroy(&ds); |
2447 | } | |
2448 | ||
0de8783a | 2449 | return flow; |
72865317 BP |
2450 | } |
2451 | ||
72865317 | 2452 | static int |
f5d317a1 DDP |
2453 | flow_put_on_pmd(struct dp_netdev_pmd_thread *pmd, |
2454 | struct netdev_flow_key *key, | |
2455 | struct match *match, | |
2456 | ovs_u128 *ufid, | |
2457 | const struct dpif_flow_put *put, | |
2458 | struct dpif_flow_stats *stats) | |
72865317 | 2459 | { |
1763b4b8 | 2460 | struct dp_netdev_flow *netdev_flow; |
f5d317a1 | 2461 | int error = 0; |
72865317 | 2462 | |
f5d317a1 DDP |
2463 | if (stats) { |
2464 | memset(stats, 0, sizeof *stats); | |
70e5ed6f JS |
2465 | } |
2466 | ||
1c1e46ed | 2467 | ovs_mutex_lock(&pmd->flow_mutex); |
f5d317a1 | 2468 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, key, NULL); |
1763b4b8 | 2469 | if (!netdev_flow) { |
89625d1e | 2470 | if (put->flags & DPIF_FP_CREATE) { |
1c1e46ed | 2471 | if (cmap_count(&pmd->flow_table) < MAX_FLOWS) { |
f5d317a1 | 2472 | dp_netdev_flow_add(pmd, match, ufid, put->actions, |
70e5ed6f | 2473 | put->actions_len); |
0de8783a | 2474 | error = 0; |
72865317 | 2475 | } else { |
5279f8fd | 2476 | error = EFBIG; |
72865317 BP |
2477 | } |
2478 | } else { | |
5279f8fd | 2479 | error = ENOENT; |
72865317 BP |
2480 | } |
2481 | } else { | |
beb75a40 | 2482 | if (put->flags & DPIF_FP_MODIFY) { |
8a4e3a85 BP |
2483 | struct dp_netdev_actions *new_actions; |
2484 | struct dp_netdev_actions *old_actions; | |
2485 | ||
2486 | new_actions = dp_netdev_actions_create(put->actions, | |
2487 | put->actions_len); | |
2488 | ||
61e7deb1 BP |
2489 | old_actions = dp_netdev_flow_get_actions(netdev_flow); |
2490 | ovsrcu_set(&netdev_flow->actions, new_actions); | |
679ba04c | 2491 | |
f5d317a1 DDP |
2492 | if (stats) { |
2493 | get_dpif_flow_stats(netdev_flow, stats); | |
a84cb64a BP |
2494 | } |
2495 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
97447f55 DDP |
2496 | /* XXX: The userspace datapath uses thread local statistics |
2497 | * (for flows), which should be updated only by the owning | |
2498 | * thread. Since we cannot write on stats memory here, | |
2499 | * we choose not to support this flag. Please note: | |
2500 | * - This feature is currently used only by dpctl commands with | |
2501 | * option --clear. | |
2502 | * - Should the need arise, this operation can be implemented | |
2503 | * by keeping a base value (to be update here) for each | |
2504 | * counter, and subtracting it before outputting the stats */ | |
2505 | error = EOPNOTSUPP; | |
72865317 | 2506 | } |
8a4e3a85 | 2507 | |
61e7deb1 | 2508 | ovsrcu_postpone(dp_netdev_actions_free, old_actions); |
2c0ea78f | 2509 | } else if (put->flags & DPIF_FP_CREATE) { |
5279f8fd | 2510 | error = EEXIST; |
2c0ea78f GS |
2511 | } else { |
2512 | /* Overlapping flow. */ | |
2513 | error = EINVAL; | |
72865317 BP |
2514 | } |
2515 | } | |
1c1e46ed | 2516 | ovs_mutex_unlock(&pmd->flow_mutex); |
5279f8fd | 2517 | return error; |
72865317 BP |
2518 | } |
2519 | ||
72865317 | 2520 | static int |
f5d317a1 | 2521 | dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put) |
72865317 BP |
2522 | { |
2523 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
beb75a40 | 2524 | struct netdev_flow_key key, mask; |
1c1e46ed | 2525 | struct dp_netdev_pmd_thread *pmd; |
f5d317a1 DDP |
2526 | struct match match; |
2527 | ovs_u128 ufid; | |
2528 | int error; | |
f0fb825a | 2529 | bool probe = put->flags & DPIF_FP_PROBE; |
72865317 | 2530 | |
f5d317a1 DDP |
2531 | if (put->stats) { |
2532 | memset(put->stats, 0, sizeof *put->stats); | |
2533 | } | |
f0fb825a EG |
2534 | error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &match.flow, |
2535 | probe); | |
f5d317a1 DDP |
2536 | if (error) { |
2537 | return error; | |
2538 | } | |
2539 | error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len, | |
2540 | put->mask, put->mask_len, | |
f0fb825a | 2541 | &match.flow, &match.wc, probe); |
f5d317a1 DDP |
2542 | if (error) { |
2543 | return error; | |
1c1e46ed AW |
2544 | } |
2545 | ||
f5d317a1 DDP |
2546 | if (put->ufid) { |
2547 | ufid = *put->ufid; | |
2548 | } else { | |
2549 | dpif_flow_hash(dpif, &match.flow, sizeof match.flow, &ufid); | |
2550 | } | |
2551 | ||
2552 | /* Must produce a netdev_flow_key for lookup. | |
beb75a40 JS |
2553 | * Use the same method as employed to create the key when adding |
2554 | * the flow to the dplcs to make sure they match. */ | |
2555 | netdev_flow_mask_init(&mask, &match); | |
2556 | netdev_flow_key_init_masked(&key, &match.flow, &mask); | |
f5d317a1 DDP |
2557 | |
2558 | if (put->pmd_id == PMD_ID_NULL) { | |
2559 | if (cmap_count(&dp->poll_threads) == 0) { | |
2560 | return EINVAL; | |
2561 | } | |
2562 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2563 | struct dpif_flow_stats pmd_stats; | |
2564 | int pmd_error; | |
2565 | ||
2566 | pmd_error = flow_put_on_pmd(pmd, &key, &match, &ufid, put, | |
2567 | &pmd_stats); | |
2568 | if (pmd_error) { | |
2569 | error = pmd_error; | |
2570 | } else if (put->stats) { | |
2571 | put->stats->n_packets += pmd_stats.n_packets; | |
2572 | put->stats->n_bytes += pmd_stats.n_bytes; | |
2573 | put->stats->used = MAX(put->stats->used, pmd_stats.used); | |
2574 | put->stats->tcp_flags |= pmd_stats.tcp_flags; | |
2575 | } | |
2576 | } | |
2577 | } else { | |
2578 | pmd = dp_netdev_get_pmd(dp, put->pmd_id); | |
2579 | if (!pmd) { | |
2580 | return EINVAL; | |
2581 | } | |
2582 | error = flow_put_on_pmd(pmd, &key, &match, &ufid, put, put->stats); | |
2583 | dp_netdev_pmd_unref(pmd); | |
2584 | } | |
2585 | ||
2586 | return error; | |
2587 | } | |
2588 | ||
2589 | static int | |
2590 | flow_del_on_pmd(struct dp_netdev_pmd_thread *pmd, | |
2591 | struct dpif_flow_stats *stats, | |
2592 | const struct dpif_flow_del *del) | |
2593 | { | |
2594 | struct dp_netdev_flow *netdev_flow; | |
2595 | int error = 0; | |
2596 | ||
1c1e46ed AW |
2597 | ovs_mutex_lock(&pmd->flow_mutex); |
2598 | netdev_flow = dp_netdev_pmd_find_flow(pmd, del->ufid, del->key, | |
2599 | del->key_len); | |
1763b4b8 | 2600 | if (netdev_flow) { |
f5d317a1 DDP |
2601 | if (stats) { |
2602 | get_dpif_flow_stats(netdev_flow, stats); | |
feebdea2 | 2603 | } |
1c1e46ed | 2604 | dp_netdev_pmd_remove_flow(pmd, netdev_flow); |
72865317 | 2605 | } else { |
5279f8fd | 2606 | error = ENOENT; |
72865317 | 2607 | } |
1c1e46ed | 2608 | ovs_mutex_unlock(&pmd->flow_mutex); |
f5d317a1 DDP |
2609 | |
2610 | return error; | |
2611 | } | |
2612 | ||
2613 | static int | |
2614 | dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del) | |
2615 | { | |
2616 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2617 | struct dp_netdev_pmd_thread *pmd; | |
2618 | int error = 0; | |
2619 | ||
2620 | if (del->stats) { | |
2621 | memset(del->stats, 0, sizeof *del->stats); | |
2622 | } | |
2623 | ||
2624 | if (del->pmd_id == PMD_ID_NULL) { | |
2625 | if (cmap_count(&dp->poll_threads) == 0) { | |
2626 | return EINVAL; | |
2627 | } | |
2628 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
2629 | struct dpif_flow_stats pmd_stats; | |
2630 | int pmd_error; | |
2631 | ||
2632 | pmd_error = flow_del_on_pmd(pmd, &pmd_stats, del); | |
2633 | if (pmd_error) { | |
2634 | error = pmd_error; | |
2635 | } else if (del->stats) { | |
2636 | del->stats->n_packets += pmd_stats.n_packets; | |
2637 | del->stats->n_bytes += pmd_stats.n_bytes; | |
2638 | del->stats->used = MAX(del->stats->used, pmd_stats.used); | |
2639 | del->stats->tcp_flags |= pmd_stats.tcp_flags; | |
2640 | } | |
2641 | } | |
2642 | } else { | |
2643 | pmd = dp_netdev_get_pmd(dp, del->pmd_id); | |
2644 | if (!pmd) { | |
2645 | return EINVAL; | |
2646 | } | |
2647 | error = flow_del_on_pmd(pmd, del->stats, del); | |
2648 | dp_netdev_pmd_unref(pmd); | |
2649 | } | |
2650 | ||
5279f8fd BP |
2651 | |
2652 | return error; | |
72865317 BP |
2653 | } |
2654 | ||
ac64794a BP |
2655 | struct dpif_netdev_flow_dump { |
2656 | struct dpif_flow_dump up; | |
1c1e46ed AW |
2657 | struct cmap_position poll_thread_pos; |
2658 | struct cmap_position flow_pos; | |
2659 | struct dp_netdev_pmd_thread *cur_pmd; | |
d2ad7ef1 JS |
2660 | int status; |
2661 | struct ovs_mutex mutex; | |
e723fd32 JS |
2662 | }; |
2663 | ||
ac64794a BP |
2664 | static struct dpif_netdev_flow_dump * |
2665 | dpif_netdev_flow_dump_cast(struct dpif_flow_dump *dump) | |
72865317 | 2666 | { |
ac64794a | 2667 | return CONTAINER_OF(dump, struct dpif_netdev_flow_dump, up); |
e723fd32 JS |
2668 | } |
2669 | ||
ac64794a | 2670 | static struct dpif_flow_dump * |
7e8b7199 PB |
2671 | dpif_netdev_flow_dump_create(const struct dpif *dpif_, bool terse, |
2672 | char *type OVS_UNUSED) | |
e723fd32 | 2673 | { |
ac64794a | 2674 | struct dpif_netdev_flow_dump *dump; |
e723fd32 | 2675 | |
1c1e46ed | 2676 | dump = xzalloc(sizeof *dump); |
ac64794a | 2677 | dpif_flow_dump_init(&dump->up, dpif_); |
64bb477f | 2678 | dump->up.terse = terse; |
ac64794a BP |
2679 | ovs_mutex_init(&dump->mutex); |
2680 | ||
2681 | return &dump->up; | |
e723fd32 JS |
2682 | } |
2683 | ||
2684 | static int | |
ac64794a | 2685 | dpif_netdev_flow_dump_destroy(struct dpif_flow_dump *dump_) |
e723fd32 | 2686 | { |
ac64794a | 2687 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); |
e723fd32 | 2688 | |
ac64794a BP |
2689 | ovs_mutex_destroy(&dump->mutex); |
2690 | free(dump); | |
704a1e09 BP |
2691 | return 0; |
2692 | } | |
2693 | ||
ac64794a BP |
2694 | struct dpif_netdev_flow_dump_thread { |
2695 | struct dpif_flow_dump_thread up; | |
2696 | struct dpif_netdev_flow_dump *dump; | |
8bb113da RW |
2697 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; |
2698 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
2699 | }; |
2700 | ||
2701 | static struct dpif_netdev_flow_dump_thread * | |
2702 | dpif_netdev_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
2703 | { | |
2704 | return CONTAINER_OF(thread, struct dpif_netdev_flow_dump_thread, up); | |
2705 | } | |
2706 | ||
2707 | static struct dpif_flow_dump_thread * | |
2708 | dpif_netdev_flow_dump_thread_create(struct dpif_flow_dump *dump_) | |
2709 | { | |
2710 | struct dpif_netdev_flow_dump *dump = dpif_netdev_flow_dump_cast(dump_); | |
2711 | struct dpif_netdev_flow_dump_thread *thread; | |
2712 | ||
2713 | thread = xmalloc(sizeof *thread); | |
2714 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
2715 | thread->dump = dump; | |
2716 | return &thread->up; | |
2717 | } | |
2718 | ||
2719 | static void | |
2720 | dpif_netdev_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) | |
2721 | { | |
2722 | struct dpif_netdev_flow_dump_thread *thread | |
2723 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2724 | ||
2725 | free(thread); | |
2726 | } | |
2727 | ||
704a1e09 | 2728 | static int |
ac64794a | 2729 | dpif_netdev_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
8bb113da | 2730 | struct dpif_flow *flows, int max_flows) |
ac64794a BP |
2731 | { |
2732 | struct dpif_netdev_flow_dump_thread *thread | |
2733 | = dpif_netdev_flow_dump_thread_cast(thread_); | |
2734 | struct dpif_netdev_flow_dump *dump = thread->dump; | |
8bb113da | 2735 | struct dp_netdev_flow *netdev_flows[FLOW_DUMP_MAX_BATCH]; |
8bb113da RW |
2736 | int n_flows = 0; |
2737 | int i; | |
14608a15 | 2738 | |
ac64794a | 2739 | ovs_mutex_lock(&dump->mutex); |
8bb113da | 2740 | if (!dump->status) { |
1c1e46ed AW |
2741 | struct dpif_netdev *dpif = dpif_netdev_cast(thread->up.dpif); |
2742 | struct dp_netdev *dp = get_dp_netdev(&dpif->dpif); | |
2743 | struct dp_netdev_pmd_thread *pmd = dump->cur_pmd; | |
2744 | int flow_limit = MIN(max_flows, FLOW_DUMP_MAX_BATCH); | |
2745 | ||
2746 | /* First call to dump_next(), extracts the first pmd thread. | |
2747 | * If there is no pmd thread, returns immediately. */ | |
2748 | if (!pmd) { | |
2749 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2750 | if (!pmd) { | |
2751 | ovs_mutex_unlock(&dump->mutex); | |
2752 | return n_flows; | |
8bb113da | 2753 | |
8bb113da | 2754 | } |
d2ad7ef1 | 2755 | } |
1c1e46ed AW |
2756 | |
2757 | do { | |
2758 | for (n_flows = 0; n_flows < flow_limit; n_flows++) { | |
2759 | struct cmap_node *node; | |
2760 | ||
2761 | node = cmap_next_position(&pmd->flow_table, &dump->flow_pos); | |
2762 | if (!node) { | |
2763 | break; | |
2764 | } | |
2765 | netdev_flows[n_flows] = CONTAINER_OF(node, | |
2766 | struct dp_netdev_flow, | |
2767 | node); | |
2768 | } | |
2769 | /* When finishing dumping the current pmd thread, moves to | |
2770 | * the next. */ | |
2771 | if (n_flows < flow_limit) { | |
2772 | memset(&dump->flow_pos, 0, sizeof dump->flow_pos); | |
2773 | dp_netdev_pmd_unref(pmd); | |
2774 | pmd = dp_netdev_pmd_get_next(dp, &dump->poll_thread_pos); | |
2775 | if (!pmd) { | |
2776 | dump->status = EOF; | |
2777 | break; | |
2778 | } | |
2779 | } | |
2780 | /* Keeps the reference to next caller. */ | |
2781 | dump->cur_pmd = pmd; | |
2782 | ||
2783 | /* If the current dump is empty, do not exit the loop, since the | |
2784 | * remaining pmds could have flows to be dumped. Just dumps again | |
2785 | * on the new 'pmd'. */ | |
2786 | } while (!n_flows); | |
8a4e3a85 | 2787 | } |
ac64794a | 2788 | ovs_mutex_unlock(&dump->mutex); |
ac64794a | 2789 | |
8bb113da RW |
2790 | for (i = 0; i < n_flows; i++) { |
2791 | struct odputil_keybuf *maskbuf = &thread->maskbuf[i]; | |
2792 | struct odputil_keybuf *keybuf = &thread->keybuf[i]; | |
2793 | struct dp_netdev_flow *netdev_flow = netdev_flows[i]; | |
2794 | struct dpif_flow *f = &flows[i]; | |
7af12bd7 | 2795 | struct ofpbuf key, mask; |
8bb113da | 2796 | |
7af12bd7 JS |
2797 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); |
2798 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
64bb477f JS |
2799 | dp_netdev_flow_to_dpif_flow(netdev_flow, &key, &mask, f, |
2800 | dump->up.terse); | |
8bb113da | 2801 | } |
feebdea2 | 2802 | |
8bb113da | 2803 | return n_flows; |
72865317 BP |
2804 | } |
2805 | ||
2806 | static int | |
758c456d | 2807 | dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute) |
65f13b50 | 2808 | OVS_NO_THREAD_SAFETY_ANALYSIS |
72865317 BP |
2809 | { |
2810 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
65f13b50 | 2811 | struct dp_netdev_pmd_thread *pmd; |
1895cc8d | 2812 | struct dp_packet_batch pp; |
72865317 | 2813 | |
cf62fa4c PS |
2814 | if (dp_packet_size(execute->packet) < ETH_HEADER_LEN || |
2815 | dp_packet_size(execute->packet) > UINT16_MAX) { | |
72865317 BP |
2816 | return EINVAL; |
2817 | } | |
2818 | ||
65f13b50 AW |
2819 | /* Tries finding the 'pmd'. If NULL is returned, that means |
2820 | * the current thread is a non-pmd thread and should use | |
b19befae | 2821 | * dp_netdev_get_pmd(dp, NON_PMD_CORE_ID). */ |
65f13b50 AW |
2822 | pmd = ovsthread_getspecific(dp->per_pmd_key); |
2823 | if (!pmd) { | |
b19befae | 2824 | pmd = dp_netdev_get_pmd(dp, NON_PMD_CORE_ID); |
546e57d4 DDP |
2825 | if (!pmd) { |
2826 | return EBUSY; | |
2827 | } | |
65f13b50 AW |
2828 | } |
2829 | ||
05267613 AZ |
2830 | if (execute->probe) { |
2831 | /* If this is part of a probe, Drop the packet, since executing | |
2832 | * the action may actually cause spurious packets be sent into | |
2833 | * the network. */ | |
2834 | return 0; | |
2835 | } | |
2836 | ||
65f13b50 AW |
2837 | /* If the current thread is non-pmd thread, acquires |
2838 | * the 'non_pmd_mutex'. */ | |
2839 | if (pmd->core_id == NON_PMD_CORE_ID) { | |
2840 | ovs_mutex_lock(&dp->non_pmd_mutex); | |
2841 | } | |
1c1e46ed | 2842 | |
36d8de17 DDP |
2843 | /* The action processing expects the RSS hash to be valid, because |
2844 | * it's always initialized at the beginning of datapath processing. | |
2845 | * In this case, though, 'execute->packet' may not have gone through | |
2846 | * the datapath at all, it may have been generated by the upper layer | |
2847 | * (OpenFlow packet-out, BFD frame, ...). */ | |
2848 | if (!dp_packet_rss_valid(execute->packet)) { | |
2849 | dp_packet_set_rss_hash(execute->packet, | |
2850 | flow_hash_5tuple(execute->flow, 0)); | |
2851 | } | |
2852 | ||
72c84bc2 | 2853 | dp_packet_batch_init_packet(&pp, execute->packet); |
66e4ad8a DDP |
2854 | dp_netdev_execute_actions(pmd, &pp, false, execute->flow, |
2855 | execute->actions, execute->actions_len, | |
2856 | time_msec()); | |
36d8de17 | 2857 | |
65f13b50 AW |
2858 | if (pmd->core_id == NON_PMD_CORE_ID) { |
2859 | ovs_mutex_unlock(&dp->non_pmd_mutex); | |
e9985d6a | 2860 | dp_netdev_pmd_unref(pmd); |
65f13b50 | 2861 | } |
8a4e3a85 | 2862 | |
758c456d | 2863 | return 0; |
72865317 BP |
2864 | } |
2865 | ||
1a0c894a BP |
2866 | static void |
2867 | dpif_netdev_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) | |
2868 | { | |
2869 | size_t i; | |
2870 | ||
2871 | for (i = 0; i < n_ops; i++) { | |
2872 | struct dpif_op *op = ops[i]; | |
2873 | ||
2874 | switch (op->type) { | |
2875 | case DPIF_OP_FLOW_PUT: | |
2876 | op->error = dpif_netdev_flow_put(dpif, &op->u.flow_put); | |
2877 | break; | |
2878 | ||
2879 | case DPIF_OP_FLOW_DEL: | |
2880 | op->error = dpif_netdev_flow_del(dpif, &op->u.flow_del); | |
2881 | break; | |
2882 | ||
2883 | case DPIF_OP_EXECUTE: | |
2884 | op->error = dpif_netdev_execute(dpif, &op->u.execute); | |
2885 | break; | |
6fe09f8c JS |
2886 | |
2887 | case DPIF_OP_FLOW_GET: | |
2888 | op->error = dpif_netdev_flow_get(dpif, &op->u.flow_get); | |
2889 | break; | |
1a0c894a BP |
2890 | } |
2891 | } | |
2892 | } | |
2893 | ||
d4f6865c DDP |
2894 | /* Applies datapath configuration from the database. Some of the changes are |
2895 | * actually applied in dpif_netdev_run(). */ | |
f2eee189 | 2896 | static int |
d4f6865c | 2897 | dpif_netdev_set_config(struct dpif *dpif, const struct smap *other_config) |
f2eee189 AW |
2898 | { |
2899 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
d4f6865c | 2900 | const char *cmask = smap_get(other_config, "pmd-cpu-mask"); |
4c30b246 CL |
2901 | unsigned long long insert_prob = |
2902 | smap_get_ullong(other_config, "emc-insert-inv-prob", | |
2903 | DEFAULT_EM_FLOW_INSERT_INV_PROB); | |
2904 | uint32_t insert_min, cur_min; | |
f2eee189 | 2905 | |
a6a426d6 IM |
2906 | if (!nullable_string_is_equal(dp->pmd_cmask, cmask)) { |
2907 | free(dp->pmd_cmask); | |
2908 | dp->pmd_cmask = nullable_xstrdup(cmask); | |
2909 | dp_netdev_request_reconfigure(dp); | |
f2eee189 AW |
2910 | } |
2911 | ||
4c30b246 CL |
2912 | atomic_read_relaxed(&dp->emc_insert_min, &cur_min); |
2913 | if (insert_prob <= UINT32_MAX) { | |
2914 | insert_min = insert_prob == 0 ? 0 : UINT32_MAX / insert_prob; | |
2915 | } else { | |
2916 | insert_min = DEFAULT_EM_FLOW_INSERT_MIN; | |
2917 | insert_prob = DEFAULT_EM_FLOW_INSERT_INV_PROB; | |
2918 | } | |
2919 | ||
2920 | if (insert_min != cur_min) { | |
2921 | atomic_store_relaxed(&dp->emc_insert_min, insert_min); | |
2922 | if (insert_min == 0) { | |
2923 | VLOG_INFO("EMC has been disabled"); | |
2924 | } else { | |
2925 | VLOG_INFO("EMC insertion probability changed to 1/%llu (~%.2f%%)", | |
2926 | insert_prob, (100 / (float)insert_prob)); | |
2927 | } | |
2928 | } | |
2929 | ||
f2eee189 AW |
2930 | return 0; |
2931 | } | |
2932 | ||
3eb67853 IM |
2933 | /* Parses affinity list and returns result in 'core_ids'. */ |
2934 | static int | |
2935 | parse_affinity_list(const char *affinity_list, unsigned *core_ids, int n_rxq) | |
2936 | { | |
2937 | unsigned i; | |
2938 | char *list, *copy, *key, *value; | |
2939 | int error = 0; | |
2940 | ||
2941 | for (i = 0; i < n_rxq; i++) { | |
51c37a56 | 2942 | core_ids[i] = OVS_CORE_UNSPEC; |
3eb67853 IM |
2943 | } |
2944 | ||
2945 | if (!affinity_list) { | |
2946 | return 0; | |
2947 | } | |
2948 | ||
2949 | list = copy = xstrdup(affinity_list); | |
2950 | ||
2951 | while (ofputil_parse_key_value(&list, &key, &value)) { | |
2952 | int rxq_id, core_id; | |
2953 | ||
2954 | if (!str_to_int(key, 0, &rxq_id) || rxq_id < 0 | |
2955 | || !str_to_int(value, 0, &core_id) || core_id < 0) { | |
2956 | error = EINVAL; | |
2957 | break; | |
2958 | } | |
2959 | ||
2960 | if (rxq_id < n_rxq) { | |
2961 | core_ids[rxq_id] = core_id; | |
2962 | } | |
2963 | } | |
2964 | ||
2965 | free(copy); | |
2966 | return error; | |
2967 | } | |
2968 | ||
2969 | /* Parses 'affinity_list' and applies configuration if it is valid. */ | |
2970 | static int | |
2971 | dpif_netdev_port_set_rxq_affinity(struct dp_netdev_port *port, | |
2972 | const char *affinity_list) | |
2973 | { | |
2974 | unsigned *core_ids, i; | |
2975 | int error = 0; | |
2976 | ||
2977 | core_ids = xmalloc(port->n_rxq * sizeof *core_ids); | |
2978 | if (parse_affinity_list(affinity_list, core_ids, port->n_rxq)) { | |
2979 | error = EINVAL; | |
2980 | goto exit; | |
2981 | } | |
2982 | ||
2983 | for (i = 0; i < port->n_rxq; i++) { | |
2984 | port->rxqs[i].core_id = core_ids[i]; | |
2985 | } | |
2986 | ||
2987 | exit: | |
2988 | free(core_ids); | |
2989 | return error; | |
2990 | } | |
2991 | ||
2992 | /* Changes the affinity of port's rx queues. The changes are actually applied | |
2993 | * in dpif_netdev_run(). */ | |
2994 | static int | |
2995 | dpif_netdev_port_set_config(struct dpif *dpif, odp_port_t port_no, | |
2996 | const struct smap *cfg) | |
2997 | { | |
2998 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
2999 | struct dp_netdev_port *port; | |
3000 | int error = 0; | |
3001 | const char *affinity_list = smap_get(cfg, "pmd-rxq-affinity"); | |
3002 | ||
3003 | ovs_mutex_lock(&dp->port_mutex); | |
3004 | error = get_port_by_number(dp, port_no, &port); | |
3005 | if (error || !netdev_is_pmd(port->netdev) | |
3006 | || nullable_string_is_equal(affinity_list, port->rxq_affinity_list)) { | |
3007 | goto unlock; | |
3008 | } | |
3009 | ||
3010 | error = dpif_netdev_port_set_rxq_affinity(port, affinity_list); | |
3011 | if (error) { | |
3012 | goto unlock; | |
3013 | } | |
3014 | free(port->rxq_affinity_list); | |
3015 | port->rxq_affinity_list = nullable_xstrdup(affinity_list); | |
3016 | ||
3017 | dp_netdev_request_reconfigure(dp); | |
3018 | unlock: | |
3019 | ovs_mutex_unlock(&dp->port_mutex); | |
3020 | return error; | |
3021 | } | |
3022 | ||
5bf93d67 EJ |
3023 | static int |
3024 | dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED, | |
3025 | uint32_t queue_id, uint32_t *priority) | |
3026 | { | |
3027 | *priority = queue_id; | |
3028 | return 0; | |
3029 | } | |
3030 | ||
72865317 | 3031 | \f |
9ff55ae2 | 3032 | /* Creates and returns a new 'struct dp_netdev_actions', whose actions are |
1401f6de | 3033 | * a copy of the 'size' bytes of 'actions' input parameters. */ |
a84cb64a BP |
3034 | struct dp_netdev_actions * |
3035 | dp_netdev_actions_create(const struct nlattr *actions, size_t size) | |
3036 | { | |
3037 | struct dp_netdev_actions *netdev_actions; | |
3038 | ||
9ff55ae2 DDP |
3039 | netdev_actions = xmalloc(sizeof *netdev_actions + size); |
3040 | memcpy(netdev_actions->actions, actions, size); | |
a84cb64a BP |
3041 | netdev_actions->size = size; |
3042 | ||
3043 | return netdev_actions; | |
3044 | } | |
3045 | ||
a84cb64a | 3046 | struct dp_netdev_actions * |
61e7deb1 | 3047 | dp_netdev_flow_get_actions(const struct dp_netdev_flow *flow) |
a84cb64a | 3048 | { |
61e7deb1 | 3049 | return ovsrcu_get(struct dp_netdev_actions *, &flow->actions); |
a84cb64a BP |
3050 | } |
3051 | ||
61e7deb1 BP |
3052 | static void |
3053 | dp_netdev_actions_free(struct dp_netdev_actions *actions) | |
a84cb64a | 3054 | { |
61e7deb1 | 3055 | free(actions); |
a84cb64a BP |
3056 | } |
3057 | \f | |
55e3ca97 DDP |
3058 | static inline unsigned long long |
3059 | cycles_counter(void) | |
3060 | { | |
3061 | #ifdef DPDK_NETDEV | |
3062 | return rte_get_tsc_cycles(); | |
3063 | #else | |
3064 | return 0; | |
3065 | #endif | |
3066 | } | |
3067 | ||
3068 | /* Fake mutex to make sure that the calls to cycles_count_* are balanced */ | |
3069 | extern struct ovs_mutex cycles_counter_fake_mutex; | |
3070 | ||
3071 | /* Start counting cycles. Must be followed by 'cycles_count_end()' */ | |
3072 | static inline void | |
3073 | cycles_count_start(struct dp_netdev_pmd_thread *pmd) | |
3074 | OVS_ACQUIRES(&cycles_counter_fake_mutex) | |
3075 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
3076 | { | |
3077 | pmd->last_cycles = cycles_counter(); | |
3078 | } | |
3079 | ||
3080 | /* Stop counting cycles and add them to the counter 'type' */ | |
3081 | static inline void | |
3082 | cycles_count_end(struct dp_netdev_pmd_thread *pmd, | |
3083 | enum pmd_cycles_counter_type type) | |
3084 | OVS_RELEASES(&cycles_counter_fake_mutex) | |
3085 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
3086 | { | |
3087 | unsigned long long interval = cycles_counter() - pmd->last_cycles; | |
3088 | ||
3089 | non_atomic_ullong_add(&pmd->cycles.n[type], interval); | |
3090 | } | |
e4cfed38 | 3091 | |
a2ac666d CL |
3092 | /* Calculate the intermediate cycle result and add to the counter 'type' */ |
3093 | static inline void | |
3094 | cycles_count_intermediate(struct dp_netdev_pmd_thread *pmd, | |
3095 | enum pmd_cycles_counter_type type) | |
3096 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
3097 | { | |
3098 | unsigned long long new_cycles = cycles_counter(); | |
3099 | unsigned long long interval = new_cycles - pmd->last_cycles; | |
3100 | pmd->last_cycles = new_cycles; | |
3101 | ||
3102 | non_atomic_ullong_add(&pmd->cycles.n[type], interval); | |
3103 | } | |
3104 | ||
3105 | static int | |
65f13b50 | 3106 | dp_netdev_process_rxq_port(struct dp_netdev_pmd_thread *pmd, |
947dc567 DDP |
3107 | struct netdev_rxq *rx, |
3108 | odp_port_t port_no) | |
e4cfed38 | 3109 | { |
1895cc8d PS |
3110 | struct dp_packet_batch batch; |
3111 | int error; | |
a2ac666d | 3112 | int batch_cnt = 0; |
e4cfed38 | 3113 | |
1895cc8d | 3114 | dp_packet_batch_init(&batch); |
947dc567 | 3115 | error = netdev_rxq_recv(rx, &batch); |
e4cfed38 | 3116 | if (!error) { |
3c33f0ff | 3117 | *recirc_depth_get() = 0; |
41ccaa24 | 3118 | |
a2ac666d | 3119 | batch_cnt = batch.count; |
947dc567 | 3120 | dp_netdev_input(pmd, &batch, port_no); |
e4cfed38 | 3121 | } else if (error != EAGAIN && error != EOPNOTSUPP) { |
3c33f0ff | 3122 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
e4cfed38 PS |
3123 | |
3124 | VLOG_ERR_RL(&rl, "error receiving data from %s: %s", | |
947dc567 | 3125 | netdev_rxq_get_name(rx), ovs_strerror(error)); |
e4cfed38 | 3126 | } |
a2ac666d CL |
3127 | |
3128 | return batch_cnt; | |
e4cfed38 PS |
3129 | } |
3130 | ||
e32971b8 DDP |
3131 | static struct tx_port * |
3132 | tx_port_lookup(const struct hmap *hmap, odp_port_t port_no) | |
3133 | { | |
3134 | struct tx_port *tx; | |
3135 | ||
3136 | HMAP_FOR_EACH_IN_BUCKET (tx, node, hash_port_no(port_no), hmap) { | |
3137 | if (tx->port->port_no == port_no) { | |
3138 | return tx; | |
3139 | } | |
3140 | } | |
3141 | ||
3142 | return NULL; | |
3143 | } | |
3144 | ||
dc36593c DDP |
3145 | static int |
3146 | port_reconfigure(struct dp_netdev_port *port) | |
3147 | { | |
3148 | struct netdev *netdev = port->netdev; | |
dc36593c DDP |
3149 | int i, err; |
3150 | ||
e32971b8 | 3151 | port->need_reconfigure = false; |
dc36593c DDP |
3152 | |
3153 | /* Closes the existing 'rxq's. */ | |
3154 | for (i = 0; i < port->n_rxq; i++) { | |
947dc567 DDP |
3155 | netdev_rxq_close(port->rxqs[i].rx); |
3156 | port->rxqs[i].rx = NULL; | |
dc36593c DDP |
3157 | } |
3158 | port->n_rxq = 0; | |
3159 | ||
050c60bf | 3160 | /* Allows 'netdev' to apply the pending configuration changes. */ |
e32971b8 DDP |
3161 | if (netdev_is_reconf_required(netdev)) { |
3162 | err = netdev_reconfigure(netdev); | |
3163 | if (err && (err != EOPNOTSUPP)) { | |
3164 | VLOG_ERR("Failed to set interface %s new configuration", | |
3165 | netdev_get_name(netdev)); | |
3166 | return err; | |
3167 | } | |
dc36593c | 3168 | } |
050c60bf | 3169 | /* If the netdev_reconfigure() above succeeds, reopens the 'rxq's. */ |
3eb67853 IM |
3170 | port->rxqs = xrealloc(port->rxqs, |
3171 | sizeof *port->rxqs * netdev_n_rxq(netdev)); | |
324c8374 IM |
3172 | /* Realloc 'used' counters for tx queues. */ |
3173 | free(port->txq_used); | |
3174 | port->txq_used = xcalloc(netdev_n_txq(netdev), sizeof *port->txq_used); | |
3175 | ||
dc36593c | 3176 | for (i = 0; i < netdev_n_rxq(netdev); i++) { |
947dc567 DDP |
3177 | port->rxqs[i].port = port; |
3178 | err = netdev_rxq_open(netdev, &port->rxqs[i].rx, i); | |
dc36593c DDP |
3179 | if (err) { |
3180 | return err; | |
3181 | } | |
3182 | port->n_rxq++; | |
3183 | } | |
3184 | ||
3eb67853 IM |
3185 | /* Parse affinity list to apply configuration for new queues. */ |
3186 | dpif_netdev_port_set_rxq_affinity(port, port->rxq_affinity_list); | |
3187 | ||
dc36593c DDP |
3188 | return 0; |
3189 | } | |
3190 | ||
e32971b8 DDP |
3191 | struct rr_numa_list { |
3192 | struct hmap numas; /* Contains 'struct rr_numa' */ | |
3193 | }; | |
3194 | ||
3195 | struct rr_numa { | |
3196 | struct hmap_node node; | |
3197 | ||
3198 | int numa_id; | |
3199 | ||
3200 | /* Non isolated pmds on numa node 'numa_id' */ | |
3201 | struct dp_netdev_pmd_thread **pmds; | |
3202 | int n_pmds; | |
3203 | ||
3204 | int cur_index; | |
3205 | }; | |
3206 | ||
3207 | static struct rr_numa * | |
3208 | rr_numa_list_lookup(struct rr_numa_list *rr, int numa_id) | |
3209 | { | |
3210 | struct rr_numa *numa; | |
3211 | ||
3212 | HMAP_FOR_EACH_WITH_HASH (numa, node, hash_int(numa_id, 0), &rr->numas) { | |
3213 | if (numa->numa_id == numa_id) { | |
3214 | return numa; | |
3215 | } | |
3216 | } | |
3217 | ||
3218 | return NULL; | |
3219 | } | |
3220 | ||
c37813fd BM |
3221 | /* Returns the next node in numa list following 'numa' in round-robin fashion. |
3222 | * Returns first node if 'numa' is a null pointer or the last node in 'rr'. | |
3223 | * Returns NULL if 'rr' numa list is empty. */ | |
3224 | static struct rr_numa * | |
3225 | rr_numa_list_next(struct rr_numa_list *rr, const struct rr_numa *numa) | |
3226 | { | |
3227 | struct hmap_node *node = NULL; | |
3228 | ||
3229 | if (numa) { | |
3230 | node = hmap_next(&rr->numas, &numa->node); | |
3231 | } | |
3232 | if (!node) { | |
3233 | node = hmap_first(&rr->numas); | |
3234 | } | |
3235 | ||
3236 | return (node) ? CONTAINER_OF(node, struct rr_numa, node) : NULL; | |
3237 | } | |
3238 | ||
e32971b8 DDP |
3239 | static void |
3240 | rr_numa_list_populate(struct dp_netdev *dp, struct rr_numa_list *rr) | |
3241 | { | |
3242 | struct dp_netdev_pmd_thread *pmd; | |
3243 | struct rr_numa *numa; | |
3244 | ||
3245 | hmap_init(&rr->numas); | |
3246 | ||
3247 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3248 | if (pmd->core_id == NON_PMD_CORE_ID || pmd->isolated) { | |
3249 | continue; | |
3250 | } | |
3251 | ||
3252 | numa = rr_numa_list_lookup(rr, pmd->numa_id); | |
3253 | if (!numa) { | |
3254 | numa = xzalloc(sizeof *numa); | |
3255 | numa->numa_id = pmd->numa_id; | |
3256 | hmap_insert(&rr->numas, &numa->node, hash_int(pmd->numa_id, 0)); | |
3257 | } | |
3258 | numa->n_pmds++; | |
3259 | numa->pmds = xrealloc(numa->pmds, numa->n_pmds * sizeof *numa->pmds); | |
3260 | numa->pmds[numa->n_pmds - 1] = pmd; | |
3261 | } | |
3262 | } | |
3263 | ||
3264 | static struct dp_netdev_pmd_thread * | |
3265 | rr_numa_get_pmd(struct rr_numa *numa) | |
3266 | { | |
3267 | return numa->pmds[numa->cur_index++ % numa->n_pmds]; | |
3268 | } | |
3269 | ||
3270 | static void | |
3271 | rr_numa_list_destroy(struct rr_numa_list *rr) | |
3272 | { | |
3273 | struct rr_numa *numa; | |
3274 | ||
3275 | HMAP_FOR_EACH_POP (numa, node, &rr->numas) { | |
3276 | free(numa->pmds); | |
3277 | free(numa); | |
3278 | } | |
3279 | hmap_destroy(&rr->numas); | |
3280 | } | |
3281 | ||
3282 | /* Assign pmds to queues. If 'pinned' is true, assign pmds to pinned | |
3283 | * queues and marks the pmds as isolated. Otherwise, assign non isolated | |
3284 | * pmds to unpinned queues. | |
3285 | * | |
3286 | * The function doesn't touch the pmd threads, it just stores the assignment | |
3287 | * in the 'pmd' member of each rxq. */ | |
3288 | static void | |
3289 | rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp->port_mutex) | |
3290 | { | |
3291 | struct dp_netdev_port *port; | |
3292 | struct rr_numa_list rr; | |
c37813fd | 3293 | struct rr_numa *non_local_numa = NULL; |
e32971b8 DDP |
3294 | |
3295 | rr_numa_list_populate(dp, &rr); | |
3296 | ||
3297 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3298 | struct rr_numa *numa; | |
3299 | int numa_id; | |
3300 | ||
3301 | if (!netdev_is_pmd(port->netdev)) { | |
3302 | continue; | |
3303 | } | |
3304 | ||
3305 | numa_id = netdev_get_numa_id(port->netdev); | |
3306 | numa = rr_numa_list_lookup(&rr, numa_id); | |
3307 | ||
3308 | for (int qid = 0; qid < port->n_rxq; qid++) { | |
3309 | struct dp_netdev_rxq *q = &port->rxqs[qid]; | |
3310 | ||
3311 | if (pinned && q->core_id != OVS_CORE_UNSPEC) { | |
3312 | struct dp_netdev_pmd_thread *pmd; | |
3313 | ||
3314 | pmd = dp_netdev_get_pmd(dp, q->core_id); | |
3315 | if (!pmd) { | |
3316 | VLOG_WARN("There is no PMD thread on core %d. Queue " | |
3317 | "%d on port \'%s\' will not be polled.", | |
3318 | q->core_id, qid, netdev_get_name(port->netdev)); | |
3319 | } else { | |
3320 | q->pmd = pmd; | |
3321 | pmd->isolated = true; | |
3322 | dp_netdev_pmd_unref(pmd); | |
3323 | } | |
3324 | } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { | |
3325 | if (!numa) { | |
c37813fd BM |
3326 | /* There are no pmds on the queue's local NUMA node. |
3327 | Round-robin on the NUMA nodes that do have pmds. */ | |
3328 | non_local_numa = rr_numa_list_next(&rr, non_local_numa); | |
3329 | if (!non_local_numa) { | |
3330 | VLOG_ERR("There is no available (non-isolated) pmd " | |
3331 | "thread for port \'%s\' queue %d. This queue " | |
3332 | "will not be polled. Is pmd-cpu-mask set to " | |
3333 | "zero? Or are all PMDs isolated to other " | |
3334 | "queues?", netdev_get_name(port->netdev), | |
3335 | qid); | |
3336 | continue; | |
3337 | } | |
3338 | q->pmd = rr_numa_get_pmd(non_local_numa); | |
3339 | VLOG_WARN("There's no available (non-isolated) pmd thread " | |
e32971b8 | 3340 | "on numa node %d. Queue %d on port \'%s\' will " |
c37813fd BM |
3341 | "be assigned to the pmd on core %d " |
3342 | "(numa node %d). Expect reduced performance.", | |
3343 | numa_id, qid, netdev_get_name(port->netdev), | |
3344 | q->pmd->core_id, q->pmd->numa_id); | |
e32971b8 | 3345 | } else { |
c37813fd BM |
3346 | /* Assign queue to the next (round-robin) PMD on it's local |
3347 | NUMA node. */ | |
e32971b8 DDP |
3348 | q->pmd = rr_numa_get_pmd(numa); |
3349 | } | |
3350 | } | |
3351 | } | |
3352 | } | |
3353 | ||
3354 | rr_numa_list_destroy(&rr); | |
3355 | } | |
3356 | ||
140dd699 IM |
3357 | static void |
3358 | reload_affected_pmds(struct dp_netdev *dp) | |
3359 | { | |
3360 | struct dp_netdev_pmd_thread *pmd; | |
3361 | ||
3362 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3363 | if (pmd->need_reload) { | |
3364 | dp_netdev_reload_pmd__(pmd); | |
3365 | pmd->need_reload = false; | |
3366 | } | |
3367 | } | |
3368 | } | |
3369 | ||
6e3c6fa4 DDP |
3370 | static void |
3371 | reconfigure_pmd_threads(struct dp_netdev *dp) | |
3372 | OVS_REQUIRES(dp->port_mutex) | |
3373 | { | |
e32971b8 DDP |
3374 | struct dp_netdev_pmd_thread *pmd; |
3375 | struct ovs_numa_dump *pmd_cores; | |
140dd699 IM |
3376 | struct ovs_numa_info_core *core; |
3377 | struct hmapx to_delete = HMAPX_INITIALIZER(&to_delete); | |
3378 | struct hmapx_node *node; | |
e32971b8 | 3379 | bool changed = false; |
140dd699 | 3380 | bool need_to_adjust_static_tx_qids = false; |
e32971b8 DDP |
3381 | |
3382 | /* The pmd threads should be started only if there's a pmd port in the | |
3383 | * datapath. If the user didn't provide any "pmd-cpu-mask", we start | |
3384 | * NR_PMD_THREADS per numa node. */ | |
3385 | if (!has_pmd_port(dp)) { | |
3386 | pmd_cores = ovs_numa_dump_n_cores_per_numa(0); | |
3387 | } else if (dp->pmd_cmask && dp->pmd_cmask[0]) { | |
3388 | pmd_cores = ovs_numa_dump_cores_with_cmask(dp->pmd_cmask); | |
3389 | } else { | |
3390 | pmd_cores = ovs_numa_dump_n_cores_per_numa(NR_PMD_THREADS); | |
3391 | } | |
3392 | ||
140dd699 IM |
3393 | /* We need to adjust 'static_tx_qid's only if we're reducing number of |
3394 | * PMD threads. Otherwise, new threads will allocate all the freed ids. */ | |
3395 | if (ovs_numa_dump_count(pmd_cores) < cmap_count(&dp->poll_threads) - 1) { | |
3396 | /* Adjustment is required to keep 'static_tx_qid's sequential and | |
3397 | * avoid possible issues, for example, imbalanced tx queue usage | |
3398 | * and unnecessary locking caused by remapping on netdev level. */ | |
3399 | need_to_adjust_static_tx_qids = true; | |
3400 | } | |
3401 | ||
3402 | /* Check for unwanted pmd threads */ | |
3403 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3404 | if (pmd->core_id == NON_PMD_CORE_ID) { | |
3405 | continue; | |
3406 | } | |
3407 | if (!ovs_numa_dump_contains_core(pmd_cores, pmd->numa_id, | |
3408 | pmd->core_id)) { | |
3409 | hmapx_add(&to_delete, pmd); | |
3410 | } else if (need_to_adjust_static_tx_qids) { | |
3411 | pmd->need_reload = true; | |
e32971b8 DDP |
3412 | } |
3413 | } | |
3414 | ||
140dd699 IM |
3415 | HMAPX_FOR_EACH (node, &to_delete) { |
3416 | pmd = (struct dp_netdev_pmd_thread *) node->data; | |
3417 | VLOG_INFO("PMD thread on numa_id: %d, core id: %2d destroyed.", | |
3418 | pmd->numa_id, pmd->core_id); | |
3419 | dp_netdev_del_pmd(dp, pmd); | |
3420 | } | |
3421 | changed = !hmapx_is_empty(&to_delete); | |
3422 | hmapx_destroy(&to_delete); | |
e32971b8 | 3423 | |
140dd699 IM |
3424 | if (need_to_adjust_static_tx_qids) { |
3425 | /* 'static_tx_qid's are not sequential now. | |
3426 | * Reload remaining threads to fix this. */ | |
3427 | reload_affected_pmds(dp); | |
3428 | } | |
e32971b8 | 3429 | |
140dd699 IM |
3430 | /* Check for required new pmd threads */ |
3431 | FOR_EACH_CORE_ON_DUMP(core, pmd_cores) { | |
3432 | pmd = dp_netdev_get_pmd(dp, core->core_id); | |
3433 | if (!pmd) { | |
3434 | pmd = xzalloc(sizeof *pmd); | |
e32971b8 | 3435 | dp_netdev_configure_pmd(pmd, dp, core->core_id, core->numa_id); |
e32971b8 | 3436 | pmd->thread = ovs_thread_create("pmd", pmd_thread_main, pmd); |
140dd699 IM |
3437 | VLOG_INFO("PMD thread on numa_id: %d, core id: %2d created.", |
3438 | pmd->numa_id, pmd->core_id); | |
3439 | changed = true; | |
3440 | } else { | |
3441 | dp_netdev_pmd_unref(pmd); | |
e32971b8 | 3442 | } |
140dd699 IM |
3443 | } |
3444 | ||
3445 | if (changed) { | |
3446 | struct ovs_numa_info_numa *numa; | |
e32971b8 DDP |
3447 | |
3448 | /* Log the number of pmd threads per numa node. */ | |
3449 | FOR_EACH_NUMA_ON_DUMP (numa, pmd_cores) { | |
140dd699 | 3450 | VLOG_INFO("There are %"PRIuSIZE" pmd threads on numa node %d", |
e32971b8 DDP |
3451 | numa->n_cores, numa->numa_id); |
3452 | } | |
3453 | } | |
3454 | ||
3455 | ovs_numa_dump_destroy(pmd_cores); | |
3456 | } | |
3457 | ||
e32971b8 DDP |
3458 | static void |
3459 | pmd_remove_stale_ports(struct dp_netdev *dp, | |
3460 | struct dp_netdev_pmd_thread *pmd) | |
3461 | OVS_EXCLUDED(pmd->port_mutex) | |
3462 | OVS_REQUIRES(dp->port_mutex) | |
3463 | { | |
3464 | struct rxq_poll *poll, *poll_next; | |
3465 | struct tx_port *tx, *tx_next; | |
3466 | ||
3467 | ovs_mutex_lock(&pmd->port_mutex); | |
3468 | HMAP_FOR_EACH_SAFE (poll, poll_next, node, &pmd->poll_list) { | |
3469 | struct dp_netdev_port *port = poll->rxq->port; | |
3470 | ||
3471 | if (port->need_reconfigure | |
3472 | || !hmap_contains(&dp->ports, &port->node)) { | |
3473 | dp_netdev_del_rxq_from_pmd(pmd, poll); | |
3474 | } | |
3475 | } | |
3476 | HMAP_FOR_EACH_SAFE (tx, tx_next, node, &pmd->tx_ports) { | |
3477 | struct dp_netdev_port *port = tx->port; | |
3478 | ||
3479 | if (port->need_reconfigure | |
3480 | || !hmap_contains(&dp->ports, &port->node)) { | |
3481 | dp_netdev_del_port_tx_from_pmd(pmd, tx); | |
3482 | } | |
3483 | } | |
3484 | ovs_mutex_unlock(&pmd->port_mutex); | |
3485 | } | |
3486 | ||
3487 | /* Must be called each time a port is added/removed or the cmask changes. | |
3488 | * This creates and destroys pmd threads, reconfigures ports, opens their | |
3489 | * rxqs and assigns all rxqs/txqs to pmd threads. */ | |
3490 | static void | |
3491 | reconfigure_datapath(struct dp_netdev *dp) | |
3492 | OVS_REQUIRES(dp->port_mutex) | |
3493 | { | |
3494 | struct dp_netdev_pmd_thread *pmd; | |
3495 | struct dp_netdev_port *port; | |
3496 | int wanted_txqs; | |
6e3c6fa4 | 3497 | |
a6a426d6 IM |
3498 | dp->last_reconfigure_seq = seq_read(dp->reconfigure_seq); |
3499 | ||
e32971b8 DDP |
3500 | /* Step 1: Adjust the pmd threads based on the datapath ports, the cores |
3501 | * on the system and the user configuration. */ | |
3502 | reconfigure_pmd_threads(dp); | |
6e3c6fa4 | 3503 | |
e32971b8 | 3504 | wanted_txqs = cmap_count(&dp->poll_threads); |
324c8374 | 3505 | |
e32971b8 DDP |
3506 | /* The number of pmd threads might have changed, or a port can be new: |
3507 | * adjust the txqs. */ | |
3508 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3509 | netdev_set_tx_multiq(port->netdev, wanted_txqs); | |
324c8374 IM |
3510 | } |
3511 | ||
e32971b8 DDP |
3512 | /* Step 2: Remove from the pmd threads ports that have been removed or |
3513 | * need reconfiguration. */ | |
3514 | ||
3515 | /* Check for all the ports that need reconfiguration. We cache this in | |
85a4f238 IM |
3516 | * 'port->need_reconfigure', because netdev_is_reconf_required() can |
3517 | * change at any time. */ | |
e32971b8 DDP |
3518 | HMAP_FOR_EACH (port, node, &dp->ports) { |
3519 | if (netdev_is_reconf_required(port->netdev)) { | |
3520 | port->need_reconfigure = true; | |
3521 | } | |
3522 | } | |
3523 | ||
3524 | /* Remove from the pmd threads all the ports that have been deleted or | |
3525 | * need reconfiguration. */ | |
3526 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3527 | pmd_remove_stale_ports(dp, pmd); | |
3528 | } | |
3529 | ||
3530 | /* Reload affected pmd threads. We must wait for the pmd threads before | |
3531 | * reconfiguring the ports, because a port cannot be reconfigured while | |
3532 | * it's being used. */ | |
3533 | reload_affected_pmds(dp); | |
3534 | ||
3535 | /* Step 3: Reconfigure ports. */ | |
3536 | ||
3537 | /* We only reconfigure the ports that we determined above, because they're | |
3538 | * not being used by any pmd thread at the moment. If a port fails to | |
3539 | * reconfigure we remove it from the datapath. */ | |
f582b6df BP |
3540 | struct dp_netdev_port *next_port; |
3541 | HMAP_FOR_EACH_SAFE (port, next_port, node, &dp->ports) { | |
dc36593c | 3542 | int err; |
6e3c6fa4 | 3543 | |
e32971b8 DDP |
3544 | if (!port->need_reconfigure) { |
3545 | continue; | |
3546 | } | |
3547 | ||
dc36593c DDP |
3548 | err = port_reconfigure(port); |
3549 | if (err) { | |
3550 | hmap_remove(&dp->ports, &port->node); | |
3551 | seq_change(dp->port_seq); | |
3552 | port_destroy(port); | |
324c8374 | 3553 | } else { |
e32971b8 | 3554 | port->dynamic_txqs = netdev_n_txq(port->netdev) < wanted_txqs; |
6e3c6fa4 DDP |
3555 | } |
3556 | } | |
e32971b8 DDP |
3557 | |
3558 | /* Step 4: Compute new rxq scheduling. We don't touch the pmd threads | |
3559 | * for now, we just update the 'pmd' pointer in each rxq to point to the | |
3560 | * wanted thread according to the scheduling policy. */ | |
3561 | ||
3562 | /* Reset all the pmd threads to non isolated. */ | |
3563 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3564 | pmd->isolated = false; | |
3565 | } | |
3566 | ||
3567 | /* Reset all the queues to unassigned */ | |
3568 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3569 | for (int i = 0; i < port->n_rxq; i++) { | |
3570 | port->rxqs[i].pmd = NULL; | |
3571 | } | |
3572 | } | |
3573 | ||
3574 | /* Add pinned queues and mark pmd threads isolated. */ | |
3575 | rxq_scheduling(dp, true); | |
3576 | ||
3577 | /* Add non-pinned queues. */ | |
3578 | rxq_scheduling(dp, false); | |
3579 | ||
3580 | /* Step 5: Remove queues not compliant with new scheduling. */ | |
3581 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3582 | struct rxq_poll *poll, *poll_next; | |
3583 | ||
3584 | ovs_mutex_lock(&pmd->port_mutex); | |
3585 | HMAP_FOR_EACH_SAFE (poll, poll_next, node, &pmd->poll_list) { | |
3586 | if (poll->rxq->pmd != pmd) { | |
3587 | dp_netdev_del_rxq_from_pmd(pmd, poll); | |
3588 | } | |
3589 | } | |
3590 | ovs_mutex_unlock(&pmd->port_mutex); | |
3591 | } | |
3592 | ||
3593 | /* Reload affected pmd threads. We must wait for the pmd threads to remove | |
3594 | * the old queues before readding them, otherwise a queue can be polled by | |
3595 | * two threads at the same time. */ | |
3596 | reload_affected_pmds(dp); | |
3597 | ||
3598 | /* Step 6: Add queues from scheduling, if they're not there already. */ | |
3599 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3600 | if (!netdev_is_pmd(port->netdev)) { | |
3601 | continue; | |
3602 | } | |
3603 | ||
3604 | for (int qid = 0; qid < port->n_rxq; qid++) { | |
3605 | struct dp_netdev_rxq *q = &port->rxqs[qid]; | |
3606 | ||
3607 | if (q->pmd) { | |
3608 | ovs_mutex_lock(&q->pmd->port_mutex); | |
3609 | dp_netdev_add_rxq_to_pmd(q->pmd, q); | |
3610 | ovs_mutex_unlock(&q->pmd->port_mutex); | |
3611 | } | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | /* Add every port to the tx cache of every pmd thread, if it's not | |
3616 | * there already and if this pmd has at least one rxq to poll. */ | |
3617 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
3618 | ovs_mutex_lock(&pmd->port_mutex); | |
3619 | if (hmap_count(&pmd->poll_list) || pmd->core_id == NON_PMD_CORE_ID) { | |
3620 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3621 | dp_netdev_add_port_tx_to_pmd(pmd, port); | |
3622 | } | |
3623 | } | |
3624 | ovs_mutex_unlock(&pmd->port_mutex); | |
3625 | } | |
3626 | ||
3627 | /* Reload affected pmd threads. */ | |
3628 | reload_affected_pmds(dp); | |
6e3c6fa4 DDP |
3629 | } |
3630 | ||
050c60bf DDP |
3631 | /* Returns true if one of the netdevs in 'dp' requires a reconfiguration */ |
3632 | static bool | |
3633 | ports_require_restart(const struct dp_netdev *dp) | |
3634 | OVS_REQUIRES(dp->port_mutex) | |
3635 | { | |
3636 | struct dp_netdev_port *port; | |
3637 | ||
3638 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
3639 | if (netdev_is_reconf_required(port->netdev)) { | |
3640 | return true; | |
3641 | } | |
3642 | } | |
3643 | ||
3644 | return false; | |
3645 | } | |
3646 | ||
a36de779 PS |
3647 | /* Return true if needs to revalidate datapath flows. */ |
3648 | static bool | |
e4cfed38 PS |
3649 | dpif_netdev_run(struct dpif *dpif) |
3650 | { | |
3651 | struct dp_netdev_port *port; | |
3652 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
546e57d4 | 3653 | struct dp_netdev_pmd_thread *non_pmd; |
a36de779 | 3654 | uint64_t new_tnl_seq; |
a2ac666d | 3655 | int process_packets = 0; |
e4cfed38 | 3656 | |
e9985d6a | 3657 | ovs_mutex_lock(&dp->port_mutex); |
546e57d4 DDP |
3658 | non_pmd = dp_netdev_get_pmd(dp, NON_PMD_CORE_ID); |
3659 | if (non_pmd) { | |
3660 | ovs_mutex_lock(&dp->non_pmd_mutex); | |
a2ac666d | 3661 | cycles_count_start(non_pmd); |
546e57d4 DDP |
3662 | HMAP_FOR_EACH (port, node, &dp->ports) { |
3663 | if (!netdev_is_pmd(port->netdev)) { | |
3664 | int i; | |
55c955bd | 3665 | |
546e57d4 | 3666 | for (i = 0; i < port->n_rxq; i++) { |
a2ac666d CL |
3667 | process_packets = |
3668 | dp_netdev_process_rxq_port(non_pmd, | |
3669 | port->rxqs[i].rx, | |
3670 | port->port_no); | |
3671 | cycles_count_intermediate(non_pmd, process_packets ? | |
3672 | PMD_CYCLES_PROCESSING | |
3673 | : PMD_CYCLES_IDLE); | |
546e57d4 | 3674 | } |
55c955bd | 3675 | } |
e4cfed38 | 3676 | } |
a2ac666d | 3677 | cycles_count_end(non_pmd, PMD_CYCLES_IDLE); |
546e57d4 DDP |
3678 | dpif_netdev_xps_revalidate_pmd(non_pmd, time_msec(), false); |
3679 | ovs_mutex_unlock(&dp->non_pmd_mutex); | |
6e3c6fa4 | 3680 | |
546e57d4 DDP |
3681 | dp_netdev_pmd_unref(non_pmd); |
3682 | } | |
1c1e46ed | 3683 | |
a6a426d6 | 3684 | if (dp_netdev_is_reconf_required(dp) || ports_require_restart(dp)) { |
e32971b8 | 3685 | reconfigure_datapath(dp); |
6e3c6fa4 DDP |
3686 | } |
3687 | ovs_mutex_unlock(&dp->port_mutex); | |
3688 | ||
53902038 | 3689 | tnl_neigh_cache_run(); |
7f9b8504 | 3690 | tnl_port_map_run(); |
a36de779 PS |
3691 | new_tnl_seq = seq_read(tnl_conf_seq); |
3692 | ||
3693 | if (dp->last_tnl_conf_seq != new_tnl_seq) { | |
3694 | dp->last_tnl_conf_seq = new_tnl_seq; | |
3695 | return true; | |
3696 | } | |
3697 | return false; | |
e4cfed38 PS |
3698 | } |
3699 | ||
3700 | static void | |
3701 | dpif_netdev_wait(struct dpif *dpif) | |
3702 | { | |
3703 | struct dp_netdev_port *port; | |
3704 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
3705 | ||
59e6d833 | 3706 | ovs_mutex_lock(&dp_netdev_mutex); |
e9985d6a DDP |
3707 | ovs_mutex_lock(&dp->port_mutex); |
3708 | HMAP_FOR_EACH (port, node, &dp->ports) { | |
050c60bf | 3709 | netdev_wait_reconf_required(port->netdev); |
55c955bd PS |
3710 | if (!netdev_is_pmd(port->netdev)) { |
3711 | int i; | |
3712 | ||
490e82af | 3713 | for (i = 0; i < port->n_rxq; i++) { |
947dc567 | 3714 | netdev_rxq_wait(port->rxqs[i].rx); |
55c955bd | 3715 | } |
e4cfed38 PS |
3716 | } |
3717 | } | |
e9985d6a | 3718 | ovs_mutex_unlock(&dp->port_mutex); |
59e6d833 | 3719 | ovs_mutex_unlock(&dp_netdev_mutex); |
a36de779 | 3720 | seq_wait(tnl_conf_seq, dp->last_tnl_conf_seq); |
e4cfed38 PS |
3721 | } |
3722 | ||
d0cca6c3 DDP |
3723 | static void |
3724 | pmd_free_cached_ports(struct dp_netdev_pmd_thread *pmd) | |
3725 | { | |
3726 | struct tx_port *tx_port_cached; | |
3727 | ||
324c8374 IM |
3728 | /* Free all used tx queue ids. */ |
3729 | dpif_netdev_xps_revalidate_pmd(pmd, 0, true); | |
3730 | ||
57eebbb4 DDP |
3731 | HMAP_FOR_EACH_POP (tx_port_cached, node, &pmd->tnl_port_cache) { |
3732 | free(tx_port_cached); | |
3733 | } | |
3734 | HMAP_FOR_EACH_POP (tx_port_cached, node, &pmd->send_port_cache) { | |
d0cca6c3 DDP |
3735 | free(tx_port_cached); |
3736 | } | |
3737 | } | |
3738 | ||
3739 | /* Copies ports from 'pmd->tx_ports' (shared with the main thread) to | |
3740 | * 'pmd->port_cache' (thread local) */ | |
3741 | static void | |
3742 | pmd_load_cached_ports(struct dp_netdev_pmd_thread *pmd) | |
3743 | OVS_REQUIRES(pmd->port_mutex) | |
3744 | { | |
3745 | struct tx_port *tx_port, *tx_port_cached; | |
3746 | ||
3747 | pmd_free_cached_ports(pmd); | |
57eebbb4 DDP |
3748 | hmap_shrink(&pmd->send_port_cache); |
3749 | hmap_shrink(&pmd->tnl_port_cache); | |
d0cca6c3 DDP |
3750 | |
3751 | HMAP_FOR_EACH (tx_port, node, &pmd->tx_ports) { | |
57eebbb4 DDP |
3752 | if (netdev_has_tunnel_push_pop(tx_port->port->netdev)) { |
3753 | tx_port_cached = xmemdup(tx_port, sizeof *tx_port_cached); | |
3754 | hmap_insert(&pmd->tnl_port_cache, &tx_port_cached->node, | |
3755 | hash_port_no(tx_port_cached->port->port_no)); | |
3756 | } | |
3757 | ||
3758 | if (netdev_n_txq(tx_port->port->netdev)) { | |
3759 | tx_port_cached = xmemdup(tx_port, sizeof *tx_port_cached); | |
3760 | hmap_insert(&pmd->send_port_cache, &tx_port_cached->node, | |
3761 | hash_port_no(tx_port_cached->port->port_no)); | |
3762 | } | |
d0cca6c3 DDP |
3763 | } |
3764 | } | |
3765 | ||
140dd699 IM |
3766 | static void |
3767 | pmd_alloc_static_tx_qid(struct dp_netdev_pmd_thread *pmd) | |
3768 | { | |
3769 | ovs_mutex_lock(&pmd->dp->tx_qid_pool_mutex); | |
3770 | if (!id_pool_alloc_id(pmd->dp->tx_qid_pool, &pmd->static_tx_qid)) { | |
3771 | VLOG_ABORT("static_tx_qid allocation failed for PMD on core %2d" | |
3772 | ", numa_id %d.", pmd->core_id, pmd->numa_id); | |
3773 | } | |
3774 | ovs_mutex_unlock(&pmd->dp->tx_qid_pool_mutex); | |
3775 | ||
3776 | VLOG_DBG("static_tx_qid = %d allocated for PMD thread on core %2d" | |
3777 | ", numa_id %d.", pmd->static_tx_qid, pmd->core_id, pmd->numa_id); | |
3778 | } | |
3779 | ||
3780 | static void | |
3781 | pmd_free_static_tx_qid(struct dp_netdev_pmd_thread *pmd) | |
3782 | { | |
3783 | ovs_mutex_lock(&pmd->dp->tx_qid_pool_mutex); | |
3784 | id_pool_free_id(pmd->dp->tx_qid_pool, pmd->static_tx_qid); | |
3785 | ovs_mutex_unlock(&pmd->dp->tx_qid_pool_mutex); | |
3786 | } | |
3787 | ||
e4cfed38 | 3788 | static int |
d0cca6c3 | 3789 | pmd_load_queues_and_ports(struct dp_netdev_pmd_thread *pmd, |
947dc567 | 3790 | struct polled_queue **ppoll_list) |
e4cfed38 | 3791 | { |
947dc567 | 3792 | struct polled_queue *poll_list = *ppoll_list; |
ae7ad0a1 IM |
3793 | struct rxq_poll *poll; |
3794 | int i; | |
e4cfed38 | 3795 | |
d0cca6c3 | 3796 | ovs_mutex_lock(&pmd->port_mutex); |
947dc567 DDP |
3797 | poll_list = xrealloc(poll_list, hmap_count(&pmd->poll_list) |
3798 | * sizeof *poll_list); | |
a1fdee13 | 3799 | |
ae7ad0a1 | 3800 | i = 0; |
947dc567 DDP |
3801 | HMAP_FOR_EACH (poll, node, &pmd->poll_list) { |
3802 | poll_list[i].rx = poll->rxq->rx; | |
3803 | poll_list[i].port_no = poll->rxq->port->port_no; | |
3804 | i++; | |
e4cfed38 | 3805 | } |
d0cca6c3 DDP |
3806 | |
3807 | pmd_load_cached_ports(pmd); | |
3808 | ||
3809 | ovs_mutex_unlock(&pmd->port_mutex); | |
e4cfed38 | 3810 | |
e4cfed38 | 3811 | *ppoll_list = poll_list; |
d42f9307 | 3812 | return i; |
e4cfed38 PS |
3813 | } |
3814 | ||
6c3eee82 | 3815 | static void * |
e4cfed38 | 3816 | pmd_thread_main(void *f_) |
6c3eee82 | 3817 | { |
65f13b50 | 3818 | struct dp_netdev_pmd_thread *pmd = f_; |
e4cfed38 | 3819 | unsigned int lc = 0; |
947dc567 | 3820 | struct polled_queue *poll_list; |
d42f9307 | 3821 | bool exiting; |
e4cfed38 PS |
3822 | int poll_cnt; |
3823 | int i; | |
a2ac666d | 3824 | int process_packets = 0; |
6c3eee82 | 3825 | |
e4cfed38 PS |
3826 | poll_list = NULL; |
3827 | ||
65f13b50 AW |
3828 | /* Stores the pmd thread's 'pmd' to 'per_pmd_key'. */ |
3829 | ovsthread_setspecific(pmd->dp->per_pmd_key, pmd); | |
6930c7e0 DDP |
3830 | ovs_numa_thread_setaffinity_core(pmd->core_id); |
3831 | dpdk_set_lcore_id(pmd->core_id); | |
d0cca6c3 | 3832 | poll_cnt = pmd_load_queues_and_ports(pmd, &poll_list); |
e215018b | 3833 | emc_cache_init(&pmd->flow_cache); |
e4cfed38 | 3834 | reload: |
140dd699 | 3835 | pmd_alloc_static_tx_qid(pmd); |
ae7ad0a1 | 3836 | |
7dd671f0 MK |
3837 | /* List port/core affinity */ |
3838 | for (i = 0; i < poll_cnt; i++) { | |
ce179f11 | 3839 | VLOG_DBG("Core %d processing port \'%s\' with queue-id %d\n", |
947dc567 | 3840 | pmd->core_id, netdev_rxq_get_name(poll_list[i].rx), |
ce179f11 | 3841 | netdev_rxq_get_queue_id(poll_list[i].rx)); |
7dd671f0 MK |
3842 | } |
3843 | ||
2788a1b1 DDP |
3844 | if (!poll_cnt) { |
3845 | while (seq_read(pmd->reload_seq) == pmd->last_reload_seq) { | |
3846 | seq_wait(pmd->reload_seq, pmd->last_reload_seq); | |
3847 | poll_block(); | |
3848 | } | |
3849 | lc = UINT_MAX; | |
3850 | } | |
3851 | ||
a2ac666d | 3852 | cycles_count_start(pmd); |
e4cfed38 | 3853 | for (;;) { |
e4cfed38 | 3854 | for (i = 0; i < poll_cnt; i++) { |
a2ac666d CL |
3855 | process_packets = |
3856 | dp_netdev_process_rxq_port(pmd, poll_list[i].rx, | |
3857 | poll_list[i].port_no); | |
3858 | cycles_count_intermediate(pmd, | |
3859 | process_packets ? PMD_CYCLES_PROCESSING | |
3860 | : PMD_CYCLES_IDLE); | |
e4cfed38 PS |
3861 | } |
3862 | ||
3863 | if (lc++ > 1024) { | |
14e3e12a | 3864 | bool reload; |
6c3eee82 | 3865 | |
e4cfed38 | 3866 | lc = 0; |
84067a4c | 3867 | |
fbe0962b | 3868 | coverage_try_clear(); |
3453b4d6 | 3869 | dp_netdev_pmd_try_optimize(pmd); |
9dede5cf FL |
3870 | if (!ovsrcu_try_quiesce()) { |
3871 | emc_cache_slow_sweep(&pmd->flow_cache); | |
3872 | } | |
84067a4c | 3873 | |
14e3e12a DDP |
3874 | atomic_read_relaxed(&pmd->reload, &reload); |
3875 | if (reload) { | |
6c3eee82 BP |
3876 | break; |
3877 | } | |
3878 | } | |
e4cfed38 | 3879 | } |
6c3eee82 | 3880 | |
a2ac666d CL |
3881 | cycles_count_end(pmd, PMD_CYCLES_IDLE); |
3882 | ||
d0cca6c3 | 3883 | poll_cnt = pmd_load_queues_and_ports(pmd, &poll_list); |
d42f9307 DDP |
3884 | exiting = latch_is_set(&pmd->exit_latch); |
3885 | /* Signal here to make sure the pmd finishes | |
3886 | * reloading the updated configuration. */ | |
3887 | dp_netdev_pmd_reload_done(pmd); | |
3888 | ||
140dd699 | 3889 | pmd_free_static_tx_qid(pmd); |
9bbf1c3d | 3890 | |
d42f9307 | 3891 | if (!exiting) { |
e4cfed38 PS |
3892 | goto reload; |
3893 | } | |
6c3eee82 | 3894 | |
e215018b | 3895 | emc_cache_uninit(&pmd->flow_cache); |
e4cfed38 | 3896 | free(poll_list); |
d0cca6c3 | 3897 | pmd_free_cached_ports(pmd); |
6c3eee82 BP |
3898 | return NULL; |
3899 | } | |
3900 | ||
6b31e073 RW |
3901 | static void |
3902 | dp_netdev_disable_upcall(struct dp_netdev *dp) | |
3903 | OVS_ACQUIRES(dp->upcall_rwlock) | |
3904 | { | |
3905 | fat_rwlock_wrlock(&dp->upcall_rwlock); | |
3906 | } | |
3907 | ||
5dddf960 JR |
3908 | \f |
3909 | /* Meters */ | |
3910 | static void | |
3911 | dpif_netdev_meter_get_features(const struct dpif * dpif OVS_UNUSED, | |
3912 | struct ofputil_meter_features *features) | |
3913 | { | |
4b27db64 JR |
3914 | features->max_meters = MAX_METERS; |
3915 | features->band_types = DP_SUPPORTED_METER_BAND_TYPES; | |
3916 | features->capabilities = DP_SUPPORTED_METER_FLAGS_MASK; | |
3917 | features->max_bands = MAX_BANDS; | |
5dddf960 JR |
3918 | features->max_color = 0; |
3919 | } | |
3920 | ||
4b27db64 JR |
3921 | /* Returns false when packet needs to be dropped. */ |
3922 | static void | |
3923 | dp_netdev_run_meter(struct dp_netdev *dp, struct dp_packet_batch *packets_, | |
3924 | uint32_t meter_id, long long int now) | |
3925 | { | |
3926 | struct dp_meter *meter; | |
3927 | struct dp_meter_band *band; | |
3928 | long long int long_delta_t; /* msec */ | |
3929 | uint32_t delta_t; /* msec */ | |
3930 | int i; | |
3931 | int cnt = packets_->count; | |
3932 | uint32_t bytes, volume; | |
3933 | int exceeded_band[NETDEV_MAX_BURST]; | |
3934 | uint32_t exceeded_rate[NETDEV_MAX_BURST]; | |
3935 | int exceeded_pkt = cnt; /* First packet that exceeded a band rate. */ | |
3936 | ||
3937 | if (meter_id >= MAX_METERS) { | |
3938 | return; | |
3939 | } | |
3940 | ||
3941 | meter_lock(dp, meter_id); | |
3942 | meter = dp->meters[meter_id]; | |
3943 | if (!meter) { | |
3944 | goto out; | |
3945 | } | |
3946 | ||
3947 | /* Initialize as negative values. */ | |
3948 | memset(exceeded_band, 0xff, cnt * sizeof *exceeded_band); | |
3949 | /* Initialize as zeroes. */ | |
3950 | memset(exceeded_rate, 0, cnt * sizeof *exceeded_rate); | |
3951 | ||
3952 | /* All packets will hit the meter at the same time. */ | |
3953 | long_delta_t = (now - meter->used); /* msec */ | |
3954 | ||
3955 | /* Make sure delta_t will not be too large, so that bucket will not | |
3956 | * wrap around below. */ | |
3957 | delta_t = (long_delta_t > (long long int)meter->max_delta_t) | |
3958 | ? meter->max_delta_t : (uint32_t)long_delta_t; | |
3959 | ||
3960 | /* Update meter stats. */ | |
3961 | meter->used = now; | |
3962 | meter->packet_count += cnt; | |
3963 | bytes = 0; | |
3964 | for (i = 0; i < cnt; i++) { | |
3965 | bytes += dp_packet_size(packets_->packets[i]); | |
3966 | } | |
3967 | meter->byte_count += bytes; | |
3968 | ||
3969 | /* Meters can operate in terms of packets per second or kilobits per | |
3970 | * second. */ | |
3971 | if (meter->flags & OFPMF13_PKTPS) { | |
3972 | /* Rate in packets/second, bucket 1/1000 packets. */ | |
3973 | /* msec * packets/sec = 1/1000 packets. */ | |
3974 | volume = cnt * 1000; /* Take 'cnt' packets from the bucket. */ | |
3975 | } else { | |
3976 | /* Rate in kbps, bucket in bits. */ | |
3977 | /* msec * kbps = bits */ | |
3978 | volume = bytes * 8; | |
3979 | } | |
3980 | ||
3981 | /* Update all bands and find the one hit with the highest rate for each | |
3982 | * packet (if any). */ | |
3983 | for (int m = 0; m < meter->n_bands; ++m) { | |
3984 | band = &meter->bands[m]; | |
3985 | ||
3986 | /* Update band's bucket. */ | |
3987 | band->bucket += delta_t * band->up.rate; | |
3988 | if (band->bucket > band->up.burst_size) { | |
3989 | band->bucket = band->up.burst_size; | |
3990 | } | |
3991 | ||
3992 | /* Drain the bucket for all the packets, if possible. */ | |
3993 | if (band->bucket >= volume) { | |
3994 | band->bucket -= volume; | |
3995 | } else { | |
3996 | int band_exceeded_pkt; | |
3997 | ||
3998 | /* Band limit hit, must process packet-by-packet. */ | |
3999 | if (meter->flags & OFPMF13_PKTPS) { | |
4000 | band_exceeded_pkt = band->bucket / 1000; | |
4001 | band->bucket %= 1000; /* Remainder stays in bucket. */ | |
4002 | ||
4003 | /* Update the exceeding band for each exceeding packet. | |
4004 | * (Only one band will be fired by a packet, and that | |
4005 | * can be different for each packet.) */ | |
4006 | for (i = band_exceeded_pkt; i < cnt; i++) { | |
4007 | if (band->up.rate > exceeded_rate[i]) { | |
4008 | exceeded_rate[i] = band->up.rate; | |
4009 | exceeded_band[i] = m; | |
4010 | } | |
4011 | } | |
4012 | } else { | |
4013 | /* Packet sizes differ, must process one-by-one. */ | |
4014 | band_exceeded_pkt = cnt; | |
4015 | for (i = 0; i < cnt; i++) { | |
4016 | uint32_t bits = dp_packet_size(packets_->packets[i]) * 8; | |
4017 | ||
4018 | if (band->bucket >= bits) { | |
4019 | band->bucket -= bits; | |
4020 | } else { | |
4021 | if (i < band_exceeded_pkt) { | |
4022 | band_exceeded_pkt = i; | |
4023 | } | |
4024 | /* Update the exceeding band for the exceeding packet. | |
4025 | * (Only one band will be fired by a packet, and that | |
4026 | * can be different for each packet.) */ | |
4027 | if (band->up.rate > exceeded_rate[i]) { | |
4028 | exceeded_rate[i] = band->up.rate; | |
4029 | exceeded_band[i] = m; | |
4030 | } | |
4031 | } | |
4032 | } | |
4033 | } | |
4034 | /* Remember the first exceeding packet. */ | |
4035 | if (exceeded_pkt > band_exceeded_pkt) { | |
4036 | exceeded_pkt = band_exceeded_pkt; | |
4037 | } | |
4038 | } | |
4039 | } | |
4040 | ||
4041 | /* Fire the highest rate band exceeded by each packet. | |
4042 | * Drop packets if needed, by swapping packet to the end that will be | |
4043 | * ignored. */ | |
4044 | const size_t size = dp_packet_batch_size(packets_); | |
4045 | struct dp_packet *packet; | |
4046 | size_t j; | |
4047 | DP_PACKET_BATCH_REFILL_FOR_EACH (j, size, packet, packets_) { | |
4048 | if (exceeded_band[j] >= 0) { | |
4049 | /* Meter drop packet. */ | |
4050 | band = &meter->bands[exceeded_band[j]]; | |
4051 | band->packet_count += 1; | |
4052 | band->byte_count += dp_packet_size(packet); | |
4053 | ||
4054 | dp_packet_delete(packet); | |
4055 | } else { | |
4056 | /* Meter accepts packet. */ | |
4057 | dp_packet_batch_refill(packets_, packet, j); | |
4058 | } | |
4059 | } | |
4060 | out: | |
4061 | meter_unlock(dp, meter_id); | |
4062 | } | |
4063 | ||
4064 | /* Meter set/get/del processing is still single-threaded. */ | |
5dddf960 | 4065 | static int |
4b27db64 JR |
4066 | dpif_netdev_meter_set(struct dpif *dpif, ofproto_meter_id *meter_id, |
4067 | struct ofputil_meter_config *config) | |
5dddf960 | 4068 | { |
4b27db64 JR |
4069 | struct dp_netdev *dp = get_dp_netdev(dpif); |
4070 | uint32_t mid = meter_id->uint32; | |
4071 | struct dp_meter *meter; | |
4072 | int i; | |
4073 | ||
4b27db64 JR |
4074 | if (mid >= MAX_METERS) { |
4075 | return EFBIG; /* Meter_id out of range. */ | |
4076 | } | |
4077 | ||
4078 | if (config->flags & ~DP_SUPPORTED_METER_FLAGS_MASK || | |
4079 | !(config->flags & (OFPMF13_KBPS | OFPMF13_PKTPS))) { | |
4080 | return EBADF; /* Unsupported flags set */ | |
4081 | } | |
4082 | /* Validate bands */ | |
4083 | if (config->n_bands == 0 || config->n_bands > MAX_BANDS) { | |
4084 | return EINVAL; /* Too many bands */ | |
4085 | } | |
4086 | for (i = 0; i < config->n_bands; ++i) { | |
4087 | switch (config->bands[i].type) { | |
4088 | case OFPMBT13_DROP: | |
4089 | break; | |
4090 | default: | |
4091 | return ENODEV; /* Unsupported band type */ | |
4092 | } | |
4093 | } | |
4094 | ||
4095 | /* Allocate meter */ | |
4096 | meter = xzalloc(sizeof *meter | |
4097 | + config->n_bands * sizeof(struct dp_meter_band)); | |
4098 | if (meter) { | |
4099 | meter->flags = config->flags; | |
4100 | meter->n_bands = config->n_bands; | |
4101 | meter->max_delta_t = 0; | |
4102 | meter->used = time_msec(); | |
4103 | ||
4104 | /* set up bands */ | |
4105 | for (i = 0; i < config->n_bands; ++i) { | |
4106 | uint32_t band_max_delta_t; | |
4107 | ||
4108 | /* Set burst size to a workable value if none specified. */ | |
4109 | if (config->bands[i].burst_size == 0) { | |
4110 | config->bands[i].burst_size = config->bands[i].rate; | |
4111 | } | |
4112 | ||
4113 | meter->bands[i].up = config->bands[i]; | |
4114 | /* Convert burst size to the bucket units: */ | |
4115 | /* pkts => 1/1000 packets, kilobits => bits. */ | |
4116 | meter->bands[i].up.burst_size *= 1000; | |
4117 | /* Initialize bucket to empty. */ | |
4118 | meter->bands[i].bucket = 0; | |
4119 | ||
4120 | /* Figure out max delta_t that is enough to fill any bucket. */ | |
4121 | band_max_delta_t | |
4122 | = meter->bands[i].up.burst_size / meter->bands[i].up.rate; | |
4123 | if (band_max_delta_t > meter->max_delta_t) { | |
4124 | meter->max_delta_t = band_max_delta_t; | |
4125 | } | |
4126 | } | |
4127 | ||
4128 | meter_lock(dp, mid); | |
4129 | dp_delete_meter(dp, mid); /* Free existing meter, if any */ | |
4130 | dp->meters[mid] = meter; | |
4131 | meter_unlock(dp, mid); | |
4132 | ||
4b27db64 JR |
4133 | return 0; |
4134 | } | |
4135 | return ENOMEM; | |
5dddf960 JR |
4136 | } |
4137 | ||
4138 | static int | |
4b27db64 JR |
4139 | dpif_netdev_meter_get(const struct dpif *dpif, |
4140 | ofproto_meter_id meter_id_, | |
4141 | struct ofputil_meter_stats *stats, uint16_t n_bands) | |
5dddf960 | 4142 | { |
4b27db64 JR |
4143 | const struct dp_netdev *dp = get_dp_netdev(dpif); |
4144 | const struct dp_meter *meter; | |
4145 | uint32_t meter_id = meter_id_.uint32; | |
4146 | ||
4147 | if (meter_id >= MAX_METERS) { | |
4148 | return EFBIG; | |
4149 | } | |
4150 | meter = dp->meters[meter_id]; | |
4151 | if (!meter) { | |
4152 | return ENOENT; | |
4153 | } | |
4154 | if (stats) { | |
4155 | int i = 0; | |
4156 | ||
4157 | meter_lock(dp, meter_id); | |
4158 | stats->packet_in_count = meter->packet_count; | |
4159 | stats->byte_in_count = meter->byte_count; | |
4160 | ||
4161 | for (i = 0; i < n_bands && i < meter->n_bands; ++i) { | |
4162 | stats->bands[i].packet_count = meter->bands[i].packet_count; | |
4163 | stats->bands[i].byte_count = meter->bands[i].byte_count; | |
4164 | } | |
4165 | meter_unlock(dp, meter_id); | |
4166 | ||
4167 | stats->n_bands = i; | |
4168 | } | |
4169 | return 0; | |
5dddf960 JR |
4170 | } |
4171 | ||
4172 | static int | |
4b27db64 JR |
4173 | dpif_netdev_meter_del(struct dpif *dpif, |
4174 | ofproto_meter_id meter_id_, | |
4175 | struct ofputil_meter_stats *stats, uint16_t n_bands) | |
5dddf960 | 4176 | { |
4b27db64 JR |
4177 | struct dp_netdev *dp = get_dp_netdev(dpif); |
4178 | int error; | |
4179 | ||
4180 | error = dpif_netdev_meter_get(dpif, meter_id_, stats, n_bands); | |
4181 | if (!error) { | |
4182 | uint32_t meter_id = meter_id_.uint32; | |
4183 | ||
4184 | meter_lock(dp, meter_id); | |
4185 | dp_delete_meter(dp, meter_id); | |
4186 | meter_unlock(dp, meter_id); | |
4b27db64 JR |
4187 | } |
4188 | return error; | |
5dddf960 JR |
4189 | } |
4190 | ||
4191 | \f | |
6b31e073 RW |
4192 | static void |
4193 | dpif_netdev_disable_upcall(struct dpif *dpif) | |
4194 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
4195 | { | |
4196 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
4197 | dp_netdev_disable_upcall(dp); | |
4198 | } | |
4199 | ||
4200 | static void | |
4201 | dp_netdev_enable_upcall(struct dp_netdev *dp) | |
4202 | OVS_RELEASES(dp->upcall_rwlock) | |
4203 | { | |
4204 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
4205 | } | |
4206 | ||
4207 | static void | |
4208 | dpif_netdev_enable_upcall(struct dpif *dpif) | |
4209 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
4210 | { | |
4211 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
4212 | dp_netdev_enable_upcall(dp); | |
4213 | } | |
4214 | ||
ae7ad0a1 | 4215 | static void |
accf8626 AW |
4216 | dp_netdev_pmd_reload_done(struct dp_netdev_pmd_thread *pmd) |
4217 | { | |
4218 | ovs_mutex_lock(&pmd->cond_mutex); | |
14e3e12a | 4219 | atomic_store_relaxed(&pmd->reload, false); |
2788a1b1 | 4220 | pmd->last_reload_seq = seq_read(pmd->reload_seq); |
accf8626 AW |
4221 | xpthread_cond_signal(&pmd->cond); |
4222 | ovs_mutex_unlock(&pmd->cond_mutex); | |
4223 | } | |
4224 | ||
1c1e46ed | 4225 | /* Finds and refs the dp_netdev_pmd_thread on core 'core_id'. Returns |
546e57d4 DDP |
4226 | * the pointer if succeeds, otherwise, NULL (it can return NULL even if |
4227 | * 'core_id' is NON_PMD_CORE_ID). | |
1c1e46ed AW |
4228 | * |
4229 | * Caller must unrefs the returned reference. */ | |
65f13b50 | 4230 | static struct dp_netdev_pmd_thread * |
bd5131ba | 4231 | dp_netdev_get_pmd(struct dp_netdev *dp, unsigned core_id) |
65f13b50 AW |
4232 | { |
4233 | struct dp_netdev_pmd_thread *pmd; | |
55847abe | 4234 | const struct cmap_node *pnode; |
65f13b50 | 4235 | |
b19befae | 4236 | pnode = cmap_find(&dp->poll_threads, hash_int(core_id, 0)); |
1c1e46ed AW |
4237 | if (!pnode) { |
4238 | return NULL; | |
4239 | } | |
65f13b50 AW |
4240 | pmd = CONTAINER_OF(pnode, struct dp_netdev_pmd_thread, node); |
4241 | ||
1c1e46ed | 4242 | return dp_netdev_pmd_try_ref(pmd) ? pmd : NULL; |
65f13b50 AW |
4243 | } |
4244 | ||
f2eee189 AW |
4245 | /* Sets the 'struct dp_netdev_pmd_thread' for non-pmd threads. */ |
4246 | static void | |
4247 | dp_netdev_set_nonpmd(struct dp_netdev *dp) | |
e9985d6a | 4248 | OVS_REQUIRES(dp->port_mutex) |
f2eee189 AW |
4249 | { |
4250 | struct dp_netdev_pmd_thread *non_pmd; | |
4251 | ||
4252 | non_pmd = xzalloc(sizeof *non_pmd); | |
00873463 | 4253 | dp_netdev_configure_pmd(non_pmd, dp, NON_PMD_CORE_ID, OVS_NUMA_UNSPEC); |
f2eee189 AW |
4254 | } |
4255 | ||
1c1e46ed AW |
4256 | /* Caller must have valid pointer to 'pmd'. */ |
4257 | static bool | |
4258 | dp_netdev_pmd_try_ref(struct dp_netdev_pmd_thread *pmd) | |
4259 | { | |
4260 | return ovs_refcount_try_ref_rcu(&pmd->ref_cnt); | |
4261 | } | |
4262 | ||
4263 | static void | |
4264 | dp_netdev_pmd_unref(struct dp_netdev_pmd_thread *pmd) | |
4265 | { | |
4266 | if (pmd && ovs_refcount_unref(&pmd->ref_cnt) == 1) { | |
4267 | ovsrcu_postpone(dp_netdev_destroy_pmd, pmd); | |
4268 | } | |
4269 | } | |
4270 | ||
4271 | /* Given cmap position 'pos', tries to ref the next node. If try_ref() | |
4272 | * fails, keeps checking for next node until reaching the end of cmap. | |
4273 | * | |
4274 | * Caller must unrefs the returned reference. */ | |
4275 | static struct dp_netdev_pmd_thread * | |
4276 | dp_netdev_pmd_get_next(struct dp_netdev *dp, struct cmap_position *pos) | |
4277 | { | |
4278 | struct dp_netdev_pmd_thread *next; | |
4279 | ||
4280 | do { | |
4281 | struct cmap_node *node; | |
4282 | ||
4283 | node = cmap_next_position(&dp->poll_threads, pos); | |
4284 | next = node ? CONTAINER_OF(node, struct dp_netdev_pmd_thread, node) | |
4285 | : NULL; | |
4286 | } while (next && !dp_netdev_pmd_try_ref(next)); | |
4287 | ||
4288 | return next; | |
4289 | } | |
4290 | ||
65f13b50 | 4291 | /* Configures the 'pmd' based on the input argument. */ |
6c3eee82 | 4292 | static void |
65f13b50 | 4293 | dp_netdev_configure_pmd(struct dp_netdev_pmd_thread *pmd, struct dp_netdev *dp, |
00873463 | 4294 | unsigned core_id, int numa_id) |
65f13b50 AW |
4295 | { |
4296 | pmd->dp = dp; | |
65f13b50 AW |
4297 | pmd->core_id = core_id; |
4298 | pmd->numa_id = numa_id; | |
e32971b8 | 4299 | pmd->need_reload = false; |
1c1e46ed AW |
4300 | |
4301 | ovs_refcount_init(&pmd->ref_cnt); | |
65f13b50 | 4302 | latch_init(&pmd->exit_latch); |
2788a1b1 DDP |
4303 | pmd->reload_seq = seq_create(); |
4304 | pmd->last_reload_seq = seq_read(pmd->reload_seq); | |
14e3e12a | 4305 | atomic_init(&pmd->reload, false); |
accf8626 AW |
4306 | xpthread_cond_init(&pmd->cond, NULL); |
4307 | ovs_mutex_init(&pmd->cond_mutex); | |
1c1e46ed | 4308 | ovs_mutex_init(&pmd->flow_mutex); |
d0cca6c3 | 4309 | ovs_mutex_init(&pmd->port_mutex); |
1c1e46ed | 4310 | cmap_init(&pmd->flow_table); |
3453b4d6 JS |
4311 | cmap_init(&pmd->classifiers); |
4312 | pmd->next_optimization = time_msec() + DPCLS_OPTIMIZATION_INTERVAL; | |
947dc567 | 4313 | hmap_init(&pmd->poll_list); |
d0cca6c3 | 4314 | hmap_init(&pmd->tx_ports); |
57eebbb4 DDP |
4315 | hmap_init(&pmd->tnl_port_cache); |
4316 | hmap_init(&pmd->send_port_cache); | |
65f13b50 AW |
4317 | /* init the 'flow_cache' since there is no |
4318 | * actual thread created for NON_PMD_CORE_ID. */ | |
4319 | if (core_id == NON_PMD_CORE_ID) { | |
4320 | emc_cache_init(&pmd->flow_cache); | |
140dd699 | 4321 | pmd_alloc_static_tx_qid(pmd); |
65f13b50 AW |
4322 | } |
4323 | cmap_insert(&dp->poll_threads, CONST_CAST(struct cmap_node *, &pmd->node), | |
4324 | hash_int(core_id, 0)); | |
4325 | } | |
4326 | ||
1c1e46ed AW |
4327 | static void |
4328 | dp_netdev_destroy_pmd(struct dp_netdev_pmd_thread *pmd) | |
4329 | { | |
3453b4d6 JS |
4330 | struct dpcls *cls; |
4331 | ||
1c1e46ed | 4332 | dp_netdev_pmd_flow_flush(pmd); |
57eebbb4 DDP |
4333 | hmap_destroy(&pmd->send_port_cache); |
4334 | hmap_destroy(&pmd->tnl_port_cache); | |
d0cca6c3 | 4335 | hmap_destroy(&pmd->tx_ports); |
947dc567 | 4336 | hmap_destroy(&pmd->poll_list); |
3453b4d6 JS |
4337 | /* All flows (including their dpcls_rules) have been deleted already */ |
4338 | CMAP_FOR_EACH (cls, node, &pmd->classifiers) { | |
4339 | dpcls_destroy(cls); | |
7c269972 | 4340 | ovsrcu_postpone(free, cls); |
3453b4d6 JS |
4341 | } |
4342 | cmap_destroy(&pmd->classifiers); | |
1c1e46ed AW |
4343 | cmap_destroy(&pmd->flow_table); |
4344 | ovs_mutex_destroy(&pmd->flow_mutex); | |
4345 | latch_destroy(&pmd->exit_latch); | |
2788a1b1 | 4346 | seq_destroy(pmd->reload_seq); |
1c1e46ed AW |
4347 | xpthread_cond_destroy(&pmd->cond); |
4348 | ovs_mutex_destroy(&pmd->cond_mutex); | |
d0cca6c3 | 4349 | ovs_mutex_destroy(&pmd->port_mutex); |
1c1e46ed AW |
4350 | free(pmd); |
4351 | } | |
4352 | ||
4353 | /* Stops the pmd thread, removes it from the 'dp->poll_threads', | |
4354 | * and unrefs the struct. */ | |
65f13b50 | 4355 | static void |
e4e74c3a | 4356 | dp_netdev_del_pmd(struct dp_netdev *dp, struct dp_netdev_pmd_thread *pmd) |
6c3eee82 | 4357 | { |
d0cca6c3 DDP |
4358 | /* NON_PMD_CORE_ID doesn't have a thread, so we don't have to synchronize, |
4359 | * but extra cleanup is necessary */ | |
65f13b50 | 4360 | if (pmd->core_id == NON_PMD_CORE_ID) { |
febf4a7a | 4361 | ovs_mutex_lock(&dp->non_pmd_mutex); |
65f13b50 | 4362 | emc_cache_uninit(&pmd->flow_cache); |
d0cca6c3 | 4363 | pmd_free_cached_ports(pmd); |
140dd699 | 4364 | pmd_free_static_tx_qid(pmd); |
febf4a7a | 4365 | ovs_mutex_unlock(&dp->non_pmd_mutex); |
65f13b50 AW |
4366 | } else { |
4367 | latch_set(&pmd->exit_latch); | |
4368 | dp_netdev_reload_pmd__(pmd); | |
65f13b50 AW |
4369 | xpthread_join(pmd->thread, NULL); |
4370 | } | |
ae7ad0a1 | 4371 | |
d0cca6c3 | 4372 | dp_netdev_pmd_clear_ports(pmd); |
ae7ad0a1 | 4373 | |
e4e74c3a AW |
4374 | /* Purges the 'pmd''s flows after stopping the thread, but before |
4375 | * destroying the flows, so that the flow stats can be collected. */ | |
4376 | if (dp->dp_purge_cb) { | |
4377 | dp->dp_purge_cb(dp->dp_purge_aux, pmd->core_id); | |
4378 | } | |
65f13b50 | 4379 | cmap_remove(&pmd->dp->poll_threads, &pmd->node, hash_int(pmd->core_id, 0)); |
1c1e46ed | 4380 | dp_netdev_pmd_unref(pmd); |
65f13b50 | 4381 | } |
6c3eee82 | 4382 | |
e32971b8 DDP |
4383 | /* Destroys all pmd threads. If 'non_pmd' is true it also destroys the non pmd |
4384 | * thread. */ | |
65f13b50 | 4385 | static void |
e32971b8 | 4386 | dp_netdev_destroy_all_pmds(struct dp_netdev *dp, bool non_pmd) |
65f13b50 AW |
4387 | { |
4388 | struct dp_netdev_pmd_thread *pmd; | |
d916785c DDP |
4389 | struct dp_netdev_pmd_thread **pmd_list; |
4390 | size_t k = 0, n_pmds; | |
4391 | ||
e32971b8 | 4392 | n_pmds = cmap_count(&dp->poll_threads); |
d916785c | 4393 | pmd_list = xcalloc(n_pmds, sizeof *pmd_list); |
65f13b50 AW |
4394 | |
4395 | CMAP_FOR_EACH (pmd, node, &dp->poll_threads) { | |
e32971b8 | 4396 | if (!non_pmd && pmd->core_id == NON_PMD_CORE_ID) { |
b9584f21 DDP |
4397 | continue; |
4398 | } | |
d916785c DDP |
4399 | /* We cannot call dp_netdev_del_pmd(), since it alters |
4400 | * 'dp->poll_threads' (while we're iterating it) and it | |
4401 | * might quiesce. */ | |
4402 | ovs_assert(k < n_pmds); | |
4403 | pmd_list[k++] = pmd; | |
6c3eee82 | 4404 | } |
d916785c DDP |
4405 | |
4406 | for (size_t i = 0; i < k; i++) { | |
4407 | dp_netdev_del_pmd(dp, pmd_list[i]); | |
4408 | } | |
4409 | free(pmd_list); | |
65f13b50 | 4410 | } |
6c3eee82 | 4411 | |
d0cca6c3 DDP |
4412 | /* Deletes all rx queues from pmd->poll_list and all the ports from |
4413 | * pmd->tx_ports. */ | |
cc245ce8 | 4414 | static void |
d0cca6c3 | 4415 | dp_netdev_pmd_clear_ports(struct dp_netdev_pmd_thread *pmd) |
cc245ce8 IM |
4416 | { |
4417 | struct rxq_poll *poll; | |
d0cca6c3 | 4418 | struct tx_port *port; |
cc245ce8 | 4419 | |
d0cca6c3 | 4420 | ovs_mutex_lock(&pmd->port_mutex); |
947dc567 | 4421 | HMAP_FOR_EACH_POP (poll, node, &pmd->poll_list) { |
cc245ce8 IM |
4422 | free(poll); |
4423 | } | |
d0cca6c3 DDP |
4424 | HMAP_FOR_EACH_POP (port, node, &pmd->tx_ports) { |
4425 | free(port); | |
4426 | } | |
4427 | ovs_mutex_unlock(&pmd->port_mutex); | |
cc245ce8 IM |
4428 | } |
4429 | ||
e32971b8 | 4430 | /* Adds rx queue to poll_list of PMD thread, if it's not there already. */ |
b68872d8 | 4431 | static void |
e32971b8 DDP |
4432 | dp_netdev_add_rxq_to_pmd(struct dp_netdev_pmd_thread *pmd, |
4433 | struct dp_netdev_rxq *rxq) | |
4434 | OVS_REQUIRES(pmd->port_mutex) | |
b68872d8 | 4435 | { |
e32971b8 DDP |
4436 | int qid = netdev_rxq_get_queue_id(rxq->rx); |
4437 | uint32_t hash = hash_2words(odp_to_u32(rxq->port->port_no), qid); | |
4438 | struct rxq_poll *poll; | |
b68872d8 | 4439 | |
e32971b8 DDP |
4440 | HMAP_FOR_EACH_WITH_HASH (poll, node, hash, &pmd->poll_list) { |
4441 | if (poll->rxq == rxq) { | |
4442 | /* 'rxq' is already polled by this thread. Do nothing. */ | |
4443 | return; | |
d0cca6c3 | 4444 | } |
cc245ce8 | 4445 | } |
cc245ce8 | 4446 | |
e32971b8 DDP |
4447 | poll = xmalloc(sizeof *poll); |
4448 | poll->rxq = rxq; | |
4449 | hmap_insert(&pmd->poll_list, &poll->node, hash); | |
b68872d8 | 4450 | |
e32971b8 | 4451 | pmd->need_reload = true; |
ae7ad0a1 IM |
4452 | } |
4453 | ||
e32971b8 | 4454 | /* Delete 'poll' from poll_list of PMD thread. */ |
ae7ad0a1 | 4455 | static void |
e32971b8 DDP |
4456 | dp_netdev_del_rxq_from_pmd(struct dp_netdev_pmd_thread *pmd, |
4457 | struct rxq_poll *poll) | |
d0cca6c3 | 4458 | OVS_REQUIRES(pmd->port_mutex) |
ae7ad0a1 | 4459 | { |
e32971b8 DDP |
4460 | hmap_remove(&pmd->poll_list, &poll->node); |
4461 | free(poll); | |
ae7ad0a1 | 4462 | |
e32971b8 | 4463 | pmd->need_reload = true; |
ae7ad0a1 IM |
4464 | } |
4465 | ||
d0cca6c3 DDP |
4466 | /* Add 'port' to the tx port cache of 'pmd', which must be reloaded for the |
4467 | * changes to take effect. */ | |
cc245ce8 | 4468 | static void |
d0cca6c3 DDP |
4469 | dp_netdev_add_port_tx_to_pmd(struct dp_netdev_pmd_thread *pmd, |
4470 | struct dp_netdev_port *port) | |
e32971b8 | 4471 | OVS_REQUIRES(pmd->port_mutex) |
d0cca6c3 | 4472 | { |
57eebbb4 DDP |
4473 | struct tx_port *tx; |
4474 | ||
e32971b8 DDP |
4475 | tx = tx_port_lookup(&pmd->tx_ports, port->port_no); |
4476 | if (tx) { | |
4477 | /* 'port' is already on this thread tx cache. Do nothing. */ | |
4478 | return; | |
4479 | } | |
4480 | ||
57eebbb4 | 4481 | tx = xzalloc(sizeof *tx); |
d0cca6c3 | 4482 | |
324c8374 IM |
4483 | tx->port = port; |
4484 | tx->qid = -1; | |
d0cca6c3 | 4485 | |
324c8374 | 4486 | hmap_insert(&pmd->tx_ports, &tx->node, hash_port_no(tx->port->port_no)); |
e32971b8 | 4487 | pmd->need_reload = true; |
d0cca6c3 DDP |
4488 | } |
4489 | ||
e32971b8 DDP |
4490 | /* Del 'tx' from the tx port cache of 'pmd', which must be reloaded for the |
4491 | * changes to take effect. */ | |
b9584f21 | 4492 | static void |
e32971b8 DDP |
4493 | dp_netdev_del_port_tx_from_pmd(struct dp_netdev_pmd_thread *pmd, |
4494 | struct tx_port *tx) | |
4495 | OVS_REQUIRES(pmd->port_mutex) | |
b9584f21 | 4496 | { |
e32971b8 DDP |
4497 | hmap_remove(&pmd->tx_ports, &tx->node); |
4498 | free(tx); | |
4499 | pmd->need_reload = true; | |
6c3eee82 BP |
4500 | } |
4501 | \f | |
b5cbbcf6 AZ |
4502 | static char * |
4503 | dpif_netdev_get_datapath_version(void) | |
4504 | { | |
4505 | return xstrdup("<built-in>"); | |
4506 | } | |
4507 | ||
72865317 | 4508 | static void |
1c1e46ed | 4509 | dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow, int cnt, int size, |
11bfdadd | 4510 | uint16_t tcp_flags, long long now) |
72865317 | 4511 | { |
eb94da30 | 4512 | uint16_t flags; |
72865317 | 4513 | |
eb94da30 DDP |
4514 | atomic_store_relaxed(&netdev_flow->stats.used, now); |
4515 | non_atomic_ullong_add(&netdev_flow->stats.packet_count, cnt); | |
4516 | non_atomic_ullong_add(&netdev_flow->stats.byte_count, size); | |
4517 | atomic_read_relaxed(&netdev_flow->stats.tcp_flags, &flags); | |
4518 | flags |= tcp_flags; | |
4519 | atomic_store_relaxed(&netdev_flow->stats.tcp_flags, flags); | |
51852a57 BP |
4520 | } |
4521 | ||
4522 | static void | |
1c1e46ed AW |
4523 | dp_netdev_count_packet(struct dp_netdev_pmd_thread *pmd, |
4524 | enum dp_stat_type type, int cnt) | |
51852a57 | 4525 | { |
eb94da30 | 4526 | non_atomic_ullong_add(&pmd->stats.n[type], cnt); |
51852a57 BP |
4527 | } |
4528 | ||
623540e4 | 4529 | static int |
e14deea0 | 4530 | dp_netdev_upcall(struct dp_netdev_pmd_thread *pmd, struct dp_packet *packet_, |
7af12bd7 | 4531 | struct flow *flow, struct flow_wildcards *wc, ovs_u128 *ufid, |
623540e4 EJ |
4532 | enum dpif_upcall_type type, const struct nlattr *userdata, |
4533 | struct ofpbuf *actions, struct ofpbuf *put_actions) | |
4534 | { | |
1c1e46ed | 4535 | struct dp_netdev *dp = pmd->dp; |
623540e4 | 4536 | |
623540e4 EJ |
4537 | if (OVS_UNLIKELY(!dp->upcall_cb)) { |
4538 | return ENODEV; | |
4539 | } | |
4540 | ||
4541 | if (OVS_UNLIKELY(!VLOG_DROP_DBG(&upcall_rl))) { | |
4542 | struct ds ds = DS_EMPTY_INITIALIZER; | |
623540e4 | 4543 | char *packet_str; |
cf62fa4c | 4544 | struct ofpbuf key; |
5262eea1 JG |
4545 | struct odp_flow_key_parms odp_parms = { |
4546 | .flow = flow, | |
1dea1435 | 4547 | .mask = wc ? &wc->masks : NULL, |
2494ccd7 | 4548 | .support = dp_netdev_support, |
5262eea1 | 4549 | }; |
623540e4 EJ |
4550 | |
4551 | ofpbuf_init(&key, 0); | |
5262eea1 | 4552 | odp_flow_key_from_flow(&odp_parms, &key); |
2482b0b0 | 4553 | packet_str = ofp_dp_packet_to_string(packet_); |
623540e4 | 4554 | |
6fd6ed71 | 4555 | odp_flow_key_format(key.data, key.size, &ds); |
623540e4 EJ |
4556 | |
4557 | VLOG_DBG("%s: %s upcall:\n%s\n%s", dp->name, | |
4558 | dpif_upcall_type_to_string(type), ds_cstr(&ds), packet_str); | |
4559 | ||
4560 | ofpbuf_uninit(&key); | |
4561 | free(packet_str); | |
6fd6ed71 | 4562 | |
623540e4 EJ |
4563 | ds_destroy(&ds); |
4564 | } | |
4565 | ||
8d8ab6c2 JG |
4566 | return dp->upcall_cb(packet_, flow, ufid, pmd->core_id, type, userdata, |
4567 | actions, wc, put_actions, dp->upcall_aux); | |
623540e4 EJ |
4568 | } |
4569 | ||
9bbf1c3d | 4570 | static inline uint32_t |
048963aa DDP |
4571 | dpif_netdev_packet_get_rss_hash(struct dp_packet *packet, |
4572 | const struct miniflow *mf) | |
9bbf1c3d | 4573 | { |
048963aa | 4574 | uint32_t hash, recirc_depth; |
9bbf1c3d | 4575 | |
f2f44f5d DDP |
4576 | if (OVS_LIKELY(dp_packet_rss_valid(packet))) { |
4577 | hash = dp_packet_get_rss_hash(packet); | |
4578 | } else { | |
9bbf1c3d | 4579 | hash = miniflow_hash_5tuple(mf, 0); |
2bc1bbd2 | 4580 | dp_packet_set_rss_hash(packet, hash); |
9bbf1c3d | 4581 | } |
048963aa DDP |
4582 | |
4583 | /* The RSS hash must account for the recirculation depth to avoid | |
4584 | * collisions in the exact match cache */ | |
4585 | recirc_depth = *recirc_depth_get_unsafe(); | |
4586 | if (OVS_UNLIKELY(recirc_depth)) { | |
4587 | hash = hash_finish(hash, recirc_depth); | |
4588 | dp_packet_set_rss_hash(packet, hash); | |
4589 | } | |
9bbf1c3d DDP |
4590 | return hash; |
4591 | } | |
4592 | ||
f7ce4811 | 4593 | struct packet_batch_per_flow { |
8cbf4f47 DDP |
4594 | unsigned int byte_count; |
4595 | uint16_t tcp_flags; | |
8cbf4f47 DDP |
4596 | struct dp_netdev_flow *flow; |
4597 | ||
1895cc8d | 4598 | struct dp_packet_batch array; |
8cbf4f47 DDP |
4599 | }; |
4600 | ||
4601 | static inline void | |
f7ce4811 PS |
4602 | packet_batch_per_flow_update(struct packet_batch_per_flow *batch, |
4603 | struct dp_packet *packet, | |
4604 | const struct miniflow *mf) | |
8cbf4f47 | 4605 | { |
cf62fa4c | 4606 | batch->byte_count += dp_packet_size(packet); |
1895cc8d PS |
4607 | batch->tcp_flags |= miniflow_get_tcp_flags(mf); |
4608 | batch->array.packets[batch->array.count++] = packet; | |
8cbf4f47 DDP |
4609 | } |
4610 | ||
4611 | static inline void | |
f7ce4811 PS |
4612 | packet_batch_per_flow_init(struct packet_batch_per_flow *batch, |
4613 | struct dp_netdev_flow *flow) | |
8cbf4f47 | 4614 | { |
11e5cf1f | 4615 | flow->batch = batch; |
8cbf4f47 | 4616 | |
11e5cf1f | 4617 | batch->flow = flow; |
1895cc8d | 4618 | dp_packet_batch_init(&batch->array); |
8cbf4f47 DDP |
4619 | batch->byte_count = 0; |
4620 | batch->tcp_flags = 0; | |
8cbf4f47 DDP |
4621 | } |
4622 | ||
4623 | static inline void | |
f7ce4811 PS |
4624 | packet_batch_per_flow_execute(struct packet_batch_per_flow *batch, |
4625 | struct dp_netdev_pmd_thread *pmd, | |
4626 | long long now) | |
8cbf4f47 DDP |
4627 | { |
4628 | struct dp_netdev_actions *actions; | |
4629 | struct dp_netdev_flow *flow = batch->flow; | |
4630 | ||
1895cc8d | 4631 | dp_netdev_flow_used(flow, batch->array.count, batch->byte_count, |
11bfdadd | 4632 | batch->tcp_flags, now); |
8cbf4f47 DDP |
4633 | |
4634 | actions = dp_netdev_flow_get_actions(flow); | |
4635 | ||
66e4ad8a | 4636 | dp_netdev_execute_actions(pmd, &batch->array, true, &flow->flow, |
324c8374 | 4637 | actions->actions, actions->size, now); |
8cbf4f47 DDP |
4638 | } |
4639 | ||
8aaa125d | 4640 | static inline void |
e14deea0 | 4641 | dp_netdev_queue_batches(struct dp_packet *pkt, |
9bbf1c3d | 4642 | struct dp_netdev_flow *flow, const struct miniflow *mf, |
47a45d86 KT |
4643 | struct packet_batch_per_flow *batches, |
4644 | size_t *n_batches) | |
9bbf1c3d | 4645 | { |
f7ce4811 | 4646 | struct packet_batch_per_flow *batch = flow->batch; |
11e5cf1f | 4647 | |
f9fe365b AZ |
4648 | if (OVS_UNLIKELY(!batch)) { |
4649 | batch = &batches[(*n_batches)++]; | |
f7ce4811 | 4650 | packet_batch_per_flow_init(batch, flow); |
9bbf1c3d DDP |
4651 | } |
4652 | ||
f7ce4811 | 4653 | packet_batch_per_flow_update(batch, pkt, mf); |
9bbf1c3d DDP |
4654 | } |
4655 | ||
9bbf1c3d | 4656 | /* Try to process all ('cnt') the 'packets' using only the exact match cache |
a90ed026 | 4657 | * 'pmd->flow_cache'. If a flow is not found for a packet 'packets[i]', the |
8aaa125d DDP |
4658 | * miniflow is copied into 'keys' and the packet pointer is moved at the |
4659 | * beginning of the 'packets' array. | |
9bbf1c3d DDP |
4660 | * |
4661 | * The function returns the number of packets that needs to be processed in the | |
4662 | * 'packets' array (they have been moved to the beginning of the vector). | |
a90ed026 | 4663 | * |
47a45d86 KT |
4664 | * If 'md_is_valid' is false, the metadata in 'packets' is not valid and must |
4665 | * be initialized by this function using 'port_no'. | |
9bbf1c3d DDP |
4666 | */ |
4667 | static inline size_t | |
72c84bc2 AZ |
4668 | emc_processing(struct dp_netdev_pmd_thread *pmd, |
4669 | struct dp_packet_batch *packets_, | |
1895cc8d | 4670 | struct netdev_flow_key *keys, |
f7ce4811 | 4671 | struct packet_batch_per_flow batches[], size_t *n_batches, |
a90ed026 | 4672 | bool md_is_valid, odp_port_t port_no) |
72865317 | 4673 | { |
65f13b50 | 4674 | struct emc_cache *flow_cache = &pmd->flow_cache; |
b89c678b | 4675 | struct netdev_flow_key *key = &keys[0]; |
72c84bc2 AZ |
4676 | size_t n_missed = 0, n_dropped = 0; |
4677 | struct dp_packet *packet; | |
4678 | const size_t size = dp_packet_batch_size(packets_); | |
f79b1ddb | 4679 | uint32_t cur_min; |
72c84bc2 | 4680 | int i; |
8cbf4f47 | 4681 | |
f79b1ddb BB |
4682 | atomic_read_relaxed(&pmd->dp->emc_insert_min, &cur_min); |
4683 | ||
72c84bc2 | 4684 | DP_PACKET_BATCH_REFILL_FOR_EACH (i, size, packet, packets_) { |
9bbf1c3d | 4685 | struct dp_netdev_flow *flow; |
9bbf1c3d | 4686 | |
5a2fed48 AZ |
4687 | if (OVS_UNLIKELY(dp_packet_size(packet) < ETH_HEADER_LEN)) { |
4688 | dp_packet_delete(packet); | |
3d88a620 | 4689 | n_dropped++; |
84d6d5eb EJ |
4690 | continue; |
4691 | } | |
8cbf4f47 | 4692 | |
72c84bc2 AZ |
4693 | if (i != size - 1) { |
4694 | struct dp_packet **packets = packets_->packets; | |
a90ed026 | 4695 | /* Prefetch next packet data and metadata. */ |
72a5e2b8 | 4696 | OVS_PREFETCH(dp_packet_data(packets[i+1])); |
a90ed026 | 4697 | pkt_metadata_prefetch_init(&packets[i+1]->md); |
72a5e2b8 DDP |
4698 | } |
4699 | ||
a90ed026 DDP |
4700 | if (!md_is_valid) { |
4701 | pkt_metadata_init(&packet->md, port_no); | |
4702 | } | |
5a2fed48 | 4703 | miniflow_extract(packet, &key->mf); |
d262ac2c | 4704 | key->len = 0; /* Not computed yet. */ |
5a2fed48 | 4705 | key->hash = dpif_netdev_packet_get_rss_hash(packet, &key->mf); |
9bbf1c3d | 4706 | |
f79b1ddb BB |
4707 | /* If EMC is disabled skip emc_lookup */ |
4708 | flow = (cur_min == 0) ? NULL: emc_lookup(flow_cache, key); | |
8aaa125d | 4709 | if (OVS_LIKELY(flow)) { |
5a2fed48 | 4710 | dp_netdev_queue_batches(packet, flow, &key->mf, batches, |
8aaa125d DDP |
4711 | n_batches); |
4712 | } else { | |
d1aa0b94 | 4713 | /* Exact match cache missed. Group missed packets together at |
72c84bc2 AZ |
4714 | * the beginning of the 'packets' array. */ |
4715 | dp_packet_batch_refill(packets_, packet, i); | |
400486f7 DDP |
4716 | /* 'key[n_missed]' contains the key of the current packet and it |
4717 | * must be returned to the caller. The next key should be extracted | |
4718 | * to 'keys[n_missed + 1]'. */ | |
4719 | key = &keys[++n_missed]; | |
9bbf1c3d DDP |
4720 | } |
4721 | } | |
4722 | ||
47a45d86 KT |
4723 | dp_netdev_count_packet(pmd, DP_STAT_EXACT_HIT, |
4724 | size - n_dropped - n_missed); | |
4f150744 | 4725 | |
72c84bc2 | 4726 | return dp_packet_batch_size(packets_); |
9bbf1c3d DDP |
4727 | } |
4728 | ||
a260d966 | 4729 | static inline void |
47a45d86 KT |
4730 | handle_packet_upcall(struct dp_netdev_pmd_thread *pmd, |
4731 | struct dp_packet *packet, | |
a260d966 PS |
4732 | const struct netdev_flow_key *key, |
4733 | struct ofpbuf *actions, struct ofpbuf *put_actions, | |
324c8374 | 4734 | int *lost_cnt, long long now) |
a260d966 PS |
4735 | { |
4736 | struct ofpbuf *add_actions; | |
4737 | struct dp_packet_batch b; | |
4738 | struct match match; | |
4739 | ovs_u128 ufid; | |
4740 | int error; | |
4741 | ||
4742 | match.tun_md.valid = false; | |
4743 | miniflow_expand(&key->mf, &match.flow); | |
4744 | ||
4745 | ofpbuf_clear(actions); | |
4746 | ofpbuf_clear(put_actions); | |
4747 | ||
4748 | dpif_flow_hash(pmd->dp->dpif, &match.flow, sizeof match.flow, &ufid); | |
4749 | error = dp_netdev_upcall(pmd, packet, &match.flow, &match.wc, | |
4750 | &ufid, DPIF_UC_MISS, NULL, actions, | |
4751 | put_actions); | |
4752 | if (OVS_UNLIKELY(error && error != ENOSPC)) { | |
4753 | dp_packet_delete(packet); | |
4754 | (*lost_cnt)++; | |
4755 | return; | |
4756 | } | |
4757 | ||
4758 | /* The Netlink encoding of datapath flow keys cannot express | |
4759 | * wildcarding the presence of a VLAN tag. Instead, a missing VLAN | |
4760 | * tag is interpreted as exact match on the fact that there is no | |
4761 | * VLAN. Unless we refactor a lot of code that translates between | |
4762 | * Netlink and struct flow representations, we have to do the same | |
4763 | * here. */ | |
f0fb825a EG |
4764 | if (!match.wc.masks.vlans[0].tci) { |
4765 | match.wc.masks.vlans[0].tci = htons(0xffff); | |
a260d966 PS |
4766 | } |
4767 | ||
4768 | /* We can't allow the packet batching in the next loop to execute | |
4769 | * the actions. Otherwise, if there are any slow path actions, | |
4770 | * we'll send the packet up twice. */ | |
72c84bc2 | 4771 | dp_packet_batch_init_packet(&b, packet); |
66e4ad8a | 4772 | dp_netdev_execute_actions(pmd, &b, true, &match.flow, |
324c8374 | 4773 | actions->data, actions->size, now); |
a260d966 PS |
4774 | |
4775 | add_actions = put_actions->size ? put_actions : actions; | |
4776 | if (OVS_LIKELY(error != ENOSPC)) { | |
4777 | struct dp_netdev_flow *netdev_flow; | |
4778 | ||
4779 | /* XXX: There's a race window where a flow covering this packet | |
4780 | * could have already been installed since we last did the flow | |
4781 | * lookup before upcall. This could be solved by moving the | |
4782 | * mutex lock outside the loop, but that's an awful long time | |
4783 | * to be locking everyone out of making flow installs. If we | |
4784 | * move to a per-core classifier, it would be reasonable. */ | |
4785 | ovs_mutex_lock(&pmd->flow_mutex); | |
3453b4d6 | 4786 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, key, NULL); |
a260d966 PS |
4787 | if (OVS_LIKELY(!netdev_flow)) { |
4788 | netdev_flow = dp_netdev_flow_add(pmd, &match, &ufid, | |
4789 | add_actions->data, | |
4790 | add_actions->size); | |
4791 | } | |
4792 | ovs_mutex_unlock(&pmd->flow_mutex); | |
4c30b246 | 4793 | emc_probabilistic_insert(pmd, key, netdev_flow); |
a260d966 PS |
4794 | } |
4795 | } | |
4796 | ||
9bbf1c3d | 4797 | static inline void |
65f13b50 | 4798 | fast_path_processing(struct dp_netdev_pmd_thread *pmd, |
1895cc8d | 4799 | struct dp_packet_batch *packets_, |
8aaa125d | 4800 | struct netdev_flow_key *keys, |
324c8374 | 4801 | struct packet_batch_per_flow batches[], size_t *n_batches, |
3453b4d6 | 4802 | odp_port_t in_port, |
324c8374 | 4803 | long long now) |
9bbf1c3d | 4804 | { |
1895cc8d | 4805 | int cnt = packets_->count; |
1a0d5831 | 4806 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
4807 | const size_t PKT_ARRAY_SIZE = cnt; |
4808 | #else | |
1a0d5831 | 4809 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 4810 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d | 4811 | #endif |
1895cc8d | 4812 | struct dp_packet **packets = packets_->packets; |
3453b4d6 | 4813 | struct dpcls *cls; |
0de8783a | 4814 | struct dpcls_rule *rules[PKT_ARRAY_SIZE]; |
65f13b50 | 4815 | struct dp_netdev *dp = pmd->dp; |
8aaa125d | 4816 | int miss_cnt = 0, lost_cnt = 0; |
3453b4d6 | 4817 | int lookup_cnt = 0, add_lookup_cnt; |
9bbf1c3d | 4818 | bool any_miss; |
8aaa125d | 4819 | size_t i; |
9bbf1c3d DDP |
4820 | |
4821 | for (i = 0; i < cnt; i++) { | |
0de8783a | 4822 | /* Key length is needed in all the cases, hash computed on demand. */ |
361d808d | 4823 | keys[i].len = netdev_flow_key_size(miniflow_n_values(&keys[i].mf)); |
9bbf1c3d | 4824 | } |
3453b4d6 JS |
4825 | /* Get the classifier for the in_port */ |
4826 | cls = dp_netdev_pmd_lookup_dpcls(pmd, in_port); | |
4827 | if (OVS_LIKELY(cls)) { | |
4828 | any_miss = !dpcls_lookup(cls, keys, rules, cnt, &lookup_cnt); | |
4829 | } else { | |
4830 | any_miss = true; | |
4831 | memset(rules, 0, sizeof(rules)); | |
4832 | } | |
623540e4 EJ |
4833 | if (OVS_UNLIKELY(any_miss) && !fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { |
4834 | uint64_t actions_stub[512 / 8], slow_stub[512 / 8]; | |
4835 | struct ofpbuf actions, put_actions; | |
623540e4 EJ |
4836 | |
4837 | ofpbuf_use_stub(&actions, actions_stub, sizeof actions_stub); | |
4838 | ofpbuf_use_stub(&put_actions, slow_stub, sizeof slow_stub); | |
4839 | ||
4840 | for (i = 0; i < cnt; i++) { | |
0de8783a | 4841 | struct dp_netdev_flow *netdev_flow; |
623540e4 | 4842 | |
0de8783a | 4843 | if (OVS_LIKELY(rules[i])) { |
623540e4 EJ |
4844 | continue; |
4845 | } | |
4846 | ||
4847 | /* It's possible that an earlier slow path execution installed | |
0de8783a | 4848 | * a rule covering this flow. In this case, it's a lot cheaper |
623540e4 | 4849 | * to catch it here than execute a miss. */ |
3453b4d6 JS |
4850 | netdev_flow = dp_netdev_pmd_lookup_flow(pmd, &keys[i], |
4851 | &add_lookup_cnt); | |
623540e4 | 4852 | if (netdev_flow) { |
3453b4d6 | 4853 | lookup_cnt += add_lookup_cnt; |
0de8783a | 4854 | rules[i] = &netdev_flow->cr; |
623540e4 EJ |
4855 | continue; |
4856 | } | |
4857 | ||
60fc3b7b | 4858 | miss_cnt++; |
324c8374 IM |
4859 | handle_packet_upcall(pmd, packets[i], &keys[i], &actions, |
4860 | &put_actions, &lost_cnt, now); | |
623540e4 EJ |
4861 | } |
4862 | ||
4863 | ofpbuf_uninit(&actions); | |
4864 | ofpbuf_uninit(&put_actions); | |
4865 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
ac8c2081 | 4866 | } else if (OVS_UNLIKELY(any_miss)) { |
ac8c2081 | 4867 | for (i = 0; i < cnt; i++) { |
0de8783a | 4868 | if (OVS_UNLIKELY(!rules[i])) { |
e14deea0 | 4869 | dp_packet_delete(packets[i]); |
8aaa125d DDP |
4870 | lost_cnt++; |
4871 | miss_cnt++; | |
ac8c2081 DDP |
4872 | } |
4873 | } | |
623540e4 | 4874 | } |
84d6d5eb | 4875 | |
8cbf4f47 | 4876 | for (i = 0; i < cnt; i++) { |
e14deea0 | 4877 | struct dp_packet *packet = packets[i]; |
84d6d5eb | 4878 | struct dp_netdev_flow *flow; |
8cbf4f47 | 4879 | |
0de8783a | 4880 | if (OVS_UNLIKELY(!rules[i])) { |
84d6d5eb EJ |
4881 | continue; |
4882 | } | |
4883 | ||
84d6d5eb | 4884 | flow = dp_netdev_flow_cast(rules[i]); |
0de8783a | 4885 | |
4c30b246 | 4886 | emc_probabilistic_insert(pmd, &keys[i], flow); |
8aaa125d | 4887 | dp_netdev_queue_batches(packet, flow, &keys[i].mf, batches, n_batches); |
8cbf4f47 DDP |
4888 | } |
4889 | ||
8aaa125d | 4890 | dp_netdev_count_packet(pmd, DP_STAT_MASKED_HIT, cnt - miss_cnt); |
3453b4d6 | 4891 | dp_netdev_count_packet(pmd, DP_STAT_LOOKUP_HIT, lookup_cnt); |
8aaa125d DDP |
4892 | dp_netdev_count_packet(pmd, DP_STAT_MISS, miss_cnt); |
4893 | dp_netdev_count_packet(pmd, DP_STAT_LOST, lost_cnt); | |
72865317 BP |
4894 | } |
4895 | ||
a90ed026 DDP |
4896 | /* Packets enter the datapath from a port (or from recirculation) here. |
4897 | * | |
4898 | * For performance reasons a caller may choose not to initialize the metadata | |
4899 | * in 'packets': in this case 'mdinit' is false and this function needs to | |
4900 | * initialize it using 'port_no'. If the metadata in 'packets' is already | |
4901 | * valid, 'md_is_valid' must be true and 'port_no' will be ignored. */ | |
adcf00ba | 4902 | static void |
a90ed026 | 4903 | dp_netdev_input__(struct dp_netdev_pmd_thread *pmd, |
1895cc8d | 4904 | struct dp_packet_batch *packets, |
a90ed026 | 4905 | bool md_is_valid, odp_port_t port_no) |
9bbf1c3d | 4906 | { |
1895cc8d | 4907 | int cnt = packets->count; |
1a0d5831 | 4908 | #if !defined(__CHECKER__) && !defined(_WIN32) |
9bbf1c3d DDP |
4909 | const size_t PKT_ARRAY_SIZE = cnt; |
4910 | #else | |
1a0d5831 | 4911 | /* Sparse or MSVC doesn't like variable length array. */ |
cd159f1a | 4912 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
9bbf1c3d | 4913 | #endif |
47a45d86 KT |
4914 | OVS_ALIGNED_VAR(CACHE_LINE_SIZE) |
4915 | struct netdev_flow_key keys[PKT_ARRAY_SIZE]; | |
f7ce4811 | 4916 | struct packet_batch_per_flow batches[PKT_ARRAY_SIZE]; |
11bfdadd | 4917 | long long now = time_msec(); |
72c84bc2 | 4918 | size_t n_batches; |
3453b4d6 | 4919 | odp_port_t in_port; |
9bbf1c3d | 4920 | |
8aaa125d | 4921 | n_batches = 0; |
72c84bc2 | 4922 | emc_processing(pmd, packets, keys, batches, &n_batches, |
a90ed026 | 4923 | md_is_valid, port_no); |
72c84bc2 | 4924 | if (!dp_packet_batch_is_empty(packets)) { |
3453b4d6 JS |
4925 | /* Get ingress port from first packet's metadata. */ |
4926 | in_port = packets->packets[0]->md.in_port.odp_port; | |
47a45d86 KT |
4927 | fast_path_processing(pmd, packets, keys, batches, &n_batches, |
4928 | in_port, now); | |
8aaa125d DDP |
4929 | } |
4930 | ||
ad9f0581 BB |
4931 | /* All the flow batches need to be reset before any call to |
4932 | * packet_batch_per_flow_execute() as it could potentially trigger | |
4933 | * recirculation. When a packet matching flow ‘j’ happens to be | |
4934 | * recirculated, the nested call to dp_netdev_input__() could potentially | |
4935 | * classify the packet as matching another flow - say 'k'. It could happen | |
4936 | * that in the previous call to dp_netdev_input__() that same flow 'k' had | |
4937 | * already its own batches[k] still waiting to be served. So if its | |
4938 | * ‘batch’ member is not reset, the recirculated packet would be wrongly | |
4939 | * appended to batches[k] of the 1st call to dp_netdev_input__(). */ | |
72c84bc2 | 4940 | size_t i; |
603f2ce0 EJ |
4941 | for (i = 0; i < n_batches; i++) { |
4942 | batches[i].flow->batch = NULL; | |
4943 | } | |
4944 | ||
8aaa125d | 4945 | for (i = 0; i < n_batches; i++) { |
f7ce4811 | 4946 | packet_batch_per_flow_execute(&batches[i], pmd, now); |
9bbf1c3d DDP |
4947 | } |
4948 | } | |
4949 | ||
a90ed026 DDP |
4950 | static void |
4951 | dp_netdev_input(struct dp_netdev_pmd_thread *pmd, | |
1895cc8d | 4952 | struct dp_packet_batch *packets, |
a90ed026 DDP |
4953 | odp_port_t port_no) |
4954 | { | |
3453b4d6 | 4955 | dp_netdev_input__(pmd, packets, false, port_no); |
a90ed026 DDP |
4956 | } |
4957 | ||
4958 | static void | |
4959 | dp_netdev_recirculate(struct dp_netdev_pmd_thread *pmd, | |
1895cc8d | 4960 | struct dp_packet_batch *packets) |
a90ed026 | 4961 | { |
3453b4d6 | 4962 | dp_netdev_input__(pmd, packets, true, 0); |
a90ed026 DDP |
4963 | } |
4964 | ||
9080a111 | 4965 | struct dp_netdev_execute_aux { |
65f13b50 | 4966 | struct dp_netdev_pmd_thread *pmd; |
324c8374 | 4967 | long long now; |
66e4ad8a | 4968 | const struct flow *flow; |
9080a111 JR |
4969 | }; |
4970 | ||
e4e74c3a AW |
4971 | static void |
4972 | dpif_netdev_register_dp_purge_cb(struct dpif *dpif, dp_purge_callback *cb, | |
4973 | void *aux) | |
4974 | { | |
4975 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
4976 | dp->dp_purge_aux = aux; | |
4977 | dp->dp_purge_cb = cb; | |
4978 | } | |
4979 | ||
6b31e073 | 4980 | static void |
623540e4 EJ |
4981 | dpif_netdev_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, |
4982 | void *aux) | |
6b31e073 RW |
4983 | { |
4984 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
623540e4 | 4985 | dp->upcall_aux = aux; |
6b31e073 RW |
4986 | dp->upcall_cb = cb; |
4987 | } | |
4988 | ||
324c8374 IM |
4989 | static void |
4990 | dpif_netdev_xps_revalidate_pmd(const struct dp_netdev_pmd_thread *pmd, | |
4991 | long long now, bool purge) | |
4992 | { | |
4993 | struct tx_port *tx; | |
4994 | struct dp_netdev_port *port; | |
4995 | long long interval; | |
4996 | ||
57eebbb4 | 4997 | HMAP_FOR_EACH (tx, node, &pmd->send_port_cache) { |
9f7a3035 | 4998 | if (!tx->port->dynamic_txqs) { |
324c8374 IM |
4999 | continue; |
5000 | } | |
5001 | interval = now - tx->last_used; | |
5002 | if (tx->qid >= 0 && (purge || interval >= XPS_TIMEOUT_MS)) { | |
5003 | port = tx->port; | |
5004 | ovs_mutex_lock(&port->txq_used_mutex); | |
5005 | port->txq_used[tx->qid]--; | |
5006 | ovs_mutex_unlock(&port->txq_used_mutex); | |
5007 | tx->qid = -1; | |
5008 | } | |
5009 | } | |
5010 | } | |
5011 | ||
5012 | static int | |
5013 | dpif_netdev_xps_get_tx_qid(const struct dp_netdev_pmd_thread *pmd, | |
5014 | struct tx_port *tx, long long now) | |
5015 | { | |
5016 | struct dp_netdev_port *port; | |
5017 | long long interval; | |
5018 | int i, min_cnt, min_qid; | |
5019 | ||
5020 | if (OVS_UNLIKELY(!now)) { | |
5021 | now = time_msec(); | |
5022 | } | |
5023 | ||
5024 | interval = now - tx->last_used; | |
5025 | tx->last_used = now; | |
5026 | ||
5027 | if (OVS_LIKELY(tx->qid >= 0 && interval < XPS_TIMEOUT_MS)) { | |
5028 | return tx->qid; | |
5029 | } | |
5030 | ||
5031 | port = tx->port; | |
5032 | ||
5033 | ovs_mutex_lock(&port->txq_used_mutex); | |
5034 | if (tx->qid >= 0) { | |
5035 | port->txq_used[tx->qid]--; | |
5036 | tx->qid = -1; | |
5037 | } | |
5038 | ||
5039 | min_cnt = -1; | |
5040 | min_qid = 0; | |
5041 | for (i = 0; i < netdev_n_txq(port->netdev); i++) { | |
5042 | if (port->txq_used[i] < min_cnt || min_cnt == -1) { | |
5043 | min_cnt = port->txq_used[i]; | |
5044 | min_qid = i; | |
5045 | } | |
5046 | } | |
5047 | ||
5048 | port->txq_used[min_qid]++; | |
5049 | tx->qid = min_qid; | |
5050 | ||
5051 | ovs_mutex_unlock(&port->txq_used_mutex); | |
5052 | ||
5053 | dpif_netdev_xps_revalidate_pmd(pmd, now, false); | |
5054 | ||
5055 | VLOG_DBG("Core %d: New TX queue ID %d for port \'%s\'.", | |
5056 | pmd->core_id, tx->qid, netdev_get_name(tx->port->netdev)); | |
5057 | return min_qid; | |
5058 | } | |
5059 | ||
d0cca6c3 | 5060 | static struct tx_port * |
57eebbb4 DDP |
5061 | pmd_tnl_port_cache_lookup(const struct dp_netdev_pmd_thread *pmd, |
5062 | odp_port_t port_no) | |
5063 | { | |
5064 | return tx_port_lookup(&pmd->tnl_port_cache, port_no); | |
5065 | } | |
5066 | ||
5067 | static struct tx_port * | |
5068 | pmd_send_port_cache_lookup(const struct dp_netdev_pmd_thread *pmd, | |
5069 | odp_port_t port_no) | |
d0cca6c3 | 5070 | { |
57eebbb4 | 5071 | return tx_port_lookup(&pmd->send_port_cache, port_no); |
d0cca6c3 DDP |
5072 | } |
5073 | ||
a36de779 | 5074 | static int |
d0cca6c3 | 5075 | push_tnl_action(const struct dp_netdev_pmd_thread *pmd, |
1895cc8d PS |
5076 | const struct nlattr *attr, |
5077 | struct dp_packet_batch *batch) | |
a36de779 | 5078 | { |
d0cca6c3 | 5079 | struct tx_port *tun_port; |
a36de779 | 5080 | const struct ovs_action_push_tnl *data; |
4c742796 | 5081 | int err; |
a36de779 PS |
5082 | |
5083 | data = nl_attr_get(attr); | |
5084 | ||
81765c00 | 5085 | tun_port = pmd_tnl_port_cache_lookup(pmd, data->tnl_port); |
a36de779 | 5086 | if (!tun_port) { |
4c742796 PS |
5087 | err = -EINVAL; |
5088 | goto error; | |
a36de779 | 5089 | } |
324c8374 | 5090 | err = netdev_push_header(tun_port->port->netdev, batch, data); |
4c742796 PS |
5091 | if (!err) { |
5092 | return 0; | |
5093 | } | |
5094 | error: | |
5095 | dp_packet_delete_batch(batch, true); | |
5096 | return err; | |
a36de779 PS |
5097 | } |
5098 | ||
66525ef3 PS |
5099 | static void |
5100 | dp_execute_userspace_action(struct dp_netdev_pmd_thread *pmd, | |
5101 | struct dp_packet *packet, bool may_steal, | |
5102 | struct flow *flow, ovs_u128 *ufid, | |
5103 | struct ofpbuf *actions, | |
324c8374 | 5104 | const struct nlattr *userdata, long long now) |
66525ef3 PS |
5105 | { |
5106 | struct dp_packet_batch b; | |
5107 | int error; | |
5108 | ||
5109 | ofpbuf_clear(actions); | |
5110 | ||
5111 | error = dp_netdev_upcall(pmd, packet, flow, NULL, ufid, | |
5112 | DPIF_UC_ACTION, userdata, actions, | |
5113 | NULL); | |
5114 | if (!error || error == ENOSPC) { | |
72c84bc2 | 5115 | dp_packet_batch_init_packet(&b, packet); |
66e4ad8a | 5116 | dp_netdev_execute_actions(pmd, &b, may_steal, flow, |
324c8374 | 5117 | actions->data, actions->size, now); |
66525ef3 PS |
5118 | } else if (may_steal) { |
5119 | dp_packet_delete(packet); | |
5120 | } | |
5121 | } | |
5122 | ||
a36de779 | 5123 | static void |
1895cc8d | 5124 | dp_execute_cb(void *aux_, struct dp_packet_batch *packets_, |
09f9da0b | 5125 | const struct nlattr *a, bool may_steal) |
4b27db64 | 5126 | OVS_NO_THREAD_SAFETY_ANALYSIS |
9080a111 JR |
5127 | { |
5128 | struct dp_netdev_execute_aux *aux = aux_; | |
623540e4 | 5129 | uint32_t *depth = recirc_depth_get(); |
28e2fa02 DDP |
5130 | struct dp_netdev_pmd_thread *pmd = aux->pmd; |
5131 | struct dp_netdev *dp = pmd->dp; | |
09f9da0b | 5132 | int type = nl_attr_type(a); |
324c8374 | 5133 | long long now = aux->now; |
d0cca6c3 | 5134 | struct tx_port *p; |
9080a111 | 5135 | |
09f9da0b JR |
5136 | switch ((enum ovs_action_attr)type) { |
5137 | case OVS_ACTION_ATTR_OUTPUT: | |
57eebbb4 | 5138 | p = pmd_send_port_cache_lookup(pmd, nl_attr_get_odp_port(a)); |
26a5075b | 5139 | if (OVS_LIKELY(p)) { |
347ba9bb | 5140 | int tx_qid; |
324c8374 | 5141 | bool dynamic_txqs; |
347ba9bb | 5142 | |
324c8374 IM |
5143 | dynamic_txqs = p->port->dynamic_txqs; |
5144 | if (dynamic_txqs) { | |
5145 | tx_qid = dpif_netdev_xps_get_tx_qid(pmd, p, now); | |
5146 | } else { | |
82d765f6 | 5147 | tx_qid = pmd->static_tx_qid; |
324c8374 | 5148 | } |
347ba9bb | 5149 | |
324c8374 IM |
5150 | netdev_send(p->port->netdev, tx_qid, packets_, may_steal, |
5151 | dynamic_txqs); | |
ac8c2081 | 5152 | return; |
8a4e3a85 | 5153 | } |
09f9da0b JR |
5154 | break; |
5155 | ||
a36de779 PS |
5156 | case OVS_ACTION_ATTR_TUNNEL_PUSH: |
5157 | if (*depth < MAX_RECIRC_DEPTH) { | |
aaca4fe0 | 5158 | dp_packet_batch_apply_cutlen(packets_); |
7c12dfc5 | 5159 | push_tnl_action(pmd, a, packets_); |
a36de779 PS |
5160 | return; |
5161 | } | |
5162 | break; | |
5163 | ||
5164 | case OVS_ACTION_ATTR_TUNNEL_POP: | |
5165 | if (*depth < MAX_RECIRC_DEPTH) { | |
aaca4fe0 | 5166 | struct dp_packet_batch *orig_packets_ = packets_; |
8611f9a4 | 5167 | odp_port_t portno = nl_attr_get_odp_port(a); |
a36de779 | 5168 | |
57eebbb4 | 5169 | p = pmd_tnl_port_cache_lookup(pmd, portno); |
a36de779 | 5170 | if (p) { |
1895cc8d | 5171 | struct dp_packet_batch tnl_pkt; |
a36de779 PS |
5172 | |
5173 | if (!may_steal) { | |
aaca4fe0 WT |
5174 | dp_packet_batch_clone(&tnl_pkt, packets_); |
5175 | packets_ = &tnl_pkt; | |
5176 | dp_packet_batch_reset_cutlen(orig_packets_); | |
a36de779 PS |
5177 | } |
5178 | ||
aaca4fe0 WT |
5179 | dp_packet_batch_apply_cutlen(packets_); |
5180 | ||
324c8374 | 5181 | netdev_pop_header(p->port->netdev, packets_); |
72c84bc2 | 5182 | if (dp_packet_batch_is_empty(packets_)) { |
1c8f98d9 PS |
5183 | return; |
5184 | } | |
9235b479 | 5185 | |
72c84bc2 AZ |
5186 | struct dp_packet *packet; |
5187 | DP_PACKET_BATCH_FOR_EACH (packet, packets_) { | |
5188 | packet->md.in_port.odp_port = portno; | |
a36de779 | 5189 | } |
9235b479 PS |
5190 | |
5191 | (*depth)++; | |
5192 | dp_netdev_recirculate(pmd, packets_); | |
5193 | (*depth)--; | |
a36de779 PS |
5194 | return; |
5195 | } | |
5196 | } | |
5197 | break; | |
5198 | ||
623540e4 EJ |
5199 | case OVS_ACTION_ATTR_USERSPACE: |
5200 | if (!fat_rwlock_tryrdlock(&dp->upcall_rwlock)) { | |
aaca4fe0 | 5201 | struct dp_packet_batch *orig_packets_ = packets_; |
623540e4 | 5202 | const struct nlattr *userdata; |
aaca4fe0 | 5203 | struct dp_packet_batch usr_pkt; |
623540e4 EJ |
5204 | struct ofpbuf actions; |
5205 | struct flow flow; | |
7af12bd7 | 5206 | ovs_u128 ufid; |
aaca4fe0 | 5207 | bool clone = false; |
4fc65926 | 5208 | |
623540e4 EJ |
5209 | userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA); |
5210 | ofpbuf_init(&actions, 0); | |
8cbf4f47 | 5211 | |
aaca4fe0 WT |
5212 | if (packets_->trunc) { |
5213 | if (!may_steal) { | |
5214 | dp_packet_batch_clone(&usr_pkt, packets_); | |
5215 | packets_ = &usr_pkt; | |
aaca4fe0 WT |
5216 | clone = true; |
5217 | dp_packet_batch_reset_cutlen(orig_packets_); | |
5218 | } | |
5219 | ||
5220 | dp_packet_batch_apply_cutlen(packets_); | |
5221 | } | |
5222 | ||
72c84bc2 AZ |
5223 | struct dp_packet *packet; |
5224 | DP_PACKET_BATCH_FOR_EACH (packet, packets_) { | |
5225 | flow_extract(packet, &flow); | |
7af12bd7 | 5226 | dpif_flow_hash(dp->dpif, &flow, sizeof flow, &ufid); |
72c84bc2 | 5227 | dp_execute_userspace_action(pmd, packet, may_steal, &flow, |
324c8374 | 5228 | &ufid, &actions, userdata, now); |
db73f716 | 5229 | } |
aaca4fe0 WT |
5230 | |
5231 | if (clone) { | |
5232 | dp_packet_delete_batch(packets_, true); | |
5233 | } | |
5234 | ||
623540e4 EJ |
5235 | ofpbuf_uninit(&actions); |
5236 | fat_rwlock_unlock(&dp->upcall_rwlock); | |
6b31e073 | 5237 | |
ac8c2081 DDP |
5238 | return; |
5239 | } | |
09f9da0b | 5240 | break; |
572f732a | 5241 | |
adcf00ba AZ |
5242 | case OVS_ACTION_ATTR_RECIRC: |
5243 | if (*depth < MAX_RECIRC_DEPTH) { | |
1895cc8d | 5244 | struct dp_packet_batch recirc_pkts; |
572f732a | 5245 | |
28e2fa02 | 5246 | if (!may_steal) { |
1895cc8d PS |
5247 | dp_packet_batch_clone(&recirc_pkts, packets_); |
5248 | packets_ = &recirc_pkts; | |
28e2fa02 | 5249 | } |
8cbf4f47 | 5250 | |
72c84bc2 AZ |
5251 | struct dp_packet *packet; |
5252 | DP_PACKET_BATCH_FOR_EACH (packet, packets_) { | |
5253 | packet->md.recirc_id = nl_attr_get_u32(a); | |
8cbf4f47 | 5254 | } |
28e2fa02 DDP |
5255 | |
5256 | (*depth)++; | |
1895cc8d | 5257 | dp_netdev_recirculate(pmd, packets_); |
adcf00ba AZ |
5258 | (*depth)--; |
5259 | ||
ac8c2081 | 5260 | return; |
adcf00ba | 5261 | } |
ac8c2081 DDP |
5262 | |
5263 | VLOG_WARN("Packet dropped. Max recirculation depth exceeded."); | |
572f732a | 5264 | break; |
572f732a | 5265 | |
5cf3edb3 DDP |
5266 | case OVS_ACTION_ATTR_CT: { |
5267 | const struct nlattr *b; | |
a76a37ef | 5268 | bool force = false; |
5cf3edb3 DDP |
5269 | bool commit = false; |
5270 | unsigned int left; | |
5271 | uint16_t zone = 0; | |
5272 | const char *helper = NULL; | |
5273 | const uint32_t *setmark = NULL; | |
5274 | const struct ovs_key_ct_labels *setlabel = NULL; | |
4cddb1f0 DB |
5275 | struct nat_action_info_t nat_action_info; |
5276 | struct nat_action_info_t *nat_action_info_ref = NULL; | |
5277 | bool nat_config = false; | |
5cf3edb3 DDP |
5278 | |
5279 | NL_ATTR_FOR_EACH_UNSAFE (b, left, nl_attr_get(a), | |
5280 | nl_attr_get_size(a)) { | |
5281 | enum ovs_ct_attr sub_type = nl_attr_type(b); | |
5282 | ||
5283 | switch(sub_type) { | |
b80e259f | 5284 | case OVS_CT_ATTR_FORCE_COMMIT: |
a76a37ef JR |
5285 | force = true; |
5286 | /* fall through. */ | |
5cf3edb3 DDP |
5287 | case OVS_CT_ATTR_COMMIT: |
5288 | commit = true; | |
5289 | break; | |
5290 | case OVS_CT_ATTR_ZONE: | |
5291 | zone = nl_attr_get_u16(b); | |
5292 | break; | |
5293 | case OVS_CT_ATTR_HELPER: | |
5294 | helper = nl_attr_get_string(b); | |
5295 | break; | |
5296 | case OVS_CT_ATTR_MARK: | |
5297 | setmark = nl_attr_get(b); | |
5298 | break; | |
5299 | case OVS_CT_ATTR_LABELS: | |
5300 | setlabel = nl_attr_get(b); | |
5301 | break; | |
8e83854c JR |
5302 | case OVS_CT_ATTR_EVENTMASK: |
5303 | /* Silently ignored, as userspace datapath does not generate | |
5304 | * netlink events. */ | |
5305 | break; | |
4cddb1f0 DB |
5306 | case OVS_CT_ATTR_NAT: { |
5307 | const struct nlattr *b_nest; | |
5308 | unsigned int left_nest; | |
5309 | bool ip_min_specified = false; | |
5310 | bool proto_num_min_specified = false; | |
5311 | bool ip_max_specified = false; | |
5312 | bool proto_num_max_specified = false; | |
5313 | memset(&nat_action_info, 0, sizeof nat_action_info); | |
5314 | nat_action_info_ref = &nat_action_info; | |
5315 | ||
5316 | NL_NESTED_FOR_EACH_UNSAFE (b_nest, left_nest, b) { | |
5317 | enum ovs_nat_attr sub_type_nest = nl_attr_type(b_nest); | |
5318 | ||
5319 | switch (sub_type_nest) { | |
5320 | case OVS_NAT_ATTR_SRC: | |
5321 | case OVS_NAT_ATTR_DST: | |
5322 | nat_config = true; | |
5323 | nat_action_info.nat_action |= | |
5324 | ((sub_type_nest == OVS_NAT_ATTR_SRC) | |
5325 | ? NAT_ACTION_SRC : NAT_ACTION_DST); | |
5326 | break; | |
5327 | case OVS_NAT_ATTR_IP_MIN: | |
5328 | memcpy(&nat_action_info.min_addr, | |
5329 | nl_attr_get(b_nest), | |
5330 | nl_attr_get_size(b_nest)); | |
5331 | ip_min_specified = true; | |
5332 | break; | |
5333 | case OVS_NAT_ATTR_IP_MAX: | |
5334 | memcpy(&nat_action_info.max_addr, | |
5335 | nl_attr_get(b_nest), | |
5336 | nl_attr_get_size(b_nest)); | |
5337 | ip_max_specified = true; | |
5338 | break; | |
5339 | case OVS_NAT_ATTR_PROTO_MIN: | |
5340 | nat_action_info.min_port = | |
5341 | nl_attr_get_u16(b_nest); | |
5342 | proto_num_min_specified = true; | |
5343 | break; | |
5344 | case OVS_NAT_ATTR_PROTO_MAX: | |
5345 | nat_action_info.max_port = | |
5346 | nl_attr_get_u16(b_nest); | |
5347 | proto_num_max_specified = true; | |
5348 | break; | |
5349 | case OVS_NAT_ATTR_PERSISTENT: | |
5350 | case OVS_NAT_ATTR_PROTO_HASH: | |
5351 | case OVS_NAT_ATTR_PROTO_RANDOM: | |
5352 | break; | |
5353 | case OVS_NAT_ATTR_UNSPEC: | |
5354 | case __OVS_NAT_ATTR_MAX: | |
5355 | OVS_NOT_REACHED(); | |
5356 | } | |
5357 | } | |
5358 | ||
5359 | if (ip_min_specified && !ip_max_specified) { | |
5360 | nat_action_info.max_addr = nat_action_info.min_addr; | |
5361 | } | |
5362 | if (proto_num_min_specified && !proto_num_max_specified) { | |
5363 | nat_action_info.max_port = nat_action_info.min_port; | |
5364 | } | |
5365 | if (proto_num_min_specified || proto_num_max_specified) { | |
5366 | if (nat_action_info.nat_action & NAT_ACTION_SRC) { | |
5367 | nat_action_info.nat_action |= NAT_ACTION_SRC_PORT; | |
5368 | } else if (nat_action_info.nat_action & NAT_ACTION_DST) { | |
5369 | nat_action_info.nat_action |= NAT_ACTION_DST_PORT; | |
5370 | } | |
5371 | } | |
5372 | break; | |
5373 | } | |
5cf3edb3 DDP |
5374 | case OVS_CT_ATTR_UNSPEC: |
5375 | case __OVS_CT_ATTR_MAX: | |
5376 | OVS_NOT_REACHED(); | |
5377 | } | |
5378 | } | |
5379 | ||
4cddb1f0 DB |
5380 | /* We won't be able to function properly in this case, hence |
5381 | * complain loudly. */ | |
5382 | if (nat_config && !commit) { | |
5383 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
5384 | VLOG_WARN_RL(&rl, "NAT specified without commit."); | |
5385 | } | |
5386 | ||
a76a37ef | 5387 | conntrack_execute(&dp->conntrack, packets_, aux->flow->dl_type, force, |
4cddb1f0 DB |
5388 | commit, zone, setmark, setlabel, helper, |
5389 | nat_action_info_ref); | |
07659514 | 5390 | break; |
5cf3edb3 | 5391 | } |
07659514 | 5392 | |
5dddf960 | 5393 | case OVS_ACTION_ATTR_METER: |
4b27db64 JR |
5394 | dp_netdev_run_meter(pmd->dp, packets_, nl_attr_get_u32(a), |
5395 | time_msec()); | |
5396 | break; | |
5397 | ||
09f9da0b JR |
5398 | case OVS_ACTION_ATTR_PUSH_VLAN: |
5399 | case OVS_ACTION_ATTR_POP_VLAN: | |
5400 | case OVS_ACTION_ATTR_PUSH_MPLS: | |
5401 | case OVS_ACTION_ATTR_POP_MPLS: | |
5402 | case OVS_ACTION_ATTR_SET: | |
6d670e7f | 5403 | case OVS_ACTION_ATTR_SET_MASKED: |
09f9da0b | 5404 | case OVS_ACTION_ATTR_SAMPLE: |
53e1d6f1 | 5405 | case OVS_ACTION_ATTR_HASH: |
09f9da0b | 5406 | case OVS_ACTION_ATTR_UNSPEC: |
aaca4fe0 | 5407 | case OVS_ACTION_ATTR_TRUNC: |
6fcecb85 YY |
5408 | case OVS_ACTION_ATTR_PUSH_ETH: |
5409 | case OVS_ACTION_ATTR_POP_ETH: | |
535e3acf | 5410 | case OVS_ACTION_ATTR_CLONE: |
09f9da0b JR |
5411 | case __OVS_ACTION_ATTR_MAX: |
5412 | OVS_NOT_REACHED(); | |
da546e07 | 5413 | } |
ac8c2081 | 5414 | |
1895cc8d | 5415 | dp_packet_delete_batch(packets_, may_steal); |
98403001 BP |
5416 | } |
5417 | ||
4edb9ae9 | 5418 | static void |
65f13b50 | 5419 | dp_netdev_execute_actions(struct dp_netdev_pmd_thread *pmd, |
1895cc8d | 5420 | struct dp_packet_batch *packets, |
66e4ad8a | 5421 | bool may_steal, const struct flow *flow, |
324c8374 IM |
5422 | const struct nlattr *actions, size_t actions_len, |
5423 | long long now) | |
72865317 | 5424 | { |
66e4ad8a | 5425 | struct dp_netdev_execute_aux aux = { pmd, now, flow }; |
9080a111 | 5426 | |
1895cc8d | 5427 | odp_execute_actions(&aux, packets, may_steal, actions, |
8cbf4f47 | 5428 | actions_len, dp_execute_cb); |
72865317 BP |
5429 | } |
5430 | ||
4d4e68ed DDP |
5431 | struct dp_netdev_ct_dump { |
5432 | struct ct_dpif_dump_state up; | |
5433 | struct conntrack_dump dump; | |
5434 | struct conntrack *ct; | |
5435 | struct dp_netdev *dp; | |
5436 | }; | |
5437 | ||
5438 | static int | |
5439 | dpif_netdev_ct_dump_start(struct dpif *dpif, struct ct_dpif_dump_state **dump_, | |
5440 | const uint16_t *pzone) | |
5441 | { | |
5442 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5443 | struct dp_netdev_ct_dump *dump; | |
5444 | ||
5445 | dump = xzalloc(sizeof *dump); | |
5446 | dump->dp = dp; | |
5447 | dump->ct = &dp->conntrack; | |
5448 | ||
5449 | conntrack_dump_start(&dp->conntrack, &dump->dump, pzone); | |
5450 | ||
5451 | *dump_ = &dump->up; | |
5452 | ||
5453 | return 0; | |
5454 | } | |
5455 | ||
5456 | static int | |
5457 | dpif_netdev_ct_dump_next(struct dpif *dpif OVS_UNUSED, | |
5458 | struct ct_dpif_dump_state *dump_, | |
5459 | struct ct_dpif_entry *entry) | |
5460 | { | |
5461 | struct dp_netdev_ct_dump *dump; | |
5462 | ||
5463 | INIT_CONTAINER(dump, dump_, up); | |
5464 | ||
5465 | return conntrack_dump_next(&dump->dump, entry); | |
5466 | } | |
5467 | ||
5468 | static int | |
5469 | dpif_netdev_ct_dump_done(struct dpif *dpif OVS_UNUSED, | |
5470 | struct ct_dpif_dump_state *dump_) | |
5471 | { | |
5472 | struct dp_netdev_ct_dump *dump; | |
5473 | int err; | |
5474 | ||
5475 | INIT_CONTAINER(dump, dump_, up); | |
5476 | ||
5477 | err = conntrack_dump_done(&dump->dump); | |
5478 | ||
5479 | free(dump); | |
5480 | ||
5481 | return err; | |
5482 | } | |
5483 | ||
5d9cbb4c DDP |
5484 | static int |
5485 | dpif_netdev_ct_flush(struct dpif *dpif, const uint16_t *zone) | |
5486 | { | |
5487 | struct dp_netdev *dp = get_dp_netdev(dpif); | |
5488 | ||
5489 | return conntrack_flush(&dp->conntrack, zone); | |
5490 | } | |
5491 | ||
72865317 | 5492 | const struct dpif_class dpif_netdev_class = { |
72865317 | 5493 | "netdev", |
6553d06b | 5494 | dpif_netdev_init, |
2197d7ab | 5495 | dpif_netdev_enumerate, |
0aeaabc8 | 5496 | dpif_netdev_port_open_type, |
72865317 BP |
5497 | dpif_netdev_open, |
5498 | dpif_netdev_close, | |
7dab847a | 5499 | dpif_netdev_destroy, |
e4cfed38 PS |
5500 | dpif_netdev_run, |
5501 | dpif_netdev_wait, | |
72865317 | 5502 | dpif_netdev_get_stats, |
72865317 BP |
5503 | dpif_netdev_port_add, |
5504 | dpif_netdev_port_del, | |
3eb67853 | 5505 | dpif_netdev_port_set_config, |
72865317 BP |
5506 | dpif_netdev_port_query_by_number, |
5507 | dpif_netdev_port_query_by_name, | |
98403001 | 5508 | NULL, /* port_get_pid */ |
b0ec0f27 BP |
5509 | dpif_netdev_port_dump_start, |
5510 | dpif_netdev_port_dump_next, | |
5511 | dpif_netdev_port_dump_done, | |
72865317 BP |
5512 | dpif_netdev_port_poll, |
5513 | dpif_netdev_port_poll_wait, | |
72865317 | 5514 | dpif_netdev_flow_flush, |
ac64794a BP |
5515 | dpif_netdev_flow_dump_create, |
5516 | dpif_netdev_flow_dump_destroy, | |
5517 | dpif_netdev_flow_dump_thread_create, | |
5518 | dpif_netdev_flow_dump_thread_destroy, | |
704a1e09 | 5519 | dpif_netdev_flow_dump_next, |
1a0c894a | 5520 | dpif_netdev_operate, |
6b31e073 RW |
5521 | NULL, /* recv_set */ |
5522 | NULL, /* handlers_set */ | |
d4f6865c | 5523 | dpif_netdev_set_config, |
5bf93d67 | 5524 | dpif_netdev_queue_to_priority, |
6b31e073 RW |
5525 | NULL, /* recv */ |
5526 | NULL, /* recv_wait */ | |
5527 | NULL, /* recv_purge */ | |
e4e74c3a | 5528 | dpif_netdev_register_dp_purge_cb, |
6b31e073 RW |
5529 | dpif_netdev_register_upcall_cb, |
5530 | dpif_netdev_enable_upcall, | |
5531 | dpif_netdev_disable_upcall, | |
b5cbbcf6 | 5532 | dpif_netdev_get_datapath_version, |
4d4e68ed DDP |
5533 | dpif_netdev_ct_dump_start, |
5534 | dpif_netdev_ct_dump_next, | |
5535 | dpif_netdev_ct_dump_done, | |
5d9cbb4c | 5536 | dpif_netdev_ct_flush, |
5dddf960 JR |
5537 | dpif_netdev_meter_get_features, |
5538 | dpif_netdev_meter_set, | |
5539 | dpif_netdev_meter_get, | |
5540 | dpif_netdev_meter_del, | |
72865317 | 5541 | }; |
614c4892 | 5542 | |
74cc3969 BP |
5543 | static void |
5544 | dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
5545 | const char *argv[], void *aux OVS_UNUSED) | |
5546 | { | |
e9985d6a | 5547 | struct dp_netdev_port *port; |
74cc3969 | 5548 | struct dp_netdev *dp; |
ff073a71 | 5549 | odp_port_t port_no; |
74cc3969 | 5550 | |
8a4e3a85 | 5551 | ovs_mutex_lock(&dp_netdev_mutex); |
74cc3969 BP |
5552 | dp = shash_find_data(&dp_netdevs, argv[1]); |
5553 | if (!dp || !dpif_netdev_class_is_dummy(dp->class)) { | |
8a4e3a85 | 5554 | ovs_mutex_unlock(&dp_netdev_mutex); |
74cc3969 BP |
5555 | unixctl_command_reply_error(conn, "unknown datapath or not a dummy"); |
5556 | return; | |
5557 | } | |
8a4e3a85 BP |
5558 | ovs_refcount_ref(&dp->ref_cnt); |
5559 | ovs_mutex_unlock(&dp_netdev_mutex); | |
74cc3969 | 5560 | |
59e6d833 | 5561 | ovs_mutex_lock(&dp->port_mutex); |
e9985d6a | 5562 | if (get_port_by_name(dp, argv[2], &port)) { |
74cc3969 | 5563 | unixctl_command_reply_error(conn, "unknown port"); |
8a4e3a85 | 5564 | goto exit; |
74cc3969 BP |
5565 | } |
5566 | ||
ff073a71 BP |
5567 | port_no = u32_to_odp(atoi(argv[3])); |
5568 | if (!port_no || port_no == ODPP_NONE) { | |
74cc3969 | 5569 | unixctl_command_reply_error(conn, "bad port number"); |
8a4e3a85 | 5570 | goto exit; |
74cc3969 | 5571 | } |
ff073a71 | 5572 | if (dp_netdev_lookup_port(dp, port_no)) { |
74cc3969 | 5573 | unixctl_command_reply_error(conn, "port number already in use"); |
8a4e3a85 | 5574 | goto exit; |
74cc3969 | 5575 | } |
59e6d833 | 5576 | |
e9985d6a DDP |
5577 | /* Remove port. */ |
5578 | hmap_remove(&dp->ports, &port->node); | |
e32971b8 | 5579 | reconfigure_datapath(dp); |
59e6d833 | 5580 | |
e9985d6a DDP |
5581 | /* Reinsert with new port number. */ |
5582 | port->port_no = port_no; | |
5583 | hmap_insert(&dp->ports, &port->node, hash_port_no(port_no)); | |
e32971b8 | 5584 | reconfigure_datapath(dp); |
59e6d833 | 5585 | |
d33ed218 | 5586 | seq_change(dp->port_seq); |
74cc3969 | 5587 | unixctl_command_reply(conn, NULL); |
8a4e3a85 BP |
5588 | |
5589 | exit: | |
59e6d833 | 5590 | ovs_mutex_unlock(&dp->port_mutex); |
8a4e3a85 | 5591 | dp_netdev_unref(dp); |
74cc3969 BP |
5592 | } |
5593 | ||
0cbfe35d BP |
5594 | static void |
5595 | dpif_dummy_register__(const char *type) | |
5596 | { | |
5597 | struct dpif_class *class; | |
5598 | ||
5599 | class = xmalloc(sizeof *class); | |
5600 | *class = dpif_netdev_class; | |
5601 | class->type = xstrdup(type); | |
5602 | dp_register_provider(class); | |
5603 | } | |
5604 | ||
8420c7ad BP |
5605 | static void |
5606 | dpif_dummy_override(const char *type) | |
5607 | { | |
65d43fdc YT |
5608 | int error; |
5609 | ||
5610 | /* | |
5611 | * Ignore EAFNOSUPPORT to allow --enable-dummy=system with | |
5612 | * a userland-only build. It's useful for testsuite. | |
5613 | */ | |
5614 | error = dp_unregister_provider(type); | |
5615 | if (error == 0 || error == EAFNOSUPPORT) { | |
8420c7ad BP |
5616 | dpif_dummy_register__(type); |
5617 | } | |
5618 | } | |
5619 | ||
614c4892 | 5620 | void |
8420c7ad | 5621 | dpif_dummy_register(enum dummy_level level) |
614c4892 | 5622 | { |
8420c7ad | 5623 | if (level == DUMMY_OVERRIDE_ALL) { |
0cbfe35d BP |
5624 | struct sset types; |
5625 | const char *type; | |
5626 | ||
5627 | sset_init(&types); | |
5628 | dp_enumerate_types(&types); | |
5629 | SSET_FOR_EACH (type, &types) { | |
8420c7ad | 5630 | dpif_dummy_override(type); |
0cbfe35d BP |
5631 | } |
5632 | sset_destroy(&types); | |
8420c7ad BP |
5633 | } else if (level == DUMMY_OVERRIDE_SYSTEM) { |
5634 | dpif_dummy_override("system"); | |
614c4892 | 5635 | } |
0cbfe35d BP |
5636 | |
5637 | dpif_dummy_register__("dummy"); | |
74cc3969 BP |
5638 | |
5639 | unixctl_command_register("dpif-dummy/change-port-number", | |
74467d5c | 5640 | "dp port new-number", |
74cc3969 | 5641 | 3, 3, dpif_dummy_change_port_number, NULL); |
614c4892 | 5642 | } |
0de8783a JR |
5643 | \f |
5644 | /* Datapath Classifier. */ | |
5645 | ||
5646 | /* A set of rules that all have the same fields wildcarded. */ | |
5647 | struct dpcls_subtable { | |
5648 | /* The fields are only used by writers. */ | |
5649 | struct cmap_node cmap_node OVS_GUARDED; /* Within dpcls 'subtables_map'. */ | |
5650 | ||
5651 | /* These fields are accessed by readers. */ | |
5652 | struct cmap rules; /* Contains "struct dpcls_rule"s. */ | |
3453b4d6 JS |
5653 | uint32_t hit_cnt; /* Number of match hits in subtable in current |
5654 | optimization interval. */ | |
0de8783a JR |
5655 | struct netdev_flow_key mask; /* Wildcards for fields (const). */ |
5656 | /* 'mask' must be the last field, additional space is allocated here. */ | |
5657 | }; | |
5658 | ||
5659 | /* Initializes 'cls' as a classifier that initially contains no classification | |
5660 | * rules. */ | |
5661 | static void | |
5662 | dpcls_init(struct dpcls *cls) | |
5663 | { | |
5664 | cmap_init(&cls->subtables_map); | |
da9cfca6 | 5665 | pvector_init(&cls->subtables); |
0de8783a JR |
5666 | } |
5667 | ||
5668 | static void | |
5669 | dpcls_destroy_subtable(struct dpcls *cls, struct dpcls_subtable *subtable) | |
5670 | { | |
3453b4d6 | 5671 | VLOG_DBG("Destroying subtable %p for in_port %d", subtable, cls->in_port); |
da9cfca6 | 5672 | pvector_remove(&cls->subtables, subtable); |
0de8783a JR |
5673 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, |
5674 | subtable->mask.hash); | |
5675 | cmap_destroy(&subtable->rules); | |
5676 | ovsrcu_postpone(free, subtable); | |
5677 | } | |
5678 | ||
5679 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
5680 | * caller's responsibility. | |
5681 | * May only be called after all the readers have been terminated. */ | |
5682 | static void | |
5683 | dpcls_destroy(struct dpcls *cls) | |
5684 | { | |
5685 | if (cls) { | |
5686 | struct dpcls_subtable *subtable; | |
5687 | ||
5688 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
361d808d | 5689 | ovs_assert(cmap_count(&subtable->rules) == 0); |
0de8783a JR |
5690 | dpcls_destroy_subtable(cls, subtable); |
5691 | } | |
5692 | cmap_destroy(&cls->subtables_map); | |
da9cfca6 | 5693 | pvector_destroy(&cls->subtables); |
0de8783a JR |
5694 | } |
5695 | } | |
5696 | ||
5697 | static struct dpcls_subtable * | |
5698 | dpcls_create_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
5699 | { | |
5700 | struct dpcls_subtable *subtable; | |
5701 | ||
5702 | /* Need to add one. */ | |
caeb4906 JR |
5703 | subtable = xmalloc(sizeof *subtable |
5704 | - sizeof subtable->mask.mf + mask->len); | |
0de8783a | 5705 | cmap_init(&subtable->rules); |
3453b4d6 | 5706 | subtable->hit_cnt = 0; |
0de8783a JR |
5707 | netdev_flow_key_clone(&subtable->mask, mask); |
5708 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, mask->hash); | |
3453b4d6 | 5709 | /* Add the new subtable at the end of the pvector (with no hits yet) */ |
da9cfca6 | 5710 | pvector_insert(&cls->subtables, subtable, 0); |
84dbfb2b | 5711 | VLOG_DBG("Creating %"PRIuSIZE". subtable %p for in_port %d", |
3453b4d6 | 5712 | cmap_count(&cls->subtables_map), subtable, cls->in_port); |
da9cfca6 | 5713 | pvector_publish(&cls->subtables); |
0de8783a JR |
5714 | |
5715 | return subtable; | |
5716 | } | |
5717 | ||
5718 | static inline struct dpcls_subtable * | |
5719 | dpcls_find_subtable(struct dpcls *cls, const struct netdev_flow_key *mask) | |
5720 | { | |
5721 | struct dpcls_subtable *subtable; | |
5722 | ||
5723 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, mask->hash, | |
5724 | &cls->subtables_map) { | |
5725 | if (netdev_flow_key_equal(&subtable->mask, mask)) { | |
5726 | return subtable; | |
5727 | } | |
5728 | } | |
5729 | return dpcls_create_subtable(cls, mask); | |
5730 | } | |
5731 | ||
3453b4d6 JS |
5732 | |
5733 | /* Periodically sort the dpcls subtable vectors according to hit counts */ | |
5734 | static void | |
5735 | dpcls_sort_subtable_vector(struct dpcls *cls) | |
5736 | { | |
5737 | struct pvector *pvec = &cls->subtables; | |
5738 | struct dpcls_subtable *subtable; | |
5739 | ||
5740 | PVECTOR_FOR_EACH (subtable, pvec) { | |
5741 | pvector_change_priority(pvec, subtable, subtable->hit_cnt); | |
5742 | subtable->hit_cnt = 0; | |
5743 | } | |
5744 | pvector_publish(pvec); | |
5745 | } | |
5746 | ||
5747 | static inline void | |
5748 | dp_netdev_pmd_try_optimize(struct dp_netdev_pmd_thread *pmd) | |
5749 | { | |
5750 | struct dpcls *cls; | |
5751 | long long int now = time_msec(); | |
5752 | ||
5753 | if (now > pmd->next_optimization) { | |
5754 | /* Try to obtain the flow lock to block out revalidator threads. | |
5755 | * If not possible, just try next time. */ | |
5756 | if (!ovs_mutex_trylock(&pmd->flow_mutex)) { | |
5757 | /* Optimize each classifier */ | |
5758 | CMAP_FOR_EACH (cls, node, &pmd->classifiers) { | |
5759 | dpcls_sort_subtable_vector(cls); | |
5760 | } | |
5761 | ovs_mutex_unlock(&pmd->flow_mutex); | |
5762 | /* Start new measuring interval */ | |
5763 | pmd->next_optimization = now + DPCLS_OPTIMIZATION_INTERVAL; | |
5764 | } | |
5765 | } | |
5766 | } | |
5767 | ||
0de8783a JR |
5768 | /* Insert 'rule' into 'cls'. */ |
5769 | static void | |
5770 | dpcls_insert(struct dpcls *cls, struct dpcls_rule *rule, | |
5771 | const struct netdev_flow_key *mask) | |
5772 | { | |
5773 | struct dpcls_subtable *subtable = dpcls_find_subtable(cls, mask); | |
5774 | ||
3453b4d6 | 5775 | /* Refer to subtable's mask, also for later removal. */ |
0de8783a JR |
5776 | rule->mask = &subtable->mask; |
5777 | cmap_insert(&subtable->rules, &rule->cmap_node, rule->flow.hash); | |
5778 | } | |
5779 | ||
5780 | /* Removes 'rule' from 'cls', also destructing the 'rule'. */ | |
5781 | static void | |
5782 | dpcls_remove(struct dpcls *cls, struct dpcls_rule *rule) | |
5783 | { | |
5784 | struct dpcls_subtable *subtable; | |
5785 | ||
5786 | ovs_assert(rule->mask); | |
5787 | ||
3453b4d6 | 5788 | /* Get subtable from reference in rule->mask. */ |
0de8783a | 5789 | INIT_CONTAINER(subtable, rule->mask, mask); |
0de8783a JR |
5790 | if (cmap_remove(&subtable->rules, &rule->cmap_node, rule->flow.hash) |
5791 | == 0) { | |
3453b4d6 | 5792 | /* Delete empty subtable. */ |
0de8783a | 5793 | dpcls_destroy_subtable(cls, subtable); |
da9cfca6 | 5794 | pvector_publish(&cls->subtables); |
0de8783a JR |
5795 | } |
5796 | } | |
5797 | ||
361d808d JR |
5798 | /* Returns true if 'target' satisfies 'key' in 'mask', that is, if each 1-bit |
5799 | * in 'mask' the values in 'key' and 'target' are the same. */ | |
0de8783a JR |
5800 | static inline bool |
5801 | dpcls_rule_matches_key(const struct dpcls_rule *rule, | |
5802 | const struct netdev_flow_key *target) | |
5803 | { | |
09b0fa9c JR |
5804 | const uint64_t *keyp = miniflow_get_values(&rule->flow.mf); |
5805 | const uint64_t *maskp = miniflow_get_values(&rule->mask->mf); | |
5fcff47b | 5806 | uint64_t value; |
0de8783a | 5807 | |
5fcff47b JR |
5808 | NETDEV_FLOW_KEY_FOR_EACH_IN_FLOWMAP(value, target, rule->flow.mf.map) { |
5809 | if (OVS_UNLIKELY((value & *maskp++) != *keyp++)) { | |
0de8783a JR |
5810 | return false; |
5811 | } | |
5812 | } | |
5813 | return true; | |
5814 | } | |
5815 | ||
5b1c9c78 FA |
5816 | /* For each miniflow in 'keys' performs a classifier lookup writing the result |
5817 | * into the corresponding slot in 'rules'. If a particular entry in 'keys' is | |
0de8783a JR |
5818 | * NULL it is skipped. |
5819 | * | |
5820 | * This function is optimized for use in the userspace datapath and therefore | |
5821 | * does not implement a lot of features available in the standard | |
5822 | * classifier_lookup() function. Specifically, it does not implement | |
5823 | * priorities, instead returning any rule which matches the flow. | |
5824 | * | |
5b1c9c78 | 5825 | * Returns true if all miniflows found a corresponding rule. */ |
0de8783a | 5826 | static bool |
3453b4d6 JS |
5827 | dpcls_lookup(struct dpcls *cls, const struct netdev_flow_key keys[], |
5828 | struct dpcls_rule **rules, const size_t cnt, | |
5829 | int *num_lookups_p) | |
0de8783a | 5830 | { |
5b1c9c78 | 5831 | /* The received 'cnt' miniflows are the search-keys that will be processed |
63906f18 BB |
5832 | * to find a matching entry into the available subtables. |
5833 | * The number of bits in map_type is equal to NETDEV_MAX_BURST. */ | |
5834 | typedef uint32_t map_type; | |
0de8783a | 5835 | #define MAP_BITS (sizeof(map_type) * CHAR_BIT) |
63906f18 | 5836 | BUILD_ASSERT_DECL(MAP_BITS >= NETDEV_MAX_BURST); |
0de8783a | 5837 | |
0de8783a JR |
5838 | struct dpcls_subtable *subtable; |
5839 | ||
63906f18 BB |
5840 | map_type keys_map = TYPE_MAXIMUM(map_type); /* Set all bits. */ |
5841 | map_type found_map; | |
5842 | uint32_t hashes[MAP_BITS]; | |
5843 | const struct cmap_node *nodes[MAP_BITS]; | |
5844 | ||
5845 | if (cnt != MAP_BITS) { | |
5846 | keys_map >>= MAP_BITS - cnt; /* Clear extra bits. */ | |
0de8783a JR |
5847 | } |
5848 | memset(rules, 0, cnt * sizeof *rules); | |
5849 | ||
3453b4d6 JS |
5850 | int lookups_match = 0, subtable_pos = 1; |
5851 | ||
5b1c9c78 FA |
5852 | /* The Datapath classifier - aka dpcls - is composed of subtables. |
5853 | * Subtables are dynamically created as needed when new rules are inserted. | |
5854 | * Each subtable collects rules with matches on a specific subset of packet | |
5855 | * fields as defined by the subtable's mask. We proceed to process every | |
5856 | * search-key against each subtable, but when a match is found for a | |
5857 | * search-key, the search for that key can stop because the rules are | |
5858 | * non-overlapping. */ | |
da9cfca6 | 5859 | PVECTOR_FOR_EACH (subtable, &cls->subtables) { |
63906f18 BB |
5860 | int i; |
5861 | ||
5862 | /* Compute hashes for the remaining keys. Each search-key is | |
5863 | * masked with the subtable's mask to avoid hashing the wildcarded | |
5864 | * bits. */ | |
5865 | ULLONG_FOR_EACH_1(i, keys_map) { | |
5866 | hashes[i] = netdev_flow_key_hash_in_mask(&keys[i], | |
5867 | &subtable->mask); | |
5868 | } | |
5869 | /* Lookup. */ | |
5870 | found_map = cmap_find_batch(&subtable->rules, keys_map, hashes, nodes); | |
5871 | /* Check results. When the i-th bit of found_map is set, it means | |
5872 | * that a set of nodes with a matching hash value was found for the | |
5873 | * i-th search-key. Due to possible hash collisions we need to check | |
5874 | * which of the found rules, if any, really matches our masked | |
5875 | * search-key. */ | |
5876 | ULLONG_FOR_EACH_1(i, found_map) { | |
5877 | struct dpcls_rule *rule; | |
5878 | ||
5879 | CMAP_NODE_FOR_EACH (rule, cmap_node, nodes[i]) { | |
5880 | if (OVS_LIKELY(dpcls_rule_matches_key(rule, &keys[i]))) { | |
5881 | rules[i] = rule; | |
5882 | /* Even at 20 Mpps the 32-bit hit_cnt cannot wrap | |
5883 | * within one second optimization interval. */ | |
5884 | subtable->hit_cnt++; | |
5885 | lookups_match += subtable_pos; | |
5886 | goto next; | |
0de8783a | 5887 | } |
0de8783a | 5888 | } |
63906f18 BB |
5889 | /* None of the found rules was a match. Reset the i-th bit to |
5890 | * keep searching this key in the next subtable. */ | |
5891 | ULLONG_SET0(found_map, i); /* Did not match. */ | |
5892 | next: | |
5893 | ; /* Keep Sparse happy. */ | |
0de8783a | 5894 | } |
63906f18 BB |
5895 | keys_map &= ~found_map; /* Clear the found rules. */ |
5896 | if (!keys_map) { | |
3453b4d6 JS |
5897 | if (num_lookups_p) { |
5898 | *num_lookups_p = lookups_match; | |
5899 | } | |
0de8783a JR |
5900 | return true; /* All found. */ |
5901 | } | |
3453b4d6 JS |
5902 | subtable_pos++; |
5903 | } | |
5904 | if (num_lookups_p) { | |
5905 | *num_lookups_p = lookups_match; | |
0de8783a JR |
5906 | } |
5907 | return false; /* Some misses. */ | |
5908 | } |