1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2017 6WIND S.A.
3 * Copyright 2017 Mellanox Technologies, Ltd
6 #ifndef _RTE_ETH_FAILSAFE_PRIVATE_H_
7 #define _RTE_ETH_FAILSAFE_PRIVATE_H_
12 #include <rte_atomic.h>
14 #include <rte_ethdev_driver.h>
15 #include <rte_devargs.h>
16 #include <rte_interrupts.h>
18 #define FAILSAFE_DRIVER_NAME "Fail-safe PMD"
19 #define FAILSAFE_OWNER_NAME "Fail-safe"
21 #define PMD_FAILSAFE_MAC_KVARG "mac"
22 #define PMD_FAILSAFE_HOTPLUG_POLL_KVARG "hotplug_poll"
23 #define PMD_FAILSAFE_PARAM_STRING \
25 "exec(<shell command>)," \
31 #define FAILSAFE_HOTPLUG_DEFAULT_TIMEOUT_MS 2000
33 #define FAILSAFE_MAX_ETHPORTS 2
34 #define FAILSAFE_MAX_ETHADDR 128
36 #define DEVARGS_MAXLEN 4096
38 enum rxp_service_state {
48 /* epoll file descriptor */
50 /* event vector to be used by epoll */
51 struct rte_epoll_event *evec;
56 enum rxp_service_state sstate;
62 /* next sub_device to poll */
63 struct sub_device *sdev;
64 unsigned int socket_id;
66 unsigned int enable_events:1;
67 struct rte_eth_rxq_info info;
68 rte_atomic64_t refcnt[];
74 unsigned int socket_id;
75 struct rte_eth_txq_info info;
76 rte_atomic64_t refcnt[];
80 TAILQ_ENTRY(rte_flow) next;
82 struct rte_flow *flows[FAILSAFE_MAX_ETHPORTS];
83 /* flow description for synchronization */
84 struct rte_flow_desc *fd;
96 struct rte_eth_stats stats;
101 /* Exhaustive DPDK device description */
102 struct sub_device *next;
103 struct rte_devargs devargs;
105 struct rte_device *dev;
106 struct rte_eth_dev *edev;
108 /* Device state machine */
109 enum dev_state state;
110 /* Last stats snapshot passed to user */
111 struct fs_stats stats_snapshot;
112 /* Some device are defined as a command line */
114 /* Others are retrieved through a file descriptor */
116 /* fail-safe device backreference */
117 struct rte_eth_dev *fs_dev;
118 /* flag calling for recollection */
119 volatile unsigned int remove:1;
120 /* flow isolation state */
125 struct rte_eth_dev *dev;
127 * Set of sub_devices.
128 * subs[0] is the preferred device
129 * any other is just another slave
131 struct sub_device *subs;
132 uint8_t subs_head; /* if head == tail, no subs */
133 uint8_t subs_tail; /* first invalid */
134 uint8_t subs_tx; /* current emitting device */
135 uint8_t current_probed;
137 TAILQ_HEAD(sub_flows, rte_flow) flow_list;
138 /* current number of mac_addr slots allocated. */
139 uint32_t nb_mac_addr;
140 struct ether_addr mac_addrs[FAILSAFE_MAX_ETHADDR];
141 uint32_t mac_addr_pool[FAILSAFE_MAX_ETHADDR];
142 /* current capabilities */
143 struct rte_eth_dev_info infos;
144 struct rte_eth_dev_owner my_owner; /* Unique owner. */
145 struct rte_intr_handle intr_handle; /* Port interrupt handle. */
147 * Fail-safe state machine.
148 * This level will be tracking state of the EAL and eth
149 * layer at large as defined by the user application.
150 * It will then steer the sub_devices toward the same
151 * synchronized state.
153 enum dev_state state;
154 struct rte_eth_stats stats_accumulator;
156 * Rx interrupts/events proxy.
157 * The PMD issues Rx events to the EAL on behalf of its subdevices,
158 * it does that by registering an event-fd for each of its queues with
159 * the EAL. A PMD service thread listens to all the Rx events from the
160 * subdevices, when an Rx event is issued by a subdevice it will be
161 * caught by this service with will trigger an Rx event in the
162 * appropriate failsafe Rx queue.
165 pthread_mutex_t hotplug_mutex;
166 /* Hot-plug mutex is locked by the alarm mechanism. */
167 volatile unsigned int alarm_lock:1;
168 unsigned int pending_alarm:1; /* An alarm is pending */
169 /* flow isolation state */
175 int failsafe_rx_intr_install(struct rte_eth_dev *dev);
176 void failsafe_rx_intr_uninstall(struct rte_eth_dev *dev);
177 int failsafe_rx_intr_install_subdevice(struct sub_device *sdev);
178 void failsafe_rx_intr_uninstall_subdevice(struct sub_device *sdev);
182 int failsafe_hotplug_alarm_install(struct rte_eth_dev *dev);
183 int failsafe_hotplug_alarm_cancel(struct rte_eth_dev *dev);
187 void set_burst_fn(struct rte_eth_dev *dev, int force_safe);
189 uint16_t failsafe_rx_burst(void *rxq,
190 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
191 uint16_t failsafe_tx_burst(void *txq,
192 struct rte_mbuf **tx_pkts, uint16_t nb_pkts);
194 uint16_t failsafe_rx_burst_fast(void *rxq,
195 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
196 uint16_t failsafe_tx_burst_fast(void *txq,
197 struct rte_mbuf **tx_pkts, uint16_t nb_pkts);
201 int failsafe_args_parse(struct rte_eth_dev *dev, const char *params);
202 void failsafe_args_free(struct rte_eth_dev *dev);
203 int failsafe_args_count_subdevice(struct rte_eth_dev *dev, const char *params);
204 int failsafe_args_parse_subs(struct rte_eth_dev *dev);
208 int failsafe_eal_init(struct rte_eth_dev *dev);
209 int failsafe_eal_uninit(struct rte_eth_dev *dev);
213 int failsafe_eth_dev_state_sync(struct rte_eth_dev *dev);
214 void failsafe_dev_remove(struct rte_eth_dev *dev);
215 void failsafe_stats_increment(struct rte_eth_stats *to,
216 struct rte_eth_stats *from);
217 int failsafe_eth_rmv_event_callback(uint16_t port_id,
218 enum rte_eth_event_type type,
219 void *arg, void *out);
220 int failsafe_eth_lsc_event_callback(uint16_t port_id,
221 enum rte_eth_event_type event,
222 void *cb_arg, void *out);
226 extern const char pmd_failsafe_driver_name[];
227 extern const struct eth_dev_ops failsafe_ops;
228 extern const struct rte_flow_ops fs_flow_ops;
229 extern uint64_t hotplug_poll;
230 extern int mac_from_arg;
234 /* dev: (struct rte_eth_dev *) fail-safe device */
236 ((struct fs_priv *)(dev)->data->dev_private)
238 /* sdev: (struct sub_device *) */
242 /* sdev: (struct sub_device *) */
243 #define PORT_ID(sdev) \
244 (ETH(sdev)->data->port_id)
246 /* sdev: (struct sub_device *) */
247 #define SUB_ID(sdev) \
251 * Stateful iterator construct over fail-safe sub-devices:
252 * s: (struct sub_device *), iterator
253 * i: (uint8_t), increment
254 * dev: (struct rte_eth_dev *), fail-safe ethdev
255 * state: (enum dev_state), minimum acceptable device state
257 #define FOREACH_SUBDEV_STATE(s, i, dev, state) \
258 for (s = fs_find_next((dev), 0, state, &i); \
260 s = fs_find_next((dev), i + 1, state, &i))
263 * Iterator construct over fail-safe sub-devices:
264 * s: (struct sub_device *), iterator
265 * i: (uint8_t), increment
266 * dev: (struct rte_eth_dev *), fail-safe ethdev
268 #define FOREACH_SUBDEV(s, i, dev) \
269 FOREACH_SUBDEV_STATE(s, i, dev, DEV_UNDEFINED)
271 /* dev: (struct rte_eth_dev *) fail-safe device */
272 #define PREFERRED_SUBDEV(dev) \
273 (&PRIV(dev)->subs[0])
275 /* dev: (struct rte_eth_dev *) fail-safe device */
276 #define TX_SUBDEV(dev) \
277 (PRIV(dev)->subs_tx >= PRIV(dev)->subs_tail ? NULL \
278 : (PRIV(dev)->subs[PRIV(dev)->subs_tx].state < DEV_PROBED ? NULL \
279 : &PRIV(dev)->subs[PRIV(dev)->subs_tx]))
282 * s: (struct sub_device *)
283 * ops: (struct eth_dev_ops) member
285 #define SUBOPS(s, ops) \
286 (ETH(s)->dev_ops->ops)
293 * a: (rte_atomic64_t)
295 #define FS_ATOMIC_P(a) \
296 rte_atomic64_set(&(a), 1)
299 * a: (rte_atomic64_t)
301 #define FS_ATOMIC_V(a) \
302 rte_atomic64_set(&(a), 0)
305 * s: (struct sub_device *)
308 #define FS_ATOMIC_RX(s, i) \
310 &((struct rxq *)((s)->fs_dev->data->rx_queues[i]))->refcnt[(s)->sid] \
313 * s: (struct sub_device *)
316 #define FS_ATOMIC_TX(s, i) \
318 &((struct txq *)((s)->fs_dev->data->tx_queues[i]))->refcnt[(s)->sid] \
321 #ifdef RTE_EXEC_ENV_BSDAPP
322 #define FS_THREADID_TYPE void*
323 #define FS_THREADID_FMT "p"
325 #define FS_THREADID_TYPE unsigned long
326 #define FS_THREADID_FMT "lu"
329 #define LOG__(level, m, ...) \
330 RTE_LOG(level, PMD, "net_failsafe: " m "%c", __VA_ARGS__)
331 #define LOG_(level, ...) LOG__(level, __VA_ARGS__, '\n')
332 #define DEBUG(...) LOG_(DEBUG, __VA_ARGS__)
333 #define INFO(...) LOG_(INFO, __VA_ARGS__)
334 #define WARN(...) LOG_(WARNING, __VA_ARGS__)
335 #define ERROR(...) LOG_(ERR, __VA_ARGS__)
337 /* inlined functions */
339 static inline struct sub_device *
340 fs_find_next(struct rte_eth_dev *dev,
342 enum dev_state min_state,
345 struct sub_device *subs;
348 subs = PRIV(dev)->subs;
349 tail = PRIV(dev)->subs_tail;
351 if (subs[sid].state >= min_state)
362 * Lock hot-plug mutex.
363 * is_alarm means that the caller is, for sure, the hot-plug alarm mechanism.
366 fs_lock(struct rte_eth_dev *dev, unsigned int is_alarm)
371 ret = pthread_mutex_trylock(&PRIV(dev)->hotplug_mutex);
373 DEBUG("Hot-plug mutex lock trying failed(%s), will try"
374 " again later...", strerror(ret));
377 PRIV(dev)->alarm_lock = 1;
379 ret = pthread_mutex_lock(&PRIV(dev)->hotplug_mutex);
381 ERROR("Cannot lock mutex(%s)", strerror(ret));
385 DEBUG("Hot-plug mutex was locked by thread %" FS_THREADID_FMT "%s",
386 (FS_THREADID_TYPE)pthread_self(),
387 PRIV(dev)->alarm_lock ? " by the hot-plug alarm" : "");
392 * Unlock hot-plug mutex.
393 * is_alarm means that the caller is, for sure, the hot-plug alarm mechanism.
396 fs_unlock(struct rte_eth_dev *dev, unsigned int is_alarm)
399 unsigned int prev_alarm_lock = PRIV(dev)->alarm_lock;
402 RTE_ASSERT(PRIV(dev)->alarm_lock == 1);
403 PRIV(dev)->alarm_lock = 0;
405 ret = pthread_mutex_unlock(&PRIV(dev)->hotplug_mutex);
407 ERROR("Cannot unlock hot-plug mutex(%s)", strerror(ret));
409 DEBUG("Hot-plug mutex was unlocked by thread %" FS_THREADID_FMT "%s",
410 (FS_THREADID_TYPE)pthread_self(),
411 prev_alarm_lock ? " by the hot-plug alarm" : "");
415 * Switch emitting device.
416 * If banned is set, banned must not be considered for
417 * the role of emitting device.
420 fs_switch_dev(struct rte_eth_dev *dev,
421 struct sub_device *banned)
423 struct sub_device *txd;
424 enum dev_state req_state;
426 req_state = PRIV(dev)->state;
427 txd = TX_SUBDEV(dev);
428 if (PREFERRED_SUBDEV(dev)->state >= req_state &&
429 PREFERRED_SUBDEV(dev) != banned) {
430 if (txd != PREFERRED_SUBDEV(dev) &&
432 (req_state == DEV_STARTED) ||
433 (txd && txd->state < DEV_STARTED))) {
434 DEBUG("Switching tx_dev to preferred sub_device");
435 PRIV(dev)->subs_tx = 0;
437 } else if ((txd && txd->state < req_state) ||
440 struct sub_device *sdev = NULL;
443 /* Using acceptable device */
444 FOREACH_SUBDEV_STATE(sdev, i, dev, req_state) {
447 DEBUG("Switching tx_dev to sub_device %d",
449 PRIV(dev)->subs_tx = i;
452 if (i >= PRIV(dev)->subs_tail || sdev == NULL) {
453 DEBUG("No device ready, deactivating tx_dev");
454 PRIV(dev)->subs_tx = PRIV(dev)->subs_tail;
459 set_burst_fn(dev, 0);
464 * Adjust error value and rte_errno to the fail-safe actual error value.
467 fs_err(struct sub_device *sdev, int err)
469 /* A device removal shouldn't be reported as an error. */
470 if (sdev->remove == 1 || err == -EIO)
471 return rte_errno = 0;
474 #endif /* _RTE_ETH_FAILSAFE_PRIVATE_H_ */