]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/drivers/bus/vdev/vdev.c
import 15.2.0 Octopus source
[ceph.git] / ceph / src / spdk / dpdk / drivers / bus / vdev / vdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 RehiveTech. All rights reserved.
3 */
4
5 #include <string.h>
6 #include <inttypes.h>
7 #include <stdio.h>
8 #include <stdlib.h>
9 #include <stdint.h>
10 #include <stdbool.h>
11 #include <sys/queue.h>
12
13 #include <rte_eal.h>
14 #include <rte_dev.h>
15 #include <rte_bus.h>
16 #include <rte_common.h>
17 #include <rte_devargs.h>
18 #include <rte_memory.h>
19 #include <rte_tailq.h>
20 #include <rte_spinlock.h>
21 #include <rte_string_fns.h>
22 #include <rte_errno.h>
23
24 #include "rte_bus_vdev.h"
25 #include "vdev_logs.h"
26 #include "vdev_private.h"
27
28 #define VDEV_MP_KEY "bus_vdev_mp"
29
30 int vdev_logtype_bus;
31
32 /* Forward declare to access virtual bus name */
33 static struct rte_bus rte_vdev_bus;
34
35 /** Double linked list of virtual device drivers. */
36 TAILQ_HEAD(vdev_device_list, rte_vdev_device);
37
38 static struct vdev_device_list vdev_device_list =
39 TAILQ_HEAD_INITIALIZER(vdev_device_list);
40 /* The lock needs to be recursive because a vdev can manage another vdev. */
41 static rte_spinlock_recursive_t vdev_device_list_lock =
42 RTE_SPINLOCK_RECURSIVE_INITIALIZER;
43
44 static struct vdev_driver_list vdev_driver_list =
45 TAILQ_HEAD_INITIALIZER(vdev_driver_list);
46
47 struct vdev_custom_scan {
48 TAILQ_ENTRY(vdev_custom_scan) next;
49 rte_vdev_scan_callback callback;
50 void *user_arg;
51 };
52 TAILQ_HEAD(vdev_custom_scans, vdev_custom_scan);
53 static struct vdev_custom_scans vdev_custom_scans =
54 TAILQ_HEAD_INITIALIZER(vdev_custom_scans);
55 static rte_spinlock_t vdev_custom_scan_lock = RTE_SPINLOCK_INITIALIZER;
56
57 /* register a driver */
58 void
59 rte_vdev_register(struct rte_vdev_driver *driver)
60 {
61 TAILQ_INSERT_TAIL(&vdev_driver_list, driver, next);
62 }
63
64 /* unregister a driver */
65 void
66 rte_vdev_unregister(struct rte_vdev_driver *driver)
67 {
68 TAILQ_REMOVE(&vdev_driver_list, driver, next);
69 }
70
71 int
72 rte_vdev_add_custom_scan(rte_vdev_scan_callback callback, void *user_arg)
73 {
74 struct vdev_custom_scan *custom_scan;
75
76 rte_spinlock_lock(&vdev_custom_scan_lock);
77
78 /* check if already registered */
79 TAILQ_FOREACH(custom_scan, &vdev_custom_scans, next) {
80 if (custom_scan->callback == callback &&
81 custom_scan->user_arg == user_arg)
82 break;
83 }
84
85 if (custom_scan == NULL) {
86 custom_scan = malloc(sizeof(struct vdev_custom_scan));
87 if (custom_scan != NULL) {
88 custom_scan->callback = callback;
89 custom_scan->user_arg = user_arg;
90 TAILQ_INSERT_TAIL(&vdev_custom_scans, custom_scan, next);
91 }
92 }
93
94 rte_spinlock_unlock(&vdev_custom_scan_lock);
95
96 return (custom_scan == NULL) ? -1 : 0;
97 }
98
99 int
100 rte_vdev_remove_custom_scan(rte_vdev_scan_callback callback, void *user_arg)
101 {
102 struct vdev_custom_scan *custom_scan, *tmp_scan;
103
104 rte_spinlock_lock(&vdev_custom_scan_lock);
105 TAILQ_FOREACH_SAFE(custom_scan, &vdev_custom_scans, next, tmp_scan) {
106 if (custom_scan->callback != callback ||
107 (custom_scan->user_arg != (void *)-1 &&
108 custom_scan->user_arg != user_arg))
109 continue;
110 TAILQ_REMOVE(&vdev_custom_scans, custom_scan, next);
111 free(custom_scan);
112 }
113 rte_spinlock_unlock(&vdev_custom_scan_lock);
114
115 return 0;
116 }
117
118 static int
119 vdev_parse(const char *name, void *addr)
120 {
121 struct rte_vdev_driver **out = addr;
122 struct rte_vdev_driver *driver = NULL;
123
124 TAILQ_FOREACH(driver, &vdev_driver_list, next) {
125 if (strncmp(driver->driver.name, name,
126 strlen(driver->driver.name)) == 0)
127 break;
128 if (driver->driver.alias &&
129 strncmp(driver->driver.alias, name,
130 strlen(driver->driver.alias)) == 0)
131 break;
132 }
133 if (driver != NULL &&
134 addr != NULL)
135 *out = driver;
136 return driver == NULL;
137 }
138
139 static int
140 vdev_probe_all_drivers(struct rte_vdev_device *dev)
141 {
142 const char *name;
143 struct rte_vdev_driver *driver;
144 int ret;
145
146 if (rte_dev_is_probed(&dev->device))
147 return -EEXIST;
148
149 name = rte_vdev_device_name(dev);
150 VDEV_LOG(DEBUG, "Search driver to probe device %s", name);
151
152 if (vdev_parse(name, &driver))
153 return -1;
154 ret = driver->probe(dev);
155 if (ret == 0)
156 dev->device.driver = &driver->driver;
157 return ret;
158 }
159
160 /* The caller shall be responsible for thread-safe */
161 static struct rte_vdev_device *
162 find_vdev(const char *name)
163 {
164 struct rte_vdev_device *dev;
165
166 if (!name)
167 return NULL;
168
169 TAILQ_FOREACH(dev, &vdev_device_list, next) {
170 const char *devname = rte_vdev_device_name(dev);
171
172 if (!strcmp(devname, name))
173 return dev;
174 }
175
176 return NULL;
177 }
178
179 static struct rte_devargs *
180 alloc_devargs(const char *name, const char *args)
181 {
182 struct rte_devargs *devargs;
183 int ret;
184
185 devargs = calloc(1, sizeof(*devargs));
186 if (!devargs)
187 return NULL;
188
189 devargs->bus = &rte_vdev_bus;
190 if (args)
191 devargs->args = strdup(args);
192 else
193 devargs->args = strdup("");
194
195 ret = strlcpy(devargs->name, name, sizeof(devargs->name));
196 if (ret < 0 || ret >= (int)sizeof(devargs->name)) {
197 free(devargs->args);
198 free(devargs);
199 return NULL;
200 }
201
202 return devargs;
203 }
204
205 static int
206 insert_vdev(const char *name, const char *args,
207 struct rte_vdev_device **p_dev,
208 bool init)
209 {
210 struct rte_vdev_device *dev;
211 struct rte_devargs *devargs;
212 int ret;
213
214 if (name == NULL)
215 return -EINVAL;
216
217 devargs = alloc_devargs(name, args);
218 if (!devargs)
219 return -ENOMEM;
220
221 dev = calloc(1, sizeof(*dev));
222 if (!dev) {
223 ret = -ENOMEM;
224 goto fail;
225 }
226
227 dev->device.bus = &rte_vdev_bus;
228 dev->device.numa_node = SOCKET_ID_ANY;
229 dev->device.name = devargs->name;
230
231 if (find_vdev(name)) {
232 /*
233 * A vdev is expected to have only one port.
234 * So there is no reason to try probing again,
235 * even with new arguments.
236 */
237 ret = -EEXIST;
238 goto fail;
239 }
240
241 if (init)
242 rte_devargs_insert(&devargs);
243 dev->device.devargs = devargs;
244 TAILQ_INSERT_TAIL(&vdev_device_list, dev, next);
245
246 if (p_dev)
247 *p_dev = dev;
248
249 return 0;
250 fail:
251 free(devargs->args);
252 free(devargs);
253 free(dev);
254 return ret;
255 }
256
257 int
258 rte_vdev_init(const char *name, const char *args)
259 {
260 struct rte_vdev_device *dev;
261 int ret;
262
263 rte_spinlock_recursive_lock(&vdev_device_list_lock);
264 ret = insert_vdev(name, args, &dev, true);
265 if (ret == 0) {
266 ret = vdev_probe_all_drivers(dev);
267 if (ret) {
268 if (ret > 0)
269 VDEV_LOG(ERR, "no driver found for %s", name);
270 /* If fails, remove it from vdev list */
271 TAILQ_REMOVE(&vdev_device_list, dev, next);
272 rte_devargs_remove(dev->device.devargs);
273 free(dev);
274 }
275 }
276 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
277 return ret;
278 }
279
280 static int
281 vdev_remove_driver(struct rte_vdev_device *dev)
282 {
283 const char *name = rte_vdev_device_name(dev);
284 const struct rte_vdev_driver *driver;
285
286 if (!dev->device.driver) {
287 VDEV_LOG(DEBUG, "no driver attach to device %s", name);
288 return 1;
289 }
290
291 driver = container_of(dev->device.driver, const struct rte_vdev_driver,
292 driver);
293 return driver->remove(dev);
294 }
295
296 int
297 rte_vdev_uninit(const char *name)
298 {
299 struct rte_vdev_device *dev;
300 int ret;
301
302 if (name == NULL)
303 return -EINVAL;
304
305 rte_spinlock_recursive_lock(&vdev_device_list_lock);
306
307 dev = find_vdev(name);
308 if (!dev) {
309 ret = -ENOENT;
310 goto unlock;
311 }
312
313 ret = vdev_remove_driver(dev);
314 if (ret)
315 goto unlock;
316
317 TAILQ_REMOVE(&vdev_device_list, dev, next);
318 rte_devargs_remove(dev->device.devargs);
319 free(dev);
320
321 unlock:
322 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
323 return ret;
324 }
325
326 struct vdev_param {
327 #define VDEV_SCAN_REQ 1
328 #define VDEV_SCAN_ONE 2
329 #define VDEV_SCAN_REP 3
330 int type;
331 int num;
332 char name[RTE_DEV_NAME_MAX_LEN];
333 };
334
335 static int vdev_plug(struct rte_device *dev);
336
337 /**
338 * This function works as the action for both primary and secondary process
339 * for static vdev discovery when a secondary process is booting.
340 *
341 * step 1, secondary process sends a sync request to ask for vdev in primary;
342 * step 2, primary process receives the request, and send vdevs one by one;
343 * step 3, primary process sends back reply, which indicates how many vdevs
344 * are sent.
345 */
346 static int
347 vdev_action(const struct rte_mp_msg *mp_msg, const void *peer)
348 {
349 struct rte_vdev_device *dev;
350 struct rte_mp_msg mp_resp;
351 struct vdev_param *ou = (struct vdev_param *)&mp_resp.param;
352 const struct vdev_param *in = (const struct vdev_param *)mp_msg->param;
353 const char *devname;
354 int num;
355 int ret;
356
357 strlcpy(mp_resp.name, VDEV_MP_KEY, sizeof(mp_resp.name));
358 mp_resp.len_param = sizeof(*ou);
359 mp_resp.num_fds = 0;
360
361 switch (in->type) {
362 case VDEV_SCAN_REQ:
363 ou->type = VDEV_SCAN_ONE;
364 ou->num = 1;
365 num = 0;
366
367 rte_spinlock_recursive_lock(&vdev_device_list_lock);
368 TAILQ_FOREACH(dev, &vdev_device_list, next) {
369 devname = rte_vdev_device_name(dev);
370 if (strlen(devname) == 0) {
371 VDEV_LOG(INFO, "vdev with no name is not sent");
372 continue;
373 }
374 VDEV_LOG(INFO, "send vdev, %s", devname);
375 strlcpy(ou->name, devname, RTE_DEV_NAME_MAX_LEN);
376 if (rte_mp_sendmsg(&mp_resp) < 0)
377 VDEV_LOG(ERR, "send vdev, %s, failed, %s",
378 devname, strerror(rte_errno));
379 num++;
380 }
381 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
382
383 ou->type = VDEV_SCAN_REP;
384 ou->num = num;
385 if (rte_mp_reply(&mp_resp, peer) < 0)
386 VDEV_LOG(ERR, "Failed to reply a scan request");
387 break;
388 case VDEV_SCAN_ONE:
389 VDEV_LOG(INFO, "receive vdev, %s", in->name);
390 ret = insert_vdev(in->name, NULL, NULL, false);
391 if (ret == -EEXIST)
392 VDEV_LOG(DEBUG, "device already exist, %s", in->name);
393 else if (ret < 0)
394 VDEV_LOG(ERR, "failed to add vdev, %s", in->name);
395 break;
396 default:
397 VDEV_LOG(ERR, "vdev cannot recognize this message");
398 }
399
400 return 0;
401 }
402
403 static int
404 vdev_scan(void)
405 {
406 struct rte_vdev_device *dev;
407 struct rte_devargs *devargs;
408 struct vdev_custom_scan *custom_scan;
409
410 if (rte_mp_action_register(VDEV_MP_KEY, vdev_action) < 0 &&
411 rte_errno != EEXIST) {
412 VDEV_LOG(ERR, "Failed to add vdev mp action");
413 return -1;
414 }
415
416 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
417 struct rte_mp_msg mp_req, *mp_rep;
418 struct rte_mp_reply mp_reply;
419 struct timespec ts = {.tv_sec = 5, .tv_nsec = 0};
420 struct vdev_param *req = (struct vdev_param *)mp_req.param;
421 struct vdev_param *resp;
422
423 strlcpy(mp_req.name, VDEV_MP_KEY, sizeof(mp_req.name));
424 mp_req.len_param = sizeof(*req);
425 mp_req.num_fds = 0;
426 req->type = VDEV_SCAN_REQ;
427 if (rte_mp_request_sync(&mp_req, &mp_reply, &ts) == 0 &&
428 mp_reply.nb_received == 1) {
429 mp_rep = &mp_reply.msgs[0];
430 resp = (struct vdev_param *)mp_rep->param;
431 VDEV_LOG(INFO, "Received %d vdevs", resp->num);
432 free(mp_reply.msgs);
433 } else
434 VDEV_LOG(ERR, "Failed to request vdev from primary");
435
436 /* Fall through to allow private vdevs in secondary process */
437 }
438
439 /* call custom scan callbacks if any */
440 rte_spinlock_lock(&vdev_custom_scan_lock);
441 TAILQ_FOREACH(custom_scan, &vdev_custom_scans, next) {
442 if (custom_scan->callback != NULL)
443 /*
444 * the callback should update devargs list
445 * by calling rte_devargs_insert() with
446 * devargs.bus = rte_bus_find_by_name("vdev");
447 * devargs.type = RTE_DEVTYPE_VIRTUAL;
448 * devargs.policy = RTE_DEV_WHITELISTED;
449 */
450 custom_scan->callback(custom_scan->user_arg);
451 }
452 rte_spinlock_unlock(&vdev_custom_scan_lock);
453
454 /* for virtual devices we scan the devargs_list populated via cmdline */
455 RTE_EAL_DEVARGS_FOREACH("vdev", devargs) {
456
457 dev = calloc(1, sizeof(*dev));
458 if (!dev)
459 return -1;
460
461 rte_spinlock_recursive_lock(&vdev_device_list_lock);
462
463 if (find_vdev(devargs->name)) {
464 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
465 free(dev);
466 continue;
467 }
468
469 dev->device.bus = &rte_vdev_bus;
470 dev->device.devargs = devargs;
471 dev->device.numa_node = SOCKET_ID_ANY;
472 dev->device.name = devargs->name;
473
474 TAILQ_INSERT_TAIL(&vdev_device_list, dev, next);
475
476 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
477 }
478
479 return 0;
480 }
481
482 static int
483 vdev_probe(void)
484 {
485 struct rte_vdev_device *dev;
486 int r, ret = 0;
487
488 /* call the init function for each virtual device */
489 TAILQ_FOREACH(dev, &vdev_device_list, next) {
490 /* we don't use the vdev lock here, as it's only used in DPDK
491 * initialization; and we don't want to hold such a lock when
492 * we call each driver probe.
493 */
494
495 r = vdev_probe_all_drivers(dev);
496 if (r != 0) {
497 if (r == -EEXIST)
498 continue;
499 VDEV_LOG(ERR, "failed to initialize %s device",
500 rte_vdev_device_name(dev));
501 ret = -1;
502 }
503 }
504
505 return ret;
506 }
507
508 struct rte_device *
509 rte_vdev_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
510 const void *data)
511 {
512 const struct rte_vdev_device *vstart;
513 struct rte_vdev_device *dev;
514
515 rte_spinlock_recursive_lock(&vdev_device_list_lock);
516 if (start != NULL) {
517 vstart = RTE_DEV_TO_VDEV_CONST(start);
518 dev = TAILQ_NEXT(vstart, next);
519 } else {
520 dev = TAILQ_FIRST(&vdev_device_list);
521 }
522 while (dev != NULL) {
523 if (cmp(&dev->device, data) == 0)
524 break;
525 dev = TAILQ_NEXT(dev, next);
526 }
527 rte_spinlock_recursive_unlock(&vdev_device_list_lock);
528
529 return dev ? &dev->device : NULL;
530 }
531
532 static int
533 vdev_plug(struct rte_device *dev)
534 {
535 return vdev_probe_all_drivers(RTE_DEV_TO_VDEV(dev));
536 }
537
538 static int
539 vdev_unplug(struct rte_device *dev)
540 {
541 return rte_vdev_uninit(dev->name);
542 }
543
544 static struct rte_bus rte_vdev_bus = {
545 .scan = vdev_scan,
546 .probe = vdev_probe,
547 .find_device = rte_vdev_find_device,
548 .plug = vdev_plug,
549 .unplug = vdev_unplug,
550 .parse = vdev_parse,
551 .dev_iterate = rte_vdev_dev_iterate,
552 };
553
554 RTE_REGISTER_BUS(vdev, rte_vdev_bus);
555
556 RTE_INIT(vdev_init_log)
557 {
558 vdev_logtype_bus = rte_log_register("bus.vdev");
559 if (vdev_logtype_bus >= 0)
560 rte_log_set_level(vdev_logtype_bus, RTE_LOG_NOTICE);
561 }