]> git.proxmox.com Git - ceph.git/blob - ceph/src/seastar/dpdk/drivers/bus/pci/pci_common.c
import 15.2.0 Octopus source
[ceph.git] / ceph / src / seastar / dpdk / drivers / bus / pci / pci_common.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation.
3 * Copyright 2013-2014 6WIND S.A.
4 */
5
6 #include <string.h>
7 #include <inttypes.h>
8 #include <stdint.h>
9 #include <stdbool.h>
10 #include <stdlib.h>
11 #include <stdio.h>
12 #include <sys/queue.h>
13 #include <sys/mman.h>
14
15 #include <rte_errno.h>
16 #include <rte_interrupts.h>
17 #include <rte_log.h>
18 #include <rte_bus.h>
19 #include <rte_pci.h>
20 #include <rte_bus_pci.h>
21 #include <rte_per_lcore.h>
22 #include <rte_memory.h>
23 #include <rte_eal.h>
24 #include <rte_string_fns.h>
25 #include <rte_common.h>
26 #include <rte_devargs.h>
27 #include <rte_vfio.h>
28
29 #include "private.h"
30
31
32 #define SYSFS_PCI_DEVICES "/sys/bus/pci/devices"
33
34 const char *rte_pci_get_sysfs_path(void)
35 {
36 const char *path = NULL;
37
38 path = getenv("SYSFS_PCI_DEVICES");
39 if (path == NULL)
40 return SYSFS_PCI_DEVICES;
41
42 return path;
43 }
44
45 static struct rte_devargs *pci_devargs_lookup(struct rte_pci_device *dev)
46 {
47 struct rte_devargs *devargs;
48 struct rte_pci_addr addr;
49
50 RTE_EAL_DEVARGS_FOREACH("pci", devargs) {
51 devargs->bus->parse(devargs->name, &addr);
52 if (!rte_pci_addr_cmp(&dev->addr, &addr))
53 return devargs;
54 }
55 return NULL;
56 }
57
58 void
59 pci_name_set(struct rte_pci_device *dev)
60 {
61 struct rte_devargs *devargs;
62
63 /* Each device has its internal, canonical name set. */
64 rte_pci_device_name(&dev->addr,
65 dev->name, sizeof(dev->name));
66 devargs = pci_devargs_lookup(dev);
67 dev->device.devargs = devargs;
68 /* In blacklist mode, if the device is not blacklisted, no
69 * rte_devargs exists for it.
70 */
71 if (devargs != NULL)
72 /* If an rte_devargs exists, the generic rte_device uses the
73 * given name as its name.
74 */
75 dev->device.name = dev->device.devargs->name;
76 else
77 /* Otherwise, it uses the internal, canonical form. */
78 dev->device.name = dev->name;
79 }
80
81 /*
82 * Match the PCI Driver and Device using the ID Table
83 */
84 int
85 rte_pci_match(const struct rte_pci_driver *pci_drv,
86 const struct rte_pci_device *pci_dev)
87 {
88 const struct rte_pci_id *id_table;
89
90 for (id_table = pci_drv->id_table; id_table->vendor_id != 0;
91 id_table++) {
92 /* check if device's identifiers match the driver's ones */
93 if (id_table->vendor_id != pci_dev->id.vendor_id &&
94 id_table->vendor_id != PCI_ANY_ID)
95 continue;
96 if (id_table->device_id != pci_dev->id.device_id &&
97 id_table->device_id != PCI_ANY_ID)
98 continue;
99 if (id_table->subsystem_vendor_id !=
100 pci_dev->id.subsystem_vendor_id &&
101 id_table->subsystem_vendor_id != PCI_ANY_ID)
102 continue;
103 if (id_table->subsystem_device_id !=
104 pci_dev->id.subsystem_device_id &&
105 id_table->subsystem_device_id != PCI_ANY_ID)
106 continue;
107 if (id_table->class_id != pci_dev->id.class_id &&
108 id_table->class_id != RTE_CLASS_ANY_ID)
109 continue;
110
111 return 1;
112 }
113
114 return 0;
115 }
116
117 /*
118 * If vendor/device ID match, call the probe() function of the
119 * driver.
120 */
121 static int
122 rte_pci_probe_one_driver(struct rte_pci_driver *dr,
123 struct rte_pci_device *dev)
124 {
125 int ret;
126 bool already_probed;
127 struct rte_pci_addr *loc;
128
129 if ((dr == NULL) || (dev == NULL))
130 return -EINVAL;
131
132 loc = &dev->addr;
133
134 /* The device is not blacklisted; Check if driver supports it */
135 if (!rte_pci_match(dr, dev))
136 /* Match of device and driver failed */
137 return 1;
138
139 RTE_LOG(INFO, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n",
140 loc->domain, loc->bus, loc->devid, loc->function,
141 dev->device.numa_node);
142
143 /* no initialization when blacklisted, return without error */
144 if (dev->device.devargs != NULL &&
145 dev->device.devargs->policy ==
146 RTE_DEV_BLACKLISTED) {
147 RTE_LOG(INFO, EAL, " Device is blacklisted, not"
148 " initializing\n");
149 return 1;
150 }
151
152 if (dev->device.numa_node < 0) {
153 RTE_LOG(WARNING, EAL, " Invalid NUMA socket, default to 0\n");
154 dev->device.numa_node = 0;
155 }
156
157 already_probed = rte_dev_is_probed(&dev->device);
158 if (already_probed && !(dr->drv_flags & RTE_PCI_DRV_PROBE_AGAIN)) {
159 RTE_LOG(DEBUG, EAL, "Device %s is already probed\n",
160 dev->device.name);
161 return -EEXIST;
162 }
163
164 RTE_LOG(INFO, EAL, " probe driver: %x:%x %s\n", dev->id.vendor_id,
165 dev->id.device_id, dr->driver.name);
166
167 /*
168 * reference driver structure
169 * This needs to be before rte_pci_map_device(), as it enables to use
170 * driver flags for adjusting configuration.
171 */
172 if (!already_probed)
173 dev->driver = dr;
174
175 if (!already_probed && (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)) {
176 /* map resources for devices that use igb_uio */
177 ret = rte_pci_map_device(dev);
178 if (ret != 0) {
179 dev->driver = NULL;
180 return ret;
181 }
182 }
183
184 /* call the driver probe() function */
185 ret = dr->probe(dr, dev);
186 if (already_probed)
187 return ret; /* no rollback if already succeeded earlier */
188 if (ret) {
189 dev->driver = NULL;
190 if ((dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) &&
191 /* Don't unmap if device is unsupported and
192 * driver needs mapped resources.
193 */
194 !(ret > 0 &&
195 (dr->drv_flags & RTE_PCI_DRV_KEEP_MAPPED_RES)))
196 rte_pci_unmap_device(dev);
197 } else {
198 dev->device.driver = &dr->driver;
199 }
200
201 return ret;
202 }
203
204 /*
205 * If vendor/device ID match, call the remove() function of the
206 * driver.
207 */
208 static int
209 rte_pci_detach_dev(struct rte_pci_device *dev)
210 {
211 struct rte_pci_addr *loc;
212 struct rte_pci_driver *dr;
213 int ret = 0;
214
215 if (dev == NULL)
216 return -EINVAL;
217
218 dr = dev->driver;
219 loc = &dev->addr;
220
221 RTE_LOG(DEBUG, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n",
222 loc->domain, loc->bus, loc->devid,
223 loc->function, dev->device.numa_node);
224
225 RTE_LOG(DEBUG, EAL, " remove driver: %x:%x %s\n", dev->id.vendor_id,
226 dev->id.device_id, dr->driver.name);
227
228 if (dr->remove) {
229 ret = dr->remove(dev);
230 if (ret < 0)
231 return ret;
232 }
233
234 /* clear driver structure */
235 dev->driver = NULL;
236
237 if (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)
238 /* unmap resources for devices that use igb_uio */
239 rte_pci_unmap_device(dev);
240
241 return 0;
242 }
243
244 /*
245 * If vendor/device ID match, call the probe() function of all
246 * registered driver for the given device. Return < 0 if initialization
247 * failed, return 1 if no driver is found for this device.
248 */
249 static int
250 pci_probe_all_drivers(struct rte_pci_device *dev)
251 {
252 struct rte_pci_driver *dr = NULL;
253 int rc = 0;
254
255 if (dev == NULL)
256 return -EINVAL;
257
258 FOREACH_DRIVER_ON_PCIBUS(dr) {
259 rc = rte_pci_probe_one_driver(dr, dev);
260 if (rc < 0)
261 /* negative value is an error */
262 return rc;
263 if (rc > 0)
264 /* positive value means driver doesn't support it */
265 continue;
266 return 0;
267 }
268 return 1;
269 }
270
271 /*
272 * Scan the content of the PCI bus, and call the probe() function for
273 * all registered drivers that have a matching entry in its id_table
274 * for discovered devices.
275 */
276 int
277 rte_pci_probe(void)
278 {
279 struct rte_pci_device *dev = NULL;
280 size_t probed = 0, failed = 0;
281 struct rte_devargs *devargs;
282 int probe_all = 0;
283 int ret = 0;
284
285 if (rte_pci_bus.bus.conf.scan_mode != RTE_BUS_SCAN_WHITELIST)
286 probe_all = 1;
287
288 FOREACH_DEVICE_ON_PCIBUS(dev) {
289 probed++;
290
291 devargs = dev->device.devargs;
292 /* probe all or only whitelisted devices */
293 if (probe_all)
294 ret = pci_probe_all_drivers(dev);
295 else if (devargs != NULL &&
296 devargs->policy == RTE_DEV_WHITELISTED)
297 ret = pci_probe_all_drivers(dev);
298 if (ret < 0) {
299 if (ret != -EEXIST) {
300 RTE_LOG(ERR, EAL, "Requested device "
301 PCI_PRI_FMT " cannot be used\n",
302 dev->addr.domain, dev->addr.bus,
303 dev->addr.devid, dev->addr.function);
304 rte_errno = errno;
305 failed++;
306 }
307 ret = 0;
308 }
309 }
310
311 return (probed && probed == failed) ? -1 : 0;
312 }
313
314 /* dump one device */
315 static int
316 pci_dump_one_device(FILE *f, struct rte_pci_device *dev)
317 {
318 int i;
319
320 fprintf(f, PCI_PRI_FMT, dev->addr.domain, dev->addr.bus,
321 dev->addr.devid, dev->addr.function);
322 fprintf(f, " - vendor:%x device:%x\n", dev->id.vendor_id,
323 dev->id.device_id);
324
325 for (i = 0; i != sizeof(dev->mem_resource) /
326 sizeof(dev->mem_resource[0]); i++) {
327 fprintf(f, " %16.16"PRIx64" %16.16"PRIx64"\n",
328 dev->mem_resource[i].phys_addr,
329 dev->mem_resource[i].len);
330 }
331 return 0;
332 }
333
334 /* dump devices on the bus */
335 void
336 rte_pci_dump(FILE *f)
337 {
338 struct rte_pci_device *dev = NULL;
339
340 FOREACH_DEVICE_ON_PCIBUS(dev) {
341 pci_dump_one_device(f, dev);
342 }
343 }
344
345 static int
346 pci_parse(const char *name, void *addr)
347 {
348 struct rte_pci_addr *out = addr;
349 struct rte_pci_addr pci_addr;
350 bool parse;
351
352 parse = (rte_pci_addr_parse(name, &pci_addr) == 0);
353 if (parse && addr != NULL)
354 *out = pci_addr;
355 return parse == false;
356 }
357
358 /* register a driver */
359 void
360 rte_pci_register(struct rte_pci_driver *driver)
361 {
362 TAILQ_INSERT_TAIL(&rte_pci_bus.driver_list, driver, next);
363 driver->bus = &rte_pci_bus;
364 }
365
366 /* unregister a driver */
367 void
368 rte_pci_unregister(struct rte_pci_driver *driver)
369 {
370 TAILQ_REMOVE(&rte_pci_bus.driver_list, driver, next);
371 driver->bus = NULL;
372 }
373
374 /* Add a device to PCI bus */
375 void
376 rte_pci_add_device(struct rte_pci_device *pci_dev)
377 {
378 TAILQ_INSERT_TAIL(&rte_pci_bus.device_list, pci_dev, next);
379 }
380
381 /* Insert a device into a predefined position in PCI bus */
382 void
383 rte_pci_insert_device(struct rte_pci_device *exist_pci_dev,
384 struct rte_pci_device *new_pci_dev)
385 {
386 TAILQ_INSERT_BEFORE(exist_pci_dev, new_pci_dev, next);
387 }
388
389 /* Remove a device from PCI bus */
390 static void
391 rte_pci_remove_device(struct rte_pci_device *pci_dev)
392 {
393 TAILQ_REMOVE(&rte_pci_bus.device_list, pci_dev, next);
394 }
395
396 static struct rte_device *
397 pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
398 const void *data)
399 {
400 const struct rte_pci_device *pstart;
401 struct rte_pci_device *pdev;
402
403 if (start != NULL) {
404 pstart = RTE_DEV_TO_PCI_CONST(start);
405 pdev = TAILQ_NEXT(pstart, next);
406 } else {
407 pdev = TAILQ_FIRST(&rte_pci_bus.device_list);
408 }
409 while (pdev != NULL) {
410 if (cmp(&pdev->device, data) == 0)
411 return &pdev->device;
412 pdev = TAILQ_NEXT(pdev, next);
413 }
414 return NULL;
415 }
416
417 /*
418 * find the device which encounter the failure, by iterate over all device on
419 * PCI bus to check if the memory failure address is located in the range
420 * of the BARs of the device.
421 */
422 static struct rte_pci_device *
423 pci_find_device_by_addr(const void *failure_addr)
424 {
425 struct rte_pci_device *pdev = NULL;
426 uint64_t check_point, start, end, len;
427 int i;
428
429 check_point = (uint64_t)(uintptr_t)failure_addr;
430
431 FOREACH_DEVICE_ON_PCIBUS(pdev) {
432 for (i = 0; i != RTE_DIM(pdev->mem_resource); i++) {
433 start = (uint64_t)(uintptr_t)pdev->mem_resource[i].addr;
434 len = pdev->mem_resource[i].len;
435 end = start + len;
436 if (check_point >= start && check_point < end) {
437 RTE_LOG(DEBUG, EAL, "Failure address %16.16"
438 PRIx64" belongs to device %s!\n",
439 check_point, pdev->device.name);
440 return pdev;
441 }
442 }
443 }
444 return NULL;
445 }
446
447 static int
448 pci_hot_unplug_handler(struct rte_device *dev)
449 {
450 struct rte_pci_device *pdev = NULL;
451 int ret = 0;
452
453 pdev = RTE_DEV_TO_PCI(dev);
454 if (!pdev)
455 return -1;
456
457 switch (pdev->kdrv) {
458 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
459 case RTE_KDRV_VFIO:
460 /*
461 * vfio kernel module guaranty the pci device would not be
462 * deleted until the user space release the resource, so no
463 * need to remap BARs resource here, just directly notify
464 * the req event to the user space to handle it.
465 */
466 rte_dev_event_callback_process(dev->name,
467 RTE_DEV_EVENT_REMOVE);
468 break;
469 #endif
470 case RTE_KDRV_IGB_UIO:
471 case RTE_KDRV_UIO_GENERIC:
472 case RTE_KDRV_NIC_UIO:
473 /* BARs resource is invalid, remap it to be safe. */
474 ret = pci_uio_remap_resource(pdev);
475 break;
476 default:
477 RTE_LOG(DEBUG, EAL,
478 "Not managed by a supported kernel driver, skipped\n");
479 ret = -1;
480 break;
481 }
482
483 return ret;
484 }
485
486 static int
487 pci_sigbus_handler(const void *failure_addr)
488 {
489 struct rte_pci_device *pdev = NULL;
490 int ret = 0;
491
492 pdev = pci_find_device_by_addr(failure_addr);
493 if (!pdev) {
494 /* It is a generic sigbus error, no bus would handle it. */
495 ret = 1;
496 } else {
497 /* The sigbus error is caused of hot-unplug. */
498 ret = pci_hot_unplug_handler(&pdev->device);
499 if (ret) {
500 RTE_LOG(ERR, EAL,
501 "Failed to handle hot-unplug for device %s",
502 pdev->name);
503 ret = -1;
504 }
505 }
506 return ret;
507 }
508
509 static int
510 pci_plug(struct rte_device *dev)
511 {
512 return pci_probe_all_drivers(RTE_DEV_TO_PCI(dev));
513 }
514
515 static int
516 pci_unplug(struct rte_device *dev)
517 {
518 struct rte_pci_device *pdev;
519 int ret;
520
521 pdev = RTE_DEV_TO_PCI(dev);
522 ret = rte_pci_detach_dev(pdev);
523 if (ret == 0) {
524 rte_pci_remove_device(pdev);
525 rte_devargs_remove(dev->devargs);
526 free(pdev);
527 }
528 return ret;
529 }
530
531 static int
532 pci_dma_map(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
533 {
534 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
535
536 if (!pdev || !pdev->driver) {
537 rte_errno = EINVAL;
538 return -1;
539 }
540 if (pdev->driver->dma_map)
541 return pdev->driver->dma_map(pdev, addr, iova, len);
542 /**
543 * In case driver don't provides any specific mapping
544 * try fallback to VFIO.
545 */
546 if (pdev->kdrv == RTE_KDRV_VFIO)
547 return rte_vfio_container_dma_map
548 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
549 iova, len);
550 rte_errno = ENOTSUP;
551 return -1;
552 }
553
554 static int
555 pci_dma_unmap(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
556 {
557 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
558
559 if (!pdev || !pdev->driver) {
560 rte_errno = EINVAL;
561 return -1;
562 }
563 if (pdev->driver->dma_unmap)
564 return pdev->driver->dma_unmap(pdev, addr, iova, len);
565 /**
566 * In case driver don't provides any specific mapping
567 * try fallback to VFIO.
568 */
569 if (pdev->kdrv == RTE_KDRV_VFIO)
570 return rte_vfio_container_dma_unmap
571 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
572 iova, len);
573 rte_errno = ENOTSUP;
574 return -1;
575 }
576
577 struct rte_pci_bus rte_pci_bus = {
578 .bus = {
579 .scan = rte_pci_scan,
580 .probe = rte_pci_probe,
581 .find_device = pci_find_device,
582 .plug = pci_plug,
583 .unplug = pci_unplug,
584 .parse = pci_parse,
585 .dma_map = pci_dma_map,
586 .dma_unmap = pci_dma_unmap,
587 .get_iommu_class = rte_pci_get_iommu_class,
588 .dev_iterate = rte_pci_dev_iterate,
589 .hot_unplug_handler = pci_hot_unplug_handler,
590 .sigbus_handler = pci_sigbus_handler,
591 },
592 .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list),
593 .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list),
594 };
595
596 RTE_REGISTER_BUS(pci, rte_pci_bus.bus);