1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation.
3 * Copyright 2013-2014 6WIND S.A.
12 #include <sys/queue.h>
15 #include <rte_errno.h>
16 #include <rte_interrupts.h>
20 #include <rte_bus_pci.h>
21 #include <rte_per_lcore.h>
22 #include <rte_memory.h>
24 #include <rte_string_fns.h>
25 #include <rte_common.h>
26 #include <rte_devargs.h>
32 #define SYSFS_PCI_DEVICES "/sys/bus/pci/devices"
34 const char *rte_pci_get_sysfs_path(void)
36 const char *path = NULL;
38 path = getenv("SYSFS_PCI_DEVICES");
40 return SYSFS_PCI_DEVICES;
45 static struct rte_devargs *pci_devargs_lookup(struct rte_pci_device *dev)
47 struct rte_devargs *devargs;
48 struct rte_pci_addr addr;
50 RTE_EAL_DEVARGS_FOREACH("pci", devargs) {
51 devargs->bus->parse(devargs->name, &addr);
52 if (!rte_pci_addr_cmp(&dev->addr, &addr))
59 pci_name_set(struct rte_pci_device *dev)
61 struct rte_devargs *devargs;
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.
72 /* If an rte_devargs exists, the generic rte_device uses the
73 * given name as its name.
75 dev->device.name = dev->device.devargs->name;
77 /* Otherwise, it uses the internal, canonical form. */
78 dev->device.name = dev->name;
82 * Match the PCI Driver and Device using the ID Table
85 rte_pci_match(const struct rte_pci_driver *pci_drv,
86 const struct rte_pci_device *pci_dev)
88 const struct rte_pci_id *id_table;
90 for (id_table = pci_drv->id_table; id_table->vendor_id != 0;
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)
96 if (id_table->device_id != pci_dev->id.device_id &&
97 id_table->device_id != PCI_ANY_ID)
99 if (id_table->subsystem_vendor_id !=
100 pci_dev->id.subsystem_vendor_id &&
101 id_table->subsystem_vendor_id != PCI_ANY_ID)
103 if (id_table->subsystem_device_id !=
104 pci_dev->id.subsystem_device_id &&
105 id_table->subsystem_device_id != PCI_ANY_ID)
107 if (id_table->class_id != pci_dev->id.class_id &&
108 id_table->class_id != RTE_CLASS_ANY_ID)
118 * If vendor/device ID match, call the probe() function of the
122 rte_pci_probe_one_driver(struct rte_pci_driver *dr,
123 struct rte_pci_device *dev)
127 struct rte_pci_addr *loc;
129 if ((dr == NULL) || (dev == NULL))
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 */
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);
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"
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;
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",
164 RTE_LOG(INFO, EAL, " probe driver: %x:%x %s\n", dev->id.vendor_id,
165 dev->id.device_id, dr->driver.name);
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.
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);
184 /* call the driver probe() function */
185 ret = dr->probe(dr, dev);
187 return ret; /* no rollback if already succeeded earlier */
190 if ((dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) &&
191 /* Don't unmap if device is unsupported and
192 * driver needs mapped resources.
195 (dr->drv_flags & RTE_PCI_DRV_KEEP_MAPPED_RES)))
196 rte_pci_unmap_device(dev);
198 dev->device.driver = &dr->driver;
205 * If vendor/device ID match, call the remove() function of the
209 rte_pci_detach_dev(struct rte_pci_device *dev)
211 struct rte_pci_addr *loc;
212 struct rte_pci_driver *dr;
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);
225 RTE_LOG(DEBUG, EAL, " remove driver: %x:%x %s\n", dev->id.vendor_id,
226 dev->id.device_id, dr->driver.name);
229 ret = dr->remove(dev);
234 /* clear driver structure */
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);
245 * If vendor/device ID match, call the probe() function of all
246 * registered driver for the given device. Return -1 if initialization
247 * failed, return 1 if no driver is found for this device.
250 pci_probe_all_drivers(struct rte_pci_device *dev)
252 struct rte_pci_driver *dr = NULL;
258 FOREACH_DRIVER_ON_PCIBUS(dr) {
259 rc = rte_pci_probe_one_driver(dr, dev);
261 /* negative value is an error */
264 /* positive value means driver doesn't support it */
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.
279 struct rte_pci_device *dev = NULL;
280 size_t probed = 0, failed = 0;
281 struct rte_devargs *devargs;
285 if (rte_pci_bus.bus.conf.scan_mode != RTE_BUS_SCAN_WHITELIST)
288 FOREACH_DEVICE_ON_PCIBUS(dev) {
291 devargs = dev->device.devargs;
292 /* probe all or only whitelisted devices */
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);
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);
311 return (probed && probed == failed) ? -1 : 0;
314 /* dump one device */
316 pci_dump_one_device(FILE *f, struct rte_pci_device *dev)
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,
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);
334 /* dump devices on the bus */
336 rte_pci_dump(FILE *f)
338 struct rte_pci_device *dev = NULL;
340 FOREACH_DEVICE_ON_PCIBUS(dev) {
341 pci_dump_one_device(f, dev);
346 pci_parse(const char *name, void *addr)
348 struct rte_pci_addr *out = addr;
349 struct rte_pci_addr pci_addr;
352 parse = (rte_pci_addr_parse(name, &pci_addr) == 0);
353 if (parse && addr != NULL)
355 return parse == false;
358 /* register a driver */
360 rte_pci_register(struct rte_pci_driver *driver)
362 TAILQ_INSERT_TAIL(&rte_pci_bus.driver_list, driver, next);
363 driver->bus = &rte_pci_bus;
366 /* unregister a driver */
368 rte_pci_unregister(struct rte_pci_driver *driver)
370 TAILQ_REMOVE(&rte_pci_bus.driver_list, driver, next);
374 /* Add a device to PCI bus */
376 rte_pci_add_device(struct rte_pci_device *pci_dev)
378 TAILQ_INSERT_TAIL(&rte_pci_bus.device_list, pci_dev, next);
381 /* Insert a device into a predefined position in PCI bus */
383 rte_pci_insert_device(struct rte_pci_device *exist_pci_dev,
384 struct rte_pci_device *new_pci_dev)
386 TAILQ_INSERT_BEFORE(exist_pci_dev, new_pci_dev, next);
389 /* Remove a device from PCI bus */
391 rte_pci_remove_device(struct rte_pci_device *pci_dev)
393 TAILQ_REMOVE(&rte_pci_bus.device_list, pci_dev, next);
396 static struct rte_device *
397 pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
400 const struct rte_pci_device *pstart;
401 struct rte_pci_device *pdev;
404 pstart = RTE_DEV_TO_PCI_CONST(start);
405 pdev = TAILQ_NEXT(pstart, next);
407 pdev = TAILQ_FIRST(&rte_pci_bus.device_list);
409 while (pdev != NULL) {
410 if (cmp(&pdev->device, data) == 0)
411 return &pdev->device;
412 pdev = TAILQ_NEXT(pdev, next);
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.
422 static struct rte_pci_device *
423 pci_find_device_by_addr(const void *failure_addr)
425 struct rte_pci_device *pdev = NULL;
426 uint64_t check_point, start, end, len;
429 check_point = (uint64_t)(uintptr_t)failure_addr;
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;
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);
448 pci_hot_unplug_handler(struct rte_device *dev)
450 struct rte_pci_device *pdev = NULL;
453 pdev = RTE_DEV_TO_PCI(dev);
457 switch (pdev->kdrv) {
458 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
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.
466 rte_dev_event_callback_process(dev->name,
467 RTE_DEV_EVENT_REMOVE);
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);
478 "Not managed by a supported kernel driver, skipped\n");
487 pci_sigbus_handler(const void *failure_addr)
489 struct rte_pci_device *pdev = NULL;
492 pdev = pci_find_device_by_addr(failure_addr);
494 /* It is a generic sigbus error, no bus would handle it. */
497 /* The sigbus error is caused of hot-unplug. */
498 ret = pci_hot_unplug_handler(&pdev->device);
501 "Failed to handle hot-unplug for device %s",
510 pci_plug(struct rte_device *dev)
512 return pci_probe_all_drivers(RTE_DEV_TO_PCI(dev));
516 pci_unplug(struct rte_device *dev)
518 struct rte_pci_device *pdev;
521 pdev = RTE_DEV_TO_PCI(dev);
522 ret = rte_pci_detach_dev(pdev);
524 rte_pci_remove_device(pdev);
525 rte_devargs_remove(dev->devargs);
531 struct rte_pci_bus rte_pci_bus = {
533 .scan = rte_pci_scan,
534 .probe = rte_pci_probe,
535 .find_device = pci_find_device,
537 .unplug = pci_unplug,
539 .get_iommu_class = rte_pci_get_iommu_class,
540 .dev_iterate = rte_pci_dev_iterate,
541 .hot_unplug_handler = pci_hot_unplug_handler,
542 .sigbus_handler = pci_sigbus_handler,
544 .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list),
545 .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list),
548 RTE_REGISTER_BUS(pci, rte_pci_bus.bus);