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.
172 if (!already_probed) {
173 enum rte_iova_mode dev_iova_mode;
174 enum rte_iova_mode iova_mode;
176 dev_iova_mode = pci_device_iova_mode(dr, dev);
177 iova_mode = rte_eal_iova_mode();
178 if (dev_iova_mode != RTE_IOVA_DC &&
179 dev_iova_mode != iova_mode) {
180 RTE_LOG(ERR, EAL, " Expecting '%s' IOVA mode but current mode is '%s', not initializing\n",
181 dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA",
182 iova_mode == RTE_IOVA_PA ? "PA" : "VA");
189 if (!already_probed && (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)) {
190 /* map resources for devices that use igb_uio */
191 ret = rte_pci_map_device(dev);
198 /* call the driver probe() function */
199 ret = dr->probe(dr, dev);
201 return ret; /* no rollback if already succeeded earlier */
204 if ((dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) &&
205 /* Don't unmap if device is unsupported and
206 * driver needs mapped resources.
209 (dr->drv_flags & RTE_PCI_DRV_KEEP_MAPPED_RES)))
210 rte_pci_unmap_device(dev);
212 dev->device.driver = &dr->driver;
219 * If vendor/device ID match, call the remove() function of the
223 rte_pci_detach_dev(struct rte_pci_device *dev)
225 struct rte_pci_addr *loc;
226 struct rte_pci_driver *dr;
235 RTE_LOG(DEBUG, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n",
236 loc->domain, loc->bus, loc->devid,
237 loc->function, dev->device.numa_node);
239 RTE_LOG(DEBUG, EAL, " remove driver: %x:%x %s\n", dev->id.vendor_id,
240 dev->id.device_id, dr->driver.name);
243 ret = dr->remove(dev);
248 /* clear driver structure */
251 if (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)
252 /* unmap resources for devices that use igb_uio */
253 rte_pci_unmap_device(dev);
259 * If vendor/device ID match, call the probe() function of all
260 * registered driver for the given device. Return < 0 if initialization
261 * failed, return 1 if no driver is found for this device.
264 pci_probe_all_drivers(struct rte_pci_device *dev)
266 struct rte_pci_driver *dr = NULL;
272 FOREACH_DRIVER_ON_PCIBUS(dr) {
273 rc = rte_pci_probe_one_driver(dr, dev);
275 /* negative value is an error */
278 /* positive value means driver doesn't support it */
286 * Scan the content of the PCI bus, and call the probe() function for
287 * all registered drivers that have a matching entry in its id_table
288 * for discovered devices.
293 struct rte_pci_device *dev = NULL;
294 size_t probed = 0, failed = 0;
295 struct rte_devargs *devargs;
299 if (rte_pci_bus.bus.conf.scan_mode != RTE_BUS_SCAN_WHITELIST)
302 FOREACH_DEVICE_ON_PCIBUS(dev) {
305 devargs = dev->device.devargs;
306 /* probe all or only whitelisted devices */
308 ret = pci_probe_all_drivers(dev);
309 else if (devargs != NULL &&
310 devargs->policy == RTE_DEV_WHITELISTED)
311 ret = pci_probe_all_drivers(dev);
313 if (ret != -EEXIST) {
314 RTE_LOG(ERR, EAL, "Requested device "
315 PCI_PRI_FMT " cannot be used\n",
316 dev->addr.domain, dev->addr.bus,
317 dev->addr.devid, dev->addr.function);
325 return (probed && probed == failed) ? -1 : 0;
328 /* dump one device */
330 pci_dump_one_device(FILE *f, struct rte_pci_device *dev)
334 fprintf(f, PCI_PRI_FMT, dev->addr.domain, dev->addr.bus,
335 dev->addr.devid, dev->addr.function);
336 fprintf(f, " - vendor:%x device:%x\n", dev->id.vendor_id,
339 for (i = 0; i != sizeof(dev->mem_resource) /
340 sizeof(dev->mem_resource[0]); i++) {
341 fprintf(f, " %16.16"PRIx64" %16.16"PRIx64"\n",
342 dev->mem_resource[i].phys_addr,
343 dev->mem_resource[i].len);
348 /* dump devices on the bus */
350 rte_pci_dump(FILE *f)
352 struct rte_pci_device *dev = NULL;
354 FOREACH_DEVICE_ON_PCIBUS(dev) {
355 pci_dump_one_device(f, dev);
360 pci_parse(const char *name, void *addr)
362 struct rte_pci_addr *out = addr;
363 struct rte_pci_addr pci_addr;
366 parse = (rte_pci_addr_parse(name, &pci_addr) == 0);
367 if (parse && addr != NULL)
369 return parse == false;
372 /* register a driver */
374 rte_pci_register(struct rte_pci_driver *driver)
376 TAILQ_INSERT_TAIL(&rte_pci_bus.driver_list, driver, next);
377 driver->bus = &rte_pci_bus;
380 /* unregister a driver */
382 rte_pci_unregister(struct rte_pci_driver *driver)
384 TAILQ_REMOVE(&rte_pci_bus.driver_list, driver, next);
388 /* Add a device to PCI bus */
390 rte_pci_add_device(struct rte_pci_device *pci_dev)
392 TAILQ_INSERT_TAIL(&rte_pci_bus.device_list, pci_dev, next);
395 /* Insert a device into a predefined position in PCI bus */
397 rte_pci_insert_device(struct rte_pci_device *exist_pci_dev,
398 struct rte_pci_device *new_pci_dev)
400 TAILQ_INSERT_BEFORE(exist_pci_dev, new_pci_dev, next);
403 /* Remove a device from PCI bus */
405 rte_pci_remove_device(struct rte_pci_device *pci_dev)
407 TAILQ_REMOVE(&rte_pci_bus.device_list, pci_dev, next);
410 static struct rte_device *
411 pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
414 const struct rte_pci_device *pstart;
415 struct rte_pci_device *pdev;
418 pstart = RTE_DEV_TO_PCI_CONST(start);
419 pdev = TAILQ_NEXT(pstart, next);
421 pdev = TAILQ_FIRST(&rte_pci_bus.device_list);
423 while (pdev != NULL) {
424 if (cmp(&pdev->device, data) == 0)
425 return &pdev->device;
426 pdev = TAILQ_NEXT(pdev, next);
432 * find the device which encounter the failure, by iterate over all device on
433 * PCI bus to check if the memory failure address is located in the range
434 * of the BARs of the device.
436 static struct rte_pci_device *
437 pci_find_device_by_addr(const void *failure_addr)
439 struct rte_pci_device *pdev = NULL;
440 uint64_t check_point, start, end, len;
443 check_point = (uint64_t)(uintptr_t)failure_addr;
445 FOREACH_DEVICE_ON_PCIBUS(pdev) {
446 for (i = 0; i != RTE_DIM(pdev->mem_resource); i++) {
447 start = (uint64_t)(uintptr_t)pdev->mem_resource[i].addr;
448 len = pdev->mem_resource[i].len;
450 if (check_point >= start && check_point < end) {
451 RTE_LOG(DEBUG, EAL, "Failure address %16.16"
452 PRIx64" belongs to device %s!\n",
453 check_point, pdev->device.name);
462 pci_hot_unplug_handler(struct rte_device *dev)
464 struct rte_pci_device *pdev = NULL;
467 pdev = RTE_DEV_TO_PCI(dev);
471 switch (pdev->kdrv) {
472 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
475 * vfio kernel module guaranty the pci device would not be
476 * deleted until the user space release the resource, so no
477 * need to remap BARs resource here, just directly notify
478 * the req event to the user space to handle it.
480 rte_dev_event_callback_process(dev->name,
481 RTE_DEV_EVENT_REMOVE);
484 case RTE_KDRV_IGB_UIO:
485 case RTE_KDRV_UIO_GENERIC:
486 case RTE_KDRV_NIC_UIO:
487 /* BARs resource is invalid, remap it to be safe. */
488 ret = pci_uio_remap_resource(pdev);
492 "Not managed by a supported kernel driver, skipped\n");
501 pci_sigbus_handler(const void *failure_addr)
503 struct rte_pci_device *pdev = NULL;
506 pdev = pci_find_device_by_addr(failure_addr);
508 /* It is a generic sigbus error, no bus would handle it. */
511 /* The sigbus error is caused of hot-unplug. */
512 ret = pci_hot_unplug_handler(&pdev->device);
515 "Failed to handle hot-unplug for device %s",
524 pci_plug(struct rte_device *dev)
526 return pci_probe_all_drivers(RTE_DEV_TO_PCI(dev));
530 pci_unplug(struct rte_device *dev)
532 struct rte_pci_device *pdev;
535 pdev = RTE_DEV_TO_PCI(dev);
536 ret = rte_pci_detach_dev(pdev);
538 rte_pci_remove_device(pdev);
539 rte_devargs_remove(dev->devargs);
546 pci_dma_map(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
548 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
550 if (!pdev || !pdev->driver) {
554 if (pdev->driver->dma_map)
555 return pdev->driver->dma_map(pdev, addr, iova, len);
557 * In case driver don't provides any specific mapping
558 * try fallback to VFIO.
560 if (pdev->kdrv == RTE_KDRV_VFIO)
561 return rte_vfio_container_dma_map
562 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
569 pci_dma_unmap(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
571 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
573 if (!pdev || !pdev->driver) {
577 if (pdev->driver->dma_unmap)
578 return pdev->driver->dma_unmap(pdev, addr, iova, len);
580 * In case driver don't provides any specific mapping
581 * try fallback to VFIO.
583 if (pdev->kdrv == RTE_KDRV_VFIO)
584 return rte_vfio_container_dma_unmap
585 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
592 pci_ignore_device(const struct rte_pci_device *dev)
594 struct rte_devargs *devargs = dev->device.devargs;
596 switch (rte_pci_bus.bus.conf.scan_mode) {
597 case RTE_BUS_SCAN_WHITELIST:
598 if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
601 case RTE_BUS_SCAN_UNDEFINED:
602 case RTE_BUS_SCAN_BLACKLIST:
603 if (devargs == NULL ||
604 devargs->policy != RTE_DEV_BLACKLISTED)
612 rte_pci_get_iommu_class(void)
614 enum rte_iova_mode iova_mode = RTE_IOVA_DC;
615 const struct rte_pci_device *dev;
616 const struct rte_pci_driver *drv;
617 bool devices_want_va = false;
618 bool devices_want_pa = false;
619 int iommu_no_va = -1;
621 FOREACH_DEVICE_ON_PCIBUS(dev) {
623 * We can check this only once, because the IOMMU hardware is
624 * the same for all of them.
626 if (iommu_no_va == -1)
627 iommu_no_va = pci_device_iommu_support_va(dev)
629 if (pci_ignore_device(dev))
631 if (dev->kdrv == RTE_KDRV_UNKNOWN ||
632 dev->kdrv == RTE_KDRV_NONE)
634 FOREACH_DRIVER_ON_PCIBUS(drv) {
635 enum rte_iova_mode dev_iova_mode;
637 if (!rte_pci_match(drv, dev))
640 dev_iova_mode = pci_device_iova_mode(drv, dev);
641 RTE_LOG(DEBUG, EAL, "PCI driver %s for device "
642 PCI_PRI_FMT " wants IOVA as '%s'\n",
644 dev->addr.domain, dev->addr.bus,
645 dev->addr.devid, dev->addr.function,
646 dev_iova_mode == RTE_IOVA_DC ? "DC" :
647 (dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA"));
648 if (dev_iova_mode == RTE_IOVA_PA)
649 devices_want_pa = true;
650 else if (dev_iova_mode == RTE_IOVA_VA)
651 devices_want_va = true;
654 if (iommu_no_va == 1) {
655 iova_mode = RTE_IOVA_PA;
656 if (devices_want_va) {
657 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but IOMMU does not support 'VA'.\n");
658 RTE_LOG(WARNING, EAL, "The devices that want 'VA' won't initialize.\n");
660 } else if (devices_want_va && !devices_want_pa) {
661 iova_mode = RTE_IOVA_VA;
662 } else if (devices_want_pa && !devices_want_va) {
663 iova_mode = RTE_IOVA_PA;
665 iova_mode = RTE_IOVA_DC;
666 if (devices_want_va) {
667 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but forcing 'DC' because other devices want 'PA'.\n");
668 RTE_LOG(WARNING, EAL, "Depending on the final decision by the EAL, not all devices may be able to initialize.\n");
674 struct rte_pci_bus rte_pci_bus = {
676 .scan = rte_pci_scan,
677 .probe = rte_pci_probe,
678 .find_device = pci_find_device,
680 .unplug = pci_unplug,
682 .dma_map = pci_dma_map,
683 .dma_unmap = pci_dma_unmap,
684 .get_iommu_class = rte_pci_get_iommu_class,
685 .dev_iterate = rte_pci_dev_iterate,
686 .hot_unplug_handler = pci_hot_unplug_handler,
687 .sigbus_handler = pci_sigbus_handler,
689 .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list),
690 .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list),
693 RTE_REGISTER_BUS(pci, rte_pci_bus.bus);