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(DEBUG, 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(DEBUG, 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 RTE_LOG(INFO, EAL, "Probe PCI driver: %s (%x:%x) device: "PCI_PRI_FMT" (socket %i)\n",
199 dr->driver.name, dev->id.vendor_id, dev->id.device_id,
200 loc->domain, loc->bus, loc->devid, loc->function,
201 dev->device.numa_node);
202 /* call the driver probe() function */
203 ret = dr->probe(dr, dev);
205 return ret; /* no rollback if already succeeded earlier */
208 if ((dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING) &&
209 /* Don't unmap if device is unsupported and
210 * driver needs mapped resources.
213 (dr->drv_flags & RTE_PCI_DRV_KEEP_MAPPED_RES)))
214 rte_pci_unmap_device(dev);
216 dev->device.driver = &dr->driver;
223 * If vendor/device ID match, call the remove() function of the
227 rte_pci_detach_dev(struct rte_pci_device *dev)
229 struct rte_pci_addr *loc;
230 struct rte_pci_driver *dr;
239 RTE_LOG(DEBUG, EAL, "PCI device "PCI_PRI_FMT" on NUMA socket %i\n",
240 loc->domain, loc->bus, loc->devid,
241 loc->function, dev->device.numa_node);
243 RTE_LOG(DEBUG, EAL, " remove driver: %x:%x %s\n", dev->id.vendor_id,
244 dev->id.device_id, dr->driver.name);
247 ret = dr->remove(dev);
252 /* clear driver structure */
254 dev->device.driver = NULL;
256 if (dr->drv_flags & RTE_PCI_DRV_NEED_MAPPING)
257 /* unmap resources for devices that use igb_uio */
258 rte_pci_unmap_device(dev);
264 * If vendor/device ID match, call the probe() function of all
265 * registered driver for the given device. Return < 0 if initialization
266 * failed, return 1 if no driver is found for this device.
269 pci_probe_all_drivers(struct rte_pci_device *dev)
271 struct rte_pci_driver *dr = NULL;
277 FOREACH_DRIVER_ON_PCIBUS(dr) {
278 rc = rte_pci_probe_one_driver(dr, dev);
280 /* negative value is an error */
283 /* positive value means driver doesn't support it */
291 * Scan the content of the PCI bus, and call the probe() function for
292 * all registered drivers that have a matching entry in its id_table
293 * for discovered devices.
298 struct rte_pci_device *dev = NULL;
299 size_t probed = 0, failed = 0;
300 struct rte_devargs *devargs;
304 if (rte_pci_bus.bus.conf.scan_mode != RTE_BUS_SCAN_WHITELIST)
307 FOREACH_DEVICE_ON_PCIBUS(dev) {
310 devargs = dev->device.devargs;
311 /* probe all or only whitelisted devices */
313 ret = pci_probe_all_drivers(dev);
314 else if (devargs != NULL &&
315 devargs->policy == RTE_DEV_WHITELISTED)
316 ret = pci_probe_all_drivers(dev);
318 if (ret != -EEXIST) {
319 RTE_LOG(ERR, EAL, "Requested device "
320 PCI_PRI_FMT " cannot be used\n",
321 dev->addr.domain, dev->addr.bus,
322 dev->addr.devid, dev->addr.function);
330 return (probed && probed == failed) ? -1 : 0;
333 /* dump one device */
335 pci_dump_one_device(FILE *f, struct rte_pci_device *dev)
339 fprintf(f, PCI_PRI_FMT, dev->addr.domain, dev->addr.bus,
340 dev->addr.devid, dev->addr.function);
341 fprintf(f, " - vendor:%x device:%x\n", dev->id.vendor_id,
344 for (i = 0; i != sizeof(dev->mem_resource) /
345 sizeof(dev->mem_resource[0]); i++) {
346 fprintf(f, " %16.16"PRIx64" %16.16"PRIx64"\n",
347 dev->mem_resource[i].phys_addr,
348 dev->mem_resource[i].len);
353 /* dump devices on the bus */
355 rte_pci_dump(FILE *f)
357 struct rte_pci_device *dev = NULL;
359 FOREACH_DEVICE_ON_PCIBUS(dev) {
360 pci_dump_one_device(f, dev);
365 pci_parse(const char *name, void *addr)
367 struct rte_pci_addr *out = addr;
368 struct rte_pci_addr pci_addr;
371 parse = (rte_pci_addr_parse(name, &pci_addr) == 0);
372 if (parse && addr != NULL)
374 return parse == false;
377 /* register a driver */
379 rte_pci_register(struct rte_pci_driver *driver)
381 TAILQ_INSERT_TAIL(&rte_pci_bus.driver_list, driver, next);
382 driver->bus = &rte_pci_bus;
385 /* unregister a driver */
387 rte_pci_unregister(struct rte_pci_driver *driver)
389 TAILQ_REMOVE(&rte_pci_bus.driver_list, driver, next);
393 /* Add a device to PCI bus */
395 rte_pci_add_device(struct rte_pci_device *pci_dev)
397 TAILQ_INSERT_TAIL(&rte_pci_bus.device_list, pci_dev, next);
400 /* Insert a device into a predefined position in PCI bus */
402 rte_pci_insert_device(struct rte_pci_device *exist_pci_dev,
403 struct rte_pci_device *new_pci_dev)
405 TAILQ_INSERT_BEFORE(exist_pci_dev, new_pci_dev, next);
408 /* Remove a device from PCI bus */
410 rte_pci_remove_device(struct rte_pci_device *pci_dev)
412 TAILQ_REMOVE(&rte_pci_bus.device_list, pci_dev, next);
415 static struct rte_device *
416 pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
419 const struct rte_pci_device *pstart;
420 struct rte_pci_device *pdev;
423 pstart = RTE_DEV_TO_PCI_CONST(start);
424 pdev = TAILQ_NEXT(pstart, next);
426 pdev = TAILQ_FIRST(&rte_pci_bus.device_list);
428 while (pdev != NULL) {
429 if (cmp(&pdev->device, data) == 0)
430 return &pdev->device;
431 pdev = TAILQ_NEXT(pdev, next);
437 * find the device which encounter the failure, by iterate over all device on
438 * PCI bus to check if the memory failure address is located in the range
439 * of the BARs of the device.
441 static struct rte_pci_device *
442 pci_find_device_by_addr(const void *failure_addr)
444 struct rte_pci_device *pdev = NULL;
445 uint64_t check_point, start, end, len;
448 check_point = (uint64_t)(uintptr_t)failure_addr;
450 FOREACH_DEVICE_ON_PCIBUS(pdev) {
451 for (i = 0; i != RTE_DIM(pdev->mem_resource); i++) {
452 start = (uint64_t)(uintptr_t)pdev->mem_resource[i].addr;
453 len = pdev->mem_resource[i].len;
455 if (check_point >= start && check_point < end) {
456 RTE_LOG(DEBUG, EAL, "Failure address %16.16"
457 PRIx64" belongs to device %s!\n",
458 check_point, pdev->device.name);
467 pci_hot_unplug_handler(struct rte_device *dev)
469 struct rte_pci_device *pdev = NULL;
472 pdev = RTE_DEV_TO_PCI(dev);
476 switch (pdev->kdrv) {
477 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
480 * vfio kernel module guaranty the pci device would not be
481 * deleted until the user space release the resource, so no
482 * need to remap BARs resource here, just directly notify
483 * the req event to the user space to handle it.
485 rte_dev_event_callback_process(dev->name,
486 RTE_DEV_EVENT_REMOVE);
489 case RTE_KDRV_IGB_UIO:
490 case RTE_KDRV_UIO_GENERIC:
491 case RTE_KDRV_NIC_UIO:
492 /* BARs resource is invalid, remap it to be safe. */
493 ret = pci_uio_remap_resource(pdev);
497 "Not managed by a supported kernel driver, skipped\n");
506 pci_sigbus_handler(const void *failure_addr)
508 struct rte_pci_device *pdev = NULL;
511 pdev = pci_find_device_by_addr(failure_addr);
513 /* It is a generic sigbus error, no bus would handle it. */
516 /* The sigbus error is caused of hot-unplug. */
517 ret = pci_hot_unplug_handler(&pdev->device);
520 "Failed to handle hot-unplug for device %s",
529 pci_plug(struct rte_device *dev)
531 return pci_probe_all_drivers(RTE_DEV_TO_PCI(dev));
535 pci_unplug(struct rte_device *dev)
537 struct rte_pci_device *pdev;
540 pdev = RTE_DEV_TO_PCI(dev);
541 ret = rte_pci_detach_dev(pdev);
543 rte_pci_remove_device(pdev);
544 rte_devargs_remove(dev->devargs);
551 pci_dma_map(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
553 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
555 if (!pdev || !pdev->driver) {
559 if (pdev->driver->dma_map)
560 return pdev->driver->dma_map(pdev, addr, iova, len);
562 * In case driver don't provides any specific mapping
563 * try fallback to VFIO.
565 if (pdev->kdrv == RTE_KDRV_VFIO)
566 return rte_vfio_container_dma_map
567 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
574 pci_dma_unmap(struct rte_device *dev, void *addr, uint64_t iova, size_t len)
576 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev);
578 if (!pdev || !pdev->driver) {
582 if (pdev->driver->dma_unmap)
583 return pdev->driver->dma_unmap(pdev, addr, iova, len);
585 * In case driver don't provides any specific mapping
586 * try fallback to VFIO.
588 if (pdev->kdrv == RTE_KDRV_VFIO)
589 return rte_vfio_container_dma_unmap
590 (RTE_VFIO_DEFAULT_CONTAINER_FD, (uintptr_t)addr,
597 pci_ignore_device(const struct rte_pci_device *dev)
599 struct rte_devargs *devargs = dev->device.devargs;
601 switch (rte_pci_bus.bus.conf.scan_mode) {
602 case RTE_BUS_SCAN_WHITELIST:
603 if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
606 case RTE_BUS_SCAN_UNDEFINED:
607 case RTE_BUS_SCAN_BLACKLIST:
608 if (devargs == NULL ||
609 devargs->policy != RTE_DEV_BLACKLISTED)
617 rte_pci_get_iommu_class(void)
619 enum rte_iova_mode iova_mode = RTE_IOVA_DC;
620 const struct rte_pci_device *dev;
621 const struct rte_pci_driver *drv;
622 bool devices_want_va = false;
623 bool devices_want_pa = false;
624 int iommu_no_va = -1;
626 FOREACH_DEVICE_ON_PCIBUS(dev) {
628 * We can check this only once, because the IOMMU hardware is
629 * the same for all of them.
631 if (iommu_no_va == -1)
632 iommu_no_va = pci_device_iommu_support_va(dev)
634 if (pci_ignore_device(dev))
636 if (dev->kdrv == RTE_KDRV_UNKNOWN ||
637 dev->kdrv == RTE_KDRV_NONE)
639 FOREACH_DRIVER_ON_PCIBUS(drv) {
640 enum rte_iova_mode dev_iova_mode;
642 if (!rte_pci_match(drv, dev))
645 dev_iova_mode = pci_device_iova_mode(drv, dev);
646 RTE_LOG(DEBUG, EAL, "PCI driver %s for device "
647 PCI_PRI_FMT " wants IOVA as '%s'\n",
649 dev->addr.domain, dev->addr.bus,
650 dev->addr.devid, dev->addr.function,
651 dev_iova_mode == RTE_IOVA_DC ? "DC" :
652 (dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA"));
653 if (dev_iova_mode == RTE_IOVA_PA)
654 devices_want_pa = true;
655 else if (dev_iova_mode == RTE_IOVA_VA)
656 devices_want_va = true;
659 if (iommu_no_va == 1) {
660 iova_mode = RTE_IOVA_PA;
661 if (devices_want_va) {
662 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but IOMMU does not support 'VA'.\n");
663 RTE_LOG(WARNING, EAL, "The devices that want 'VA' won't initialize.\n");
665 } else if (devices_want_va && !devices_want_pa) {
666 iova_mode = RTE_IOVA_VA;
667 } else if (devices_want_pa && !devices_want_va) {
668 iova_mode = RTE_IOVA_PA;
670 iova_mode = RTE_IOVA_DC;
671 if (devices_want_va) {
672 RTE_LOG(WARNING, EAL, "Some devices want 'VA' but forcing 'DC' because other devices want 'PA'.\n");
673 RTE_LOG(WARNING, EAL, "Depending on the final decision by the EAL, not all devices may be able to initialize.\n");
679 struct rte_pci_bus rte_pci_bus = {
681 .scan = rte_pci_scan,
683 .find_device = pci_find_device,
685 .unplug = pci_unplug,
687 .dma_map = pci_dma_map,
688 .dma_unmap = pci_dma_unmap,
689 .get_iommu_class = rte_pci_get_iommu_class,
690 .dev_iterate = rte_pci_dev_iterate,
691 .hot_unplug_handler = pci_hot_unplug_handler,
692 .sigbus_handler = pci_sigbus_handler,
694 .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list),
695 .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list),
698 RTE_REGISTER_BUS(pci, rte_pci_bus.bus);