1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
11 #include <rte_bus_pci.h>
12 #include <rte_eal_memconfig.h>
13 #include <rte_malloc.h>
14 #include <rte_devargs.h>
15 #include <rte_memcpy.h>
18 #include "eal_filesystem.h"
25 * PCI probing under linux
27 * This code is used to simulate a PCI probe by parsing information in sysfs.
28 * When a registered device matches a driver, it is then initialized with
29 * IGB_UIO driver (or doesn't initialize, if the device wasn't bound to it).
32 extern struct rte_pci_bus rte_pci_bus;
35 pci_get_kernel_driver_by_path(const char *filename, char *dri_name,
42 if (!filename || !dri_name)
45 count = readlink(filename, path, PATH_MAX);
46 if (count >= PATH_MAX)
49 /* For device does not have a driver */
55 name = strrchr(path, '/');
57 strlcpy(dri_name, name + 1, len);
66 rte_pci_map_device(struct rte_pci_device *dev)
70 /* try mapping the NIC resources using VFIO if it exists */
74 if (pci_vfio_is_enabled())
75 ret = pci_vfio_map_resource(dev);
78 case RTE_KDRV_IGB_UIO:
79 case RTE_KDRV_UIO_GENERIC:
80 if (rte_eal_using_phys_addrs()) {
81 /* map resources for devices that use uio */
82 ret = pci_uio_map_resource(dev);
87 " Not managed by a supported kernel driver, skipped\n");
95 /* Unmap pci device */
97 rte_pci_unmap_device(struct rte_pci_device *dev)
99 /* try unmapping the NIC resources using VFIO if it exists */
103 if (pci_vfio_is_enabled())
104 pci_vfio_unmap_resource(dev);
107 case RTE_KDRV_IGB_UIO:
108 case RTE_KDRV_UIO_GENERIC:
109 /* unmap resources for devices that use uio */
110 pci_uio_unmap_resource(dev);
114 " Not managed by a supported kernel driver, skipped\n");
120 find_max_end_va(const struct rte_memseg_list *msl, void *arg)
122 size_t sz = msl->len;
123 void *end_va = RTE_PTR_ADD(msl->base_va, sz);
126 if (*max_va < end_va)
132 pci_find_max_end_va(void)
136 rte_memseg_list_walk(find_max_end_va, &va);
141 /* parse one line of the "resource" sysfs file (note that the 'line'
142 * string is modified)
145 pci_parse_one_sysfs_resource(char *line, size_t len, uint64_t *phys_addr,
146 uint64_t *end_addr, uint64_t *flags)
148 union pci_resource_info {
154 char *ptrs[PCI_RESOURCE_FMT_NVAL];
157 if (rte_strsplit(line, len, res_info.ptrs, 3, ' ') != 3) {
159 "%s(): bad resource format\n", __func__);
163 *phys_addr = strtoull(res_info.phys_addr, NULL, 16);
164 *end_addr = strtoull(res_info.end_addr, NULL, 16);
165 *flags = strtoull(res_info.flags, NULL, 16);
168 "%s(): bad resource format\n", __func__);
175 /* parse the "resource" sysfs file */
177 pci_parse_sysfs_resource(const char *filename, struct rte_pci_device *dev)
182 uint64_t phys_addr, end_addr, flags;
184 f = fopen(filename, "r");
186 RTE_LOG(ERR, EAL, "Cannot open sysfs resource\n");
190 for (i = 0; i<PCI_MAX_RESOURCE; i++) {
192 if (fgets(buf, sizeof(buf), f) == NULL) {
194 "%s(): cannot read resource\n", __func__);
197 if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
198 &end_addr, &flags) < 0)
201 if (flags & IORESOURCE_MEM) {
202 dev->mem_resource[i].phys_addr = phys_addr;
203 dev->mem_resource[i].len = end_addr - phys_addr + 1;
204 /* not mapped for now */
205 dev->mem_resource[i].addr = NULL;
216 /* Scan one pci sysfs entry, and fill the devices list from it. */
218 pci_scan_one(const char *dirname, const struct rte_pci_addr *addr)
220 char filename[PATH_MAX];
222 struct rte_pci_device *dev;
223 char driver[PATH_MAX];
226 dev = malloc(sizeof(*dev));
230 memset(dev, 0, sizeof(*dev));
231 dev->device.bus = &rte_pci_bus.bus;
235 snprintf(filename, sizeof(filename), "%s/vendor", dirname);
236 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
240 dev->id.vendor_id = (uint16_t)tmp;
243 snprintf(filename, sizeof(filename), "%s/device", dirname);
244 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
248 dev->id.device_id = (uint16_t)tmp;
250 /* get subsystem_vendor id */
251 snprintf(filename, sizeof(filename), "%s/subsystem_vendor",
253 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
257 dev->id.subsystem_vendor_id = (uint16_t)tmp;
259 /* get subsystem_device id */
260 snprintf(filename, sizeof(filename), "%s/subsystem_device",
262 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
266 dev->id.subsystem_device_id = (uint16_t)tmp;
269 snprintf(filename, sizeof(filename), "%s/class",
271 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
275 /* the least 24 bits are valid: class, subclass, program interface */
276 dev->id.class_id = (uint32_t)tmp & RTE_CLASS_ANY_ID;
280 snprintf(filename, sizeof(filename), "%s/max_vfs", dirname);
281 if (!access(filename, F_OK) &&
282 eal_parse_sysfs_value(filename, &tmp) == 0)
283 dev->max_vfs = (uint16_t)tmp;
285 /* for non igb_uio driver, need kernel version >= 3.8 */
286 snprintf(filename, sizeof(filename),
287 "%s/sriov_numvfs", dirname);
288 if (!access(filename, F_OK) &&
289 eal_parse_sysfs_value(filename, &tmp) == 0)
290 dev->max_vfs = (uint16_t)tmp;
293 /* get numa node, default to 0 if not present */
294 snprintf(filename, sizeof(filename), "%s/numa_node",
297 if (access(filename, F_OK) != -1) {
298 if (eal_parse_sysfs_value(filename, &tmp) == 0)
299 dev->device.numa_node = tmp;
301 dev->device.numa_node = -1;
303 dev->device.numa_node = 0;
308 /* parse resources */
309 snprintf(filename, sizeof(filename), "%s/resource", dirname);
310 if (pci_parse_sysfs_resource(filename, dev) < 0) {
311 RTE_LOG(ERR, EAL, "%s(): cannot parse resource\n", __func__);
317 snprintf(filename, sizeof(filename), "%s/driver", dirname);
318 ret = pci_get_kernel_driver_by_path(filename, driver, sizeof(driver));
320 RTE_LOG(ERR, EAL, "Fail to get kernel driver\n");
326 if (!strcmp(driver, "vfio-pci"))
327 dev->kdrv = RTE_KDRV_VFIO;
328 else if (!strcmp(driver, "igb_uio"))
329 dev->kdrv = RTE_KDRV_IGB_UIO;
330 else if (!strcmp(driver, "uio_pci_generic"))
331 dev->kdrv = RTE_KDRV_UIO_GENERIC;
333 dev->kdrv = RTE_KDRV_UNKNOWN;
335 dev->kdrv = RTE_KDRV_NONE;
337 /* device is valid, add in list (sorted) */
338 if (TAILQ_EMPTY(&rte_pci_bus.device_list)) {
339 rte_pci_add_device(dev);
341 struct rte_pci_device *dev2;
344 TAILQ_FOREACH(dev2, &rte_pci_bus.device_list, next) {
345 ret = rte_pci_addr_cmp(&dev->addr, &dev2->addr);
350 rte_pci_insert_device(dev2, dev);
351 } else { /* already registered */
352 if (!rte_dev_is_probed(&dev2->device)) {
353 dev2->kdrv = dev->kdrv;
354 dev2->max_vfs = dev->max_vfs;
356 memmove(dev2->mem_resource,
358 sizeof(dev->mem_resource));
361 * If device is plugged and driver is
362 * probed already, (This happens when
363 * we call rte_dev_probe which will
364 * scan all device on the bus) we don't
365 * need to do anything here unless...
367 if (dev2->kdrv != dev->kdrv ||
368 dev2->max_vfs != dev->max_vfs)
370 * This should not happens.
371 * But it is still possible if
372 * we unbind a device from
373 * vfio or uio before hotplug
374 * remove and rebind it with
375 * a different configure.
376 * So we just print out the
379 RTE_LOG(ERR, EAL, "Unexpected device scan at %s!\n",
387 rte_pci_add_device(dev);
394 pci_update_device(const struct rte_pci_addr *addr)
396 char filename[PATH_MAX];
398 snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
399 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
402 return pci_scan_one(filename, addr);
406 * split up a pci address into its constituent parts.
409 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
411 /* first split on ':' */
419 char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
422 char *buf_copy = strndup(buf, bufsize);
423 if (buf_copy == NULL)
426 if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
429 /* final split is on '.' between devid and function */
430 splitaddr.function = strchr(splitaddr.devid,'.');
431 if (splitaddr.function == NULL)
433 *splitaddr.function++ = '\0';
435 /* now convert to int values */
437 addr->domain = strtoul(splitaddr.domain, NULL, 16);
438 addr->bus = strtoul(splitaddr.bus, NULL, 16);
439 addr->devid = strtoul(splitaddr.devid, NULL, 16);
440 addr->function = strtoul(splitaddr.function, NULL, 10);
444 free(buf_copy); /* free the copy made with strdup */
452 * Scan the content of the PCI bus, and the devices in the devices
460 char dirname[PATH_MAX];
461 struct rte_pci_addr addr;
463 /* for debug purposes, PCI can be disabled */
464 if (!rte_eal_has_pci())
468 if (!pci_vfio_is_enabled())
469 RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
472 dir = opendir(rte_pci_get_sysfs_path());
474 RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
475 __func__, strerror(errno));
479 while ((e = readdir(dir)) != NULL) {
480 if (e->d_name[0] == '.')
483 if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
486 snprintf(dirname, sizeof(dirname), "%s/%s",
487 rte_pci_get_sysfs_path(), e->d_name);
489 if (pci_scan_one(dirname, &addr) < 0)
501 * Is pci device bound to any kdrv
504 pci_one_device_is_bound(void)
506 struct rte_pci_device *dev = NULL;
509 FOREACH_DEVICE_ON_PCIBUS(dev) {
510 if (dev->kdrv == RTE_KDRV_UNKNOWN ||
511 dev->kdrv == RTE_KDRV_NONE) {
522 * Any one of the device bound to uio
525 pci_one_device_bound_uio(void)
527 struct rte_pci_device *dev = NULL;
528 struct rte_devargs *devargs;
531 FOREACH_DEVICE_ON_PCIBUS(dev) {
532 devargs = dev->device.devargs;
535 switch (rte_pci_bus.bus.conf.scan_mode) {
536 case RTE_BUS_SCAN_WHITELIST:
537 if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
540 case RTE_BUS_SCAN_UNDEFINED:
541 case RTE_BUS_SCAN_BLACKLIST:
542 if (devargs == NULL ||
543 devargs->policy != RTE_DEV_BLACKLISTED)
551 if (dev->kdrv == RTE_KDRV_IGB_UIO ||
552 dev->kdrv == RTE_KDRV_UIO_GENERIC) {
560 * Any one of the device has iova as va
563 pci_one_device_has_iova_va(void)
565 struct rte_pci_device *dev = NULL;
566 struct rte_pci_driver *drv = NULL;
568 FOREACH_DRIVER_ON_PCIBUS(drv) {
569 if (drv && drv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) {
570 FOREACH_DEVICE_ON_PCIBUS(dev) {
571 if (dev->kdrv == RTE_KDRV_VFIO &&
572 rte_pci_match(drv, dev))
580 #if defined(RTE_ARCH_X86)
582 pci_one_device_iommu_support_va(struct rte_pci_device *dev)
584 #define VTD_CAP_MGAW_SHIFT 16
585 #define VTD_CAP_MGAW_MASK (0x3fULL << VTD_CAP_MGAW_SHIFT)
586 #define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
587 struct rte_pci_addr *addr = &dev->addr;
588 char filename[PATH_MAX];
590 uint64_t mgaw, vtd_cap_reg = 0;
592 snprintf(filename, sizeof(filename),
593 "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
594 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
596 if (access(filename, F_OK) == -1) {
597 /* We don't have an Intel IOMMU, assume VA supported*/
601 /* We have an intel IOMMU */
602 fp = fopen(filename, "r");
604 RTE_LOG(ERR, EAL, "%s(): can't open %s\n", __func__, filename);
608 if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
609 RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
616 mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
619 * Assuming there is no limitation by now. We can not know at this point
620 * because the memory has not been initialized yet. Setting the dma mask
621 * will force a check once memory initialization is done. We can not do
622 * a fallback to IOVA PA now, but if the dma check fails, the error
623 * message should advice for using '--iova-mode pa' if IOVA VA is the
626 rte_mem_set_dma_mask(mgaw);
629 #elif defined(RTE_ARCH_PPC_64)
631 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
637 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
644 * All devices IOMMUs support VA as IOVA
647 pci_devices_iommu_support_va(void)
649 struct rte_pci_device *dev = NULL;
650 struct rte_pci_driver *drv = NULL;
652 FOREACH_DRIVER_ON_PCIBUS(drv) {
653 FOREACH_DEVICE_ON_PCIBUS(dev) {
654 if (!rte_pci_match(drv, dev))
657 * just one PCI device needs to be checked out because
658 * the IOMMU hardware is the same for all of them.
660 return pci_one_device_iommu_support_va(dev);
667 * Get iommu class of PCI devices on the bus.
670 rte_pci_get_iommu_class(void)
673 bool is_vfio_noiommu_enabled = true;
678 is_bound = pci_one_device_is_bound();
682 has_iova_va = pci_one_device_has_iova_va();
683 is_bound_uio = pci_one_device_bound_uio();
684 iommu_no_va = !pci_devices_iommu_support_va();
686 is_vfio_noiommu_enabled = rte_vfio_noiommu_is_enabled() == true ?
690 if (has_iova_va && !is_bound_uio && !is_vfio_noiommu_enabled &&
695 RTE_LOG(WARNING, EAL, "Some devices want iova as va but pa will be used because.. ");
696 if (is_vfio_noiommu_enabled)
697 RTE_LOG(WARNING, EAL, "vfio-noiommu mode configured\n");
699 RTE_LOG(WARNING, EAL, "few device bound to UIO\n");
701 RTE_LOG(WARNING, EAL, "IOMMU does not support IOVA as VA\n");
707 /* Read PCI config space. */
708 int rte_pci_read_config(const struct rte_pci_device *device,
709 void *buf, size_t len, off_t offset)
711 char devname[RTE_DEV_NAME_MAX_LEN] = "";
712 const struct rte_intr_handle *intr_handle = &device->intr_handle;
714 switch (device->kdrv) {
715 case RTE_KDRV_IGB_UIO:
716 case RTE_KDRV_UIO_GENERIC:
717 return pci_uio_read_config(intr_handle, buf, len, offset);
720 return pci_vfio_read_config(intr_handle, buf, len, offset);
723 rte_pci_device_name(&device->addr, devname,
724 RTE_DEV_NAME_MAX_LEN);
726 "Unknown driver type for %s\n", devname);
731 /* Write PCI config space. */
732 int rte_pci_write_config(const struct rte_pci_device *device,
733 const void *buf, size_t len, off_t offset)
735 char devname[RTE_DEV_NAME_MAX_LEN] = "";
736 const struct rte_intr_handle *intr_handle = &device->intr_handle;
738 switch (device->kdrv) {
739 case RTE_KDRV_IGB_UIO:
740 case RTE_KDRV_UIO_GENERIC:
741 return pci_uio_write_config(intr_handle, buf, len, offset);
744 return pci_vfio_write_config(intr_handle, buf, len, offset);
747 rte_pci_device_name(&device->addr, devname,
748 RTE_DEV_NAME_MAX_LEN);
750 "Unknown driver type for %s\n", devname);
755 #if defined(RTE_ARCH_X86)
757 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
758 struct rte_pci_ioport *p)
767 snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
768 dev->addr.domain, dev->addr.bus,
769 dev->addr.devid, dev->addr.function);
771 fp = fopen("/proc/ioports", "r");
773 RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
777 while (getdelim(&line, &linesz, '\n', fp) > 0) {
782 n = strcspn(ptr, ":");
786 while (*left && isspace(*left))
789 if (!strncmp(left, pci_id, strlen(pci_id))) {
792 while (*ptr && isspace(*ptr))
795 sscanf(ptr, "%04hx-%04hx", &start, &end);
808 RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
815 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
816 struct rte_pci_ioport *p)
823 if (pci_vfio_is_enabled())
824 ret = pci_vfio_ioport_map(dev, bar, p);
827 case RTE_KDRV_IGB_UIO:
828 ret = pci_uio_ioport_map(dev, bar, p);
830 case RTE_KDRV_UIO_GENERIC:
831 #if defined(RTE_ARCH_X86)
832 ret = pci_ioport_map(dev, bar, p);
834 ret = pci_uio_ioport_map(dev, bar, p);
838 #if defined(RTE_ARCH_X86)
839 ret = pci_ioport_map(dev, bar, p);
853 rte_pci_ioport_read(struct rte_pci_ioport *p,
854 void *data, size_t len, off_t offset)
856 switch (p->dev->kdrv) {
859 pci_vfio_ioport_read(p, data, len, offset);
862 case RTE_KDRV_IGB_UIO:
863 pci_uio_ioport_read(p, data, len, offset);
865 case RTE_KDRV_UIO_GENERIC:
866 pci_uio_ioport_read(p, data, len, offset);
869 #if defined(RTE_ARCH_X86)
870 pci_uio_ioport_read(p, data, len, offset);
879 rte_pci_ioport_write(struct rte_pci_ioport *p,
880 const void *data, size_t len, off_t offset)
882 switch (p->dev->kdrv) {
885 pci_vfio_ioport_write(p, data, len, offset);
888 case RTE_KDRV_IGB_UIO:
889 pci_uio_ioport_write(p, data, len, offset);
891 case RTE_KDRV_UIO_GENERIC:
892 pci_uio_ioport_write(p, data, len, offset);
895 #if defined(RTE_ARCH_X86)
896 pci_uio_ioport_write(p, data, len, offset);
905 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
909 switch (p->dev->kdrv) {
912 if (pci_vfio_is_enabled())
913 ret = pci_vfio_ioport_unmap(p);
916 case RTE_KDRV_IGB_UIO:
917 ret = pci_uio_ioport_unmap(p);
919 case RTE_KDRV_UIO_GENERIC:
920 #if defined(RTE_ARCH_X86)
923 ret = pci_uio_ioport_unmap(p);
927 #if defined(RTE_ARCH_X86)