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_private.h"
19 #include "eal_filesystem.h"
26 * PCI probing under linux
28 * This code is used to simulate a PCI probe by parsing information in sysfs.
29 * When a registered device matches a driver, it is then initialized with
30 * IGB_UIO driver (or doesn't initialize, if the device wasn't bound to it).
33 extern struct rte_pci_bus rte_pci_bus;
36 pci_get_kernel_driver_by_path(const char *filename, char *dri_name,
43 if (!filename || !dri_name)
46 count = readlink(filename, path, PATH_MAX);
47 if (count >= PATH_MAX)
50 /* For device does not have a driver */
56 name = strrchr(path, '/');
58 strlcpy(dri_name, name + 1, len);
67 rte_pci_map_device(struct rte_pci_device *dev)
71 /* try mapping the NIC resources using VFIO if it exists */
75 if (pci_vfio_is_enabled())
76 ret = pci_vfio_map_resource(dev);
79 case RTE_KDRV_IGB_UIO:
80 case RTE_KDRV_UIO_GENERIC:
81 if (rte_eal_using_phys_addrs()) {
82 /* map resources for devices that use uio */
83 ret = pci_uio_map_resource(dev);
88 " Not managed by a supported kernel driver, skipped\n");
96 /* Unmap pci device */
98 rte_pci_unmap_device(struct rte_pci_device *dev)
100 /* try unmapping the NIC resources using VFIO if it exists */
104 if (pci_vfio_is_enabled())
105 pci_vfio_unmap_resource(dev);
108 case RTE_KDRV_IGB_UIO:
109 case RTE_KDRV_UIO_GENERIC:
110 /* unmap resources for devices that use uio */
111 pci_uio_unmap_resource(dev);
115 " Not managed by a supported kernel driver, skipped\n");
121 find_max_end_va(const struct rte_memseg_list *msl, void *arg)
123 size_t sz = msl->memseg_arr.len * msl->page_sz;
124 void *end_va = RTE_PTR_ADD(msl->base_va, sz);
127 if (*max_va < end_va)
133 pci_find_max_end_va(void)
137 rte_memseg_list_walk(find_max_end_va, &va);
142 /* parse one line of the "resource" sysfs file (note that the 'line'
143 * string is modified)
146 pci_parse_one_sysfs_resource(char *line, size_t len, uint64_t *phys_addr,
147 uint64_t *end_addr, uint64_t *flags)
149 union pci_resource_info {
155 char *ptrs[PCI_RESOURCE_FMT_NVAL];
158 if (rte_strsplit(line, len, res_info.ptrs, 3, ' ') != 3) {
160 "%s(): bad resource format\n", __func__);
164 *phys_addr = strtoull(res_info.phys_addr, NULL, 16);
165 *end_addr = strtoull(res_info.end_addr, NULL, 16);
166 *flags = strtoull(res_info.flags, NULL, 16);
169 "%s(): bad resource format\n", __func__);
176 /* parse the "resource" sysfs file */
178 pci_parse_sysfs_resource(const char *filename, struct rte_pci_device *dev)
183 uint64_t phys_addr, end_addr, flags;
185 f = fopen(filename, "r");
187 RTE_LOG(ERR, EAL, "Cannot open sysfs resource\n");
191 for (i = 0; i<PCI_MAX_RESOURCE; i++) {
193 if (fgets(buf, sizeof(buf), f) == NULL) {
195 "%s(): cannot read resource\n", __func__);
198 if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
199 &end_addr, &flags) < 0)
202 if (flags & IORESOURCE_MEM) {
203 dev->mem_resource[i].phys_addr = phys_addr;
204 dev->mem_resource[i].len = end_addr - phys_addr + 1;
205 /* not mapped for now */
206 dev->mem_resource[i].addr = NULL;
217 /* Scan one pci sysfs entry, and fill the devices list from it. */
219 pci_scan_one(const char *dirname, const struct rte_pci_addr *addr)
221 char filename[PATH_MAX];
223 struct rte_pci_device *dev;
224 char driver[PATH_MAX];
227 dev = malloc(sizeof(*dev));
231 memset(dev, 0, sizeof(*dev));
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 dev2->kdrv = dev->kdrv;
353 dev2->max_vfs = dev->max_vfs;
355 memmove(dev2->mem_resource, dev->mem_resource,
356 sizeof(dev->mem_resource));
362 rte_pci_add_device(dev);
369 pci_update_device(const struct rte_pci_addr *addr)
371 char filename[PATH_MAX];
373 snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
374 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
377 return pci_scan_one(filename, addr);
381 * split up a pci address into its constituent parts.
384 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
386 /* first split on ':' */
394 char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
397 char *buf_copy = strndup(buf, bufsize);
398 if (buf_copy == NULL)
401 if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
404 /* final split is on '.' between devid and function */
405 splitaddr.function = strchr(splitaddr.devid,'.');
406 if (splitaddr.function == NULL)
408 *splitaddr.function++ = '\0';
410 /* now convert to int values */
412 addr->domain = strtoul(splitaddr.domain, NULL, 16);
413 addr->bus = strtoul(splitaddr.bus, NULL, 16);
414 addr->devid = strtoul(splitaddr.devid, NULL, 16);
415 addr->function = strtoul(splitaddr.function, NULL, 10);
419 free(buf_copy); /* free the copy made with strdup */
427 * Scan the content of the PCI bus, and the devices in the devices
435 char dirname[PATH_MAX];
436 struct rte_pci_addr addr;
438 /* for debug purposes, PCI can be disabled */
439 if (!rte_eal_has_pci())
443 if (!pci_vfio_is_enabled())
444 RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
447 dir = opendir(rte_pci_get_sysfs_path());
449 RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
450 __func__, strerror(errno));
454 while ((e = readdir(dir)) != NULL) {
455 if (e->d_name[0] == '.')
458 if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
461 snprintf(dirname, sizeof(dirname), "%s/%s",
462 rte_pci_get_sysfs_path(), e->d_name);
464 if (pci_scan_one(dirname, &addr) < 0)
476 * Is pci device bound to any kdrv
479 pci_one_device_is_bound(void)
481 struct rte_pci_device *dev = NULL;
484 FOREACH_DEVICE_ON_PCIBUS(dev) {
485 if (dev->kdrv == RTE_KDRV_UNKNOWN ||
486 dev->kdrv == RTE_KDRV_NONE) {
497 * Any one of the device bound to uio
500 pci_one_device_bound_uio(void)
502 struct rte_pci_device *dev = NULL;
503 struct rte_devargs *devargs;
506 FOREACH_DEVICE_ON_PCIBUS(dev) {
507 devargs = dev->device.devargs;
510 switch (rte_pci_bus.bus.conf.scan_mode) {
511 case RTE_BUS_SCAN_WHITELIST:
512 if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
515 case RTE_BUS_SCAN_UNDEFINED:
516 case RTE_BUS_SCAN_BLACKLIST:
517 if (devargs == NULL ||
518 devargs->policy != RTE_DEV_BLACKLISTED)
526 if (dev->kdrv == RTE_KDRV_IGB_UIO ||
527 dev->kdrv == RTE_KDRV_UIO_GENERIC) {
535 * Any one of the device has iova as va
538 pci_one_device_has_iova_va(void)
540 struct rte_pci_device *dev = NULL;
541 struct rte_pci_driver *drv = NULL;
543 FOREACH_DRIVER_ON_PCIBUS(drv) {
544 if (drv && drv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) {
545 FOREACH_DEVICE_ON_PCIBUS(dev) {
546 if (dev->kdrv == RTE_KDRV_VFIO &&
547 rte_pci_match(drv, dev))
555 #if defined(RTE_ARCH_X86)
557 pci_one_device_iommu_support_va(struct rte_pci_device *dev)
559 #define VTD_CAP_MGAW_SHIFT 16
560 #define VTD_CAP_MGAW_MASK (0x3fULL << VTD_CAP_MGAW_SHIFT)
561 #define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
562 struct rte_pci_addr *addr = &dev->addr;
563 char filename[PATH_MAX];
565 uint64_t mgaw, vtd_cap_reg = 0;
567 snprintf(filename, sizeof(filename),
568 "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
569 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
571 if (access(filename, F_OK) == -1) {
572 /* We don't have an Intel IOMMU, assume VA supported*/
576 /* We have an intel IOMMU */
577 fp = fopen(filename, "r");
579 RTE_LOG(ERR, EAL, "%s(): can't open %s\n", __func__, filename);
583 if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
584 RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
591 mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
592 if (mgaw < X86_VA_WIDTH)
597 #elif defined(RTE_ARCH_PPC_64)
599 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
605 pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
612 * All devices IOMMUs support VA as IOVA
615 pci_devices_iommu_support_va(void)
617 struct rte_pci_device *dev = NULL;
618 struct rte_pci_driver *drv = NULL;
620 FOREACH_DRIVER_ON_PCIBUS(drv) {
621 FOREACH_DEVICE_ON_PCIBUS(dev) {
622 if (!rte_pci_match(drv, dev))
624 if (!pci_one_device_iommu_support_va(dev))
632 * Get iommu class of PCI devices on the bus.
635 rte_pci_get_iommu_class(void)
638 bool is_vfio_noiommu_enabled = true;
643 is_bound = pci_one_device_is_bound();
647 has_iova_va = pci_one_device_has_iova_va();
648 is_bound_uio = pci_one_device_bound_uio();
649 iommu_no_va = !pci_devices_iommu_support_va();
651 is_vfio_noiommu_enabled = rte_vfio_noiommu_is_enabled() == true ?
655 if (has_iova_va && !is_bound_uio && !is_vfio_noiommu_enabled &&
660 RTE_LOG(WARNING, EAL, "Some devices want iova as va but pa will be used because.. ");
661 if (is_vfio_noiommu_enabled)
662 RTE_LOG(WARNING, EAL, "vfio-noiommu mode configured\n");
664 RTE_LOG(WARNING, EAL, "few device bound to UIO\n");
666 RTE_LOG(WARNING, EAL, "IOMMU does not support IOVA as VA\n");
672 /* Read PCI config space. */
673 int rte_pci_read_config(const struct rte_pci_device *device,
674 void *buf, size_t len, off_t offset)
676 const struct rte_intr_handle *intr_handle = &device->intr_handle;
678 switch (intr_handle->type) {
679 case RTE_INTR_HANDLE_UIO:
680 case RTE_INTR_HANDLE_UIO_INTX:
681 return pci_uio_read_config(intr_handle, buf, len, offset);
684 case RTE_INTR_HANDLE_VFIO_MSIX:
685 case RTE_INTR_HANDLE_VFIO_MSI:
686 case RTE_INTR_HANDLE_VFIO_LEGACY:
687 return pci_vfio_read_config(intr_handle, buf, len, offset);
691 "Unknown handle type of fd %d\n",
697 /* Write PCI config space. */
698 int rte_pci_write_config(const struct rte_pci_device *device,
699 const void *buf, size_t len, off_t offset)
701 const struct rte_intr_handle *intr_handle = &device->intr_handle;
703 switch (intr_handle->type) {
704 case RTE_INTR_HANDLE_UIO:
705 case RTE_INTR_HANDLE_UIO_INTX:
706 return pci_uio_write_config(intr_handle, buf, len, offset);
709 case RTE_INTR_HANDLE_VFIO_MSIX:
710 case RTE_INTR_HANDLE_VFIO_MSI:
711 case RTE_INTR_HANDLE_VFIO_LEGACY:
712 return pci_vfio_write_config(intr_handle, buf, len, offset);
716 "Unknown handle type of fd %d\n",
722 #if defined(RTE_ARCH_X86)
724 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
725 struct rte_pci_ioport *p)
734 snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
735 dev->addr.domain, dev->addr.bus,
736 dev->addr.devid, dev->addr.function);
738 fp = fopen("/proc/ioports", "r");
740 RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
744 while (getdelim(&line, &linesz, '\n', fp) > 0) {
749 n = strcspn(ptr, ":");
753 while (*left && isspace(*left))
756 if (!strncmp(left, pci_id, strlen(pci_id))) {
759 while (*ptr && isspace(*ptr))
762 sscanf(ptr, "%04hx-%04hx", &start, &end);
775 RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
782 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
783 struct rte_pci_ioport *p)
790 if (pci_vfio_is_enabled())
791 ret = pci_vfio_ioport_map(dev, bar, p);
794 case RTE_KDRV_IGB_UIO:
795 ret = pci_uio_ioport_map(dev, bar, p);
797 case RTE_KDRV_UIO_GENERIC:
798 #if defined(RTE_ARCH_X86)
799 ret = pci_ioport_map(dev, bar, p);
801 ret = pci_uio_ioport_map(dev, bar, p);
805 #if defined(RTE_ARCH_X86)
806 ret = pci_ioport_map(dev, bar, p);
820 rte_pci_ioport_read(struct rte_pci_ioport *p,
821 void *data, size_t len, off_t offset)
823 switch (p->dev->kdrv) {
826 pci_vfio_ioport_read(p, data, len, offset);
829 case RTE_KDRV_IGB_UIO:
830 pci_uio_ioport_read(p, data, len, offset);
832 case RTE_KDRV_UIO_GENERIC:
833 pci_uio_ioport_read(p, data, len, offset);
836 #if defined(RTE_ARCH_X86)
837 pci_uio_ioport_read(p, data, len, offset);
846 rte_pci_ioport_write(struct rte_pci_ioport *p,
847 const void *data, size_t len, off_t offset)
849 switch (p->dev->kdrv) {
852 pci_vfio_ioport_write(p, data, len, offset);
855 case RTE_KDRV_IGB_UIO:
856 pci_uio_ioport_write(p, data, len, offset);
858 case RTE_KDRV_UIO_GENERIC:
859 pci_uio_ioport_write(p, data, len, offset);
862 #if defined(RTE_ARCH_X86)
863 pci_uio_ioport_write(p, data, len, offset);
872 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
876 switch (p->dev->kdrv) {
879 if (pci_vfio_is_enabled())
880 ret = pci_vfio_ioport_unmap(p);
883 case RTE_KDRV_IGB_UIO:
884 ret = pci_uio_ioport_unmap(p);
886 case RTE_KDRV_UIO_GENERIC:
887 #if defined(RTE_ARCH_X86)
890 ret = pci_uio_ioport_unmap(p);
894 #if defined(RTE_ARCH_X86)