1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
11 #include <rte_bus_pci.h>
12 #include <rte_malloc.h>
13 #include <rte_devargs.h>
14 #include <rte_memcpy.h>
17 #include "eal_filesystem.h"
24 * PCI probing under linux
26 * This code is used to simulate a PCI probe by parsing information in sysfs.
27 * When a registered device matches a driver, it is then initialized with
28 * IGB_UIO driver (or doesn't initialize, if the device wasn't bound to it).
31 extern struct rte_pci_bus rte_pci_bus;
34 pci_get_kernel_driver_by_path(const char *filename, char *dri_name,
41 if (!filename || !dri_name)
44 count = readlink(filename, path, PATH_MAX);
45 if (count >= PATH_MAX)
48 /* For device does not have a driver */
54 name = strrchr(path, '/');
56 strlcpy(dri_name, name + 1, len);
65 rte_pci_map_device(struct rte_pci_device *dev)
69 /* try mapping the NIC resources using VFIO if it exists */
71 case RTE_PCI_KDRV_VFIO:
73 if (pci_vfio_is_enabled())
74 ret = pci_vfio_map_resource(dev);
77 case RTE_PCI_KDRV_IGB_UIO:
78 case RTE_PCI_KDRV_UIO_GENERIC:
79 if (rte_eal_using_phys_addrs()) {
80 /* map resources for devices that use uio */
81 ret = pci_uio_map_resource(dev);
86 " Not managed by a supported kernel driver, skipped\n");
94 /* Unmap pci device */
96 rte_pci_unmap_device(struct rte_pci_device *dev)
98 /* try unmapping the NIC resources using VFIO if it exists */
100 case RTE_PCI_KDRV_VFIO:
102 if (pci_vfio_is_enabled())
103 pci_vfio_unmap_resource(dev);
106 case RTE_PCI_KDRV_IGB_UIO:
107 case RTE_PCI_KDRV_UIO_GENERIC:
108 /* unmap resources for devices that use uio */
109 pci_uio_unmap_resource(dev);
113 " Not managed by a supported kernel driver, skipped\n");
119 find_max_end_va(const struct rte_memseg_list *msl, void *arg)
121 size_t sz = msl->len;
122 void *end_va = RTE_PTR_ADD(msl->base_va, sz);
125 if (*max_va < end_va)
131 pci_find_max_end_va(void)
135 rte_memseg_list_walk(find_max_end_va, &va);
140 /* parse one line of the "resource" sysfs file (note that the 'line'
141 * string is modified)
144 pci_parse_one_sysfs_resource(char *line, size_t len, uint64_t *phys_addr,
145 uint64_t *end_addr, uint64_t *flags)
147 union pci_resource_info {
153 char *ptrs[PCI_RESOURCE_FMT_NVAL];
156 if (rte_strsplit(line, len, res_info.ptrs, 3, ' ') != 3) {
158 "%s(): bad resource format\n", __func__);
162 *phys_addr = strtoull(res_info.phys_addr, NULL, 16);
163 *end_addr = strtoull(res_info.end_addr, NULL, 16);
164 *flags = strtoull(res_info.flags, NULL, 16);
167 "%s(): bad resource format\n", __func__);
174 /* parse the "resource" sysfs file */
176 pci_parse_sysfs_resource(const char *filename, struct rte_pci_device *dev)
181 uint64_t phys_addr, end_addr, flags;
183 f = fopen(filename, "r");
185 RTE_LOG(ERR, EAL, "Cannot open sysfs resource\n");
189 for (i = 0; i<PCI_MAX_RESOURCE; i++) {
191 if (fgets(buf, sizeof(buf), f) == NULL) {
193 "%s(): cannot read resource\n", __func__);
196 if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
197 &end_addr, &flags) < 0)
200 if (flags & IORESOURCE_MEM) {
201 dev->mem_resource[i].phys_addr = phys_addr;
202 dev->mem_resource[i].len = end_addr - phys_addr + 1;
203 /* not mapped for now */
204 dev->mem_resource[i].addr = NULL;
215 /* Scan one pci sysfs entry, and fill the devices list from it. */
217 pci_scan_one(const char *dirname, const struct rte_pci_addr *addr)
219 char filename[PATH_MAX];
221 struct rte_pci_device *dev;
222 char driver[PATH_MAX];
225 dev = malloc(sizeof(*dev));
229 memset(dev, 0, sizeof(*dev));
230 dev->device.bus = &rte_pci_bus.bus;
234 snprintf(filename, sizeof(filename), "%s/vendor", dirname);
235 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
239 dev->id.vendor_id = (uint16_t)tmp;
242 snprintf(filename, sizeof(filename), "%s/device", dirname);
243 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
247 dev->id.device_id = (uint16_t)tmp;
249 /* get subsystem_vendor id */
250 snprintf(filename, sizeof(filename), "%s/subsystem_vendor",
252 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
256 dev->id.subsystem_vendor_id = (uint16_t)tmp;
258 /* get subsystem_device id */
259 snprintf(filename, sizeof(filename), "%s/subsystem_device",
261 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
265 dev->id.subsystem_device_id = (uint16_t)tmp;
268 snprintf(filename, sizeof(filename), "%s/class",
270 if (eal_parse_sysfs_value(filename, &tmp) < 0) {
274 /* the least 24 bits are valid: class, subclass, program interface */
275 dev->id.class_id = (uint32_t)tmp & RTE_CLASS_ANY_ID;
279 snprintf(filename, sizeof(filename), "%s/max_vfs", dirname);
280 if (!access(filename, F_OK) &&
281 eal_parse_sysfs_value(filename, &tmp) == 0)
282 dev->max_vfs = (uint16_t)tmp;
284 /* for non igb_uio driver, need kernel version >= 3.8 */
285 snprintf(filename, sizeof(filename),
286 "%s/sriov_numvfs", dirname);
287 if (!access(filename, F_OK) &&
288 eal_parse_sysfs_value(filename, &tmp) == 0)
289 dev->max_vfs = (uint16_t)tmp;
292 /* get numa node, default to 0 if not present */
293 snprintf(filename, sizeof(filename), "%s/numa_node",
296 if (access(filename, F_OK) != -1) {
297 if (eal_parse_sysfs_value(filename, &tmp) == 0)
298 dev->device.numa_node = tmp;
300 dev->device.numa_node = -1;
302 dev->device.numa_node = 0;
307 /* parse resources */
308 snprintf(filename, sizeof(filename), "%s/resource", dirname);
309 if (pci_parse_sysfs_resource(filename, dev) < 0) {
310 RTE_LOG(ERR, EAL, "%s(): cannot parse resource\n", __func__);
316 snprintf(filename, sizeof(filename), "%s/driver", dirname);
317 ret = pci_get_kernel_driver_by_path(filename, driver, sizeof(driver));
319 RTE_LOG(ERR, EAL, "Fail to get kernel driver\n");
325 if (!strcmp(driver, "vfio-pci"))
326 dev->kdrv = RTE_PCI_KDRV_VFIO;
327 else if (!strcmp(driver, "igb_uio"))
328 dev->kdrv = RTE_PCI_KDRV_IGB_UIO;
329 else if (!strcmp(driver, "uio_pci_generic"))
330 dev->kdrv = RTE_PCI_KDRV_UIO_GENERIC;
332 dev->kdrv = RTE_PCI_KDRV_UNKNOWN;
334 dev->kdrv = RTE_PCI_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",
381 else if (dev2->device.devargs !=
382 dev->device.devargs) {
383 rte_devargs_remove(dev2->device.devargs);
392 rte_pci_add_device(dev);
399 pci_update_device(const struct rte_pci_addr *addr)
401 char filename[PATH_MAX];
403 snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
404 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
407 return pci_scan_one(filename, addr);
411 * split up a pci address into its constituent parts.
414 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
416 /* first split on ':' */
424 char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
427 char *buf_copy = strndup(buf, bufsize);
428 if (buf_copy == NULL)
431 if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
434 /* final split is on '.' between devid and function */
435 splitaddr.function = strchr(splitaddr.devid,'.');
436 if (splitaddr.function == NULL)
438 *splitaddr.function++ = '\0';
440 /* now convert to int values */
442 addr->domain = strtoul(splitaddr.domain, NULL, 16);
443 addr->bus = strtoul(splitaddr.bus, NULL, 16);
444 addr->devid = strtoul(splitaddr.devid, NULL, 16);
445 addr->function = strtoul(splitaddr.function, NULL, 10);
449 free(buf_copy); /* free the copy made with strdup */
457 * Scan the content of the PCI bus, and the devices in the devices
465 char dirname[PATH_MAX];
466 struct rte_pci_addr addr;
468 /* for debug purposes, PCI can be disabled */
469 if (!rte_eal_has_pci())
473 if (!pci_vfio_is_enabled())
474 RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
477 dir = opendir(rte_pci_get_sysfs_path());
479 RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
480 __func__, strerror(errno));
484 while ((e = readdir(dir)) != NULL) {
485 if (e->d_name[0] == '.')
488 if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
491 if (rte_pci_ignore_device(&addr))
494 snprintf(dirname, sizeof(dirname), "%s/%s",
495 rte_pci_get_sysfs_path(), e->d_name);
497 if (pci_scan_one(dirname, &addr) < 0)
508 #if defined(RTE_ARCH_X86)
510 pci_device_iommu_support_va(const struct rte_pci_device *dev)
512 #define VTD_CAP_MGAW_SHIFT 16
513 #define VTD_CAP_MGAW_MASK (0x3fULL << VTD_CAP_MGAW_SHIFT)
514 const struct rte_pci_addr *addr = &dev->addr;
515 char filename[PATH_MAX];
517 uint64_t mgaw, vtd_cap_reg = 0;
519 snprintf(filename, sizeof(filename),
520 "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
521 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
524 fp = fopen(filename, "r");
526 /* We don't have an Intel IOMMU, assume VA supported */
530 RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
531 __func__, filename, strerror(errno));
535 /* We have an Intel IOMMU */
536 if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
537 RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
544 mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
547 * Assuming there is no limitation by now. We can not know at this point
548 * because the memory has not been initialized yet. Setting the dma mask
549 * will force a check once memory initialization is done. We can not do
550 * a fallback to IOVA PA now, but if the dma check fails, the error
551 * message should advice for using '--iova-mode pa' if IOVA VA is the
554 rte_mem_set_dma_mask(mgaw);
557 #elif defined(RTE_ARCH_PPC_64)
559 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
562 * IOMMU is always present on a PowerNV host (IOMMUv2).
563 * IOMMU is also present in a KVM/QEMU VM (IOMMUv1) but is not
564 * currently supported by DPDK. Test for our current environment
565 * and report VA support as appropriate.
570 char filename[PATH_MAX] = "/proc/cpuinfo";
571 FILE *fp = fopen(filename, "r");
575 RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
576 __func__, filename, strerror(errno));
580 /* Check for a PowerNV platform */
581 while (getline(&line, &len, fp) != -1) {
582 if (strstr(line, "platform") != NULL)
585 if (strstr(line, "PowerNV") != NULL) {
586 RTE_LOG(DEBUG, EAL, "Running on a PowerNV system\n");
598 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
605 pci_device_iova_mode(const struct rte_pci_driver *pdrv,
606 const struct rte_pci_device *pdev)
608 enum rte_iova_mode iova_mode = RTE_IOVA_DC;
610 switch (pdev->kdrv) {
611 case RTE_PCI_KDRV_VFIO: {
613 static int is_vfio_noiommu_enabled = -1;
615 if (is_vfio_noiommu_enabled == -1) {
616 if (rte_vfio_noiommu_is_enabled() == 1)
617 is_vfio_noiommu_enabled = 1;
619 is_vfio_noiommu_enabled = 0;
621 if (is_vfio_noiommu_enabled != 0)
622 iova_mode = RTE_IOVA_PA;
623 else if ((pdrv->drv_flags & RTE_PCI_DRV_NEED_IOVA_AS_VA) != 0)
624 iova_mode = RTE_IOVA_VA;
629 case RTE_PCI_KDRV_IGB_UIO:
630 case RTE_PCI_KDRV_UIO_GENERIC:
631 iova_mode = RTE_IOVA_PA;
635 if ((pdrv->drv_flags & RTE_PCI_DRV_NEED_IOVA_AS_VA) != 0)
636 iova_mode = RTE_IOVA_VA;
642 /* Read PCI config space. */
643 int rte_pci_read_config(const struct rte_pci_device *device,
644 void *buf, size_t len, off_t offset)
646 char devname[RTE_DEV_NAME_MAX_LEN] = "";
647 const struct rte_intr_handle *intr_handle = &device->intr_handle;
649 switch (device->kdrv) {
650 case RTE_PCI_KDRV_IGB_UIO:
651 case RTE_PCI_KDRV_UIO_GENERIC:
652 return pci_uio_read_config(intr_handle, buf, len, offset);
654 case RTE_PCI_KDRV_VFIO:
655 return pci_vfio_read_config(intr_handle, buf, len, offset);
658 rte_pci_device_name(&device->addr, devname,
659 RTE_DEV_NAME_MAX_LEN);
661 "Unknown driver type for %s\n", devname);
666 /* Write PCI config space. */
667 int rte_pci_write_config(const struct rte_pci_device *device,
668 const void *buf, size_t len, off_t offset)
670 char devname[RTE_DEV_NAME_MAX_LEN] = "";
671 const struct rte_intr_handle *intr_handle = &device->intr_handle;
673 switch (device->kdrv) {
674 case RTE_PCI_KDRV_IGB_UIO:
675 case RTE_PCI_KDRV_UIO_GENERIC:
676 return pci_uio_write_config(intr_handle, buf, len, offset);
678 case RTE_PCI_KDRV_VFIO:
679 return pci_vfio_write_config(intr_handle, buf, len, offset);
682 rte_pci_device_name(&device->addr, devname,
683 RTE_DEV_NAME_MAX_LEN);
685 "Unknown driver type for %s\n", devname);
690 #if defined(RTE_ARCH_X86)
692 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
693 struct rte_pci_ioport *p)
702 if (rte_eal_iopl_init() != 0) {
703 RTE_LOG(ERR, EAL, "%s(): insufficient ioport permissions for PCI device %s\n",
704 __func__, dev->name);
708 snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
709 dev->addr.domain, dev->addr.bus,
710 dev->addr.devid, dev->addr.function);
712 fp = fopen("/proc/ioports", "r");
714 RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
718 while (getdelim(&line, &linesz, '\n', fp) > 0) {
723 n = strcspn(ptr, ":");
727 while (*left && isspace(*left))
730 if (!strncmp(left, pci_id, strlen(pci_id))) {
733 while (*ptr && isspace(*ptr))
736 sscanf(ptr, "%04hx-%04hx", &start, &end);
749 RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
756 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
757 struct rte_pci_ioport *p)
763 case RTE_PCI_KDRV_VFIO:
764 if (pci_vfio_is_enabled())
765 ret = pci_vfio_ioport_map(dev, bar, p);
768 case RTE_PCI_KDRV_IGB_UIO:
769 ret = pci_uio_ioport_map(dev, bar, p);
771 case RTE_PCI_KDRV_UIO_GENERIC:
772 #if defined(RTE_ARCH_X86)
773 ret = pci_ioport_map(dev, bar, p);
775 ret = pci_uio_ioport_map(dev, bar, p);
789 rte_pci_ioport_read(struct rte_pci_ioport *p,
790 void *data, size_t len, off_t offset)
792 switch (p->dev->kdrv) {
794 case RTE_PCI_KDRV_VFIO:
795 pci_vfio_ioport_read(p, data, len, offset);
798 case RTE_PCI_KDRV_IGB_UIO:
799 pci_uio_ioport_read(p, data, len, offset);
801 case RTE_PCI_KDRV_UIO_GENERIC:
802 pci_uio_ioport_read(p, data, len, offset);
810 rte_pci_ioport_write(struct rte_pci_ioport *p,
811 const void *data, size_t len, off_t offset)
813 switch (p->dev->kdrv) {
815 case RTE_PCI_KDRV_VFIO:
816 pci_vfio_ioport_write(p, data, len, offset);
819 case RTE_PCI_KDRV_IGB_UIO:
820 pci_uio_ioport_write(p, data, len, offset);
822 case RTE_PCI_KDRV_UIO_GENERIC:
823 pci_uio_ioport_write(p, data, len, offset);
831 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
835 switch (p->dev->kdrv) {
837 case RTE_PCI_KDRV_VFIO:
838 if (pci_vfio_is_enabled())
839 ret = pci_vfio_ioport_unmap(p);
842 case RTE_PCI_KDRV_IGB_UIO:
843 ret = pci_uio_ioport_unmap(p);
845 case RTE_PCI_KDRV_UIO_GENERIC:
846 #if defined(RTE_ARCH_X86)
849 ret = pci_uio_ioport_unmap(p);