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;
332 else if (!strcmp(driver, "mlx4_core") ||
333 !strcmp(driver, "mlx5_core"))
334 dev->kdrv = RTE_KDRV_NIC_MLX;
336 dev->kdrv = RTE_KDRV_UNKNOWN;
338 dev->kdrv = RTE_KDRV_NONE;
340 /* device is valid, add in list (sorted) */
341 if (TAILQ_EMPTY(&rte_pci_bus.device_list)) {
342 rte_pci_add_device(dev);
344 struct rte_pci_device *dev2;
347 TAILQ_FOREACH(dev2, &rte_pci_bus.device_list, next) {
348 ret = rte_pci_addr_cmp(&dev->addr, &dev2->addr);
353 rte_pci_insert_device(dev2, dev);
354 } else { /* already registered */
355 if (!rte_dev_is_probed(&dev2->device)) {
356 dev2->kdrv = dev->kdrv;
357 dev2->max_vfs = dev->max_vfs;
359 memmove(dev2->mem_resource,
361 sizeof(dev->mem_resource));
364 * If device is plugged and driver is
365 * probed already, (This happens when
366 * we call rte_dev_probe which will
367 * scan all device on the bus) we don't
368 * need to do anything here unless...
370 if (dev2->kdrv != dev->kdrv ||
371 dev2->max_vfs != dev->max_vfs)
373 * This should not happens.
374 * But it is still possible if
375 * we unbind a device from
376 * vfio or uio before hotplug
377 * remove and rebind it with
378 * a different configure.
379 * So we just print out the
382 RTE_LOG(ERR, EAL, "Unexpected device scan at %s!\n",
390 rte_pci_add_device(dev);
397 pci_update_device(const struct rte_pci_addr *addr)
399 char filename[PATH_MAX];
401 snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
402 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
405 return pci_scan_one(filename, addr);
409 * split up a pci address into its constituent parts.
412 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
414 /* first split on ':' */
422 char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
425 char *buf_copy = strndup(buf, bufsize);
426 if (buf_copy == NULL)
429 if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
432 /* final split is on '.' between devid and function */
433 splitaddr.function = strchr(splitaddr.devid,'.');
434 if (splitaddr.function == NULL)
436 *splitaddr.function++ = '\0';
438 /* now convert to int values */
440 addr->domain = strtoul(splitaddr.domain, NULL, 16);
441 addr->bus = strtoul(splitaddr.bus, NULL, 16);
442 addr->devid = strtoul(splitaddr.devid, NULL, 16);
443 addr->function = strtoul(splitaddr.function, NULL, 10);
447 free(buf_copy); /* free the copy made with strdup */
455 * Scan the content of the PCI bus, and the devices in the devices
463 char dirname[PATH_MAX];
464 struct rte_pci_addr addr;
466 /* for debug purposes, PCI can be disabled */
467 if (!rte_eal_has_pci())
471 if (!pci_vfio_is_enabled())
472 RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
475 dir = opendir(rte_pci_get_sysfs_path());
477 RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
478 __func__, strerror(errno));
482 while ((e = readdir(dir)) != NULL) {
483 if (e->d_name[0] == '.')
486 if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
489 snprintf(dirname, sizeof(dirname), "%s/%s",
490 rte_pci_get_sysfs_path(), e->d_name);
492 if (pci_scan_one(dirname, &addr) < 0)
503 #if defined(RTE_ARCH_X86)
505 pci_one_device_iommu_support_va(const struct rte_pci_device *dev)
507 #define VTD_CAP_MGAW_SHIFT 16
508 #define VTD_CAP_MGAW_MASK (0x3fULL << VTD_CAP_MGAW_SHIFT)
509 #define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
510 const struct rte_pci_addr *addr = &dev->addr;
511 char filename[PATH_MAX];
513 uint64_t mgaw, vtd_cap_reg = 0;
515 snprintf(filename, sizeof(filename),
516 "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
517 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
519 if (access(filename, F_OK) == -1) {
520 /* We don't have an Intel IOMMU, assume VA supported*/
524 /* We have an intel IOMMU */
525 fp = fopen(filename, "r");
527 RTE_LOG(ERR, EAL, "%s(): can't open %s\n", __func__, filename);
531 if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
532 RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
539 mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
542 * Assuming there is no limitation by now. We can not know at this point
543 * because the memory has not been initialized yet. Setting the dma mask
544 * will force a check once memory initialization is done. We can not do
545 * a fallback to IOVA PA now, but if the dma check fails, the error
546 * message should advice for using '--iova-mode pa' if IOVA VA is the
549 rte_mem_set_dma_mask(mgaw);
552 #elif defined(RTE_ARCH_PPC_64)
554 pci_one_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
560 pci_one_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
567 pci_device_iova_mode(const struct rte_pci_driver *pdrv,
568 const struct rte_pci_device *pdev)
570 enum rte_iova_mode iova_mode = RTE_IOVA_DC;
571 static int iommu_no_va = -1;
573 switch (pdev->kdrv) {
574 case RTE_KDRV_VFIO: {
576 static int is_vfio_noiommu_enabled = -1;
578 if (is_vfio_noiommu_enabled == -1) {
579 if (rte_vfio_noiommu_is_enabled() == 1)
580 is_vfio_noiommu_enabled = 1;
582 is_vfio_noiommu_enabled = 0;
584 if ((pdrv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) == 0) {
585 iova_mode = RTE_IOVA_PA;
586 } else if (is_vfio_noiommu_enabled != 0) {
587 RTE_LOG(DEBUG, EAL, "Forcing to 'PA', vfio-noiommu mode configured\n");
588 iova_mode = RTE_IOVA_PA;
594 case RTE_KDRV_NIC_MLX:
595 if ((pdrv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) == 0)
596 iova_mode = RTE_IOVA_PA;
599 case RTE_KDRV_IGB_UIO:
600 case RTE_KDRV_UIO_GENERIC:
601 iova_mode = RTE_IOVA_PA;
605 RTE_LOG(DEBUG, EAL, "Unsupported kernel driver? Defaulting to IOVA as 'PA'\n");
606 iova_mode = RTE_IOVA_PA;
610 if (iova_mode != RTE_IOVA_PA) {
612 * We can check this only once, because the IOMMU hardware is
613 * the same for all of them.
615 if (iommu_no_va == -1)
616 iommu_no_va = pci_one_device_iommu_support_va(pdev)
618 if (iommu_no_va != 0) {
619 RTE_LOG(DEBUG, EAL, "Forcing to 'PA', IOMMU does not support IOVA as 'VA'\n");
620 iova_mode = RTE_IOVA_PA;
626 /* Read PCI config space. */
627 int rte_pci_read_config(const struct rte_pci_device *device,
628 void *buf, size_t len, off_t offset)
630 char devname[RTE_DEV_NAME_MAX_LEN] = "";
631 const struct rte_intr_handle *intr_handle = &device->intr_handle;
633 switch (device->kdrv) {
634 case RTE_KDRV_IGB_UIO:
635 case RTE_KDRV_UIO_GENERIC:
636 return pci_uio_read_config(intr_handle, buf, len, offset);
639 return pci_vfio_read_config(intr_handle, buf, len, offset);
642 rte_pci_device_name(&device->addr, devname,
643 RTE_DEV_NAME_MAX_LEN);
645 "Unknown driver type for %s\n", devname);
650 /* Write PCI config space. */
651 int rte_pci_write_config(const struct rte_pci_device *device,
652 const void *buf, size_t len, off_t offset)
654 char devname[RTE_DEV_NAME_MAX_LEN] = "";
655 const struct rte_intr_handle *intr_handle = &device->intr_handle;
657 switch (device->kdrv) {
658 case RTE_KDRV_IGB_UIO:
659 case RTE_KDRV_UIO_GENERIC:
660 return pci_uio_write_config(intr_handle, buf, len, offset);
663 return pci_vfio_write_config(intr_handle, buf, len, offset);
666 rte_pci_device_name(&device->addr, devname,
667 RTE_DEV_NAME_MAX_LEN);
669 "Unknown driver type for %s\n", devname);
674 #if defined(RTE_ARCH_X86)
676 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
677 struct rte_pci_ioport *p)
686 snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
687 dev->addr.domain, dev->addr.bus,
688 dev->addr.devid, dev->addr.function);
690 fp = fopen("/proc/ioports", "r");
692 RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
696 while (getdelim(&line, &linesz, '\n', fp) > 0) {
701 n = strcspn(ptr, ":");
705 while (*left && isspace(*left))
708 if (!strncmp(left, pci_id, strlen(pci_id))) {
711 while (*ptr && isspace(*ptr))
714 sscanf(ptr, "%04hx-%04hx", &start, &end);
727 RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
734 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
735 struct rte_pci_ioport *p)
742 if (pci_vfio_is_enabled())
743 ret = pci_vfio_ioport_map(dev, bar, p);
746 case RTE_KDRV_IGB_UIO:
747 ret = pci_uio_ioport_map(dev, bar, p);
749 case RTE_KDRV_UIO_GENERIC:
750 #if defined(RTE_ARCH_X86)
751 ret = pci_ioport_map(dev, bar, p);
753 ret = pci_uio_ioport_map(dev, bar, p);
757 #if defined(RTE_ARCH_X86)
758 ret = pci_ioport_map(dev, bar, p);
772 rte_pci_ioport_read(struct rte_pci_ioport *p,
773 void *data, size_t len, off_t offset)
775 switch (p->dev->kdrv) {
778 pci_vfio_ioport_read(p, data, len, offset);
781 case RTE_KDRV_IGB_UIO:
782 pci_uio_ioport_read(p, data, len, offset);
784 case RTE_KDRV_UIO_GENERIC:
785 pci_uio_ioport_read(p, data, len, offset);
788 #if defined(RTE_ARCH_X86)
789 pci_uio_ioport_read(p, data, len, offset);
798 rte_pci_ioport_write(struct rte_pci_ioport *p,
799 const void *data, size_t len, off_t offset)
801 switch (p->dev->kdrv) {
804 pci_vfio_ioport_write(p, data, len, offset);
807 case RTE_KDRV_IGB_UIO:
808 pci_uio_ioport_write(p, data, len, offset);
810 case RTE_KDRV_UIO_GENERIC:
811 pci_uio_ioport_write(p, data, len, offset);
814 #if defined(RTE_ARCH_X86)
815 pci_uio_ioport_write(p, data, len, offset);
824 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
828 switch (p->dev->kdrv) {
831 if (pci_vfio_is_enabled())
832 ret = pci_vfio_ioport_unmap(p);
835 case RTE_KDRV_IGB_UIO:
836 ret = pci_uio_ioport_unmap(p);
838 case RTE_KDRV_UIO_GENERIC:
839 #if defined(RTE_ARCH_X86)
842 ret = pci_uio_ioport_unmap(p);
846 #if defined(RTE_ARCH_X86)