xref: /dpdk/drivers/bus/pci/linux/pci.c (revision 8809f78c7dd9f33a44a4f89c58fc91ded34296ed)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #include <string.h>
6 #include <dirent.h>
7 
8 #include <rte_log.h>
9 #include <rte_bus.h>
10 #include <rte_pci.h>
11 #include <rte_bus_pci.h>
12 #include <rte_malloc.h>
13 #include <rte_devargs.h>
14 #include <rte_memcpy.h>
15 #include <rte_vfio.h>
16 
17 #include "eal_filesystem.h"
18 
19 #include "private.h"
20 #include "pci_init.h"
21 
22 /**
23  * @file
24  * PCI probing under linux
25  *
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).
29  */
30 
31 extern struct rte_pci_bus rte_pci_bus;
32 
33 static int
34 pci_get_kernel_driver_by_path(const char *filename, char *dri_name,
35 			      size_t len)
36 {
37 	int count;
38 	char path[PATH_MAX];
39 	char *name;
40 
41 	if (!filename || !dri_name)
42 		return -1;
43 
44 	count = readlink(filename, path, PATH_MAX);
45 	if (count >= PATH_MAX)
46 		return -1;
47 
48 	/* For device does not have a driver */
49 	if (count < 0)
50 		return 1;
51 
52 	path[count] = '\0';
53 
54 	name = strrchr(path, '/');
55 	if (name) {
56 		strlcpy(dri_name, name + 1, len);
57 		return 0;
58 	}
59 
60 	return -1;
61 }
62 
63 /* Map pci device */
64 int
65 rte_pci_map_device(struct rte_pci_device *dev)
66 {
67 	int ret = -1;
68 
69 	/* try mapping the NIC resources using VFIO if it exists */
70 	switch (dev->kdrv) {
71 	case RTE_PCI_KDRV_VFIO:
72 #ifdef VFIO_PRESENT
73 		if (pci_vfio_is_enabled())
74 			ret = pci_vfio_map_resource(dev);
75 #endif
76 		break;
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);
82 		}
83 		break;
84 	default:
85 		RTE_LOG(DEBUG, EAL,
86 			"  Not managed by a supported kernel driver, skipped\n");
87 		ret = 1;
88 		break;
89 	}
90 
91 	return ret;
92 }
93 
94 /* Unmap pci device */
95 void
96 rte_pci_unmap_device(struct rte_pci_device *dev)
97 {
98 	/* try unmapping the NIC resources using VFIO if it exists */
99 	switch (dev->kdrv) {
100 	case RTE_PCI_KDRV_VFIO:
101 #ifdef VFIO_PRESENT
102 		if (pci_vfio_is_enabled())
103 			pci_vfio_unmap_resource(dev);
104 #endif
105 		break;
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);
110 		break;
111 	default:
112 		RTE_LOG(DEBUG, EAL,
113 			"  Not managed by a supported kernel driver, skipped\n");
114 		break;
115 	}
116 }
117 
118 static int
119 find_max_end_va(const struct rte_memseg_list *msl, void *arg)
120 {
121 	size_t sz = msl->len;
122 	void *end_va = RTE_PTR_ADD(msl->base_va, sz);
123 	void **max_va = arg;
124 
125 	if (*max_va < end_va)
126 		*max_va = end_va;
127 	return 0;
128 }
129 
130 void *
131 pci_find_max_end_va(void)
132 {
133 	void *va = NULL;
134 
135 	rte_memseg_list_walk(find_max_end_va, &va);
136 	return va;
137 }
138 
139 
140 /* parse one line of the "resource" sysfs file (note that the 'line'
141  * string is modified)
142  */
143 int
144 pci_parse_one_sysfs_resource(char *line, size_t len, uint64_t *phys_addr,
145 	uint64_t *end_addr, uint64_t *flags)
146 {
147 	union pci_resource_info {
148 		struct {
149 			char *phys_addr;
150 			char *end_addr;
151 			char *flags;
152 		};
153 		char *ptrs[PCI_RESOURCE_FMT_NVAL];
154 	} res_info;
155 
156 	if (rte_strsplit(line, len, res_info.ptrs, 3, ' ') != 3) {
157 		RTE_LOG(ERR, EAL,
158 			"%s(): bad resource format\n", __func__);
159 		return -1;
160 	}
161 	errno = 0;
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);
165 	if (errno != 0) {
166 		RTE_LOG(ERR, EAL,
167 			"%s(): bad resource format\n", __func__);
168 		return -1;
169 	}
170 
171 	return 0;
172 }
173 
174 /* parse the "resource" sysfs file */
175 static int
176 pci_parse_sysfs_resource(const char *filename, struct rte_pci_device *dev)
177 {
178 	FILE *f;
179 	char buf[BUFSIZ];
180 	int i;
181 	uint64_t phys_addr, end_addr, flags;
182 
183 	f = fopen(filename, "r");
184 	if (f == NULL) {
185 		RTE_LOG(ERR, EAL, "Cannot open sysfs resource\n");
186 		return -1;
187 	}
188 
189 	for (i = 0; i<PCI_MAX_RESOURCE; i++) {
190 
191 		if (fgets(buf, sizeof(buf), f) == NULL) {
192 			RTE_LOG(ERR, EAL,
193 				"%s(): cannot read resource\n", __func__);
194 			goto error;
195 		}
196 		if (pci_parse_one_sysfs_resource(buf, sizeof(buf), &phys_addr,
197 				&end_addr, &flags) < 0)
198 			goto error;
199 
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;
205 		}
206 	}
207 	fclose(f);
208 	return 0;
209 
210 error:
211 	fclose(f);
212 	return -1;
213 }
214 
215 /* Scan one pci sysfs entry, and fill the devices list from it. */
216 static int
217 pci_scan_one(const char *dirname, const struct rte_pci_addr *addr)
218 {
219 	char filename[PATH_MAX];
220 	unsigned long tmp;
221 	struct rte_pci_device *dev;
222 	char driver[PATH_MAX];
223 	int ret;
224 
225 	dev = malloc(sizeof(*dev));
226 	if (dev == NULL)
227 		return -1;
228 
229 	memset(dev, 0, sizeof(*dev));
230 	dev->device.bus = &rte_pci_bus.bus;
231 	dev->addr = *addr;
232 
233 	/* get vendor id */
234 	snprintf(filename, sizeof(filename), "%s/vendor", dirname);
235 	if (eal_parse_sysfs_value(filename, &tmp) < 0) {
236 		free(dev);
237 		return -1;
238 	}
239 	dev->id.vendor_id = (uint16_t)tmp;
240 
241 	/* get device id */
242 	snprintf(filename, sizeof(filename), "%s/device", dirname);
243 	if (eal_parse_sysfs_value(filename, &tmp) < 0) {
244 		free(dev);
245 		return -1;
246 	}
247 	dev->id.device_id = (uint16_t)tmp;
248 
249 	/* get subsystem_vendor id */
250 	snprintf(filename, sizeof(filename), "%s/subsystem_vendor",
251 		 dirname);
252 	if (eal_parse_sysfs_value(filename, &tmp) < 0) {
253 		free(dev);
254 		return -1;
255 	}
256 	dev->id.subsystem_vendor_id = (uint16_t)tmp;
257 
258 	/* get subsystem_device id */
259 	snprintf(filename, sizeof(filename), "%s/subsystem_device",
260 		 dirname);
261 	if (eal_parse_sysfs_value(filename, &tmp) < 0) {
262 		free(dev);
263 		return -1;
264 	}
265 	dev->id.subsystem_device_id = (uint16_t)tmp;
266 
267 	/* get class_id */
268 	snprintf(filename, sizeof(filename), "%s/class",
269 		 dirname);
270 	if (eal_parse_sysfs_value(filename, &tmp) < 0) {
271 		free(dev);
272 		return -1;
273 	}
274 	/* the least 24 bits are valid: class, subclass, program interface */
275 	dev->id.class_id = (uint32_t)tmp & RTE_CLASS_ANY_ID;
276 
277 	/* get max_vfs */
278 	dev->max_vfs = 0;
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;
283 	else {
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;
290 	}
291 
292 	/* get numa node, default to 0 if not present */
293 	snprintf(filename, sizeof(filename), "%s/numa_node",
294 		 dirname);
295 
296 	if (access(filename, F_OK) != -1) {
297 		if (eal_parse_sysfs_value(filename, &tmp) == 0)
298 			dev->device.numa_node = tmp;
299 		else
300 			dev->device.numa_node = -1;
301 	} else {
302 		dev->device.numa_node = 0;
303 	}
304 
305 	pci_name_set(dev);
306 
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__);
311 		free(dev);
312 		return -1;
313 	}
314 
315 	/* parse driver */
316 	snprintf(filename, sizeof(filename), "%s/driver", dirname);
317 	ret = pci_get_kernel_driver_by_path(filename, driver, sizeof(driver));
318 	if (ret < 0) {
319 		RTE_LOG(ERR, EAL, "Fail to get kernel driver\n");
320 		free(dev);
321 		return -1;
322 	}
323 
324 	if (!ret) {
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;
331 		else
332 			dev->kdrv = RTE_PCI_KDRV_UNKNOWN;
333 	} else {
334 		dev->kdrv = RTE_PCI_KDRV_NONE;
335 		return 0;
336 	}
337 	/* device is valid, add in list (sorted) */
338 	if (TAILQ_EMPTY(&rte_pci_bus.device_list)) {
339 		rte_pci_add_device(dev);
340 	} else {
341 		struct rte_pci_device *dev2;
342 		int ret;
343 
344 		TAILQ_FOREACH(dev2, &rte_pci_bus.device_list, next) {
345 			ret = rte_pci_addr_cmp(&dev->addr, &dev2->addr);
346 			if (ret > 0)
347 				continue;
348 
349 			if (ret < 0) {
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;
355 					dev2->id = dev->id;
356 					pci_name_set(dev2);
357 					memmove(dev2->mem_resource,
358 						dev->mem_resource,
359 						sizeof(dev->mem_resource));
360 				} else {
361 					/**
362 					 * If device is plugged and driver is
363 					 * probed already, (This happens when
364 					 * we call rte_dev_probe which will
365 					 * scan all device on the bus) we don't
366 					 * need to do anything here unless...
367 					 **/
368 					if (dev2->kdrv != dev->kdrv ||
369 						dev2->max_vfs != dev->max_vfs ||
370 						memcmp(&dev2->id, &dev->id, sizeof(dev2->id)))
371 						/*
372 						 * This should not happens.
373 						 * But it is still possible if
374 						 * we unbind a device from
375 						 * vfio or uio before hotplug
376 						 * remove and rebind it with
377 						 * a different configure.
378 						 * So we just print out the
379 						 * error as an alarm.
380 						 */
381 						RTE_LOG(ERR, EAL, "Unexpected device scan at %s!\n",
382 							filename);
383 					else if (dev2->device.devargs !=
384 						 dev->device.devargs) {
385 						rte_devargs_remove(dev2->device.devargs);
386 						pci_name_set(dev2);
387 					}
388 				}
389 				free(dev);
390 			}
391 			return 0;
392 		}
393 
394 		rte_pci_add_device(dev);
395 	}
396 
397 	return 0;
398 }
399 
400 /*
401  * split up a pci address into its constituent parts.
402  */
403 static int
404 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
405 {
406 	/* first split on ':' */
407 	union splitaddr {
408 		struct {
409 			char *domain;
410 			char *bus;
411 			char *devid;
412 			char *function;
413 		};
414 		char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
415 	} splitaddr;
416 
417 	char *buf_copy = strndup(buf, bufsize);
418 	if (buf_copy == NULL)
419 		return -1;
420 
421 	if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
422 			!= PCI_FMT_NVAL - 1)
423 		goto error;
424 	/* final split is on '.' between devid and function */
425 	splitaddr.function = strchr(splitaddr.devid,'.');
426 	if (splitaddr.function == NULL)
427 		goto error;
428 	*splitaddr.function++ = '\0';
429 
430 	/* now convert to int values */
431 	errno = 0;
432 	addr->domain = strtoul(splitaddr.domain, NULL, 16);
433 	addr->bus = strtoul(splitaddr.bus, NULL, 16);
434 	addr->devid = strtoul(splitaddr.devid, NULL, 16);
435 	addr->function = strtoul(splitaddr.function, NULL, 10);
436 	if (errno != 0)
437 		goto error;
438 
439 	free(buf_copy); /* free the copy made with strdup */
440 	return 0;
441 error:
442 	free(buf_copy);
443 	return -1;
444 }
445 
446 /*
447  * Scan the content of the PCI bus, and the devices in the devices
448  * list
449  */
450 int
451 rte_pci_scan(void)
452 {
453 	struct dirent *e;
454 	DIR *dir;
455 	char dirname[PATH_MAX];
456 	struct rte_pci_addr addr;
457 
458 	/* for debug purposes, PCI can be disabled */
459 	if (!rte_eal_has_pci())
460 		return 0;
461 
462 #ifdef VFIO_PRESENT
463 	if (!pci_vfio_is_enabled())
464 		RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
465 #endif
466 
467 	dir = opendir(rte_pci_get_sysfs_path());
468 	if (dir == NULL) {
469 		RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
470 			__func__, strerror(errno));
471 		return -1;
472 	}
473 
474 	while ((e = readdir(dir)) != NULL) {
475 		if (e->d_name[0] == '.')
476 			continue;
477 
478 		if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
479 			continue;
480 
481 		if (rte_pci_ignore_device(&addr))
482 			continue;
483 
484 		snprintf(dirname, sizeof(dirname), "%s/%s",
485 				rte_pci_get_sysfs_path(), e->d_name);
486 
487 		if (pci_scan_one(dirname, &addr) < 0)
488 			goto error;
489 	}
490 	closedir(dir);
491 	return 0;
492 
493 error:
494 	closedir(dir);
495 	return -1;
496 }
497 
498 #if defined(RTE_ARCH_X86)
499 bool
500 pci_device_iommu_support_va(const struct rte_pci_device *dev)
501 {
502 #define VTD_CAP_MGAW_SHIFT	16
503 #define VTD_CAP_MGAW_MASK	(0x3fULL << VTD_CAP_MGAW_SHIFT)
504 	const struct rte_pci_addr *addr = &dev->addr;
505 	char filename[PATH_MAX];
506 	FILE *fp;
507 	uint64_t mgaw, vtd_cap_reg = 0;
508 
509 	snprintf(filename, sizeof(filename),
510 		 "%s/" PCI_PRI_FMT "/iommu/intel-iommu/cap",
511 		 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
512 		 addr->function);
513 
514 	fp = fopen(filename, "r");
515 	if (fp == NULL) {
516 		/* We don't have an Intel IOMMU, assume VA supported */
517 		if (errno == ENOENT)
518 			return true;
519 
520 		RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
521 			__func__, filename, strerror(errno));
522 		return false;
523 	}
524 
525 	/* We have an Intel IOMMU */
526 	if (fscanf(fp, "%" PRIx64, &vtd_cap_reg) != 1) {
527 		RTE_LOG(ERR, EAL, "%s(): can't read %s\n", __func__, filename);
528 		fclose(fp);
529 		return false;
530 	}
531 
532 	fclose(fp);
533 
534 	mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
535 
536 	/*
537 	 * Assuming there is no limitation by now. We can not know at this point
538 	 * because the memory has not been initialized yet. Setting the dma mask
539 	 * will force a check once memory initialization is done. We can not do
540 	 * a fallback to IOVA PA now, but if the dma check fails, the error
541 	 * message should advice for using '--iova-mode pa' if IOVA VA is the
542 	 * current mode.
543 	 */
544 	rte_mem_set_dma_mask(mgaw);
545 	return true;
546 }
547 #elif defined(RTE_ARCH_PPC_64)
548 bool
549 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
550 {
551 	/*
552 	 * IOMMU is always present on a PowerNV host (IOMMUv2).
553 	 * IOMMU is also present in a KVM/QEMU VM (IOMMUv1) but is not
554 	 * currently supported by DPDK. Test for our current environment
555 	 * and report VA support as appropriate.
556 	 */
557 
558 	char *line = NULL;
559 	size_t len = 0;
560 	char filename[PATH_MAX] = "/proc/cpuinfo";
561 	FILE *fp = fopen(filename, "r");
562 	bool ret = false;
563 
564 	if (fp == NULL) {
565 		RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
566 			__func__, filename, strerror(errno));
567 		return ret;
568 	}
569 
570 	/* Check for a PowerNV platform */
571 	while (getline(&line, &len, fp) != -1) {
572 		if (strstr(line, "platform") != NULL)
573 			continue;
574 
575 		if (strstr(line, "PowerNV") != NULL) {
576 			RTE_LOG(DEBUG, EAL, "Running on a PowerNV system\n");
577 			ret = true;
578 			break;
579 		}
580 	}
581 
582 	free(line);
583 	fclose(fp);
584 	return ret;
585 }
586 #else
587 bool
588 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
589 {
590 	return true;
591 }
592 #endif
593 
594 enum rte_iova_mode
595 pci_device_iova_mode(const struct rte_pci_driver *pdrv,
596 		     const struct rte_pci_device *pdev)
597 {
598 	enum rte_iova_mode iova_mode = RTE_IOVA_DC;
599 
600 	switch (pdev->kdrv) {
601 	case RTE_PCI_KDRV_VFIO: {
602 #ifdef VFIO_PRESENT
603 		static int is_vfio_noiommu_enabled = -1;
604 
605 		if (is_vfio_noiommu_enabled == -1) {
606 			if (rte_vfio_noiommu_is_enabled() == 1)
607 				is_vfio_noiommu_enabled = 1;
608 			else
609 				is_vfio_noiommu_enabled = 0;
610 		}
611 		if (is_vfio_noiommu_enabled != 0)
612 			iova_mode = RTE_IOVA_PA;
613 		else if ((pdrv->drv_flags & RTE_PCI_DRV_NEED_IOVA_AS_VA) != 0)
614 			iova_mode = RTE_IOVA_VA;
615 #endif
616 		break;
617 	}
618 
619 	case RTE_PCI_KDRV_IGB_UIO:
620 	case RTE_PCI_KDRV_UIO_GENERIC:
621 		iova_mode = RTE_IOVA_PA;
622 		break;
623 
624 	default:
625 		if ((pdrv->drv_flags & RTE_PCI_DRV_NEED_IOVA_AS_VA) != 0)
626 			iova_mode = RTE_IOVA_VA;
627 		break;
628 	}
629 	return iova_mode;
630 }
631 
632 /* Read PCI config space. */
633 int rte_pci_read_config(const struct rte_pci_device *device,
634 		void *buf, size_t len, off_t offset)
635 {
636 	char devname[RTE_DEV_NAME_MAX_LEN] = "";
637 	const struct rte_intr_handle *intr_handle = &device->intr_handle;
638 
639 	switch (device->kdrv) {
640 	case RTE_PCI_KDRV_IGB_UIO:
641 	case RTE_PCI_KDRV_UIO_GENERIC:
642 		return pci_uio_read_config(intr_handle, buf, len, offset);
643 #ifdef VFIO_PRESENT
644 	case RTE_PCI_KDRV_VFIO:
645 		return pci_vfio_read_config(intr_handle, buf, len, offset);
646 #endif
647 	default:
648 		rte_pci_device_name(&device->addr, devname,
649 				    RTE_DEV_NAME_MAX_LEN);
650 		RTE_LOG(ERR, EAL,
651 			"Unknown driver type for %s\n", devname);
652 		return -1;
653 	}
654 }
655 
656 /* Write PCI config space. */
657 int rte_pci_write_config(const struct rte_pci_device *device,
658 		const void *buf, size_t len, off_t offset)
659 {
660 	char devname[RTE_DEV_NAME_MAX_LEN] = "";
661 	const struct rte_intr_handle *intr_handle = &device->intr_handle;
662 
663 	switch (device->kdrv) {
664 	case RTE_PCI_KDRV_IGB_UIO:
665 	case RTE_PCI_KDRV_UIO_GENERIC:
666 		return pci_uio_write_config(intr_handle, buf, len, offset);
667 #ifdef VFIO_PRESENT
668 	case RTE_PCI_KDRV_VFIO:
669 		return pci_vfio_write_config(intr_handle, buf, len, offset);
670 #endif
671 	default:
672 		rte_pci_device_name(&device->addr, devname,
673 				    RTE_DEV_NAME_MAX_LEN);
674 		RTE_LOG(ERR, EAL,
675 			"Unknown driver type for %s\n", devname);
676 		return -1;
677 	}
678 }
679 
680 #if defined(RTE_ARCH_X86)
681 static int
682 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
683 		struct rte_pci_ioport *p)
684 {
685 	uint16_t start, end;
686 	FILE *fp;
687 	char *line = NULL;
688 	char pci_id[16];
689 	int found = 0;
690 	size_t linesz;
691 
692 	if (rte_eal_iopl_init() != 0) {
693 		RTE_LOG(ERR, EAL, "%s(): insufficient ioport permissions for PCI device %s\n",
694 			__func__, dev->name);
695 		return -1;
696 	}
697 
698 	snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
699 		 dev->addr.domain, dev->addr.bus,
700 		 dev->addr.devid, dev->addr.function);
701 
702 	fp = fopen("/proc/ioports", "r");
703 	if (fp == NULL) {
704 		RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
705 		return -1;
706 	}
707 
708 	while (getdelim(&line, &linesz, '\n', fp) > 0) {
709 		char *ptr = line;
710 		char *left;
711 		int n;
712 
713 		n = strcspn(ptr, ":");
714 		ptr[n] = 0;
715 		left = &ptr[n + 1];
716 
717 		while (*left && isspace(*left))
718 			left++;
719 
720 		if (!strncmp(left, pci_id, strlen(pci_id))) {
721 			found = 1;
722 
723 			while (*ptr && isspace(*ptr))
724 				ptr++;
725 
726 			sscanf(ptr, "%04hx-%04hx", &start, &end);
727 
728 			break;
729 		}
730 	}
731 
732 	free(line);
733 	fclose(fp);
734 
735 	if (!found)
736 		return -1;
737 
738 	p->base = start;
739 	RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
740 
741 	return 0;
742 }
743 #endif
744 
745 int
746 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
747 		struct rte_pci_ioport *p)
748 {
749 	int ret = -1;
750 
751 	switch (dev->kdrv) {
752 #ifdef VFIO_PRESENT
753 	case RTE_PCI_KDRV_VFIO:
754 		if (pci_vfio_is_enabled())
755 			ret = pci_vfio_ioport_map(dev, bar, p);
756 		break;
757 #endif
758 	case RTE_PCI_KDRV_IGB_UIO:
759 		ret = pci_uio_ioport_map(dev, bar, p);
760 		break;
761 	case RTE_PCI_KDRV_UIO_GENERIC:
762 #if defined(RTE_ARCH_X86)
763 		ret = pci_ioport_map(dev, bar, p);
764 #else
765 		ret = pci_uio_ioport_map(dev, bar, p);
766 #endif
767 		break;
768 	default:
769 		break;
770 	}
771 
772 	if (!ret)
773 		p->dev = dev;
774 
775 	return ret;
776 }
777 
778 void
779 rte_pci_ioport_read(struct rte_pci_ioport *p,
780 		void *data, size_t len, off_t offset)
781 {
782 	switch (p->dev->kdrv) {
783 #ifdef VFIO_PRESENT
784 	case RTE_PCI_KDRV_VFIO:
785 		pci_vfio_ioport_read(p, data, len, offset);
786 		break;
787 #endif
788 	case RTE_PCI_KDRV_IGB_UIO:
789 		pci_uio_ioport_read(p, data, len, offset);
790 		break;
791 	case RTE_PCI_KDRV_UIO_GENERIC:
792 		pci_uio_ioport_read(p, data, len, offset);
793 		break;
794 	default:
795 		break;
796 	}
797 }
798 
799 void
800 rte_pci_ioport_write(struct rte_pci_ioport *p,
801 		const void *data, size_t len, off_t offset)
802 {
803 	switch (p->dev->kdrv) {
804 #ifdef VFIO_PRESENT
805 	case RTE_PCI_KDRV_VFIO:
806 		pci_vfio_ioport_write(p, data, len, offset);
807 		break;
808 #endif
809 	case RTE_PCI_KDRV_IGB_UIO:
810 		pci_uio_ioport_write(p, data, len, offset);
811 		break;
812 	case RTE_PCI_KDRV_UIO_GENERIC:
813 		pci_uio_ioport_write(p, data, len, offset);
814 		break;
815 	default:
816 		break;
817 	}
818 }
819 
820 int
821 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
822 {
823 	int ret = -1;
824 
825 	switch (p->dev->kdrv) {
826 #ifdef VFIO_PRESENT
827 	case RTE_PCI_KDRV_VFIO:
828 		if (pci_vfio_is_enabled())
829 			ret = pci_vfio_ioport_unmap(p);
830 		break;
831 #endif
832 	case RTE_PCI_KDRV_IGB_UIO:
833 		ret = pci_uio_ioport_unmap(p);
834 		break;
835 	case RTE_PCI_KDRV_UIO_GENERIC:
836 #if defined(RTE_ARCH_X86)
837 		ret = 0;
838 #else
839 		ret = pci_uio_ioport_unmap(p);
840 #endif
841 		break;
842 	default:
843 		break;
844 	}
845 
846 	return ret;
847 }
848