xref: /dpdk/drivers/bus/pci/linux/pci.c (revision f5057be340e44f3edc0fe90fa875eb89a4c49b4f)
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 					pci_name_set(dev2);
356 					memmove(dev2->mem_resource,
357 						dev->mem_resource,
358 						sizeof(dev->mem_resource));
359 				} else {
360 					/**
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...
366 					 **/
367 					if (dev2->kdrv != dev->kdrv ||
368 						dev2->max_vfs != dev->max_vfs)
369 						/*
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
377 						 * error as an alarm.
378 						 */
379 						RTE_LOG(ERR, EAL, "Unexpected device scan at %s!\n",
380 							filename);
381 					else if (dev2->device.devargs !=
382 						 dev->device.devargs) {
383 						rte_devargs_remove(dev2->device.devargs);
384 						pci_name_set(dev2);
385 					}
386 				}
387 				free(dev);
388 			}
389 			return 0;
390 		}
391 
392 		rte_pci_add_device(dev);
393 	}
394 
395 	return 0;
396 }
397 
398 int
399 pci_update_device(const struct rte_pci_addr *addr)
400 {
401 	char filename[PATH_MAX];
402 
403 	snprintf(filename, sizeof(filename), "%s/" PCI_PRI_FMT,
404 		 rte_pci_get_sysfs_path(), addr->domain, addr->bus, addr->devid,
405 		 addr->function);
406 
407 	return pci_scan_one(filename, addr);
408 }
409 
410 /*
411  * split up a pci address into its constituent parts.
412  */
413 static int
414 parse_pci_addr_format(const char *buf, int bufsize, struct rte_pci_addr *addr)
415 {
416 	/* first split on ':' */
417 	union splitaddr {
418 		struct {
419 			char *domain;
420 			char *bus;
421 			char *devid;
422 			char *function;
423 		};
424 		char *str[PCI_FMT_NVAL]; /* last element-separator is "." not ":" */
425 	} splitaddr;
426 
427 	char *buf_copy = strndup(buf, bufsize);
428 	if (buf_copy == NULL)
429 		return -1;
430 
431 	if (rte_strsplit(buf_copy, bufsize, splitaddr.str, PCI_FMT_NVAL, ':')
432 			!= PCI_FMT_NVAL - 1)
433 		goto error;
434 	/* final split is on '.' between devid and function */
435 	splitaddr.function = strchr(splitaddr.devid,'.');
436 	if (splitaddr.function == NULL)
437 		goto error;
438 	*splitaddr.function++ = '\0';
439 
440 	/* now convert to int values */
441 	errno = 0;
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);
446 	if (errno != 0)
447 		goto error;
448 
449 	free(buf_copy); /* free the copy made with strdup */
450 	return 0;
451 error:
452 	free(buf_copy);
453 	return -1;
454 }
455 
456 /*
457  * Scan the content of the PCI bus, and the devices in the devices
458  * list
459  */
460 int
461 rte_pci_scan(void)
462 {
463 	struct dirent *e;
464 	DIR *dir;
465 	char dirname[PATH_MAX];
466 	struct rte_pci_addr addr;
467 
468 	/* for debug purposes, PCI can be disabled */
469 	if (!rte_eal_has_pci())
470 		return 0;
471 
472 #ifdef VFIO_PRESENT
473 	if (!pci_vfio_is_enabled())
474 		RTE_LOG(DEBUG, EAL, "VFIO PCI modules not loaded\n");
475 #endif
476 
477 	dir = opendir(rte_pci_get_sysfs_path());
478 	if (dir == NULL) {
479 		RTE_LOG(ERR, EAL, "%s(): opendir failed: %s\n",
480 			__func__, strerror(errno));
481 		return -1;
482 	}
483 
484 	while ((e = readdir(dir)) != NULL) {
485 		if (e->d_name[0] == '.')
486 			continue;
487 
488 		if (parse_pci_addr_format(e->d_name, sizeof(e->d_name), &addr) != 0)
489 			continue;
490 
491 		if (rte_pci_ignore_device(&addr))
492 			continue;
493 
494 		snprintf(dirname, sizeof(dirname), "%s/%s",
495 				rte_pci_get_sysfs_path(), e->d_name);
496 
497 		if (pci_scan_one(dirname, &addr) < 0)
498 			goto error;
499 	}
500 	closedir(dir);
501 	return 0;
502 
503 error:
504 	closedir(dir);
505 	return -1;
506 }
507 
508 #if defined(RTE_ARCH_X86)
509 bool
510 pci_device_iommu_support_va(const struct rte_pci_device *dev)
511 {
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];
516 	FILE *fp;
517 	uint64_t mgaw, vtd_cap_reg = 0;
518 
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,
522 		 addr->function);
523 
524 	fp = fopen(filename, "r");
525 	if (fp == NULL) {
526 		/* We don't have an Intel IOMMU, assume VA supported */
527 		if (errno == ENOENT)
528 			return true;
529 
530 		RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
531 			__func__, filename, strerror(errno));
532 		return false;
533 	}
534 
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);
538 		fclose(fp);
539 		return false;
540 	}
541 
542 	fclose(fp);
543 
544 	mgaw = ((vtd_cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
545 
546 	/*
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
552 	 * current mode.
553 	 */
554 	rte_mem_set_dma_mask(mgaw);
555 	return true;
556 }
557 #elif defined(RTE_ARCH_PPC_64)
558 bool
559 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
560 {
561 	/*
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.
566 	 */
567 
568 	char *line = NULL;
569 	size_t len = 0;
570 	char filename[PATH_MAX] = "/proc/cpuinfo";
571 	FILE *fp = fopen(filename, "r");
572 	bool ret = false;
573 
574 	if (fp == NULL) {
575 		RTE_LOG(ERR, EAL, "%s(): can't open %s: %s\n",
576 			__func__, filename, strerror(errno));
577 		return ret;
578 	}
579 
580 	/* Check for a PowerNV platform */
581 	while (getline(&line, &len, fp) != -1) {
582 		if (strstr(line, "platform") != NULL)
583 			continue;
584 
585 		if (strstr(line, "PowerNV") != NULL) {
586 			RTE_LOG(DEBUG, EAL, "Running on a PowerNV system\n");
587 			ret = true;
588 			break;
589 		}
590 	}
591 
592 	free(line);
593 	fclose(fp);
594 	return ret;
595 }
596 #else
597 bool
598 pci_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
599 {
600 	return true;
601 }
602 #endif
603 
604 enum rte_iova_mode
605 pci_device_iova_mode(const struct rte_pci_driver *pdrv,
606 		     const struct rte_pci_device *pdev)
607 {
608 	enum rte_iova_mode iova_mode = RTE_IOVA_DC;
609 
610 	switch (pdev->kdrv) {
611 	case RTE_PCI_KDRV_VFIO: {
612 #ifdef VFIO_PRESENT
613 		static int is_vfio_noiommu_enabled = -1;
614 
615 		if (is_vfio_noiommu_enabled == -1) {
616 			if (rte_vfio_noiommu_is_enabled() == 1)
617 				is_vfio_noiommu_enabled = 1;
618 			else
619 				is_vfio_noiommu_enabled = 0;
620 		}
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;
625 #endif
626 		break;
627 	}
628 
629 	case RTE_PCI_KDRV_IGB_UIO:
630 	case RTE_PCI_KDRV_UIO_GENERIC:
631 		iova_mode = RTE_IOVA_PA;
632 		break;
633 
634 	default:
635 		if ((pdrv->drv_flags & RTE_PCI_DRV_NEED_IOVA_AS_VA) != 0)
636 			iova_mode = RTE_IOVA_VA;
637 		break;
638 	}
639 	return iova_mode;
640 }
641 
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)
645 {
646 	char devname[RTE_DEV_NAME_MAX_LEN] = "";
647 	const struct rte_intr_handle *intr_handle = &device->intr_handle;
648 
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);
653 #ifdef VFIO_PRESENT
654 	case RTE_PCI_KDRV_VFIO:
655 		return pci_vfio_read_config(intr_handle, buf, len, offset);
656 #endif
657 	default:
658 		rte_pci_device_name(&device->addr, devname,
659 				    RTE_DEV_NAME_MAX_LEN);
660 		RTE_LOG(ERR, EAL,
661 			"Unknown driver type for %s\n", devname);
662 		return -1;
663 	}
664 }
665 
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)
669 {
670 	char devname[RTE_DEV_NAME_MAX_LEN] = "";
671 	const struct rte_intr_handle *intr_handle = &device->intr_handle;
672 
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);
677 #ifdef VFIO_PRESENT
678 	case RTE_PCI_KDRV_VFIO:
679 		return pci_vfio_write_config(intr_handle, buf, len, offset);
680 #endif
681 	default:
682 		rte_pci_device_name(&device->addr, devname,
683 				    RTE_DEV_NAME_MAX_LEN);
684 		RTE_LOG(ERR, EAL,
685 			"Unknown driver type for %s\n", devname);
686 		return -1;
687 	}
688 }
689 
690 #if defined(RTE_ARCH_X86)
691 static int
692 pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
693 		struct rte_pci_ioport *p)
694 {
695 	uint16_t start, end;
696 	FILE *fp;
697 	char *line = NULL;
698 	char pci_id[16];
699 	int found = 0;
700 	size_t linesz;
701 
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);
705 		return -1;
706 	}
707 
708 	snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
709 		 dev->addr.domain, dev->addr.bus,
710 		 dev->addr.devid, dev->addr.function);
711 
712 	fp = fopen("/proc/ioports", "r");
713 	if (fp == NULL) {
714 		RTE_LOG(ERR, EAL, "%s(): can't open ioports\n", __func__);
715 		return -1;
716 	}
717 
718 	while (getdelim(&line, &linesz, '\n', fp) > 0) {
719 		char *ptr = line;
720 		char *left;
721 		int n;
722 
723 		n = strcspn(ptr, ":");
724 		ptr[n] = 0;
725 		left = &ptr[n + 1];
726 
727 		while (*left && isspace(*left))
728 			left++;
729 
730 		if (!strncmp(left, pci_id, strlen(pci_id))) {
731 			found = 1;
732 
733 			while (*ptr && isspace(*ptr))
734 				ptr++;
735 
736 			sscanf(ptr, "%04hx-%04hx", &start, &end);
737 
738 			break;
739 		}
740 	}
741 
742 	free(line);
743 	fclose(fp);
744 
745 	if (!found)
746 		return -1;
747 
748 	p->base = start;
749 	RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);
750 
751 	return 0;
752 }
753 #endif
754 
755 int
756 rte_pci_ioport_map(struct rte_pci_device *dev, int bar,
757 		struct rte_pci_ioport *p)
758 {
759 	int ret = -1;
760 
761 	switch (dev->kdrv) {
762 #ifdef VFIO_PRESENT
763 	case RTE_PCI_KDRV_VFIO:
764 		if (pci_vfio_is_enabled())
765 			ret = pci_vfio_ioport_map(dev, bar, p);
766 		break;
767 #endif
768 	case RTE_PCI_KDRV_IGB_UIO:
769 		ret = pci_uio_ioport_map(dev, bar, p);
770 		break;
771 	case RTE_PCI_KDRV_UIO_GENERIC:
772 #if defined(RTE_ARCH_X86)
773 		ret = pci_ioport_map(dev, bar, p);
774 #else
775 		ret = pci_uio_ioport_map(dev, bar, p);
776 #endif
777 		break;
778 	default:
779 		break;
780 	}
781 
782 	if (!ret)
783 		p->dev = dev;
784 
785 	return ret;
786 }
787 
788 void
789 rte_pci_ioport_read(struct rte_pci_ioport *p,
790 		void *data, size_t len, off_t offset)
791 {
792 	switch (p->dev->kdrv) {
793 #ifdef VFIO_PRESENT
794 	case RTE_PCI_KDRV_VFIO:
795 		pci_vfio_ioport_read(p, data, len, offset);
796 		break;
797 #endif
798 	case RTE_PCI_KDRV_IGB_UIO:
799 		pci_uio_ioport_read(p, data, len, offset);
800 		break;
801 	case RTE_PCI_KDRV_UIO_GENERIC:
802 		pci_uio_ioport_read(p, data, len, offset);
803 		break;
804 	default:
805 		break;
806 	}
807 }
808 
809 void
810 rte_pci_ioport_write(struct rte_pci_ioport *p,
811 		const void *data, size_t len, off_t offset)
812 {
813 	switch (p->dev->kdrv) {
814 #ifdef VFIO_PRESENT
815 	case RTE_PCI_KDRV_VFIO:
816 		pci_vfio_ioport_write(p, data, len, offset);
817 		break;
818 #endif
819 	case RTE_PCI_KDRV_IGB_UIO:
820 		pci_uio_ioport_write(p, data, len, offset);
821 		break;
822 	case RTE_PCI_KDRV_UIO_GENERIC:
823 		pci_uio_ioport_write(p, data, len, offset);
824 		break;
825 	default:
826 		break;
827 	}
828 }
829 
830 int
831 rte_pci_ioport_unmap(struct rte_pci_ioport *p)
832 {
833 	int ret = -1;
834 
835 	switch (p->dev->kdrv) {
836 #ifdef VFIO_PRESENT
837 	case RTE_PCI_KDRV_VFIO:
838 		if (pci_vfio_is_enabled())
839 			ret = pci_vfio_ioport_unmap(p);
840 		break;
841 #endif
842 	case RTE_PCI_KDRV_IGB_UIO:
843 		ret = pci_uio_ioport_unmap(p);
844 		break;
845 	case RTE_PCI_KDRV_UIO_GENERIC:
846 #if defined(RTE_ARCH_X86)
847 		ret = 0;
848 #else
849 		ret = pci_uio_ioport_unmap(p);
850 #endif
851 		break;
852 	default:
853 		break;
854 	}
855 
856 	return ret;
857 }
858