xref: /linux/tools/testing/selftests/vfio/lib/vfio_pci_device.c (revision 118e073ef6a3446862ada27bdca2b8a53447f428)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <dirent.h>
3 #include <fcntl.h>
4 #include <libgen.h>
5 #include <stdlib.h>
6 #include <string.h>
7 #include <unistd.h>
8 
9 #include <sys/eventfd.h>
10 #include <sys/ioctl.h>
11 #include <sys/mman.h>
12 
13 #include <linux/limits.h>
14 #include <linux/mman.h>
15 #include <linux/types.h>
16 #include <linux/vfio.h>
17 
18 #include "../../../kselftest.h"
19 #include <vfio_util.h>
20 
21 #define VFIO_DEV_PATH	"/dev/vfio/vfio"
22 #define PCI_SYSFS_PATH	"/sys/bus/pci/devices"
23 
24 #define ioctl_assert(_fd, _op, _arg) do {						       \
25 	void *__arg = (_arg);								       \
26 	int __ret = ioctl((_fd), (_op), (__arg));					       \
27 	VFIO_ASSERT_EQ(__ret, 0, "ioctl(%s, %s, %s) returned %d\n", #_fd, #_op, #_arg, __ret); \
28 } while (0)
29 
30 iova_t __to_iova(struct vfio_pci_device *device, void *vaddr)
31 {
32 	struct vfio_dma_region *region;
33 
34 	list_for_each_entry(region, &device->dma_regions, link) {
35 		if (vaddr < region->vaddr)
36 			continue;
37 
38 		if (vaddr >= region->vaddr + region->size)
39 			continue;
40 
41 		return region->iova + (vaddr - region->vaddr);
42 	}
43 
44 	return INVALID_IOVA;
45 }
46 
47 iova_t to_iova(struct vfio_pci_device *device, void *vaddr)
48 {
49 	iova_t iova;
50 
51 	iova = __to_iova(device, vaddr);
52 	VFIO_ASSERT_NE(iova, INVALID_IOVA, "%p is not mapped into device.\n", vaddr);
53 
54 	return iova;
55 }
56 
57 static void vfio_pci_irq_set(struct vfio_pci_device *device,
58 			     u32 index, u32 vector, u32 count, int *fds)
59 {
60 	u8 buf[sizeof(struct vfio_irq_set) + sizeof(int) * count] = {};
61 	struct vfio_irq_set *irq = (void *)&buf;
62 	int *irq_fds = (void *)&irq->data;
63 
64 	irq->argsz = sizeof(buf);
65 	irq->flags = VFIO_IRQ_SET_ACTION_TRIGGER;
66 	irq->index = index;
67 	irq->start = vector;
68 	irq->count = count;
69 
70 	if (count) {
71 		irq->flags |= VFIO_IRQ_SET_DATA_EVENTFD;
72 		memcpy(irq_fds, fds, sizeof(int) * count);
73 	} else {
74 		irq->flags |= VFIO_IRQ_SET_DATA_NONE;
75 	}
76 
77 	ioctl_assert(device->fd, VFIO_DEVICE_SET_IRQS, irq);
78 }
79 
80 void vfio_pci_irq_trigger(struct vfio_pci_device *device, u32 index, u32 vector)
81 {
82 	struct vfio_irq_set irq = {
83 		.argsz = sizeof(irq),
84 		.flags = VFIO_IRQ_SET_ACTION_TRIGGER | VFIO_IRQ_SET_DATA_NONE,
85 		.index = index,
86 		.start = vector,
87 		.count = 1,
88 	};
89 
90 	ioctl_assert(device->fd, VFIO_DEVICE_SET_IRQS, &irq);
91 }
92 
93 static void check_supported_irq_index(u32 index)
94 {
95 	/* VFIO selftests only supports MSI and MSI-x for now. */
96 	VFIO_ASSERT_TRUE(index == VFIO_PCI_MSI_IRQ_INDEX ||
97 			 index == VFIO_PCI_MSIX_IRQ_INDEX,
98 			 "Unsupported IRQ index: %u\n", index);
99 }
100 
101 void vfio_pci_irq_enable(struct vfio_pci_device *device, u32 index, u32 vector,
102 			 int count)
103 {
104 	int i;
105 
106 	check_supported_irq_index(index);
107 
108 	for (i = vector; i < vector + count; i++) {
109 		VFIO_ASSERT_LT(device->msi_eventfds[i], 0);
110 		device->msi_eventfds[i] = eventfd(0, 0);
111 		VFIO_ASSERT_GE(device->msi_eventfds[i], 0);
112 	}
113 
114 	vfio_pci_irq_set(device, index, vector, count, device->msi_eventfds + vector);
115 }
116 
117 void vfio_pci_irq_disable(struct vfio_pci_device *device, u32 index)
118 {
119 	int i;
120 
121 	check_supported_irq_index(index);
122 
123 	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++) {
124 		if (device->msi_eventfds[i] < 0)
125 			continue;
126 
127 		VFIO_ASSERT_EQ(close(device->msi_eventfds[i]), 0);
128 		device->msi_eventfds[i] = -1;
129 	}
130 
131 	vfio_pci_irq_set(device, index, 0, 0, NULL);
132 }
133 
134 static void vfio_pci_irq_get(struct vfio_pci_device *device, u32 index,
135 			     struct vfio_irq_info *irq_info)
136 {
137 	irq_info->argsz = sizeof(*irq_info);
138 	irq_info->index = index;
139 
140 	ioctl_assert(device->fd, VFIO_DEVICE_GET_IRQ_INFO, irq_info);
141 }
142 
143 void vfio_pci_dma_map(struct vfio_pci_device *device,
144 		      struct vfio_dma_region *region)
145 {
146 	struct vfio_iommu_type1_dma_map map = {
147 		.argsz = sizeof(map),
148 		.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
149 		.vaddr = (u64)region->vaddr,
150 		.iova = region->iova,
151 		.size = region->size,
152 	};
153 
154 	ioctl_assert(device->container_fd, VFIO_IOMMU_MAP_DMA, &map);
155 
156 	list_add(&region->link, &device->dma_regions);
157 }
158 
159 void vfio_pci_dma_unmap(struct vfio_pci_device *device,
160 			struct vfio_dma_region *region)
161 {
162 	struct vfio_iommu_type1_dma_unmap unmap = {
163 		.argsz = sizeof(unmap),
164 		.iova = region->iova,
165 		.size = region->size,
166 	};
167 
168 	ioctl_assert(device->container_fd, VFIO_IOMMU_UNMAP_DMA, &unmap);
169 
170 	list_del(&region->link);
171 }
172 
173 static void vfio_pci_region_get(struct vfio_pci_device *device, int index,
174 				struct vfio_region_info *info)
175 {
176 	memset(info, 0, sizeof(*info));
177 
178 	info->argsz = sizeof(*info);
179 	info->index = index;
180 
181 	ioctl_assert(device->fd, VFIO_DEVICE_GET_REGION_INFO, info);
182 }
183 
184 static void vfio_pci_bar_map(struct vfio_pci_device *device, int index)
185 {
186 	struct vfio_pci_bar *bar = &device->bars[index];
187 	int prot = 0;
188 
189 	VFIO_ASSERT_LT(index, PCI_STD_NUM_BARS);
190 	VFIO_ASSERT_NULL(bar->vaddr);
191 	VFIO_ASSERT_TRUE(bar->info.flags & VFIO_REGION_INFO_FLAG_MMAP);
192 
193 	if (bar->info.flags & VFIO_REGION_INFO_FLAG_READ)
194 		prot |= PROT_READ;
195 	if (bar->info.flags & VFIO_REGION_INFO_FLAG_WRITE)
196 		prot |= PROT_WRITE;
197 
198 	bar->vaddr = mmap(NULL, bar->info.size, prot, MAP_FILE | MAP_SHARED,
199 			  device->fd, bar->info.offset);
200 	VFIO_ASSERT_NE(bar->vaddr, MAP_FAILED);
201 }
202 
203 static void vfio_pci_bar_unmap(struct vfio_pci_device *device, int index)
204 {
205 	struct vfio_pci_bar *bar = &device->bars[index];
206 
207 	VFIO_ASSERT_LT(index, PCI_STD_NUM_BARS);
208 	VFIO_ASSERT_NOT_NULL(bar->vaddr);
209 
210 	VFIO_ASSERT_EQ(munmap(bar->vaddr, bar->info.size), 0);
211 	bar->vaddr = NULL;
212 }
213 
214 static void vfio_pci_bar_unmap_all(struct vfio_pci_device *device)
215 {
216 	int i;
217 
218 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
219 		if (device->bars[i].vaddr)
220 			vfio_pci_bar_unmap(device, i);
221 	}
222 }
223 
224 void vfio_pci_config_access(struct vfio_pci_device *device, bool write,
225 			    size_t config, size_t size, void *data)
226 {
227 	struct vfio_region_info *config_space = &device->config_space;
228 	int ret;
229 
230 	if (write)
231 		ret = pwrite(device->fd, data, size, config_space->offset + config);
232 	else
233 		ret = pread(device->fd, data, size, config_space->offset + config);
234 
235 	VFIO_ASSERT_EQ(ret, size, "Failed to %s PCI config space: 0x%lx\n",
236 		       write ? "write to" : "read from", config);
237 }
238 
239 void vfio_pci_device_reset(struct vfio_pci_device *device)
240 {
241 	ioctl_assert(device->fd, VFIO_DEVICE_RESET, NULL);
242 }
243 
244 static unsigned int vfio_pci_get_group_from_dev(const char *bdf)
245 {
246 	char dev_iommu_group_path[PATH_MAX] = {0};
247 	char sysfs_path[PATH_MAX] = {0};
248 	unsigned int group;
249 	int ret;
250 
251 	snprintf(sysfs_path, PATH_MAX, "%s/%s/iommu_group", PCI_SYSFS_PATH, bdf);
252 
253 	ret = readlink(sysfs_path, dev_iommu_group_path, sizeof(dev_iommu_group_path));
254 	VFIO_ASSERT_NE(ret, -1, "Failed to get the IOMMU group for device: %s\n", bdf);
255 
256 	ret = sscanf(basename(dev_iommu_group_path), "%u", &group);
257 	VFIO_ASSERT_EQ(ret, 1, "Failed to get the IOMMU group for device: %s\n", bdf);
258 
259 	return group;
260 }
261 
262 static void vfio_pci_container_setup(struct vfio_pci_device *device)
263 {
264 	int version;
265 
266 	device->container_fd = open(VFIO_DEV_PATH, O_RDWR);
267 	VFIO_ASSERT_GE(device->container_fd, 0, "open(%s) failed\n", VFIO_DEV_PATH);
268 
269 	version = ioctl(device->container_fd, VFIO_GET_API_VERSION);
270 	VFIO_ASSERT_EQ(version, VFIO_API_VERSION);
271 }
272 
273 static void vfio_pci_group_setup(struct vfio_pci_device *device, const char *bdf)
274 {
275 	struct vfio_group_status group_status = {
276 		.argsz = sizeof(group_status),
277 	};
278 	char group_path[32];
279 	int group;
280 
281 	group = vfio_pci_get_group_from_dev(bdf);
282 	snprintf(group_path, sizeof(group_path), "/dev/vfio/%d", group);
283 
284 	device->group_fd = open(group_path, O_RDWR);
285 	VFIO_ASSERT_GE(device->group_fd, 0, "open(%s) failed\n", group_path);
286 
287 	ioctl_assert(device->group_fd, VFIO_GROUP_GET_STATUS, &group_status);
288 	VFIO_ASSERT_TRUE(group_status.flags & VFIO_GROUP_FLAGS_VIABLE);
289 
290 	ioctl_assert(device->group_fd, VFIO_GROUP_SET_CONTAINER, &device->container_fd);
291 }
292 
293 static void vfio_pci_iommu_setup(struct vfio_pci_device *device, unsigned long iommu_type)
294 {
295 	int ret;
296 
297 	INIT_LIST_HEAD(&device->dma_regions);
298 
299 	ret = ioctl(device->container_fd, VFIO_CHECK_EXTENSION, iommu_type);
300 	VFIO_ASSERT_GT(ret, 0, "VFIO IOMMU type %lu not supported\n", iommu_type);
301 
302 	ioctl_assert(device->container_fd, VFIO_SET_IOMMU, (void *)iommu_type);
303 }
304 
305 static void vfio_pci_device_setup(struct vfio_pci_device *device, const char *bdf)
306 {
307 	int i;
308 
309 	device->fd = ioctl(device->group_fd, VFIO_GROUP_GET_DEVICE_FD, bdf);
310 	VFIO_ASSERT_GE(device->fd, 0);
311 
312 	device->info.argsz = sizeof(device->info);
313 	ioctl_assert(device->fd, VFIO_DEVICE_GET_INFO, &device->info);
314 
315 	vfio_pci_region_get(device, VFIO_PCI_CONFIG_REGION_INDEX, &device->config_space);
316 
317 	/* Sanity check VFIO does not advertise mmap for config space */
318 	VFIO_ASSERT_TRUE(!(device->config_space.flags & VFIO_REGION_INFO_FLAG_MMAP),
319 			 "PCI config space should not support mmap()\n");
320 
321 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
322 		struct vfio_pci_bar *bar = device->bars + i;
323 
324 		vfio_pci_region_get(device, i, &bar->info);
325 		if (bar->info.flags & VFIO_REGION_INFO_FLAG_MMAP)
326 			vfio_pci_bar_map(device, i);
327 	}
328 
329 	vfio_pci_irq_get(device, VFIO_PCI_MSI_IRQ_INDEX, &device->msi_info);
330 	vfio_pci_irq_get(device, VFIO_PCI_MSIX_IRQ_INDEX, &device->msix_info);
331 
332 	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++)
333 		device->msi_eventfds[i] = -1;
334 }
335 
336 const char *vfio_pci_get_cdev_path(const char *bdf)
337 {
338 	char dir_path[PATH_MAX];
339 	struct dirent *entry;
340 	char *cdev_path;
341 	DIR *dir;
342 
343 	cdev_path = calloc(PATH_MAX, 1);
344 	VFIO_ASSERT_NOT_NULL(cdev_path);
345 
346 	snprintf(dir_path, sizeof(dir_path), "/sys/bus/pci/devices/%s/vfio-dev/", bdf);
347 
348 	dir = opendir(dir_path);
349 	VFIO_ASSERT_NOT_NULL(dir, "Failed to open directory %s\n", dir_path);
350 
351 	while ((entry = readdir(dir)) != NULL) {
352 		/* Find the file that starts with "vfio" */
353 		if (strncmp("vfio", entry->d_name, 4))
354 			continue;
355 
356 		snprintf(cdev_path, PATH_MAX, "/dev/vfio/devices/%s", entry->d_name);
357 		break;
358 	}
359 
360 	VFIO_ASSERT_NE(cdev_path[0], 0, "Failed to find vfio cdev file.\n");
361 	VFIO_ASSERT_EQ(closedir(dir), 0);
362 
363 	return cdev_path;
364 }
365 
366 struct vfio_pci_device *vfio_pci_device_init(const char *bdf, int iommu_type)
367 {
368 	struct vfio_pci_device *device;
369 
370 	device = calloc(1, sizeof(*device));
371 	VFIO_ASSERT_NOT_NULL(device);
372 
373 	vfio_pci_container_setup(device);
374 	vfio_pci_group_setup(device, bdf);
375 	vfio_pci_iommu_setup(device, iommu_type);
376 	vfio_pci_device_setup(device, bdf);
377 
378 	vfio_pci_driver_probe(device);
379 
380 	return device;
381 }
382 
383 void vfio_pci_device_cleanup(struct vfio_pci_device *device)
384 {
385 	int i;
386 
387 	if (device->driver.initialized)
388 		vfio_pci_driver_remove(device);
389 
390 	vfio_pci_bar_unmap_all(device);
391 
392 	VFIO_ASSERT_EQ(close(device->fd), 0);
393 
394 	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++) {
395 		if (device->msi_eventfds[i] < 0)
396 			continue;
397 
398 		VFIO_ASSERT_EQ(close(device->msi_eventfds[i]), 0);
399 	}
400 
401 	VFIO_ASSERT_EQ(close(device->group_fd), 0);
402 	VFIO_ASSERT_EQ(close(device->container_fd), 0);
403 
404 	free(device);
405 }
406 
407 static bool is_bdf(const char *str)
408 {
409 	unsigned int s, b, d, f;
410 	int length, count;
411 
412 	count = sscanf(str, "%4x:%2x:%2x.%2x%n", &s, &b, &d, &f, &length);
413 	return count == 4 && length == strlen(str);
414 }
415 
416 const char *vfio_selftests_get_bdf(int *argc, char *argv[])
417 {
418 	char *bdf;
419 
420 	if (*argc > 1 && is_bdf(argv[*argc - 1]))
421 		return argv[--(*argc)];
422 
423 	bdf = getenv("VFIO_SELFTESTS_BDF");
424 	if (bdf) {
425 		VFIO_ASSERT_TRUE(is_bdf(bdf), "Invalid BDF: %s\n", bdf);
426 		return bdf;
427 	}
428 
429 	fprintf(stderr, "Unable to determine which device to use, skipping test.\n");
430 	fprintf(stderr, "\n");
431 	fprintf(stderr, "To pass the device address via environment variable:\n");
432 	fprintf(stderr, "\n");
433 	fprintf(stderr, "    export VFIO_SELFTESTS_BDF=segment:bus:device.function\n");
434 	fprintf(stderr, "    %s [options]\n", argv[0]);
435 	fprintf(stderr, "\n");
436 	fprintf(stderr, "To pass the device address via argv:\n");
437 	fprintf(stderr, "\n");
438 	fprintf(stderr, "    %s [options] segment:bus:device.function\n", argv[0]);
439 	fprintf(stderr, "\n");
440 	exit(KSFT_SKIP);
441 }
442