vfio_pci_device.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2#include <dirent.h>
  3#include <fcntl.h>
  4#include <libgen.h>
  5#include <stdint.h>
  6#include <stdlib.h>
  7#include <string.h>
  8#include <unistd.h>
  9
 10#include <sys/eventfd.h>
 11#include <sys/ioctl.h>
 12#include <sys/mman.h>
 13
 14#include <linux/iommufd.h>
 15#include <linux/limits.h>
 16#include <linux/mman.h>
 17#include <linux/overflow.h>
 18#include <linux/types.h>
 19#include <linux/vfio.h>
 20
 21#include "kselftest.h"
 22#include <libvfio.h>
 23
 24#define PCI_SYSFS_PATH	"/sys/bus/pci/devices"
 25
 26static void vfio_pci_irq_set(struct vfio_pci_device *device,
 27			     u32 index, u32 vector, u32 count, int *fds)
 28{
 29	u8 buf[sizeof(struct vfio_irq_set) + sizeof(int) * count] = {};
 30	struct vfio_irq_set *irq = (void *)&buf;
 31	int *irq_fds = (void *)&irq->data;
 32
 33	irq->argsz = sizeof(buf);
 34	irq->flags = VFIO_IRQ_SET_ACTION_TRIGGER;
 35	irq->index = index;
 36	irq->start = vector;
 37	irq->count = count;
 38
 39	if (count) {
 40		irq->flags |= VFIO_IRQ_SET_DATA_EVENTFD;
 41		memcpy(irq_fds, fds, sizeof(int) * count);
 42	} else {
 43		irq->flags |= VFIO_IRQ_SET_DATA_NONE;
 44	}
 45
 46	ioctl_assert(device->fd, VFIO_DEVICE_SET_IRQS, irq);
 47}
 48
 49void vfio_pci_irq_trigger(struct vfio_pci_device *device, u32 index, u32 vector)
 50{
 51	struct vfio_irq_set irq = {
 52		.argsz = sizeof(irq),
 53		.flags = VFIO_IRQ_SET_ACTION_TRIGGER | VFIO_IRQ_SET_DATA_NONE,
 54		.index = index,
 55		.start = vector,
 56		.count = 1,
 57	};
 58
 59	ioctl_assert(device->fd, VFIO_DEVICE_SET_IRQS, &irq);
 60}
 61
 62static void check_supported_irq_index(u32 index)
 63{
 64	/* VFIO selftests only supports MSI and MSI-x for now. */
 65	VFIO_ASSERT_TRUE(index == VFIO_PCI_MSI_IRQ_INDEX ||
 66			 index == VFIO_PCI_MSIX_IRQ_INDEX,
 67			 "Unsupported IRQ index: %u\n", index);
 68}
 69
 70void vfio_pci_irq_enable(struct vfio_pci_device *device, u32 index, u32 vector,
 71			 int count)
 72{
 73	int i;
 74
 75	check_supported_irq_index(index);
 76
 77	for (i = vector; i < vector + count; i++) {
 78		VFIO_ASSERT_LT(device->msi_eventfds[i], 0);
 79		device->msi_eventfds[i] = eventfd(0, 0);
 80		VFIO_ASSERT_GE(device->msi_eventfds[i], 0);
 81	}
 82
 83	vfio_pci_irq_set(device, index, vector, count, device->msi_eventfds + vector);
 84}
 85
 86void vfio_pci_irq_disable(struct vfio_pci_device *device, u32 index)
 87{
 88	int i;
 89
 90	check_supported_irq_index(index);
 91
 92	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++) {
 93		if (device->msi_eventfds[i] < 0)
 94			continue;
 95
 96		VFIO_ASSERT_EQ(close(device->msi_eventfds[i]), 0);
 97		device->msi_eventfds[i] = -1;
 98	}
 99
100	vfio_pci_irq_set(device, index, 0, 0, NULL);
101}
102
103static void vfio_pci_irq_get(struct vfio_pci_device *device, u32 index,
104			     struct vfio_irq_info *irq_info)
105{
106	irq_info->argsz = sizeof(*irq_info);
107	irq_info->index = index;
108
109	ioctl_assert(device->fd, VFIO_DEVICE_GET_IRQ_INFO, irq_info);
110}
111
112static void vfio_pci_region_get(struct vfio_pci_device *device, int index,
113				struct vfio_region_info *info)
114{
115	memset(info, 0, sizeof(*info));
116
117	info->argsz = sizeof(*info);
118	info->index = index;
119
120	ioctl_assert(device->fd, VFIO_DEVICE_GET_REGION_INFO, info);
121}
122
123static void vfio_pci_bar_map(struct vfio_pci_device *device, int index)
124{
125	struct vfio_pci_bar *bar = &device->bars[index];
126	int prot = 0;
127
128	VFIO_ASSERT_LT(index, PCI_STD_NUM_BARS);
129	VFIO_ASSERT_NULL(bar->vaddr);
130	VFIO_ASSERT_TRUE(bar->info.flags & VFIO_REGION_INFO_FLAG_MMAP);
131
132	if (bar->info.flags & VFIO_REGION_INFO_FLAG_READ)
133		prot |= PROT_READ;
134	if (bar->info.flags & VFIO_REGION_INFO_FLAG_WRITE)
135		prot |= PROT_WRITE;
136
137	bar->vaddr = mmap(NULL, bar->info.size, prot, MAP_FILE | MAP_SHARED,
138			  device->fd, bar->info.offset);
139	VFIO_ASSERT_NE(bar->vaddr, MAP_FAILED);
140}
141
142static void vfio_pci_bar_unmap(struct vfio_pci_device *device, int index)
143{
144	struct vfio_pci_bar *bar = &device->bars[index];
145
146	VFIO_ASSERT_LT(index, PCI_STD_NUM_BARS);
147	VFIO_ASSERT_NOT_NULL(bar->vaddr);
148
149	VFIO_ASSERT_EQ(munmap(bar->vaddr, bar->info.size), 0);
150	bar->vaddr = NULL;
151}
152
153static void vfio_pci_bar_unmap_all(struct vfio_pci_device *device)
154{
155	int i;
156
157	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
158		if (device->bars[i].vaddr)
159			vfio_pci_bar_unmap(device, i);
160	}
161}
162
163void vfio_pci_config_access(struct vfio_pci_device *device, bool write,
164			    size_t config, size_t size, void *data)
165{
166	struct vfio_region_info *config_space = &device->config_space;
167	int ret;
168
169	if (write)
170		ret = pwrite(device->fd, data, size, config_space->offset + config);
171	else
172		ret = pread(device->fd, data, size, config_space->offset + config);
173
174	VFIO_ASSERT_EQ(ret, size, "Failed to %s PCI config space: 0x%lx\n",
175		       write ? "write to" : "read from", config);
176}
177
178void vfio_pci_device_reset(struct vfio_pci_device *device)
179{
180	ioctl_assert(device->fd, VFIO_DEVICE_RESET, NULL);
181}
182
183static unsigned int vfio_pci_get_group_from_dev(const char *bdf)
184{
185	char dev_iommu_group_path[PATH_MAX] = {0};
186	char sysfs_path[PATH_MAX] = {0};
187	unsigned int group;
188	int ret;
189
190	snprintf(sysfs_path, PATH_MAX, "%s/%s/iommu_group", PCI_SYSFS_PATH, bdf);
191
192	ret = readlink(sysfs_path, dev_iommu_group_path, sizeof(dev_iommu_group_path));
193	VFIO_ASSERT_NE(ret, -1, "Failed to get the IOMMU group for device: %s\n", bdf);
194
195	ret = sscanf(basename(dev_iommu_group_path), "%u", &group);
196	VFIO_ASSERT_EQ(ret, 1, "Failed to get the IOMMU group for device: %s\n", bdf);
197
198	return group;
199}
200
201static void vfio_pci_group_setup(struct vfio_pci_device *device, const char *bdf)
202{
203	struct vfio_group_status group_status = {
204		.argsz = sizeof(group_status),
205	};
206	char group_path[32];
207	int group;
208
209	group = vfio_pci_get_group_from_dev(bdf);
210	snprintf(group_path, sizeof(group_path), "/dev/vfio/%d", group);
211
212	device->group_fd = open(group_path, O_RDWR);
213	VFIO_ASSERT_GE(device->group_fd, 0, "open(%s) failed\n", group_path);
214
215	ioctl_assert(device->group_fd, VFIO_GROUP_GET_STATUS, &group_status);
216	VFIO_ASSERT_TRUE(group_status.flags & VFIO_GROUP_FLAGS_VIABLE);
217
218	ioctl_assert(device->group_fd, VFIO_GROUP_SET_CONTAINER, &device->iommu->container_fd);
219}
220
221static void vfio_pci_container_setup(struct vfio_pci_device *device, const char *bdf)
222{
223	struct iommu *iommu = device->iommu;
224	unsigned long iommu_type = iommu->mode->iommu_type;
225	int ret;
226
227	vfio_pci_group_setup(device, bdf);
228
229	ret = ioctl(iommu->container_fd, VFIO_CHECK_EXTENSION, iommu_type);
230	VFIO_ASSERT_GT(ret, 0, "VFIO IOMMU type %lu not supported\n", iommu_type);
231
232	/*
233	 * Allow multiple threads to race to set the IOMMU type on the
234	 * container. The first will succeed and the rest should fail
235	 * because the IOMMU type is already set.
236	 */
237	(void)ioctl(iommu->container_fd, VFIO_SET_IOMMU, (void *)iommu_type);
238
239	device->fd = ioctl(device->group_fd, VFIO_GROUP_GET_DEVICE_FD, bdf);
240	VFIO_ASSERT_GE(device->fd, 0);
241}
242
243static void vfio_pci_device_setup(struct vfio_pci_device *device)
244{
245	int i;
246
247	device->info.argsz = sizeof(device->info);
248	ioctl_assert(device->fd, VFIO_DEVICE_GET_INFO, &device->info);
249
250	vfio_pci_region_get(device, VFIO_PCI_CONFIG_REGION_INDEX, &device->config_space);
251
252	/* Sanity check VFIO does not advertise mmap for config space */
253	VFIO_ASSERT_TRUE(!(device->config_space.flags & VFIO_REGION_INFO_FLAG_MMAP),
254			 "PCI config space should not support mmap()\n");
255
256	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
257		struct vfio_pci_bar *bar = device->bars + i;
258
259		vfio_pci_region_get(device, i, &bar->info);
260		if (bar->info.flags & VFIO_REGION_INFO_FLAG_MMAP)
261			vfio_pci_bar_map(device, i);
262	}
263
264	vfio_pci_irq_get(device, VFIO_PCI_MSI_IRQ_INDEX, &device->msi_info);
265	vfio_pci_irq_get(device, VFIO_PCI_MSIX_IRQ_INDEX, &device->msix_info);
266
267	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++)
268		device->msi_eventfds[i] = -1;
269}
270
271const char *vfio_pci_get_cdev_path(const char *bdf)
272{
273	char dir_path[PATH_MAX];
274	struct dirent *entry;
275	char *cdev_path;
276	DIR *dir;
277
278	cdev_path = calloc(PATH_MAX, 1);
279	VFIO_ASSERT_NOT_NULL(cdev_path);
280
281	snprintf(dir_path, sizeof(dir_path), "/sys/bus/pci/devices/%s/vfio-dev/", bdf);
282
283	dir = opendir(dir_path);
284	VFIO_ASSERT_NOT_NULL(dir, "Failed to open directory %s\n", dir_path);
285
286	while ((entry = readdir(dir)) != NULL) {
287		/* Find the file that starts with "vfio" */
288		if (strncmp("vfio", entry->d_name, 4))
289			continue;
290
291		snprintf(cdev_path, PATH_MAX, "/dev/vfio/devices/%s", entry->d_name);
292		break;
293	}
294
295	VFIO_ASSERT_NE(cdev_path[0], 0, "Failed to find vfio cdev file.\n");
296	VFIO_ASSERT_EQ(closedir(dir), 0);
297
298	return cdev_path;
299}
300
301static void vfio_device_bind_iommufd(int device_fd, int iommufd)
302{
303	struct vfio_device_bind_iommufd args = {
304		.argsz = sizeof(args),
305		.iommufd = iommufd,
306	};
307
308	ioctl_assert(device_fd, VFIO_DEVICE_BIND_IOMMUFD, &args);
309}
310
311static void vfio_device_attach_iommufd_pt(int device_fd, u32 pt_id)
312{
313	struct vfio_device_attach_iommufd_pt args = {
314		.argsz = sizeof(args),
315		.pt_id = pt_id,
316	};
317
318	ioctl_assert(device_fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &args);
319}
320
321static void vfio_pci_iommufd_setup(struct vfio_pci_device *device, const char *bdf)
322{
323	const char *cdev_path = vfio_pci_get_cdev_path(bdf);
324
325	device->fd = open(cdev_path, O_RDWR);
326	VFIO_ASSERT_GE(device->fd, 0);
327	free((void *)cdev_path);
328
329	vfio_device_bind_iommufd(device->fd, device->iommu->iommufd);
330	vfio_device_attach_iommufd_pt(device->fd, device->iommu->ioas_id);
331}
332
333struct vfio_pci_device *vfio_pci_device_init(const char *bdf, struct iommu *iommu)
334{
335	struct vfio_pci_device *device;
336
337	device = calloc(1, sizeof(*device));
338	VFIO_ASSERT_NOT_NULL(device);
339
340	VFIO_ASSERT_NOT_NULL(iommu);
341	device->iommu = iommu;
342	device->bdf = bdf;
343
344	if (iommu->mode->container_path)
345		vfio_pci_container_setup(device, bdf);
346	else
347		vfio_pci_iommufd_setup(device, bdf);
348
349	vfio_pci_device_setup(device);
350	vfio_pci_driver_probe(device);
351
352	return device;
353}
354
355void vfio_pci_device_cleanup(struct vfio_pci_device *device)
356{
357	int i;
358
359	if (device->driver.initialized)
360		vfio_pci_driver_remove(device);
361
362	vfio_pci_bar_unmap_all(device);
363
364	VFIO_ASSERT_EQ(close(device->fd), 0);
365
366	for (i = 0; i < ARRAY_SIZE(device->msi_eventfds); i++) {
367		if (device->msi_eventfds[i] < 0)
368			continue;
369
370		VFIO_ASSERT_EQ(close(device->msi_eventfds[i]), 0);
371	}
372
373	if (device->group_fd)
374		VFIO_ASSERT_EQ(close(device->group_fd), 0);
375
376	free(device);
377}