]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * vfio based device assignment support | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Alex Williamson <alex.williamson@redhat.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | * Based on qemu-kvm device-assignment: | |
13 | * Adapted for KVM by Qumranet. | |
14 | * Copyright (c) 2007, Neocleus, Alex Novik (alex@neocleus.com) | |
15 | * Copyright (c) 2007, Neocleus, Guy Zana (guy@neocleus.com) | |
16 | * Copyright (C) 2008, Qumranet, Amit Shah (amit.shah@qumranet.com) | |
17 | * Copyright (C) 2008, Red Hat, Amit Shah (amit.shah@redhat.com) | |
18 | * Copyright (C) 2008, IBM, Muli Ben-Yehuda (muli@il.ibm.com) | |
19 | */ | |
20 | ||
21 | #include <dirent.h> | |
22 | #include <unistd.h> | |
23 | #include <sys/ioctl.h> | |
24 | #include <sys/mman.h> | |
25 | #include <sys/stat.h> | |
26 | #include <sys/types.h> | |
27 | #include <linux/vfio.h> | |
28 | ||
29 | #include "config.h" | |
30 | #include "event_notifier.h" | |
31 | #include "exec/address-spaces.h" | |
32 | #include "kvm.h" | |
33 | #include "exec/memory.h" | |
34 | #include "pci/msi.h" | |
35 | #include "pci/msix.h" | |
36 | #include "pci/pci.h" | |
37 | #include "qemu-common.h" | |
38 | #include "qemu-error.h" | |
39 | #include "qemu-queue.h" | |
40 | #include "range.h" | |
41 | ||
42 | /* #define DEBUG_VFIO */ | |
43 | #ifdef DEBUG_VFIO | |
44 | #define DPRINTF(fmt, ...) \ | |
45 | do { fprintf(stderr, "vfio: " fmt, ## __VA_ARGS__); } while (0) | |
46 | #else | |
47 | #define DPRINTF(fmt, ...) \ | |
48 | do { } while (0) | |
49 | #endif | |
50 | ||
51 | typedef struct VFIOBAR { | |
52 | off_t fd_offset; /* offset of BAR within device fd */ | |
53 | int fd; /* device fd, allows us to pass VFIOBAR as opaque data */ | |
54 | MemoryRegion mem; /* slow, read/write access */ | |
55 | MemoryRegion mmap_mem; /* direct mapped access */ | |
56 | void *mmap; | |
57 | size_t size; | |
58 | uint32_t flags; /* VFIO region flags (rd/wr/mmap) */ | |
59 | uint8_t nr; /* cache the BAR number for debug */ | |
60 | } VFIOBAR; | |
61 | ||
62 | typedef struct VFIOINTx { | |
63 | bool pending; /* interrupt pending */ | |
64 | bool kvm_accel; /* set when QEMU bypass through KVM enabled */ | |
65 | uint8_t pin; /* which pin to pull for qemu_set_irq */ | |
66 | EventNotifier interrupt; /* eventfd triggered on interrupt */ | |
67 | EventNotifier unmask; /* eventfd for unmask on QEMU bypass */ | |
68 | PCIINTxRoute route; /* routing info for QEMU bypass */ | |
69 | uint32_t mmap_timeout; /* delay to re-enable mmaps after interrupt */ | |
70 | QEMUTimer *mmap_timer; /* enable mmaps after periods w/o interrupts */ | |
71 | } VFIOINTx; | |
72 | ||
73 | struct VFIODevice; | |
74 | ||
75 | typedef struct VFIOMSIVector { | |
76 | EventNotifier interrupt; /* eventfd triggered on interrupt */ | |
77 | struct VFIODevice *vdev; /* back pointer to device */ | |
78 | int virq; /* KVM irqchip route for QEMU bypass */ | |
79 | bool use; | |
80 | } VFIOMSIVector; | |
81 | ||
82 | enum { | |
83 | VFIO_INT_NONE = 0, | |
84 | VFIO_INT_INTx = 1, | |
85 | VFIO_INT_MSI = 2, | |
86 | VFIO_INT_MSIX = 3, | |
87 | }; | |
88 | ||
89 | struct VFIOGroup; | |
90 | ||
91 | typedef struct VFIOContainer { | |
92 | int fd; /* /dev/vfio/vfio, empowered by the attached groups */ | |
93 | struct { | |
94 | /* enable abstraction to support various iommu backends */ | |
95 | union { | |
96 | MemoryListener listener; /* Used by type1 iommu */ | |
97 | }; | |
98 | void (*release)(struct VFIOContainer *); | |
99 | } iommu_data; | |
100 | QLIST_HEAD(, VFIOGroup) group_list; | |
101 | QLIST_ENTRY(VFIOContainer) next; | |
102 | } VFIOContainer; | |
103 | ||
104 | /* Cache of MSI-X setup plus extra mmap and memory region for split BAR map */ | |
105 | typedef struct VFIOMSIXInfo { | |
106 | uint8_t table_bar; | |
107 | uint8_t pba_bar; | |
108 | uint16_t entries; | |
109 | uint32_t table_offset; | |
110 | uint32_t pba_offset; | |
111 | MemoryRegion mmap_mem; | |
112 | void *mmap; | |
113 | } VFIOMSIXInfo; | |
114 | ||
115 | typedef struct VFIODevice { | |
116 | PCIDevice pdev; | |
117 | int fd; | |
118 | VFIOINTx intx; | |
119 | unsigned int config_size; | |
120 | off_t config_offset; /* Offset of config space region within device fd */ | |
121 | unsigned int rom_size; | |
122 | off_t rom_offset; /* Offset of ROM region within device fd */ | |
123 | int msi_cap_size; | |
124 | VFIOMSIVector *msi_vectors; | |
125 | VFIOMSIXInfo *msix; | |
126 | int nr_vectors; /* Number of MSI/MSIX vectors currently in use */ | |
127 | int interrupt; /* Current interrupt type */ | |
128 | VFIOBAR bars[PCI_NUM_REGIONS - 1]; /* No ROM */ | |
129 | PCIHostDeviceAddress host; | |
130 | QLIST_ENTRY(VFIODevice) next; | |
131 | struct VFIOGroup *group; | |
132 | bool reset_works; | |
133 | } VFIODevice; | |
134 | ||
135 | typedef struct VFIOGroup { | |
136 | int fd; | |
137 | int groupid; | |
138 | VFIOContainer *container; | |
139 | QLIST_HEAD(, VFIODevice) device_list; | |
140 | QLIST_ENTRY(VFIOGroup) next; | |
141 | QLIST_ENTRY(VFIOGroup) container_next; | |
142 | } VFIOGroup; | |
143 | ||
144 | #define MSIX_CAP_LENGTH 12 | |
145 | ||
146 | static QLIST_HEAD(, VFIOContainer) | |
147 | container_list = QLIST_HEAD_INITIALIZER(container_list); | |
148 | ||
149 | static QLIST_HEAD(, VFIOGroup) | |
150 | group_list = QLIST_HEAD_INITIALIZER(group_list); | |
151 | ||
152 | static void vfio_disable_interrupts(VFIODevice *vdev); | |
153 | static uint32_t vfio_pci_read_config(PCIDevice *pdev, uint32_t addr, int len); | |
154 | static void vfio_mmap_set_enabled(VFIODevice *vdev, bool enabled); | |
155 | ||
156 | /* | |
157 | * Common VFIO interrupt disable | |
158 | */ | |
159 | static void vfio_disable_irqindex(VFIODevice *vdev, int index) | |
160 | { | |
161 | struct vfio_irq_set irq_set = { | |
162 | .argsz = sizeof(irq_set), | |
163 | .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER, | |
164 | .index = index, | |
165 | .start = 0, | |
166 | .count = 0, | |
167 | }; | |
168 | ||
169 | ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, &irq_set); | |
170 | } | |
171 | ||
172 | /* | |
173 | * INTx | |
174 | */ | |
175 | static void vfio_unmask_intx(VFIODevice *vdev) | |
176 | { | |
177 | struct vfio_irq_set irq_set = { | |
178 | .argsz = sizeof(irq_set), | |
179 | .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK, | |
180 | .index = VFIO_PCI_INTX_IRQ_INDEX, | |
181 | .start = 0, | |
182 | .count = 1, | |
183 | }; | |
184 | ||
185 | ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, &irq_set); | |
186 | } | |
187 | ||
188 | #ifdef CONFIG_KVM /* Unused outside of CONFIG_KVM code */ | |
189 | static void vfio_mask_intx(VFIODevice *vdev) | |
190 | { | |
191 | struct vfio_irq_set irq_set = { | |
192 | .argsz = sizeof(irq_set), | |
193 | .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_MASK, | |
194 | .index = VFIO_PCI_INTX_IRQ_INDEX, | |
195 | .start = 0, | |
196 | .count = 1, | |
197 | }; | |
198 | ||
199 | ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, &irq_set); | |
200 | } | |
201 | #endif | |
202 | ||
203 | /* | |
204 | * Disabling BAR mmaping can be slow, but toggling it around INTx can | |
205 | * also be a huge overhead. We try to get the best of both worlds by | |
206 | * waiting until an interrupt to disable mmaps (subsequent transitions | |
207 | * to the same state are effectively no overhead). If the interrupt has | |
208 | * been serviced and the time gap is long enough, we re-enable mmaps for | |
209 | * performance. This works well for things like graphics cards, which | |
210 | * may not use their interrupt at all and are penalized to an unusable | |
211 | * level by read/write BAR traps. Other devices, like NICs, have more | |
212 | * regular interrupts and see much better latency by staying in non-mmap | |
213 | * mode. We therefore set the default mmap_timeout such that a ping | |
214 | * is just enough to keep the mmap disabled. Users can experiment with | |
215 | * other options with the x-intx-mmap-timeout-ms parameter (a value of | |
216 | * zero disables the timer). | |
217 | */ | |
218 | static void vfio_intx_mmap_enable(void *opaque) | |
219 | { | |
220 | VFIODevice *vdev = opaque; | |
221 | ||
222 | if (vdev->intx.pending) { | |
223 | qemu_mod_timer(vdev->intx.mmap_timer, | |
224 | qemu_get_clock_ms(vm_clock) + vdev->intx.mmap_timeout); | |
225 | return; | |
226 | } | |
227 | ||
228 | vfio_mmap_set_enabled(vdev, true); | |
229 | } | |
230 | ||
231 | static void vfio_intx_interrupt(void *opaque) | |
232 | { | |
233 | VFIODevice *vdev = opaque; | |
234 | ||
235 | if (!event_notifier_test_and_clear(&vdev->intx.interrupt)) { | |
236 | return; | |
237 | } | |
238 | ||
239 | DPRINTF("%s(%04x:%02x:%02x.%x) Pin %c\n", __func__, vdev->host.domain, | |
240 | vdev->host.bus, vdev->host.slot, vdev->host.function, | |
241 | 'A' + vdev->intx.pin); | |
242 | ||
243 | vdev->intx.pending = true; | |
244 | qemu_set_irq(vdev->pdev.irq[vdev->intx.pin], 1); | |
245 | vfio_mmap_set_enabled(vdev, false); | |
246 | if (vdev->intx.mmap_timeout) { | |
247 | qemu_mod_timer(vdev->intx.mmap_timer, | |
248 | qemu_get_clock_ms(vm_clock) + vdev->intx.mmap_timeout); | |
249 | } | |
250 | } | |
251 | ||
252 | static void vfio_eoi(VFIODevice *vdev) | |
253 | { | |
254 | if (!vdev->intx.pending) { | |
255 | return; | |
256 | } | |
257 | ||
258 | DPRINTF("%s(%04x:%02x:%02x.%x) EOI\n", __func__, vdev->host.domain, | |
259 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
260 | ||
261 | vdev->intx.pending = false; | |
262 | qemu_set_irq(vdev->pdev.irq[vdev->intx.pin], 0); | |
263 | vfio_unmask_intx(vdev); | |
264 | } | |
265 | ||
266 | static void vfio_enable_intx_kvm(VFIODevice *vdev) | |
267 | { | |
268 | #ifdef CONFIG_KVM | |
269 | struct kvm_irqfd irqfd = { | |
270 | .fd = event_notifier_get_fd(&vdev->intx.interrupt), | |
271 | .gsi = vdev->intx.route.irq, | |
272 | .flags = KVM_IRQFD_FLAG_RESAMPLE, | |
273 | }; | |
274 | struct vfio_irq_set *irq_set; | |
275 | int ret, argsz; | |
276 | int32_t *pfd; | |
277 | ||
278 | if (!kvm_irqfds_enabled() || | |
279 | vdev->intx.route.mode != PCI_INTX_ENABLED || | |
280 | !kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) { | |
281 | return; | |
282 | } | |
283 | ||
284 | /* Get to a known interrupt state */ | |
285 | qemu_set_fd_handler(irqfd.fd, NULL, NULL, vdev); | |
286 | vfio_mask_intx(vdev); | |
287 | vdev->intx.pending = false; | |
288 | qemu_set_irq(vdev->pdev.irq[vdev->intx.pin], 0); | |
289 | ||
290 | /* Get an eventfd for resample/unmask */ | |
291 | if (event_notifier_init(&vdev->intx.unmask, 0)) { | |
292 | error_report("vfio: Error: event_notifier_init failed eoi\n"); | |
293 | goto fail; | |
294 | } | |
295 | ||
296 | /* KVM triggers it, VFIO listens for it */ | |
297 | irqfd.resamplefd = event_notifier_get_fd(&vdev->intx.unmask); | |
298 | ||
299 | if (kvm_vm_ioctl(kvm_state, KVM_IRQFD, &irqfd)) { | |
300 | error_report("vfio: Error: Failed to setup resample irqfd: %m\n"); | |
301 | goto fail_irqfd; | |
302 | } | |
303 | ||
304 | argsz = sizeof(*irq_set) + sizeof(*pfd); | |
305 | ||
306 | irq_set = g_malloc0(argsz); | |
307 | irq_set->argsz = argsz; | |
308 | irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_UNMASK; | |
309 | irq_set->index = VFIO_PCI_INTX_IRQ_INDEX; | |
310 | irq_set->start = 0; | |
311 | irq_set->count = 1; | |
312 | pfd = (int32_t *)&irq_set->data; | |
313 | ||
314 | *pfd = irqfd.resamplefd; | |
315 | ||
316 | ret = ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set); | |
317 | g_free(irq_set); | |
318 | if (ret) { | |
319 | error_report("vfio: Error: Failed to setup INTx unmask fd: %m\n"); | |
320 | goto fail_vfio; | |
321 | } | |
322 | ||
323 | /* Let'em rip */ | |
324 | vfio_unmask_intx(vdev); | |
325 | ||
326 | vdev->intx.kvm_accel = true; | |
327 | ||
328 | DPRINTF("%s(%04x:%02x:%02x.%x) KVM INTx accel enabled\n", | |
329 | __func__, vdev->host.domain, vdev->host.bus, | |
330 | vdev->host.slot, vdev->host.function); | |
331 | ||
332 | return; | |
333 | ||
334 | fail_vfio: | |
335 | irqfd.flags = KVM_IRQFD_FLAG_DEASSIGN; | |
336 | kvm_vm_ioctl(kvm_state, KVM_IRQFD, &irqfd); | |
337 | fail_irqfd: | |
338 | event_notifier_cleanup(&vdev->intx.unmask); | |
339 | fail: | |
340 | qemu_set_fd_handler(irqfd.fd, vfio_intx_interrupt, NULL, vdev); | |
341 | vfio_unmask_intx(vdev); | |
342 | #endif | |
343 | } | |
344 | ||
345 | static void vfio_disable_intx_kvm(VFIODevice *vdev) | |
346 | { | |
347 | #ifdef CONFIG_KVM | |
348 | struct kvm_irqfd irqfd = { | |
349 | .fd = event_notifier_get_fd(&vdev->intx.interrupt), | |
350 | .gsi = vdev->intx.route.irq, | |
351 | .flags = KVM_IRQFD_FLAG_DEASSIGN, | |
352 | }; | |
353 | ||
354 | if (!vdev->intx.kvm_accel) { | |
355 | return; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Get to a known state, hardware masked, QEMU ready to accept new | |
360 | * interrupts, QEMU IRQ de-asserted. | |
361 | */ | |
362 | vfio_mask_intx(vdev); | |
363 | vdev->intx.pending = false; | |
364 | qemu_set_irq(vdev->pdev.irq[vdev->intx.pin], 0); | |
365 | ||
366 | /* Tell KVM to stop listening for an INTx irqfd */ | |
367 | if (kvm_vm_ioctl(kvm_state, KVM_IRQFD, &irqfd)) { | |
368 | error_report("vfio: Error: Failed to disable INTx irqfd: %m\n"); | |
369 | } | |
370 | ||
371 | /* We only need to close the eventfd for VFIO to cleanup the kernel side */ | |
372 | event_notifier_cleanup(&vdev->intx.unmask); | |
373 | ||
374 | /* QEMU starts listening for interrupt events. */ | |
375 | qemu_set_fd_handler(irqfd.fd, vfio_intx_interrupt, NULL, vdev); | |
376 | ||
377 | vdev->intx.kvm_accel = false; | |
378 | ||
379 | /* If we've missed an event, let it re-fire through QEMU */ | |
380 | vfio_unmask_intx(vdev); | |
381 | ||
382 | DPRINTF("%s(%04x:%02x:%02x.%x) KVM INTx accel disabled\n", | |
383 | __func__, vdev->host.domain, vdev->host.bus, | |
384 | vdev->host.slot, vdev->host.function); | |
385 | #endif | |
386 | } | |
387 | ||
388 | static void vfio_update_irq(PCIDevice *pdev) | |
389 | { | |
390 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
391 | PCIINTxRoute route; | |
392 | ||
393 | if (vdev->interrupt != VFIO_INT_INTx) { | |
394 | return; | |
395 | } | |
396 | ||
397 | route = pci_device_route_intx_to_irq(&vdev->pdev, vdev->intx.pin); | |
398 | ||
399 | if (!pci_intx_route_changed(&vdev->intx.route, &route)) { | |
400 | return; /* Nothing changed */ | |
401 | } | |
402 | ||
403 | DPRINTF("%s(%04x:%02x:%02x.%x) IRQ moved %d -> %d\n", __func__, | |
404 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
405 | vdev->host.function, vdev->intx.route.irq, route.irq); | |
406 | ||
407 | vfio_disable_intx_kvm(vdev); | |
408 | ||
409 | vdev->intx.route = route; | |
410 | ||
411 | if (route.mode != PCI_INTX_ENABLED) { | |
412 | return; | |
413 | } | |
414 | ||
415 | vfio_enable_intx_kvm(vdev); | |
416 | ||
417 | /* Re-enable the interrupt in cased we missed an EOI */ | |
418 | vfio_eoi(vdev); | |
419 | } | |
420 | ||
421 | static int vfio_enable_intx(VFIODevice *vdev) | |
422 | { | |
423 | uint8_t pin = vfio_pci_read_config(&vdev->pdev, PCI_INTERRUPT_PIN, 1); | |
424 | int ret, argsz; | |
425 | struct vfio_irq_set *irq_set; | |
426 | int32_t *pfd; | |
427 | ||
428 | if (!pin) { | |
429 | return 0; | |
430 | } | |
431 | ||
432 | vfio_disable_interrupts(vdev); | |
433 | ||
434 | vdev->intx.pin = pin - 1; /* Pin A (1) -> irq[0] */ | |
435 | ||
436 | #ifdef CONFIG_KVM | |
437 | /* | |
438 | * Only conditional to avoid generating error messages on platforms | |
439 | * where we won't actually use the result anyway. | |
440 | */ | |
441 | if (kvm_irqfds_enabled() && | |
442 | kvm_check_extension(kvm_state, KVM_CAP_IRQFD_RESAMPLE)) { | |
443 | vdev->intx.route = pci_device_route_intx_to_irq(&vdev->pdev, | |
444 | vdev->intx.pin); | |
445 | } | |
446 | #endif | |
447 | ||
448 | ret = event_notifier_init(&vdev->intx.interrupt, 0); | |
449 | if (ret) { | |
450 | error_report("vfio: Error: event_notifier_init failed\n"); | |
451 | return ret; | |
452 | } | |
453 | ||
454 | argsz = sizeof(*irq_set) + sizeof(*pfd); | |
455 | ||
456 | irq_set = g_malloc0(argsz); | |
457 | irq_set->argsz = argsz; | |
458 | irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER; | |
459 | irq_set->index = VFIO_PCI_INTX_IRQ_INDEX; | |
460 | irq_set->start = 0; | |
461 | irq_set->count = 1; | |
462 | pfd = (int32_t *)&irq_set->data; | |
463 | ||
464 | *pfd = event_notifier_get_fd(&vdev->intx.interrupt); | |
465 | qemu_set_fd_handler(*pfd, vfio_intx_interrupt, NULL, vdev); | |
466 | ||
467 | ret = ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set); | |
468 | g_free(irq_set); | |
469 | if (ret) { | |
470 | error_report("vfio: Error: Failed to setup INTx fd: %m\n"); | |
471 | qemu_set_fd_handler(*pfd, NULL, NULL, vdev); | |
472 | event_notifier_cleanup(&vdev->intx.interrupt); | |
473 | return -errno; | |
474 | } | |
475 | ||
476 | vfio_enable_intx_kvm(vdev); | |
477 | ||
478 | vdev->interrupt = VFIO_INT_INTx; | |
479 | ||
480 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
481 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
482 | ||
483 | return 0; | |
484 | } | |
485 | ||
486 | static void vfio_disable_intx(VFIODevice *vdev) | |
487 | { | |
488 | int fd; | |
489 | ||
490 | qemu_del_timer(vdev->intx.mmap_timer); | |
491 | vfio_disable_intx_kvm(vdev); | |
492 | vfio_disable_irqindex(vdev, VFIO_PCI_INTX_IRQ_INDEX); | |
493 | vdev->intx.pending = false; | |
494 | qemu_set_irq(vdev->pdev.irq[vdev->intx.pin], 0); | |
495 | vfio_mmap_set_enabled(vdev, true); | |
496 | ||
497 | fd = event_notifier_get_fd(&vdev->intx.interrupt); | |
498 | qemu_set_fd_handler(fd, NULL, NULL, vdev); | |
499 | event_notifier_cleanup(&vdev->intx.interrupt); | |
500 | ||
501 | vdev->interrupt = VFIO_INT_NONE; | |
502 | ||
503 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
504 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
505 | } | |
506 | ||
507 | /* | |
508 | * MSI/X | |
509 | */ | |
510 | static void vfio_msi_interrupt(void *opaque) | |
511 | { | |
512 | VFIOMSIVector *vector = opaque; | |
513 | VFIODevice *vdev = vector->vdev; | |
514 | int nr = vector - vdev->msi_vectors; | |
515 | ||
516 | if (!event_notifier_test_and_clear(&vector->interrupt)) { | |
517 | return; | |
518 | } | |
519 | ||
520 | DPRINTF("%s(%04x:%02x:%02x.%x) vector %d\n", __func__, | |
521 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
522 | vdev->host.function, nr); | |
523 | ||
524 | if (vdev->interrupt == VFIO_INT_MSIX) { | |
525 | msix_notify(&vdev->pdev, nr); | |
526 | } else if (vdev->interrupt == VFIO_INT_MSI) { | |
527 | msi_notify(&vdev->pdev, nr); | |
528 | } else { | |
529 | error_report("vfio: MSI interrupt receieved, but not enabled?\n"); | |
530 | } | |
531 | } | |
532 | ||
533 | static int vfio_enable_vectors(VFIODevice *vdev, bool msix) | |
534 | { | |
535 | struct vfio_irq_set *irq_set; | |
536 | int ret = 0, i, argsz; | |
537 | int32_t *fds; | |
538 | ||
539 | argsz = sizeof(*irq_set) + (vdev->nr_vectors * sizeof(*fds)); | |
540 | ||
541 | irq_set = g_malloc0(argsz); | |
542 | irq_set->argsz = argsz; | |
543 | irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER; | |
544 | irq_set->index = msix ? VFIO_PCI_MSIX_IRQ_INDEX : VFIO_PCI_MSI_IRQ_INDEX; | |
545 | irq_set->start = 0; | |
546 | irq_set->count = vdev->nr_vectors; | |
547 | fds = (int32_t *)&irq_set->data; | |
548 | ||
549 | for (i = 0; i < vdev->nr_vectors; i++) { | |
550 | if (!vdev->msi_vectors[i].use) { | |
551 | fds[i] = -1; | |
552 | continue; | |
553 | } | |
554 | ||
555 | fds[i] = event_notifier_get_fd(&vdev->msi_vectors[i].interrupt); | |
556 | } | |
557 | ||
558 | ret = ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set); | |
559 | ||
560 | g_free(irq_set); | |
561 | ||
562 | return ret; | |
563 | } | |
564 | ||
565 | static int vfio_msix_vector_use(PCIDevice *pdev, | |
566 | unsigned int nr, MSIMessage msg) | |
567 | { | |
568 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
569 | VFIOMSIVector *vector; | |
570 | int ret; | |
571 | ||
572 | DPRINTF("%s(%04x:%02x:%02x.%x) vector %d used\n", __func__, | |
573 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
574 | vdev->host.function, nr); | |
575 | ||
576 | vector = &vdev->msi_vectors[nr]; | |
577 | vector->vdev = vdev; | |
578 | vector->use = true; | |
579 | ||
580 | msix_vector_use(pdev, nr); | |
581 | ||
582 | if (event_notifier_init(&vector->interrupt, 0)) { | |
583 | error_report("vfio: Error: event_notifier_init failed\n"); | |
584 | } | |
585 | ||
586 | /* | |
587 | * Attempt to enable route through KVM irqchip, | |
588 | * default to userspace handling if unavailable. | |
589 | */ | |
590 | vector->virq = kvm_irqchip_add_msi_route(kvm_state, msg); | |
591 | if (vector->virq < 0 || | |
592 | kvm_irqchip_add_irqfd_notifier(kvm_state, &vector->interrupt, | |
593 | vector->virq) < 0) { | |
594 | if (vector->virq >= 0) { | |
595 | kvm_irqchip_release_virq(kvm_state, vector->virq); | |
596 | vector->virq = -1; | |
597 | } | |
598 | qemu_set_fd_handler(event_notifier_get_fd(&vector->interrupt), | |
599 | vfio_msi_interrupt, NULL, vector); | |
600 | } | |
601 | ||
602 | /* | |
603 | * We don't want to have the host allocate all possible MSI vectors | |
604 | * for a device if they're not in use, so we shutdown and incrementally | |
605 | * increase them as needed. | |
606 | */ | |
607 | if (vdev->nr_vectors < nr + 1) { | |
608 | vfio_disable_irqindex(vdev, VFIO_PCI_MSIX_IRQ_INDEX); | |
609 | vdev->nr_vectors = nr + 1; | |
610 | ret = vfio_enable_vectors(vdev, true); | |
611 | if (ret) { | |
612 | error_report("vfio: failed to enable vectors, %d\n", ret); | |
613 | } | |
614 | } else { | |
615 | int argsz; | |
616 | struct vfio_irq_set *irq_set; | |
617 | int32_t *pfd; | |
618 | ||
619 | argsz = sizeof(*irq_set) + sizeof(*pfd); | |
620 | ||
621 | irq_set = g_malloc0(argsz); | |
622 | irq_set->argsz = argsz; | |
623 | irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | | |
624 | VFIO_IRQ_SET_ACTION_TRIGGER; | |
625 | irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX; | |
626 | irq_set->start = nr; | |
627 | irq_set->count = 1; | |
628 | pfd = (int32_t *)&irq_set->data; | |
629 | ||
630 | *pfd = event_notifier_get_fd(&vector->interrupt); | |
631 | ||
632 | ret = ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set); | |
633 | g_free(irq_set); | |
634 | if (ret) { | |
635 | error_report("vfio: failed to modify vector, %d\n", ret); | |
636 | } | |
637 | } | |
638 | ||
639 | return 0; | |
640 | } | |
641 | ||
642 | static void vfio_msix_vector_release(PCIDevice *pdev, unsigned int nr) | |
643 | { | |
644 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
645 | VFIOMSIVector *vector = &vdev->msi_vectors[nr]; | |
646 | int argsz; | |
647 | struct vfio_irq_set *irq_set; | |
648 | int32_t *pfd; | |
649 | ||
650 | DPRINTF("%s(%04x:%02x:%02x.%x) vector %d released\n", __func__, | |
651 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
652 | vdev->host.function, nr); | |
653 | ||
654 | /* | |
655 | * XXX What's the right thing to do here? This turns off the interrupt | |
656 | * completely, but do we really just want to switch the interrupt to | |
657 | * bouncing through userspace and let msix.c drop it? Not sure. | |
658 | */ | |
659 | msix_vector_unuse(pdev, nr); | |
660 | ||
661 | argsz = sizeof(*irq_set) + sizeof(*pfd); | |
662 | ||
663 | irq_set = g_malloc0(argsz); | |
664 | irq_set->argsz = argsz; | |
665 | irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | | |
666 | VFIO_IRQ_SET_ACTION_TRIGGER; | |
667 | irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX; | |
668 | irq_set->start = nr; | |
669 | irq_set->count = 1; | |
670 | pfd = (int32_t *)&irq_set->data; | |
671 | ||
672 | *pfd = -1; | |
673 | ||
674 | ioctl(vdev->fd, VFIO_DEVICE_SET_IRQS, irq_set); | |
675 | ||
676 | g_free(irq_set); | |
677 | ||
678 | if (vector->virq < 0) { | |
679 | qemu_set_fd_handler(event_notifier_get_fd(&vector->interrupt), | |
680 | NULL, NULL, NULL); | |
681 | } else { | |
682 | kvm_irqchip_remove_irqfd_notifier(kvm_state, &vector->interrupt, | |
683 | vector->virq); | |
684 | kvm_irqchip_release_virq(kvm_state, vector->virq); | |
685 | vector->virq = -1; | |
686 | } | |
687 | ||
688 | event_notifier_cleanup(&vector->interrupt); | |
689 | vector->use = false; | |
690 | } | |
691 | ||
692 | static void vfio_enable_msix(VFIODevice *vdev) | |
693 | { | |
694 | vfio_disable_interrupts(vdev); | |
695 | ||
696 | vdev->msi_vectors = g_malloc0(vdev->msix->entries * sizeof(VFIOMSIVector)); | |
697 | ||
698 | vdev->interrupt = VFIO_INT_MSIX; | |
699 | ||
700 | if (msix_set_vector_notifiers(&vdev->pdev, vfio_msix_vector_use, | |
701 | vfio_msix_vector_release)) { | |
702 | error_report("vfio: msix_set_vector_notifiers failed\n"); | |
703 | } | |
704 | ||
705 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
706 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
707 | } | |
708 | ||
709 | static void vfio_enable_msi(VFIODevice *vdev) | |
710 | { | |
711 | int ret, i; | |
712 | ||
713 | vfio_disable_interrupts(vdev); | |
714 | ||
715 | vdev->nr_vectors = msi_nr_vectors_allocated(&vdev->pdev); | |
716 | retry: | |
717 | vdev->msi_vectors = g_malloc0(vdev->nr_vectors * sizeof(VFIOMSIVector)); | |
718 | ||
719 | for (i = 0; i < vdev->nr_vectors; i++) { | |
720 | MSIMessage msg; | |
721 | VFIOMSIVector *vector = &vdev->msi_vectors[i]; | |
722 | ||
723 | vector->vdev = vdev; | |
724 | vector->use = true; | |
725 | ||
726 | if (event_notifier_init(&vector->interrupt, 0)) { | |
727 | error_report("vfio: Error: event_notifier_init failed\n"); | |
728 | } | |
729 | ||
730 | msg = msi_get_message(&vdev->pdev, i); | |
731 | ||
732 | /* | |
733 | * Attempt to enable route through KVM irqchip, | |
734 | * default to userspace handling if unavailable. | |
735 | */ | |
736 | vector->virq = kvm_irqchip_add_msi_route(kvm_state, msg); | |
737 | if (vector->virq < 0 || | |
738 | kvm_irqchip_add_irqfd_notifier(kvm_state, &vector->interrupt, | |
739 | vector->virq) < 0) { | |
740 | qemu_set_fd_handler(event_notifier_get_fd(&vector->interrupt), | |
741 | vfio_msi_interrupt, NULL, vector); | |
742 | } | |
743 | } | |
744 | ||
745 | ret = vfio_enable_vectors(vdev, false); | |
746 | if (ret) { | |
747 | if (ret < 0) { | |
748 | error_report("vfio: Error: Failed to setup MSI fds: %m\n"); | |
749 | } else if (ret != vdev->nr_vectors) { | |
750 | error_report("vfio: Error: Failed to enable %d " | |
751 | "MSI vectors, retry with %d\n", vdev->nr_vectors, ret); | |
752 | } | |
753 | ||
754 | for (i = 0; i < vdev->nr_vectors; i++) { | |
755 | VFIOMSIVector *vector = &vdev->msi_vectors[i]; | |
756 | if (vector->virq >= 0) { | |
757 | kvm_irqchip_remove_irqfd_notifier(kvm_state, &vector->interrupt, | |
758 | vector->virq); | |
759 | kvm_irqchip_release_virq(kvm_state, vector->virq); | |
760 | vector->virq = -1; | |
761 | } else { | |
762 | qemu_set_fd_handler(event_notifier_get_fd(&vector->interrupt), | |
763 | NULL, NULL, NULL); | |
764 | } | |
765 | event_notifier_cleanup(&vector->interrupt); | |
766 | } | |
767 | ||
768 | g_free(vdev->msi_vectors); | |
769 | ||
770 | if (ret > 0 && ret != vdev->nr_vectors) { | |
771 | vdev->nr_vectors = ret; | |
772 | goto retry; | |
773 | } | |
774 | vdev->nr_vectors = 0; | |
775 | ||
776 | return; | |
777 | } | |
778 | ||
779 | vdev->interrupt = VFIO_INT_MSI; | |
780 | ||
781 | DPRINTF("%s(%04x:%02x:%02x.%x) Enabled %d MSI vectors\n", __func__, | |
782 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
783 | vdev->host.function, vdev->nr_vectors); | |
784 | } | |
785 | ||
786 | static void vfio_disable_msi_common(VFIODevice *vdev) | |
787 | { | |
788 | g_free(vdev->msi_vectors); | |
789 | vdev->msi_vectors = NULL; | |
790 | vdev->nr_vectors = 0; | |
791 | vdev->interrupt = VFIO_INT_NONE; | |
792 | ||
793 | vfio_enable_intx(vdev); | |
794 | } | |
795 | ||
796 | static void vfio_disable_msix(VFIODevice *vdev) | |
797 | { | |
798 | msix_unset_vector_notifiers(&vdev->pdev); | |
799 | ||
800 | if (vdev->nr_vectors) { | |
801 | vfio_disable_irqindex(vdev, VFIO_PCI_MSIX_IRQ_INDEX); | |
802 | } | |
803 | ||
804 | vfio_disable_msi_common(vdev); | |
805 | ||
806 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
807 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
808 | } | |
809 | ||
810 | static void vfio_disable_msi(VFIODevice *vdev) | |
811 | { | |
812 | int i; | |
813 | ||
814 | vfio_disable_irqindex(vdev, VFIO_PCI_MSI_IRQ_INDEX); | |
815 | ||
816 | for (i = 0; i < vdev->nr_vectors; i++) { | |
817 | VFIOMSIVector *vector = &vdev->msi_vectors[i]; | |
818 | ||
819 | if (!vector->use) { | |
820 | continue; | |
821 | } | |
822 | ||
823 | if (vector->virq >= 0) { | |
824 | kvm_irqchip_remove_irqfd_notifier(kvm_state, | |
825 | &vector->interrupt, vector->virq); | |
826 | kvm_irqchip_release_virq(kvm_state, vector->virq); | |
827 | vector->virq = -1; | |
828 | } else { | |
829 | qemu_set_fd_handler(event_notifier_get_fd(&vector->interrupt), | |
830 | NULL, NULL, NULL); | |
831 | } | |
832 | ||
833 | event_notifier_cleanup(&vector->interrupt); | |
834 | } | |
835 | ||
836 | vfio_disable_msi_common(vdev); | |
837 | ||
838 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
839 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
840 | } | |
841 | ||
842 | /* | |
843 | * IO Port/MMIO - Beware of the endians, VFIO is always little endian | |
844 | */ | |
845 | static void vfio_bar_write(void *opaque, hwaddr addr, | |
846 | uint64_t data, unsigned size) | |
847 | { | |
848 | VFIOBAR *bar = opaque; | |
849 | union { | |
850 | uint8_t byte; | |
851 | uint16_t word; | |
852 | uint32_t dword; | |
853 | uint64_t qword; | |
854 | } buf; | |
855 | ||
856 | switch (size) { | |
857 | case 1: | |
858 | buf.byte = data; | |
859 | break; | |
860 | case 2: | |
861 | buf.word = cpu_to_le16(data); | |
862 | break; | |
863 | case 4: | |
864 | buf.dword = cpu_to_le32(data); | |
865 | break; | |
866 | default: | |
867 | hw_error("vfio: unsupported write size, %d bytes\n", size); | |
868 | break; | |
869 | } | |
870 | ||
871 | if (pwrite(bar->fd, &buf, size, bar->fd_offset + addr) != size) { | |
872 | error_report("%s(,0x%"HWADDR_PRIx", 0x%"PRIx64", %d) failed: %m\n", | |
873 | __func__, addr, data, size); | |
874 | } | |
875 | ||
876 | DPRINTF("%s(BAR%d+0x%"HWADDR_PRIx", 0x%"PRIx64", %d)\n", | |
877 | __func__, bar->nr, addr, data, size); | |
878 | ||
879 | /* | |
880 | * A read or write to a BAR always signals an INTx EOI. This will | |
881 | * do nothing if not pending (including not in INTx mode). We assume | |
882 | * that a BAR access is in response to an interrupt and that BAR | |
883 | * accesses will service the interrupt. Unfortunately, we don't know | |
884 | * which access will service the interrupt, so we're potentially | |
885 | * getting quite a few host interrupts per guest interrupt. | |
886 | */ | |
887 | vfio_eoi(container_of(bar, VFIODevice, bars[bar->nr])); | |
888 | } | |
889 | ||
890 | static uint64_t vfio_bar_read(void *opaque, | |
891 | hwaddr addr, unsigned size) | |
892 | { | |
893 | VFIOBAR *bar = opaque; | |
894 | union { | |
895 | uint8_t byte; | |
896 | uint16_t word; | |
897 | uint32_t dword; | |
898 | uint64_t qword; | |
899 | } buf; | |
900 | uint64_t data = 0; | |
901 | ||
902 | if (pread(bar->fd, &buf, size, bar->fd_offset + addr) != size) { | |
903 | error_report("%s(,0x%"HWADDR_PRIx", %d) failed: %m\n", | |
904 | __func__, addr, size); | |
905 | return (uint64_t)-1; | |
906 | } | |
907 | ||
908 | switch (size) { | |
909 | case 1: | |
910 | data = buf.byte; | |
911 | break; | |
912 | case 2: | |
913 | data = le16_to_cpu(buf.word); | |
914 | break; | |
915 | case 4: | |
916 | data = le32_to_cpu(buf.dword); | |
917 | break; | |
918 | default: | |
919 | hw_error("vfio: unsupported read size, %d bytes\n", size); | |
920 | break; | |
921 | } | |
922 | ||
923 | DPRINTF("%s(BAR%d+0x%"HWADDR_PRIx", %d) = 0x%"PRIx64"\n", | |
924 | __func__, bar->nr, addr, size, data); | |
925 | ||
926 | /* Same as write above */ | |
927 | vfio_eoi(container_of(bar, VFIODevice, bars[bar->nr])); | |
928 | ||
929 | return data; | |
930 | } | |
931 | ||
932 | static const MemoryRegionOps vfio_bar_ops = { | |
933 | .read = vfio_bar_read, | |
934 | .write = vfio_bar_write, | |
935 | .endianness = DEVICE_LITTLE_ENDIAN, | |
936 | }; | |
937 | ||
938 | /* | |
939 | * PCI config space | |
940 | */ | |
941 | static uint32_t vfio_pci_read_config(PCIDevice *pdev, uint32_t addr, int len) | |
942 | { | |
943 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
944 | uint32_t val = 0; | |
945 | ||
946 | /* | |
947 | * We only need QEMU PCI config support for the ROM BAR, the MSI and MSIX | |
948 | * capabilities, and the multifunction bit below. We let VFIO handle | |
949 | * virtualizing everything else. Performance is not a concern here. | |
950 | */ | |
951 | if (ranges_overlap(addr, len, PCI_ROM_ADDRESS, 4) || | |
952 | (pdev->cap_present & QEMU_PCI_CAP_MSIX && | |
953 | ranges_overlap(addr, len, pdev->msix_cap, MSIX_CAP_LENGTH)) || | |
954 | (pdev->cap_present & QEMU_PCI_CAP_MSI && | |
955 | ranges_overlap(addr, len, pdev->msi_cap, vdev->msi_cap_size))) { | |
956 | ||
957 | val = pci_default_read_config(pdev, addr, len); | |
958 | } else { | |
959 | if (pread(vdev->fd, &val, len, vdev->config_offset + addr) != len) { | |
960 | error_report("%s(%04x:%02x:%02x.%x, 0x%x, 0x%x) failed: %m\n", | |
961 | __func__, vdev->host.domain, vdev->host.bus, | |
962 | vdev->host.slot, vdev->host.function, addr, len); | |
963 | return -errno; | |
964 | } | |
965 | val = le32_to_cpu(val); | |
966 | } | |
967 | ||
968 | /* Multifunction bit is virualized in QEMU */ | |
969 | if (unlikely(ranges_overlap(addr, len, PCI_HEADER_TYPE, 1))) { | |
970 | uint32_t mask = PCI_HEADER_TYPE_MULTI_FUNCTION; | |
971 | ||
972 | if (len == 4) { | |
973 | mask <<= 16; | |
974 | } | |
975 | ||
976 | if (pdev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) { | |
977 | val |= mask; | |
978 | } else { | |
979 | val &= ~mask; | |
980 | } | |
981 | } | |
982 | ||
983 | DPRINTF("%s(%04x:%02x:%02x.%x, @0x%x, len=0x%x) %x\n", __func__, | |
984 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
985 | vdev->host.function, addr, len, val); | |
986 | ||
987 | return val; | |
988 | } | |
989 | ||
990 | static void vfio_pci_write_config(PCIDevice *pdev, uint32_t addr, | |
991 | uint32_t val, int len) | |
992 | { | |
993 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
994 | uint32_t val_le = cpu_to_le32(val); | |
995 | ||
996 | DPRINTF("%s(%04x:%02x:%02x.%x, @0x%x, 0x%x, len=0x%x)\n", __func__, | |
997 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
998 | vdev->host.function, addr, val, len); | |
999 | ||
1000 | /* Write everything to VFIO, let it filter out what we can't write */ | |
1001 | if (pwrite(vdev->fd, &val_le, len, vdev->config_offset + addr) != len) { | |
1002 | error_report("%s(%04x:%02x:%02x.%x, 0x%x, 0x%x, 0x%x) failed: %m\n", | |
1003 | __func__, vdev->host.domain, vdev->host.bus, | |
1004 | vdev->host.slot, vdev->host.function, addr, val, len); | |
1005 | } | |
1006 | ||
1007 | /* Write standard header bits to emulation */ | |
1008 | if (addr < PCI_CONFIG_HEADER_SIZE) { | |
1009 | pci_default_write_config(pdev, addr, val, len); | |
1010 | return; | |
1011 | } | |
1012 | ||
1013 | /* MSI/MSI-X Enabling/Disabling */ | |
1014 | if (pdev->cap_present & QEMU_PCI_CAP_MSI && | |
1015 | ranges_overlap(addr, len, pdev->msi_cap, vdev->msi_cap_size)) { | |
1016 | int is_enabled, was_enabled = msi_enabled(pdev); | |
1017 | ||
1018 | pci_default_write_config(pdev, addr, val, len); | |
1019 | ||
1020 | is_enabled = msi_enabled(pdev); | |
1021 | ||
1022 | if (!was_enabled && is_enabled) { | |
1023 | vfio_enable_msi(vdev); | |
1024 | } else if (was_enabled && !is_enabled) { | |
1025 | vfio_disable_msi(vdev); | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | if (pdev->cap_present & QEMU_PCI_CAP_MSIX && | |
1030 | ranges_overlap(addr, len, pdev->msix_cap, MSIX_CAP_LENGTH)) { | |
1031 | int is_enabled, was_enabled = msix_enabled(pdev); | |
1032 | ||
1033 | pci_default_write_config(pdev, addr, val, len); | |
1034 | ||
1035 | is_enabled = msix_enabled(pdev); | |
1036 | ||
1037 | if (!was_enabled && is_enabled) { | |
1038 | vfio_enable_msix(vdev); | |
1039 | } else if (was_enabled && !is_enabled) { | |
1040 | vfio_disable_msix(vdev); | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * DMA - Mapping and unmapping for the "type1" IOMMU interface used on x86 | |
1047 | */ | |
1048 | static int vfio_dma_unmap(VFIOContainer *container, | |
1049 | hwaddr iova, ram_addr_t size) | |
1050 | { | |
1051 | struct vfio_iommu_type1_dma_unmap unmap = { | |
1052 | .argsz = sizeof(unmap), | |
1053 | .flags = 0, | |
1054 | .iova = iova, | |
1055 | .size = size, | |
1056 | }; | |
1057 | ||
1058 | if (ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, &unmap)) { | |
1059 | DPRINTF("VFIO_UNMAP_DMA: %d\n", -errno); | |
1060 | return -errno; | |
1061 | } | |
1062 | ||
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | static int vfio_dma_map(VFIOContainer *container, hwaddr iova, | |
1067 | ram_addr_t size, void *vaddr, bool readonly) | |
1068 | { | |
1069 | struct vfio_iommu_type1_dma_map map = { | |
1070 | .argsz = sizeof(map), | |
1071 | .flags = VFIO_DMA_MAP_FLAG_READ, | |
1072 | .vaddr = (__u64)(uintptr_t)vaddr, | |
1073 | .iova = iova, | |
1074 | .size = size, | |
1075 | }; | |
1076 | ||
1077 | if (!readonly) { | |
1078 | map.flags |= VFIO_DMA_MAP_FLAG_WRITE; | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Try the mapping, if it fails with EBUSY, unmap the region and try | |
1083 | * again. This shouldn't be necessary, but we sometimes see it in | |
1084 | * the the VGA ROM space. | |
1085 | */ | |
1086 | if (ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0 || | |
1087 | (errno == EBUSY && vfio_dma_unmap(container, iova, size) == 0 && | |
1088 | ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0)) { | |
1089 | return 0; | |
1090 | } | |
1091 | ||
1092 | DPRINTF("VFIO_MAP_DMA: %d\n", -errno); | |
1093 | return -errno; | |
1094 | } | |
1095 | ||
1096 | static bool vfio_listener_skipped_section(MemoryRegionSection *section) | |
1097 | { | |
1098 | return !memory_region_is_ram(section->mr); | |
1099 | } | |
1100 | ||
1101 | static void vfio_listener_region_add(MemoryListener *listener, | |
1102 | MemoryRegionSection *section) | |
1103 | { | |
1104 | VFIOContainer *container = container_of(listener, VFIOContainer, | |
1105 | iommu_data.listener); | |
1106 | hwaddr iova, end; | |
1107 | void *vaddr; | |
1108 | int ret; | |
1109 | ||
1110 | if (vfio_listener_skipped_section(section)) { | |
1111 | DPRINTF("vfio: SKIPPING region_add %"HWADDR_PRIx" - %"PRIx64"\n", | |
1112 | section->offset_within_address_space, | |
1113 | section->offset_within_address_space + section->size - 1); | |
1114 | return; | |
1115 | } | |
1116 | ||
1117 | if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) != | |
1118 | (section->offset_within_region & ~TARGET_PAGE_MASK))) { | |
1119 | error_report("%s received unaligned region\n", __func__); | |
1120 | return; | |
1121 | } | |
1122 | ||
1123 | iova = TARGET_PAGE_ALIGN(section->offset_within_address_space); | |
1124 | end = (section->offset_within_address_space + section->size) & | |
1125 | TARGET_PAGE_MASK; | |
1126 | ||
1127 | if (iova >= end) { | |
1128 | return; | |
1129 | } | |
1130 | ||
1131 | vaddr = memory_region_get_ram_ptr(section->mr) + | |
1132 | section->offset_within_region + | |
1133 | (iova - section->offset_within_address_space); | |
1134 | ||
1135 | DPRINTF("vfio: region_add %"HWADDR_PRIx" - %"HWADDR_PRIx" [%p]\n", | |
1136 | iova, end - 1, vaddr); | |
1137 | ||
1138 | ret = vfio_dma_map(container, iova, end - iova, vaddr, section->readonly); | |
1139 | if (ret) { | |
1140 | error_report("vfio_dma_map(%p, 0x%"HWADDR_PRIx", " | |
1141 | "0x%"HWADDR_PRIx", %p) = %d (%m)\n", | |
1142 | container, iova, end - iova, vaddr, ret); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | static void vfio_listener_region_del(MemoryListener *listener, | |
1147 | MemoryRegionSection *section) | |
1148 | { | |
1149 | VFIOContainer *container = container_of(listener, VFIOContainer, | |
1150 | iommu_data.listener); | |
1151 | hwaddr iova, end; | |
1152 | int ret; | |
1153 | ||
1154 | if (vfio_listener_skipped_section(section)) { | |
1155 | DPRINTF("vfio: SKIPPING region_del %"HWADDR_PRIx" - %"PRIx64"\n", | |
1156 | section->offset_within_address_space, | |
1157 | section->offset_within_address_space + section->size - 1); | |
1158 | return; | |
1159 | } | |
1160 | ||
1161 | if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) != | |
1162 | (section->offset_within_region & ~TARGET_PAGE_MASK))) { | |
1163 | error_report("%s received unaligned region\n", __func__); | |
1164 | return; | |
1165 | } | |
1166 | ||
1167 | iova = TARGET_PAGE_ALIGN(section->offset_within_address_space); | |
1168 | end = (section->offset_within_address_space + section->size) & | |
1169 | TARGET_PAGE_MASK; | |
1170 | ||
1171 | if (iova >= end) { | |
1172 | return; | |
1173 | } | |
1174 | ||
1175 | DPRINTF("vfio: region_del %"HWADDR_PRIx" - %"HWADDR_PRIx"\n", | |
1176 | iova, end - 1); | |
1177 | ||
1178 | ret = vfio_dma_unmap(container, iova, end - iova); | |
1179 | if (ret) { | |
1180 | error_report("vfio_dma_unmap(%p, 0x%"HWADDR_PRIx", " | |
1181 | "0x%"HWADDR_PRIx") = %d (%m)\n", | |
1182 | container, iova, end - iova, ret); | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | static MemoryListener vfio_memory_listener = { | |
1187 | .region_add = vfio_listener_region_add, | |
1188 | .region_del = vfio_listener_region_del, | |
1189 | }; | |
1190 | ||
1191 | static void vfio_listener_release(VFIOContainer *container) | |
1192 | { | |
1193 | memory_listener_unregister(&container->iommu_data.listener); | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Interrupt setup | |
1198 | */ | |
1199 | static void vfio_disable_interrupts(VFIODevice *vdev) | |
1200 | { | |
1201 | switch (vdev->interrupt) { | |
1202 | case VFIO_INT_INTx: | |
1203 | vfio_disable_intx(vdev); | |
1204 | break; | |
1205 | case VFIO_INT_MSI: | |
1206 | vfio_disable_msi(vdev); | |
1207 | break; | |
1208 | case VFIO_INT_MSIX: | |
1209 | vfio_disable_msix(vdev); | |
1210 | break; | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | static int vfio_setup_msi(VFIODevice *vdev, int pos) | |
1215 | { | |
1216 | uint16_t ctrl; | |
1217 | bool msi_64bit, msi_maskbit; | |
1218 | int ret, entries; | |
1219 | ||
1220 | if (pread(vdev->fd, &ctrl, sizeof(ctrl), | |
1221 | vdev->config_offset + pos + PCI_CAP_FLAGS) != sizeof(ctrl)) { | |
1222 | return -errno; | |
1223 | } | |
1224 | ctrl = le16_to_cpu(ctrl); | |
1225 | ||
1226 | msi_64bit = !!(ctrl & PCI_MSI_FLAGS_64BIT); | |
1227 | msi_maskbit = !!(ctrl & PCI_MSI_FLAGS_MASKBIT); | |
1228 | entries = 1 << ((ctrl & PCI_MSI_FLAGS_QMASK) >> 1); | |
1229 | ||
1230 | DPRINTF("%04x:%02x:%02x.%x PCI MSI CAP @0x%x\n", vdev->host.domain, | |
1231 | vdev->host.bus, vdev->host.slot, vdev->host.function, pos); | |
1232 | ||
1233 | ret = msi_init(&vdev->pdev, pos, entries, msi_64bit, msi_maskbit); | |
1234 | if (ret < 0) { | |
1235 | if (ret == -ENOTSUP) { | |
1236 | return 0; | |
1237 | } | |
1238 | error_report("vfio: msi_init failed\n"); | |
1239 | return ret; | |
1240 | } | |
1241 | vdev->msi_cap_size = 0xa + (msi_maskbit ? 0xa : 0) + (msi_64bit ? 0x4 : 0); | |
1242 | ||
1243 | return 0; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * We don't have any control over how pci_add_capability() inserts | |
1248 | * capabilities into the chain. In order to setup MSI-X we need a | |
1249 | * MemoryRegion for the BAR. In order to setup the BAR and not | |
1250 | * attempt to mmap the MSI-X table area, which VFIO won't allow, we | |
1251 | * need to first look for where the MSI-X table lives. So we | |
1252 | * unfortunately split MSI-X setup across two functions. | |
1253 | */ | |
1254 | static int vfio_early_setup_msix(VFIODevice *vdev) | |
1255 | { | |
1256 | uint8_t pos; | |
1257 | uint16_t ctrl; | |
1258 | uint32_t table, pba; | |
1259 | ||
1260 | pos = pci_find_capability(&vdev->pdev, PCI_CAP_ID_MSIX); | |
1261 | if (!pos) { | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | if (pread(vdev->fd, &ctrl, sizeof(ctrl), | |
1266 | vdev->config_offset + pos + PCI_CAP_FLAGS) != sizeof(ctrl)) { | |
1267 | return -errno; | |
1268 | } | |
1269 | ||
1270 | if (pread(vdev->fd, &table, sizeof(table), | |
1271 | vdev->config_offset + pos + PCI_MSIX_TABLE) != sizeof(table)) { | |
1272 | return -errno; | |
1273 | } | |
1274 | ||
1275 | if (pread(vdev->fd, &pba, sizeof(pba), | |
1276 | vdev->config_offset + pos + PCI_MSIX_PBA) != sizeof(pba)) { | |
1277 | return -errno; | |
1278 | } | |
1279 | ||
1280 | ctrl = le16_to_cpu(ctrl); | |
1281 | table = le32_to_cpu(table); | |
1282 | pba = le32_to_cpu(pba); | |
1283 | ||
1284 | vdev->msix = g_malloc0(sizeof(*(vdev->msix))); | |
1285 | vdev->msix->table_bar = table & PCI_MSIX_FLAGS_BIRMASK; | |
1286 | vdev->msix->table_offset = table & ~PCI_MSIX_FLAGS_BIRMASK; | |
1287 | vdev->msix->pba_bar = pba & PCI_MSIX_FLAGS_BIRMASK; | |
1288 | vdev->msix->pba_offset = pba & ~PCI_MSIX_FLAGS_BIRMASK; | |
1289 | vdev->msix->entries = (ctrl & PCI_MSIX_FLAGS_QSIZE) + 1; | |
1290 | ||
1291 | DPRINTF("%04x:%02x:%02x.%x " | |
1292 | "PCI MSI-X CAP @0x%x, BAR %d, offset 0x%x, entries %d\n", | |
1293 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
1294 | vdev->host.function, pos, vdev->msix->table_bar, | |
1295 | vdev->msix->table_offset, vdev->msix->entries); | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
1300 | static int vfio_setup_msix(VFIODevice *vdev, int pos) | |
1301 | { | |
1302 | int ret; | |
1303 | ||
1304 | ret = msix_init(&vdev->pdev, vdev->msix->entries, | |
1305 | &vdev->bars[vdev->msix->table_bar].mem, | |
1306 | vdev->msix->table_bar, vdev->msix->table_offset, | |
1307 | &vdev->bars[vdev->msix->pba_bar].mem, | |
1308 | vdev->msix->pba_bar, vdev->msix->pba_offset, pos); | |
1309 | if (ret < 0) { | |
1310 | if (ret == -ENOTSUP) { | |
1311 | return 0; | |
1312 | } | |
1313 | error_report("vfio: msix_init failed\n"); | |
1314 | return ret; | |
1315 | } | |
1316 | ||
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | static void vfio_teardown_msi(VFIODevice *vdev) | |
1321 | { | |
1322 | msi_uninit(&vdev->pdev); | |
1323 | ||
1324 | if (vdev->msix) { | |
1325 | msix_uninit(&vdev->pdev, &vdev->bars[vdev->msix->table_bar].mem, | |
1326 | &vdev->bars[vdev->msix->pba_bar].mem); | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Resource setup | |
1332 | */ | |
1333 | static void vfio_mmap_set_enabled(VFIODevice *vdev, bool enabled) | |
1334 | { | |
1335 | int i; | |
1336 | ||
1337 | for (i = 0; i < PCI_ROM_SLOT; i++) { | |
1338 | VFIOBAR *bar = &vdev->bars[i]; | |
1339 | ||
1340 | if (!bar->size) { | |
1341 | continue; | |
1342 | } | |
1343 | ||
1344 | memory_region_set_enabled(&bar->mmap_mem, enabled); | |
1345 | if (vdev->msix && vdev->msix->table_bar == i) { | |
1346 | memory_region_set_enabled(&vdev->msix->mmap_mem, enabled); | |
1347 | } | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | static void vfio_unmap_bar(VFIODevice *vdev, int nr) | |
1352 | { | |
1353 | VFIOBAR *bar = &vdev->bars[nr]; | |
1354 | ||
1355 | if (!bar->size) { | |
1356 | return; | |
1357 | } | |
1358 | ||
1359 | memory_region_del_subregion(&bar->mem, &bar->mmap_mem); | |
1360 | munmap(bar->mmap, memory_region_size(&bar->mmap_mem)); | |
1361 | ||
1362 | if (vdev->msix && vdev->msix->table_bar == nr) { | |
1363 | memory_region_del_subregion(&bar->mem, &vdev->msix->mmap_mem); | |
1364 | munmap(vdev->msix->mmap, memory_region_size(&vdev->msix->mmap_mem)); | |
1365 | } | |
1366 | ||
1367 | memory_region_destroy(&bar->mem); | |
1368 | } | |
1369 | ||
1370 | static int vfio_mmap_bar(VFIOBAR *bar, MemoryRegion *mem, MemoryRegion *submem, | |
1371 | void **map, size_t size, off_t offset, | |
1372 | const char *name) | |
1373 | { | |
1374 | int ret = 0; | |
1375 | ||
1376 | if (size && bar->flags & VFIO_REGION_INFO_FLAG_MMAP) { | |
1377 | int prot = 0; | |
1378 | ||
1379 | if (bar->flags & VFIO_REGION_INFO_FLAG_READ) { | |
1380 | prot |= PROT_READ; | |
1381 | } | |
1382 | ||
1383 | if (bar->flags & VFIO_REGION_INFO_FLAG_WRITE) { | |
1384 | prot |= PROT_WRITE; | |
1385 | } | |
1386 | ||
1387 | *map = mmap(NULL, size, prot, MAP_SHARED, | |
1388 | bar->fd, bar->fd_offset + offset); | |
1389 | if (*map == MAP_FAILED) { | |
1390 | *map = NULL; | |
1391 | ret = -errno; | |
1392 | goto empty_region; | |
1393 | } | |
1394 | ||
1395 | memory_region_init_ram_ptr(submem, name, size, *map); | |
1396 | } else { | |
1397 | empty_region: | |
1398 | /* Create a zero sized sub-region to make cleanup easy. */ | |
1399 | memory_region_init(submem, name, 0); | |
1400 | } | |
1401 | ||
1402 | memory_region_add_subregion(mem, offset, submem); | |
1403 | ||
1404 | return ret; | |
1405 | } | |
1406 | ||
1407 | static void vfio_map_bar(VFIODevice *vdev, int nr) | |
1408 | { | |
1409 | VFIOBAR *bar = &vdev->bars[nr]; | |
1410 | unsigned size = bar->size; | |
1411 | char name[64]; | |
1412 | uint32_t pci_bar; | |
1413 | uint8_t type; | |
1414 | int ret; | |
1415 | ||
1416 | /* Skip both unimplemented BARs and the upper half of 64bit BARS. */ | |
1417 | if (!size) { | |
1418 | return; | |
1419 | } | |
1420 | ||
1421 | snprintf(name, sizeof(name), "VFIO %04x:%02x:%02x.%x BAR %d", | |
1422 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
1423 | vdev->host.function, nr); | |
1424 | ||
1425 | /* Determine what type of BAR this is for registration */ | |
1426 | ret = pread(vdev->fd, &pci_bar, sizeof(pci_bar), | |
1427 | vdev->config_offset + PCI_BASE_ADDRESS_0 + (4 * nr)); | |
1428 | if (ret != sizeof(pci_bar)) { | |
1429 | error_report("vfio: Failed to read BAR %d (%m)\n", nr); | |
1430 | return; | |
1431 | } | |
1432 | ||
1433 | pci_bar = le32_to_cpu(pci_bar); | |
1434 | type = pci_bar & (pci_bar & PCI_BASE_ADDRESS_SPACE_IO ? | |
1435 | ~PCI_BASE_ADDRESS_IO_MASK : ~PCI_BASE_ADDRESS_MEM_MASK); | |
1436 | ||
1437 | /* A "slow" read/write mapping underlies all BARs */ | |
1438 | memory_region_init_io(&bar->mem, &vfio_bar_ops, bar, name, size); | |
1439 | pci_register_bar(&vdev->pdev, nr, type, &bar->mem); | |
1440 | ||
1441 | /* | |
1442 | * We can't mmap areas overlapping the MSIX vector table, so we | |
1443 | * potentially insert a direct-mapped subregion before and after it. | |
1444 | */ | |
1445 | if (vdev->msix && vdev->msix->table_bar == nr) { | |
1446 | size = vdev->msix->table_offset & TARGET_PAGE_MASK; | |
1447 | } | |
1448 | ||
1449 | strncat(name, " mmap", sizeof(name) - strlen(name) - 1); | |
1450 | if (vfio_mmap_bar(bar, &bar->mem, | |
1451 | &bar->mmap_mem, &bar->mmap, size, 0, name)) { | |
1452 | error_report("%s unsupported. Performance may be slow\n", name); | |
1453 | } | |
1454 | ||
1455 | if (vdev->msix && vdev->msix->table_bar == nr) { | |
1456 | unsigned start; | |
1457 | ||
1458 | start = TARGET_PAGE_ALIGN(vdev->msix->table_offset + | |
1459 | (vdev->msix->entries * PCI_MSIX_ENTRY_SIZE)); | |
1460 | ||
1461 | size = start < bar->size ? bar->size - start : 0; | |
1462 | strncat(name, " msix-hi", sizeof(name) - strlen(name) - 1); | |
1463 | /* VFIOMSIXInfo contains another MemoryRegion for this mapping */ | |
1464 | if (vfio_mmap_bar(bar, &bar->mem, &vdev->msix->mmap_mem, | |
1465 | &vdev->msix->mmap, size, start, name)) { | |
1466 | error_report("%s unsupported. Performance may be slow\n", name); | |
1467 | } | |
1468 | } | |
1469 | } | |
1470 | ||
1471 | static void vfio_map_bars(VFIODevice *vdev) | |
1472 | { | |
1473 | int i; | |
1474 | ||
1475 | for (i = 0; i < PCI_ROM_SLOT; i++) { | |
1476 | vfio_map_bar(vdev, i); | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | static void vfio_unmap_bars(VFIODevice *vdev) | |
1481 | { | |
1482 | int i; | |
1483 | ||
1484 | for (i = 0; i < PCI_ROM_SLOT; i++) { | |
1485 | vfio_unmap_bar(vdev, i); | |
1486 | } | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * General setup | |
1491 | */ | |
1492 | static uint8_t vfio_std_cap_max_size(PCIDevice *pdev, uint8_t pos) | |
1493 | { | |
1494 | uint8_t tmp, next = 0xff; | |
1495 | ||
1496 | for (tmp = pdev->config[PCI_CAPABILITY_LIST]; tmp; | |
1497 | tmp = pdev->config[tmp + 1]) { | |
1498 | if (tmp > pos && tmp < next) { | |
1499 | next = tmp; | |
1500 | } | |
1501 | } | |
1502 | ||
1503 | return next - pos; | |
1504 | } | |
1505 | ||
1506 | static int vfio_add_std_cap(VFIODevice *vdev, uint8_t pos) | |
1507 | { | |
1508 | PCIDevice *pdev = &vdev->pdev; | |
1509 | uint8_t cap_id, next, size; | |
1510 | int ret; | |
1511 | ||
1512 | cap_id = pdev->config[pos]; | |
1513 | next = pdev->config[pos + 1]; | |
1514 | ||
1515 | /* | |
1516 | * If it becomes important to configure capabilities to their actual | |
1517 | * size, use this as the default when it's something we don't recognize. | |
1518 | * Since QEMU doesn't actually handle many of the config accesses, | |
1519 | * exact size doesn't seem worthwhile. | |
1520 | */ | |
1521 | size = vfio_std_cap_max_size(pdev, pos); | |
1522 | ||
1523 | /* | |
1524 | * pci_add_capability always inserts the new capability at the head | |
1525 | * of the chain. Therefore to end up with a chain that matches the | |
1526 | * physical device, we insert from the end by making this recursive. | |
1527 | * This is also why we pre-caclulate size above as cached config space | |
1528 | * will be changed as we unwind the stack. | |
1529 | */ | |
1530 | if (next) { | |
1531 | ret = vfio_add_std_cap(vdev, next); | |
1532 | if (ret) { | |
1533 | return ret; | |
1534 | } | |
1535 | } else { | |
1536 | pdev->config[PCI_CAPABILITY_LIST] = 0; /* Begin the rebuild */ | |
1537 | } | |
1538 | ||
1539 | switch (cap_id) { | |
1540 | case PCI_CAP_ID_MSI: | |
1541 | ret = vfio_setup_msi(vdev, pos); | |
1542 | break; | |
1543 | case PCI_CAP_ID_MSIX: | |
1544 | ret = vfio_setup_msix(vdev, pos); | |
1545 | break; | |
1546 | default: | |
1547 | ret = pci_add_capability(pdev, cap_id, pos, size); | |
1548 | break; | |
1549 | } | |
1550 | ||
1551 | if (ret < 0) { | |
1552 | error_report("vfio: %04x:%02x:%02x.%x Error adding PCI capability " | |
1553 | "0x%x[0x%x]@0x%x: %d\n", vdev->host.domain, | |
1554 | vdev->host.bus, vdev->host.slot, vdev->host.function, | |
1555 | cap_id, size, pos, ret); | |
1556 | return ret; | |
1557 | } | |
1558 | ||
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | static int vfio_add_capabilities(VFIODevice *vdev) | |
1563 | { | |
1564 | PCIDevice *pdev = &vdev->pdev; | |
1565 | ||
1566 | if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST) || | |
1567 | !pdev->config[PCI_CAPABILITY_LIST]) { | |
1568 | return 0; /* Nothing to add */ | |
1569 | } | |
1570 | ||
1571 | return vfio_add_std_cap(vdev, pdev->config[PCI_CAPABILITY_LIST]); | |
1572 | } | |
1573 | ||
1574 | static int vfio_load_rom(VFIODevice *vdev) | |
1575 | { | |
1576 | uint64_t size = vdev->rom_size; | |
1577 | char name[32]; | |
1578 | off_t off = 0, voff = vdev->rom_offset; | |
1579 | ssize_t bytes; | |
1580 | void *ptr; | |
1581 | ||
1582 | /* If loading ROM from file, pci handles it */ | |
1583 | if (vdev->pdev.romfile || !vdev->pdev.rom_bar || !size) { | |
1584 | return 0; | |
1585 | } | |
1586 | ||
1587 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
1588 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
1589 | ||
1590 | snprintf(name, sizeof(name), "vfio[%04x:%02x:%02x.%x].rom", | |
1591 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
1592 | vdev->host.function); | |
1593 | memory_region_init_ram(&vdev->pdev.rom, name, size); | |
1594 | ptr = memory_region_get_ram_ptr(&vdev->pdev.rom); | |
1595 | memset(ptr, 0xff, size); | |
1596 | ||
1597 | while (size) { | |
1598 | bytes = pread(vdev->fd, ptr + off, size, voff + off); | |
1599 | if (bytes == 0) { | |
1600 | break; /* expect that we could get back less than the ROM BAR */ | |
1601 | } else if (bytes > 0) { | |
1602 | off += bytes; | |
1603 | size -= bytes; | |
1604 | } else { | |
1605 | if (errno == EINTR || errno == EAGAIN) { | |
1606 | continue; | |
1607 | } | |
1608 | error_report("vfio: Error reading device ROM: %m\n"); | |
1609 | memory_region_destroy(&vdev->pdev.rom); | |
1610 | return -errno; | |
1611 | } | |
1612 | } | |
1613 | ||
1614 | pci_register_bar(&vdev->pdev, PCI_ROM_SLOT, 0, &vdev->pdev.rom); | |
1615 | vdev->pdev.has_rom = true; | |
1616 | return 0; | |
1617 | } | |
1618 | ||
1619 | static int vfio_connect_container(VFIOGroup *group) | |
1620 | { | |
1621 | VFIOContainer *container; | |
1622 | int ret, fd; | |
1623 | ||
1624 | if (group->container) { | |
1625 | return 0; | |
1626 | } | |
1627 | ||
1628 | QLIST_FOREACH(container, &container_list, next) { | |
1629 | if (!ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &container->fd)) { | |
1630 | group->container = container; | |
1631 | QLIST_INSERT_HEAD(&container->group_list, group, container_next); | |
1632 | return 0; | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | fd = qemu_open("/dev/vfio/vfio", O_RDWR); | |
1637 | if (fd < 0) { | |
1638 | error_report("vfio: failed to open /dev/vfio/vfio: %m\n"); | |
1639 | return -errno; | |
1640 | } | |
1641 | ||
1642 | ret = ioctl(fd, VFIO_GET_API_VERSION); | |
1643 | if (ret != VFIO_API_VERSION) { | |
1644 | error_report("vfio: supported vfio version: %d, " | |
1645 | "reported version: %d\n", VFIO_API_VERSION, ret); | |
1646 | close(fd); | |
1647 | return -EINVAL; | |
1648 | } | |
1649 | ||
1650 | container = g_malloc0(sizeof(*container)); | |
1651 | container->fd = fd; | |
1652 | ||
1653 | if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1_IOMMU)) { | |
1654 | ret = ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &fd); | |
1655 | if (ret) { | |
1656 | error_report("vfio: failed to set group container: %m\n"); | |
1657 | g_free(container); | |
1658 | close(fd); | |
1659 | return -errno; | |
1660 | } | |
1661 | ||
1662 | ret = ioctl(fd, VFIO_SET_IOMMU, VFIO_TYPE1_IOMMU); | |
1663 | if (ret) { | |
1664 | error_report("vfio: failed to set iommu for container: %m\n"); | |
1665 | g_free(container); | |
1666 | close(fd); | |
1667 | return -errno; | |
1668 | } | |
1669 | ||
1670 | container->iommu_data.listener = vfio_memory_listener; | |
1671 | container->iommu_data.release = vfio_listener_release; | |
1672 | ||
1673 | memory_listener_register(&container->iommu_data.listener, &address_space_memory); | |
1674 | } else { | |
1675 | error_report("vfio: No available IOMMU models\n"); | |
1676 | g_free(container); | |
1677 | close(fd); | |
1678 | return -EINVAL; | |
1679 | } | |
1680 | ||
1681 | QLIST_INIT(&container->group_list); | |
1682 | QLIST_INSERT_HEAD(&container_list, container, next); | |
1683 | ||
1684 | group->container = container; | |
1685 | QLIST_INSERT_HEAD(&container->group_list, group, container_next); | |
1686 | ||
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | static void vfio_disconnect_container(VFIOGroup *group) | |
1691 | { | |
1692 | VFIOContainer *container = group->container; | |
1693 | ||
1694 | if (ioctl(group->fd, VFIO_GROUP_UNSET_CONTAINER, &container->fd)) { | |
1695 | error_report("vfio: error disconnecting group %d from container\n", | |
1696 | group->groupid); | |
1697 | } | |
1698 | ||
1699 | QLIST_REMOVE(group, container_next); | |
1700 | group->container = NULL; | |
1701 | ||
1702 | if (QLIST_EMPTY(&container->group_list)) { | |
1703 | if (container->iommu_data.release) { | |
1704 | container->iommu_data.release(container); | |
1705 | } | |
1706 | QLIST_REMOVE(container, next); | |
1707 | DPRINTF("vfio_disconnect_container: close container->fd\n"); | |
1708 | close(container->fd); | |
1709 | g_free(container); | |
1710 | } | |
1711 | } | |
1712 | ||
1713 | static VFIOGroup *vfio_get_group(int groupid) | |
1714 | { | |
1715 | VFIOGroup *group; | |
1716 | char path[32]; | |
1717 | struct vfio_group_status status = { .argsz = sizeof(status) }; | |
1718 | ||
1719 | QLIST_FOREACH(group, &group_list, next) { | |
1720 | if (group->groupid == groupid) { | |
1721 | return group; | |
1722 | } | |
1723 | } | |
1724 | ||
1725 | group = g_malloc0(sizeof(*group)); | |
1726 | ||
1727 | snprintf(path, sizeof(path), "/dev/vfio/%d", groupid); | |
1728 | group->fd = qemu_open(path, O_RDWR); | |
1729 | if (group->fd < 0) { | |
1730 | error_report("vfio: error opening %s: %m\n", path); | |
1731 | g_free(group); | |
1732 | return NULL; | |
1733 | } | |
1734 | ||
1735 | if (ioctl(group->fd, VFIO_GROUP_GET_STATUS, &status)) { | |
1736 | error_report("vfio: error getting group status: %m\n"); | |
1737 | close(group->fd); | |
1738 | g_free(group); | |
1739 | return NULL; | |
1740 | } | |
1741 | ||
1742 | if (!(status.flags & VFIO_GROUP_FLAGS_VIABLE)) { | |
1743 | error_report("vfio: error, group %d is not viable, please ensure " | |
1744 | "all devices within the iommu_group are bound to their " | |
1745 | "vfio bus driver.\n", groupid); | |
1746 | close(group->fd); | |
1747 | g_free(group); | |
1748 | return NULL; | |
1749 | } | |
1750 | ||
1751 | group->groupid = groupid; | |
1752 | QLIST_INIT(&group->device_list); | |
1753 | ||
1754 | if (vfio_connect_container(group)) { | |
1755 | error_report("vfio: failed to setup container for group %d\n", groupid); | |
1756 | close(group->fd); | |
1757 | g_free(group); | |
1758 | return NULL; | |
1759 | } | |
1760 | ||
1761 | QLIST_INSERT_HEAD(&group_list, group, next); | |
1762 | ||
1763 | return group; | |
1764 | } | |
1765 | ||
1766 | static void vfio_put_group(VFIOGroup *group) | |
1767 | { | |
1768 | if (!QLIST_EMPTY(&group->device_list)) { | |
1769 | return; | |
1770 | } | |
1771 | ||
1772 | vfio_disconnect_container(group); | |
1773 | QLIST_REMOVE(group, next); | |
1774 | DPRINTF("vfio_put_group: close group->fd\n"); | |
1775 | close(group->fd); | |
1776 | g_free(group); | |
1777 | } | |
1778 | ||
1779 | static int vfio_get_device(VFIOGroup *group, const char *name, VFIODevice *vdev) | |
1780 | { | |
1781 | struct vfio_device_info dev_info = { .argsz = sizeof(dev_info) }; | |
1782 | struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) }; | |
1783 | int ret, i; | |
1784 | ||
1785 | ret = ioctl(group->fd, VFIO_GROUP_GET_DEVICE_FD, name); | |
1786 | if (ret < 0) { | |
1787 | error_report("vfio: error getting device %s from group %d: %m\n", | |
1788 | name, group->groupid); | |
1789 | error_report("Verify all devices in group %d are bound to vfio-pci " | |
1790 | "or pci-stub and not already in use\n", group->groupid); | |
1791 | return ret; | |
1792 | } | |
1793 | ||
1794 | vdev->fd = ret; | |
1795 | vdev->group = group; | |
1796 | QLIST_INSERT_HEAD(&group->device_list, vdev, next); | |
1797 | ||
1798 | /* Sanity check device */ | |
1799 | ret = ioctl(vdev->fd, VFIO_DEVICE_GET_INFO, &dev_info); | |
1800 | if (ret) { | |
1801 | error_report("vfio: error getting device info: %m\n"); | |
1802 | goto error; | |
1803 | } | |
1804 | ||
1805 | DPRINTF("Device %s flags: %u, regions: %u, irgs: %u\n", name, | |
1806 | dev_info.flags, dev_info.num_regions, dev_info.num_irqs); | |
1807 | ||
1808 | if (!(dev_info.flags & VFIO_DEVICE_FLAGS_PCI)) { | |
1809 | error_report("vfio: Um, this isn't a PCI device\n"); | |
1810 | goto error; | |
1811 | } | |
1812 | ||
1813 | vdev->reset_works = !!(dev_info.flags & VFIO_DEVICE_FLAGS_RESET); | |
1814 | if (!vdev->reset_works) { | |
1815 | error_report("Warning, device %s does not support reset\n", name); | |
1816 | } | |
1817 | ||
1818 | if (dev_info.num_regions != VFIO_PCI_NUM_REGIONS) { | |
1819 | error_report("vfio: unexpected number of io regions %u\n", | |
1820 | dev_info.num_regions); | |
1821 | goto error; | |
1822 | } | |
1823 | ||
1824 | if (dev_info.num_irqs != VFIO_PCI_NUM_IRQS) { | |
1825 | error_report("vfio: unexpected number of irqs %u\n", dev_info.num_irqs); | |
1826 | goto error; | |
1827 | } | |
1828 | ||
1829 | for (i = VFIO_PCI_BAR0_REGION_INDEX; i < VFIO_PCI_ROM_REGION_INDEX; i++) { | |
1830 | reg_info.index = i; | |
1831 | ||
1832 | ret = ioctl(vdev->fd, VFIO_DEVICE_GET_REGION_INFO, ®_info); | |
1833 | if (ret) { | |
1834 | error_report("vfio: Error getting region %d info: %m\n", i); | |
1835 | goto error; | |
1836 | } | |
1837 | ||
1838 | DPRINTF("Device %s region %d:\n", name, i); | |
1839 | DPRINTF(" size: 0x%lx, offset: 0x%lx, flags: 0x%lx\n", | |
1840 | (unsigned long)reg_info.size, (unsigned long)reg_info.offset, | |
1841 | (unsigned long)reg_info.flags); | |
1842 | ||
1843 | vdev->bars[i].flags = reg_info.flags; | |
1844 | vdev->bars[i].size = reg_info.size; | |
1845 | vdev->bars[i].fd_offset = reg_info.offset; | |
1846 | vdev->bars[i].fd = vdev->fd; | |
1847 | vdev->bars[i].nr = i; | |
1848 | } | |
1849 | ||
1850 | reg_info.index = VFIO_PCI_ROM_REGION_INDEX; | |
1851 | ||
1852 | ret = ioctl(vdev->fd, VFIO_DEVICE_GET_REGION_INFO, ®_info); | |
1853 | if (ret) { | |
1854 | error_report("vfio: Error getting ROM info: %m\n"); | |
1855 | goto error; | |
1856 | } | |
1857 | ||
1858 | DPRINTF("Device %s ROM:\n", name); | |
1859 | DPRINTF(" size: 0x%lx, offset: 0x%lx, flags: 0x%lx\n", | |
1860 | (unsigned long)reg_info.size, (unsigned long)reg_info.offset, | |
1861 | (unsigned long)reg_info.flags); | |
1862 | ||
1863 | vdev->rom_size = reg_info.size; | |
1864 | vdev->rom_offset = reg_info.offset; | |
1865 | ||
1866 | reg_info.index = VFIO_PCI_CONFIG_REGION_INDEX; | |
1867 | ||
1868 | ret = ioctl(vdev->fd, VFIO_DEVICE_GET_REGION_INFO, ®_info); | |
1869 | if (ret) { | |
1870 | error_report("vfio: Error getting config info: %m\n"); | |
1871 | goto error; | |
1872 | } | |
1873 | ||
1874 | DPRINTF("Device %s config:\n", name); | |
1875 | DPRINTF(" size: 0x%lx, offset: 0x%lx, flags: 0x%lx\n", | |
1876 | (unsigned long)reg_info.size, (unsigned long)reg_info.offset, | |
1877 | (unsigned long)reg_info.flags); | |
1878 | ||
1879 | vdev->config_size = reg_info.size; | |
1880 | vdev->config_offset = reg_info.offset; | |
1881 | ||
1882 | error: | |
1883 | if (ret) { | |
1884 | QLIST_REMOVE(vdev, next); | |
1885 | vdev->group = NULL; | |
1886 | close(vdev->fd); | |
1887 | } | |
1888 | return ret; | |
1889 | } | |
1890 | ||
1891 | static void vfio_put_device(VFIODevice *vdev) | |
1892 | { | |
1893 | QLIST_REMOVE(vdev, next); | |
1894 | vdev->group = NULL; | |
1895 | DPRINTF("vfio_put_device: close vdev->fd\n"); | |
1896 | close(vdev->fd); | |
1897 | if (vdev->msix) { | |
1898 | g_free(vdev->msix); | |
1899 | vdev->msix = NULL; | |
1900 | } | |
1901 | } | |
1902 | ||
1903 | static int vfio_initfn(PCIDevice *pdev) | |
1904 | { | |
1905 | VFIODevice *pvdev, *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
1906 | VFIOGroup *group; | |
1907 | char path[PATH_MAX], iommu_group_path[PATH_MAX], *group_name; | |
1908 | ssize_t len; | |
1909 | struct stat st; | |
1910 | int groupid; | |
1911 | int ret; | |
1912 | ||
1913 | /* Check that the host device exists */ | |
1914 | snprintf(path, sizeof(path), | |
1915 | "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/", | |
1916 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
1917 | vdev->host.function); | |
1918 | if (stat(path, &st) < 0) { | |
1919 | error_report("vfio: error: no such host device: %s\n", path); | |
1920 | return -errno; | |
1921 | } | |
1922 | ||
1923 | strncat(path, "iommu_group", sizeof(path) - strlen(path) - 1); | |
1924 | ||
1925 | len = readlink(path, iommu_group_path, PATH_MAX); | |
1926 | if (len <= 0) { | |
1927 | error_report("vfio: error no iommu_group for device\n"); | |
1928 | return -errno; | |
1929 | } | |
1930 | ||
1931 | iommu_group_path[len] = 0; | |
1932 | group_name = basename(iommu_group_path); | |
1933 | ||
1934 | if (sscanf(group_name, "%d", &groupid) != 1) { | |
1935 | error_report("vfio: error reading %s: %m\n", path); | |
1936 | return -errno; | |
1937 | } | |
1938 | ||
1939 | DPRINTF("%s(%04x:%02x:%02x.%x) group %d\n", __func__, vdev->host.domain, | |
1940 | vdev->host.bus, vdev->host.slot, vdev->host.function, groupid); | |
1941 | ||
1942 | group = vfio_get_group(groupid); | |
1943 | if (!group) { | |
1944 | error_report("vfio: failed to get group %d\n", groupid); | |
1945 | return -ENOENT; | |
1946 | } | |
1947 | ||
1948 | snprintf(path, sizeof(path), "%04x:%02x:%02x.%01x", | |
1949 | vdev->host.domain, vdev->host.bus, vdev->host.slot, | |
1950 | vdev->host.function); | |
1951 | ||
1952 | QLIST_FOREACH(pvdev, &group->device_list, next) { | |
1953 | if (pvdev->host.domain == vdev->host.domain && | |
1954 | pvdev->host.bus == vdev->host.bus && | |
1955 | pvdev->host.slot == vdev->host.slot && | |
1956 | pvdev->host.function == vdev->host.function) { | |
1957 | ||
1958 | error_report("vfio: error: device %s is already attached\n", path); | |
1959 | vfio_put_group(group); | |
1960 | return -EBUSY; | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | ret = vfio_get_device(group, path, vdev); | |
1965 | if (ret) { | |
1966 | error_report("vfio: failed to get device %s\n", path); | |
1967 | vfio_put_group(group); | |
1968 | return ret; | |
1969 | } | |
1970 | ||
1971 | /* Get a copy of config space */ | |
1972 | ret = pread(vdev->fd, vdev->pdev.config, | |
1973 | MIN(pci_config_size(&vdev->pdev), vdev->config_size), | |
1974 | vdev->config_offset); | |
1975 | if (ret < (int)MIN(pci_config_size(&vdev->pdev), vdev->config_size)) { | |
1976 | ret = ret < 0 ? -errno : -EFAULT; | |
1977 | error_report("vfio: Failed to read device config space\n"); | |
1978 | goto out_put; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Clear host resource mapping info. If we choose not to register a | |
1983 | * BAR, such as might be the case with the option ROM, we can get | |
1984 | * confusing, unwritable, residual addresses from the host here. | |
1985 | */ | |
1986 | memset(&vdev->pdev.config[PCI_BASE_ADDRESS_0], 0, 24); | |
1987 | memset(&vdev->pdev.config[PCI_ROM_ADDRESS], 0, 4); | |
1988 | ||
1989 | vfio_load_rom(vdev); | |
1990 | ||
1991 | ret = vfio_early_setup_msix(vdev); | |
1992 | if (ret) { | |
1993 | goto out_put; | |
1994 | } | |
1995 | ||
1996 | vfio_map_bars(vdev); | |
1997 | ||
1998 | ret = vfio_add_capabilities(vdev); | |
1999 | if (ret) { | |
2000 | goto out_teardown; | |
2001 | } | |
2002 | ||
2003 | if (vfio_pci_read_config(&vdev->pdev, PCI_INTERRUPT_PIN, 1)) { | |
2004 | vdev->intx.mmap_timer = qemu_new_timer_ms(vm_clock, | |
2005 | vfio_intx_mmap_enable, vdev); | |
2006 | pci_device_set_intx_routing_notifier(&vdev->pdev, vfio_update_irq); | |
2007 | ret = vfio_enable_intx(vdev); | |
2008 | if (ret) { | |
2009 | goto out_teardown; | |
2010 | } | |
2011 | } | |
2012 | ||
2013 | return 0; | |
2014 | ||
2015 | out_teardown: | |
2016 | pci_device_set_intx_routing_notifier(&vdev->pdev, NULL); | |
2017 | vfio_teardown_msi(vdev); | |
2018 | vfio_unmap_bars(vdev); | |
2019 | out_put: | |
2020 | vfio_put_device(vdev); | |
2021 | vfio_put_group(group); | |
2022 | return ret; | |
2023 | } | |
2024 | ||
2025 | static void vfio_exitfn(PCIDevice *pdev) | |
2026 | { | |
2027 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
2028 | VFIOGroup *group = vdev->group; | |
2029 | ||
2030 | pci_device_set_intx_routing_notifier(&vdev->pdev, NULL); | |
2031 | vfio_disable_interrupts(vdev); | |
2032 | if (vdev->intx.mmap_timer) { | |
2033 | qemu_free_timer(vdev->intx.mmap_timer); | |
2034 | } | |
2035 | vfio_teardown_msi(vdev); | |
2036 | vfio_unmap_bars(vdev); | |
2037 | vfio_put_device(vdev); | |
2038 | vfio_put_group(group); | |
2039 | } | |
2040 | ||
2041 | static void vfio_pci_reset(DeviceState *dev) | |
2042 | { | |
2043 | PCIDevice *pdev = DO_UPCAST(PCIDevice, qdev, dev); | |
2044 | VFIODevice *vdev = DO_UPCAST(VFIODevice, pdev, pdev); | |
2045 | uint16_t cmd; | |
2046 | ||
2047 | DPRINTF("%s(%04x:%02x:%02x.%x)\n", __func__, vdev->host.domain, | |
2048 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
2049 | ||
2050 | vfio_disable_interrupts(vdev); | |
2051 | ||
2052 | /* | |
2053 | * Stop any ongoing DMA by disconecting I/O, MMIO, and bus master. | |
2054 | * Also put INTx Disable in known state. | |
2055 | */ | |
2056 | cmd = vfio_pci_read_config(pdev, PCI_COMMAND, 2); | |
2057 | cmd &= ~(PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER | | |
2058 | PCI_COMMAND_INTX_DISABLE); | |
2059 | vfio_pci_write_config(pdev, PCI_COMMAND, cmd, 2); | |
2060 | ||
2061 | if (vdev->reset_works) { | |
2062 | if (ioctl(vdev->fd, VFIO_DEVICE_RESET)) { | |
2063 | error_report("vfio: Error unable to reset physical device " | |
2064 | "(%04x:%02x:%02x.%x): %m\n", vdev->host.domain, | |
2065 | vdev->host.bus, vdev->host.slot, vdev->host.function); | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | vfio_enable_intx(vdev); | |
2070 | } | |
2071 | ||
2072 | static Property vfio_pci_dev_properties[] = { | |
2073 | DEFINE_PROP_PCI_HOST_DEVADDR("host", VFIODevice, host), | |
2074 | DEFINE_PROP_UINT32("x-intx-mmap-timeout-ms", VFIODevice, | |
2075 | intx.mmap_timeout, 1100), | |
2076 | /* | |
2077 | * TODO - support passed fds... is this necessary? | |
2078 | * DEFINE_PROP_STRING("vfiofd", VFIODevice, vfiofd_name), | |
2079 | * DEFINE_PROP_STRING("vfiogroupfd, VFIODevice, vfiogroupfd_name), | |
2080 | */ | |
2081 | DEFINE_PROP_END_OF_LIST(), | |
2082 | }; | |
2083 | ||
2084 | static const VMStateDescription vfio_pci_vmstate = { | |
2085 | .name = "vfio-pci", | |
2086 | .unmigratable = 1, | |
2087 | }; | |
2088 | ||
2089 | static void vfio_pci_dev_class_init(ObjectClass *klass, void *data) | |
2090 | { | |
2091 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2092 | PCIDeviceClass *pdc = PCI_DEVICE_CLASS(klass); | |
2093 | ||
2094 | dc->reset = vfio_pci_reset; | |
2095 | dc->props = vfio_pci_dev_properties; | |
2096 | dc->vmsd = &vfio_pci_vmstate; | |
2097 | dc->desc = "VFIO-based PCI device assignment"; | |
2098 | pdc->init = vfio_initfn; | |
2099 | pdc->exit = vfio_exitfn; | |
2100 | pdc->config_read = vfio_pci_read_config; | |
2101 | pdc->config_write = vfio_pci_write_config; | |
2102 | } | |
2103 | ||
2104 | static const TypeInfo vfio_pci_dev_info = { | |
2105 | .name = "vfio-pci", | |
2106 | .parent = TYPE_PCI_DEVICE, | |
2107 | .instance_size = sizeof(VFIODevice), | |
2108 | .class_init = vfio_pci_dev_class_init, | |
2109 | }; | |
2110 | ||
2111 | static void register_vfio_pci_dev_type(void) | |
2112 | { | |
2113 | type_register_static(&vfio_pci_dev_info); | |
2114 | } | |
2115 | ||
2116 | type_init(register_vfio_pci_dev_type) |