?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / vfio / pci / vfio_pci.c
blob66783a37f450cc5df7e7935e983c0f2365b73115
1 /*
2 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
3 * Author: Alex Williamson <alex.williamson@redhat.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
8 *
9 * Derived from original vfio:
10 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
11 * Author: Tom Lyon, pugs@cisco.com
12 */
13
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16 #include <linux/device.h>
17 #include <linux/eventfd.h>
18 #include <linux/file.h>
19 #include <linux/interrupt.h>
20 #include <linux/iommu.h>
21 #include <linux/module.h>
22 #include <linux/mutex.h>
23 #include <linux/notifier.h>
24 #include <linux/pci.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/slab.h>
27 #include <linux/types.h>
28 #include <linux/uaccess.h>
29 #include <linux/vfio.h>
30 #include <linux/vgaarb.h>
31 #include <linux/nospec.h>
32
33 #include "vfio_pci_private.h"
34
35 #define DRIVER_VERSION "0.2"
36 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>"
37 #define DRIVER_DESC "VFIO PCI - User Level meta-driver"
38
39 static char ids[1024] __initdata;
40 module_param_string(ids, ids, sizeof(ids), 0);
41 MODULE_PARM_DESC(ids, "Initial PCI IDs to add to the vfio driver, format is \"vendor:device[:subvendor[:subdevice[:class[:class_mask]]]]\" and multiple comma separated entries can be specified");
42
43 static bool nointxmask;
44 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
45 MODULE_PARM_DESC(nointxmask,
46 "Disable support for PCI 2.3 style INTx masking. If this resolves problems for specific devices, report lspci -vvvxxx to linux-pci@vger.kernel.org so the device can be fixed automatically via the broken_intx_masking flag.");
47
48 #ifdef CONFIG_VFIO_PCI_VGA
49 static bool disable_vga;
50 module_param(disable_vga, bool, S_IRUGO);
51 MODULE_PARM_DESC(disable_vga, "Disable VGA resource access through vfio-pci");
52 #endif
53
54 static bool disable_idle_d3;
55 module_param(disable_idle_d3, bool, S_IRUGO | S_IWUSR);
56 MODULE_PARM_DESC(disable_idle_d3,
57 "Disable using the PCI D3 low power state for idle, unused devices");
58
59 static DEFINE_MUTEX(driver_lock);
60
61 static inline bool vfio_vga_disabled(void)
62 {
63 #ifdef CONFIG_VFIO_PCI_VGA
64 return disable_vga;
65 #else
66 return true;
67 #endif
68 }
69
70 /*
71 * Our VGA arbiter participation is limited since we don't know anything
72 * about the device itself. However, if the device is the only VGA device
73 * downstream of a bridge and VFIO VGA support is disabled, then we can
74 * safely return legacy VGA IO and memory as not decoded since the user
75 * has no way to get to it and routing can be disabled externally at the
76 * bridge.
77 */
78 static unsigned int vfio_pci_set_vga_decode(void *opaque, bool single_vga)
79 {
80 struct vfio_pci_device *vdev = opaque;
81 struct pci_dev *tmp = NULL, *pdev = vdev->pdev;
82 unsigned char max_busnr;
83 unsigned int decodes;
84
85 if (single_vga || !vfio_vga_disabled() || pci_is_root_bus(pdev->bus))
86 return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
87 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
88
89 max_busnr = pci_bus_max_busnr(pdev->bus);
90 decodes = VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
91
92 while ((tmp = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, tmp)) != NULL) {
93 if (tmp == pdev ||
94 pci_domain_nr(tmp->bus) != pci_domain_nr(pdev->bus) ||
95 pci_is_root_bus(tmp->bus))
96 continue;
97
98 if (tmp->bus->number >= pdev->bus->number &&
99 tmp->bus->number <= max_busnr) {
100 pci_dev_put(tmp);
101 decodes |= VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
102 break;
103 }
104 }
105
106 return decodes;
107 }
108
109 static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
110 {
111 return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
112 }
113
114 static void vfio_pci_probe_mmaps(struct vfio_pci_device *vdev)
115 {
116 struct resource *res;
117 int bar;
118 struct vfio_pci_dummy_resource *dummy_res;
119
120 INIT_LIST_HEAD(&vdev->dummy_resources_list);
121
122 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
123 res = vdev->pdev->resource + bar;
124
125 if (!IS_ENABLED(CONFIG_VFIO_PCI_MMAP))
126 goto no_mmap;
127
128 if (!(res->flags & IORESOURCE_MEM))
129 goto no_mmap;
130
131 /*
132 * The PCI core shouldn't set up a resource with a
133 * type but zero size. But there may be bugs that
134 * cause us to do that.
135 */
136 if (!resource_size(res))
137 goto no_mmap;
138
139 if (resource_size(res) >= PAGE_SIZE) {
140 vdev->bar_mmap_supported[bar] = true;
141 continue;
142 }
143
144 if (!(res->start & ~PAGE_MASK)) {
145 /*
146 * Add a dummy resource to reserve the remainder
147 * of the exclusive page in case that hot-add
148 * device's bar is assigned into it.
149 */
150 dummy_res = kzalloc(sizeof(*dummy_res), GFP_KERNEL);
151 if (dummy_res == NULL)
152 goto no_mmap;
153
154 dummy_res->resource.name = "vfio sub-page reserved";
155 dummy_res->resource.start = res->end + 1;
156 dummy_res->resource.end = res->start + PAGE_SIZE - 1;
157 dummy_res->resource.flags = res->flags;
158 if (request_resource(res->parent,
159 &dummy_res->resource)) {
160 kfree(dummy_res);
161 goto no_mmap;
162 }
163 dummy_res->index = bar;
164 list_add(&dummy_res->res_next,
165 &vdev->dummy_resources_list);
166 vdev->bar_mmap_supported[bar] = true;
167 continue;
168 }
169 /*
170 * Here we don't handle the case when the BAR is not page
171 * aligned because we can't expect the BAR will be
172 * assigned into the same location in a page in guest
173 * when we passthrough the BAR. And it's hard to access
174 * this BAR in userspace because we have no way to get
175 * the BAR's location in a page.
176 */
177 no_mmap:
178 vdev->bar_mmap_supported[bar] = false;
179 }
180 }
181
182 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
183 static void vfio_pci_disable(struct vfio_pci_device *vdev);
184
185 /*
186 * INTx masking requires the ability to disable INTx signaling via PCI_COMMAND
187 * _and_ the ability detect when the device is asserting INTx via PCI_STATUS.
188 * If a device implements the former but not the latter we would typically
189 * expect broken_intx_masking be set and require an exclusive interrupt.
190 * However since we do have control of the device's ability to assert INTx,
191 * we can instead pretend that the device does not implement INTx, virtualizing
192 * the pin register to report zero and maintaining DisINTx set on the host.
193 */
194 static bool vfio_pci_nointx(struct pci_dev *pdev)
195 {
196 switch (pdev->vendor) {
197 case PCI_VENDOR_ID_INTEL:
198 switch (pdev->device) {
199 /* All i40e (XL710/X710/XXV710) 10/20/25/40GbE NICs */
200 case 0x1572:
201 case 0x1574:
202 case 0x1580 ... 0x1581:
203 case 0x1583 ... 0x158b:
204 case 0x37d0 ... 0x37d2:
205 return true;
206 default:
207 return false;
208 }
209 }
210
211 return false;
212 }
213
214 static int vfio_pci_enable(struct vfio_pci_device *vdev)
215 {
216 struct pci_dev *pdev = vdev->pdev;
217 int ret;
218 u16 cmd;
219 u8 msix_pos;
220
221 pci_set_power_state(pdev, PCI_D0);
222
223 /* Don't allow our initial saved state to include busmaster */
224 pci_clear_master(pdev);
225
226 ret = pci_enable_device(pdev);
227 if (ret)
228 return ret;
229
230 /* If reset fails because of the device lock, fail this path entirely */
231 ret = pci_try_reset_function(pdev);
232 if (ret == -EAGAIN) {
233 pci_disable_device(pdev);
234 return ret;
235 }
236
237 vdev->reset_works = !ret;
238 pci_save_state(pdev);
239 vdev->pci_saved_state = pci_store_saved_state(pdev);
240 if (!vdev->pci_saved_state)
241 pr_debug("%s: Couldn't store %s saved state\n",
242 __func__, dev_name(&pdev->dev));
243
244 if (likely(!nointxmask)) {
245 if (vfio_pci_nointx(pdev)) {
246 dev_info(&pdev->dev, "Masking broken INTx support\n");
247 vdev->nointx = true;
248 pci_intx(pdev, 0);
249 } else
250 vdev->pci_2_3 = pci_intx_mask_supported(pdev);
251 }
252
253 pci_read_config_word(pdev, PCI_COMMAND, &cmd);
254 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) {
255 cmd &= ~PCI_COMMAND_INTX_DISABLE;
256 pci_write_config_word(pdev, PCI_COMMAND, cmd);
257 }
258
259 ret = vfio_config_init(vdev);
260 if (ret) {
261 kfree(vdev->pci_saved_state);
262 vdev->pci_saved_state = NULL;
263 pci_disable_device(pdev);
264 return ret;
265 }
266
267 msix_pos = pdev->msix_cap;
268 if (msix_pos) {
269 u16 flags;
270 u32 table;
271
272 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags);
273 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table);
274
275 vdev->msix_bar = table & PCI_MSIX_TABLE_BIR;
276 vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET;
277 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16;
278 } else
279 vdev->msix_bar = 0xFF;
280
281 if (!vfio_vga_disabled() && vfio_pci_is_vga(pdev))
282 vdev->has_vga = true;
283
284
285 if (vfio_pci_is_vga(pdev) &&
286 pdev->vendor == PCI_VENDOR_ID_INTEL &&
287 IS_ENABLED(CONFIG_VFIO_PCI_IGD)) {
288 ret = vfio_pci_igd_init(vdev);
289 if (ret) {
290 dev_warn(&vdev->pdev->dev,
291 "Failed to setup Intel IGD regions\n");
292 vfio_pci_disable(vdev);
293 return ret;
294 }
295 }
296
297 vfio_pci_probe_mmaps(vdev);
298
299 return 0;
300 }
301
302 static void vfio_pci_disable(struct vfio_pci_device *vdev)
303 {
304 struct pci_dev *pdev = vdev->pdev;
305 struct vfio_pci_dummy_resource *dummy_res, *tmp;
306 struct vfio_pci_ioeventfd *ioeventfd, *ioeventfd_tmp;
307 int i, bar;
308
309 /* Stop the device from further DMA */
310 pci_clear_master(pdev);
311
312 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE |
313 VFIO_IRQ_SET_ACTION_TRIGGER,
314 vdev->irq_type, 0, 0, NULL);
315
316 /* Device closed, don't need mutex here */
317 list_for_each_entry_safe(ioeventfd, ioeventfd_tmp,
318 &vdev->ioeventfds_list, next) {
319 vfio_virqfd_disable(&ioeventfd->virqfd);
320 list_del(&ioeventfd->next);
321 kfree(ioeventfd);
322 }
323 vdev->ioeventfds_nr = 0;
324
325 vdev->virq_disabled = false;
326
327 for (i = 0; i < vdev->num_regions; i++)
328 vdev->region[i].ops->release(vdev, &vdev->region[i]);
329
330 vdev->num_regions = 0;
331 kfree(vdev->region);
332 vdev->region = NULL; /* don't krealloc a freed pointer */
333
334 vfio_config_free(vdev);
335
336 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
337 if (!vdev->barmap[bar])
338 continue;
339 pci_iounmap(pdev, vdev->barmap[bar]);
340 pci_release_selected_regions(pdev, 1 << bar);
341 vdev->barmap[bar] = NULL;
342 }
343
344 list_for_each_entry_safe(dummy_res, tmp,
345 &vdev->dummy_resources_list, res_next) {
346 list_del(&dummy_res->res_next);
347 release_resource(&dummy_res->resource);
348 kfree(dummy_res);
349 }
350
351 vdev->needs_reset = true;
352
353 /*
354 * If we have saved state, restore it. If we can reset the device,
355 * even better. Resetting with current state seems better than
356 * nothing, but saving and restoring current state without reset
357 * is just busy work.
358 */
359 if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) {
360 pr_info("%s: Couldn't reload %s saved state\n",
361 __func__, dev_name(&pdev->dev));
362
363 if (!vdev->reset_works)
364 goto out;
365
366 pci_save_state(pdev);
367 }
368
369 /*
370 * Disable INTx and MSI, presumably to avoid spurious interrupts
371 * during reset. Stolen from pci_reset_function()
372 */
373 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
374
375 /*
376 * Try to get the locks ourselves to prevent a deadlock. The
377 * success of this is dependent on being able to lock the device,
378 * which is not always possible.
379 * We can not use the "try" reset interface here, which will
380 * overwrite the previously restored configuration information.
381 */
382 if (vdev->reset_works && pci_cfg_access_trylock(pdev)) {
383 if (device_trylock(&pdev->dev)) {
384 if (!__pci_reset_function_locked(pdev))
385 vdev->needs_reset = false;
386 device_unlock(&pdev->dev);
387 }
388 pci_cfg_access_unlock(pdev);
389 }
390
391 pci_restore_state(pdev);
392 out:
393 pci_disable_device(pdev);
394
395 vfio_pci_try_bus_reset(vdev);
396
397 if (!disable_idle_d3)
398 pci_set_power_state(pdev, PCI_D3hot);
399 }
400
401 static void vfio_pci_release(void *device_data)
402 {
403 struct vfio_pci_device *vdev = device_data;
404
405 mutex_lock(&driver_lock);
406
407 if (!(--vdev->refcnt)) {
408 vfio_spapr_pci_eeh_release(vdev->pdev);
409 vfio_pci_disable(vdev);
410 }
411
412 mutex_unlock(&driver_lock);
413
414 module_put(THIS_MODULE);
415 }
416
417 static int vfio_pci_open(void *device_data)
418 {
419 struct vfio_pci_device *vdev = device_data;
420 int ret = 0;
421
422 if (!try_module_get(THIS_MODULE))
423 return -ENODEV;
424
425 mutex_lock(&driver_lock);
426
427 if (!vdev->refcnt) {
428 ret = vfio_pci_enable(vdev);
429 if (ret)
430 goto error;
431
432 vfio_spapr_pci_eeh_open(vdev->pdev);
433 }
434 vdev->refcnt++;
435 error:
436 mutex_unlock(&driver_lock);
437 if (ret)
438 module_put(THIS_MODULE);
439 return ret;
440 }
441
442 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type)
443 {
444 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) {
445 u8 pin;
446
447 if (!IS_ENABLED(CONFIG_VFIO_PCI_INTX) ||
448 vdev->nointx || vdev->pdev->is_virtfn)
449 return 0;
450
451 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin);
452
453 return pin ? 1 : 0;
454 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) {
455 u8 pos;
456 u16 flags;
457
458 pos = vdev->pdev->msi_cap;
459 if (pos) {
460 pci_read_config_word(vdev->pdev,
461 pos + PCI_MSI_FLAGS, &flags);
462 return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1);
463 }
464 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) {
465 u8 pos;
466 u16 flags;
467
468 pos = vdev->pdev->msix_cap;
469 if (pos) {
470 pci_read_config_word(vdev->pdev,
471 pos + PCI_MSIX_FLAGS, &flags);
472
473 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1;
474 }
475 } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) {
476 if (pci_is_pcie(vdev->pdev))
477 return 1;
478 } else if (irq_type == VFIO_PCI_REQ_IRQ_INDEX) {
479 return 1;
480 }
481
482 return 0;
483 }
484
485 static int vfio_pci_count_devs(struct pci_dev *pdev, void *data)
486 {
487 (*(int *)data)++;
488 return 0;
489 }
490
491 struct vfio_pci_fill_info {
492 int max;
493 int cur;
494 struct vfio_pci_dependent_device *devices;
495 };
496
497 static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data)
498 {
499 struct vfio_pci_fill_info *fill = data;
500 struct iommu_group *iommu_group;
501
502 if (fill->cur == fill->max)
503 return -EAGAIN; /* Something changed, try again */
504
505 iommu_group = iommu_group_get(&pdev->dev);
506 if (!iommu_group)
507 return -EPERM; /* Cannot reset non-isolated devices */
508
509 fill->devices[fill->cur].group_id = iommu_group_id(iommu_group);
510 fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus);
511 fill->devices[fill->cur].bus = pdev->bus->number;
512 fill->devices[fill->cur].devfn = pdev->devfn;
513 fill->cur++;
514 iommu_group_put(iommu_group);
515 return 0;
516 }
517
518 struct vfio_pci_group_entry {
519 struct vfio_group *group;
520 int id;
521 };
522
523 struct vfio_pci_group_info {
524 int count;
525 struct vfio_pci_group_entry *groups;
526 };
527
528 static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data)
529 {
530 struct vfio_pci_group_info *info = data;
531 struct iommu_group *group;
532 int id, i;
533
534 group = iommu_group_get(&pdev->dev);
535 if (!group)
536 return -EPERM;
537
538 id = iommu_group_id(group);
539
540 for (i = 0; i < info->count; i++)
541 if (info->groups[i].id == id)
542 break;
543
544 iommu_group_put(group);
545
546 return (i == info->count) ? -EINVAL : 0;
547 }
548
549 static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot)
550 {
551 for (; pdev; pdev = pdev->bus->self)
552 if (pdev->bus == slot->bus)
553 return (pdev->slot == slot);
554 return false;
555 }
556
557 struct vfio_pci_walk_info {
558 int (*fn)(struct pci_dev *, void *data);
559 void *data;
560 struct pci_dev *pdev;
561 bool slot;
562 int ret;
563 };
564
565 static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data)
566 {
567 struct vfio_pci_walk_info *walk = data;
568
569 if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot))
570 walk->ret = walk->fn(pdev, walk->data);
571
572 return walk->ret;
573 }
574
575 static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev,
576 int (*fn)(struct pci_dev *,
577 void *data), void *data,
578 bool slot)
579 {
580 struct vfio_pci_walk_info walk = {
581 .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0,
582 };
583
584 pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk);
585
586 return walk.ret;
587 }
588
589 static int msix_mmappable_cap(struct vfio_pci_device *vdev,
590 struct vfio_info_cap *caps)
591 {
592 struct vfio_info_cap_header header = {
593 .id = VFIO_REGION_INFO_CAP_MSIX_MAPPABLE,
594 .version = 1
595 };
596
597 return vfio_info_add_capability(caps, &header, sizeof(header));
598 }
599
600 int vfio_pci_register_dev_region(struct vfio_pci_device *vdev,
601 unsigned int type, unsigned int subtype,
602 const struct vfio_pci_regops *ops,
603 size_t size, u32 flags, void *data)
604 {
605 struct vfio_pci_region *region;
606
607 region = krealloc(vdev->region,
608 (vdev->num_regions + 1) * sizeof(*region),
609 GFP_KERNEL);
610 if (!region)
611 return -ENOMEM;
612
613 vdev->region = region;
614 vdev->region[vdev->num_regions].type = type;
615 vdev->region[vdev->num_regions].subtype = subtype;
616 vdev->region[vdev->num_regions].ops = ops;
617 vdev->region[vdev->num_regions].size = size;
618 vdev->region[vdev->num_regions].flags = flags;
619 vdev->region[vdev->num_regions].data = data;
620
621 vdev->num_regions++;
622
623 return 0;
624 }
625
626 static long vfio_pci_ioctl(void *device_data,
627 unsigned int cmd, unsigned long arg)
628 {
629 struct vfio_pci_device *vdev = device_data;
630 unsigned long minsz;
631
632 if (cmd == VFIO_DEVICE_GET_INFO) {
633 struct vfio_device_info info;
634
635 minsz = offsetofend(struct vfio_device_info, num_irqs);
636
637 if (copy_from_user(&info, (void __user *)arg, minsz))
638 return -EFAULT;
639
640 if (info.argsz < minsz)
641 return -EINVAL;
642
643 info.flags = VFIO_DEVICE_FLAGS_PCI;
644
645 if (vdev->reset_works)
646 info.flags |= VFIO_DEVICE_FLAGS_RESET;
647
648 info.num_regions = VFIO_PCI_NUM_REGIONS + vdev->num_regions;
649 info.num_irqs = VFIO_PCI_NUM_IRQS;
650
651 return copy_to_user((void __user *)arg, &info, minsz) ?
652 -EFAULT : 0;
653
654 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
655 struct pci_dev *pdev = vdev->pdev;
656 struct vfio_region_info info;
657 struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
658 int i, ret;
659
660 minsz = offsetofend(struct vfio_region_info, offset);
661
662 if (copy_from_user(&info, (void __user *)arg, minsz))
663 return -EFAULT;
664
665 if (info.argsz < minsz)
666 return -EINVAL;
667
668 switch (info.index) {
669 case VFIO_PCI_CONFIG_REGION_INDEX:
670 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
671 info.size = pdev->cfg_size;
672 info.flags = VFIO_REGION_INFO_FLAG_READ |
673 VFIO_REGION_INFO_FLAG_WRITE;
674 break;
675 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
676 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
677 info.size = pci_resource_len(pdev, info.index);
678 if (!info.size) {
679 info.flags = 0;
680 break;
681 }
682
683 info.flags = VFIO_REGION_INFO_FLAG_READ |
684 VFIO_REGION_INFO_FLAG_WRITE;
685 if (vdev->bar_mmap_supported[info.index]) {
686 info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
687 if (info.index == vdev->msix_bar) {
688 ret = msix_mmappable_cap(vdev, &caps);
689 if (ret)
690 return ret;
691 }
692 }
693
694 break;
695 case VFIO_PCI_ROM_REGION_INDEX:
696 {
697 void __iomem *io;
698 size_t size;
699 u16 orig_cmd;
700
701 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
702 info.flags = 0;
703
704 /* Report the BAR size, not the ROM size */
705 info.size = pci_resource_len(pdev, info.index);
706 if (!info.size) {
707 /* Shadow ROMs appear as PCI option ROMs */
708 if (pdev->resource[PCI_ROM_RESOURCE].flags &
709 IORESOURCE_ROM_SHADOW)
710 info.size = 0x20000;
711 else
712 break;
713 }
714
715 /*
716 * Is it really there? Enable memory decode for
717 * implicit access in pci_map_rom().
718 */
719 pci_read_config_word(pdev, PCI_COMMAND, &orig_cmd);
720 pci_write_config_word(pdev, PCI_COMMAND,
721 orig_cmd | PCI_COMMAND_MEMORY);
722
723 io = pci_map_rom(pdev, &size);
724 if (io) {
725 info.flags = VFIO_REGION_INFO_FLAG_READ;
726 pci_unmap_rom(pdev, io);
727 } else {
728 info.size = 0;
729 }
730
731 pci_write_config_word(pdev, PCI_COMMAND, orig_cmd);
732 break;
733 }
734 case VFIO_PCI_VGA_REGION_INDEX:
735 if (!vdev->has_vga)
736 return -EINVAL;
737
738 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
739 info.size = 0xc0000;
740 info.flags = VFIO_REGION_INFO_FLAG_READ |
741 VFIO_REGION_INFO_FLAG_WRITE;
742
743 break;
744 default:
745 {
746 struct vfio_region_info_cap_type cap_type = {
747 .header.id = VFIO_REGION_INFO_CAP_TYPE,
748 .header.version = 1 };
749
750 if (info.index >=
751 VFIO_PCI_NUM_REGIONS + vdev->num_regions)
752 return -EINVAL;
753 info.index = array_index_nospec(info.index,
754 VFIO_PCI_NUM_REGIONS +
755 vdev->num_regions);
756
757 i = info.index - VFIO_PCI_NUM_REGIONS;
758
759 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
760 info.size = vdev->region[i].size;
761 info.flags = vdev->region[i].flags;
762
763 cap_type.type = vdev->region[i].type;
764 cap_type.subtype = vdev->region[i].subtype;
765
766 ret = vfio_info_add_capability(&caps, &cap_type.header,
767 sizeof(cap_type));
768 if (ret)
769 return ret;
770
771 }
772 }
773
774 if (caps.size) {
775 info.flags |= VFIO_REGION_INFO_FLAG_CAPS;
776 if (info.argsz < sizeof(info) + caps.size) {
777 info.argsz = sizeof(info) + caps.size;
778 info.cap_offset = 0;
779 } else {
780 vfio_info_cap_shift(&caps, sizeof(info));
781 if (copy_to_user((void __user *)arg +
782 sizeof(info), caps.buf,
783 caps.size)) {
784 kfree(caps.buf);
785 return -EFAULT;
786 }
787 info.cap_offset = sizeof(info);
788 }
789
790 kfree(caps.buf);
791 }
792
793 return copy_to_user((void __user *)arg, &info, minsz) ?
794 -EFAULT : 0;
795
796 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
797 struct vfio_irq_info info;
798
799 minsz = offsetofend(struct vfio_irq_info, count);
800
801 if (copy_from_user(&info, (void __user *)arg, minsz))
802 return -EFAULT;
803
804 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
805 return -EINVAL;
806
807 switch (info.index) {
808 case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX:
809 case VFIO_PCI_REQ_IRQ_INDEX:
810 break;
811 case VFIO_PCI_ERR_IRQ_INDEX:
812 if (pci_is_pcie(vdev->pdev))
813 break;
814 /* fall through */
815 default:
816 return -EINVAL;
817 }
818
819 info.flags = VFIO_IRQ_INFO_EVENTFD;
820
821 info.count = vfio_pci_get_irq_count(vdev, info.index);
822
823 if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
824 info.flags |= (VFIO_IRQ_INFO_MASKABLE |
825 VFIO_IRQ_INFO_AUTOMASKED);
826 else
827 info.flags |= VFIO_IRQ_INFO_NORESIZE;
828
829 return copy_to_user((void __user *)arg, &info, minsz) ?
830 -EFAULT : 0;
831
832 } else if (cmd == VFIO_DEVICE_SET_IRQS) {
833 struct vfio_irq_set hdr;
834 u8 *data = NULL;
835 int max, ret = 0;
836 size_t data_size = 0;
837
838 minsz = offsetofend(struct vfio_irq_set, count);
839
840 if (copy_from_user(&hdr, (void __user *)arg, minsz))
841 return -EFAULT;
842
843 max = vfio_pci_get_irq_count(vdev, hdr.index);
844
845 ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
846 VFIO_PCI_NUM_IRQS, &data_size);
847 if (ret)
848 return ret;
849
850 if (data_size) {
851 data = memdup_user((void __user *)(arg + minsz),
852 data_size);
853 if (IS_ERR(data))
854 return PTR_ERR(data);
855 }
856
857 mutex_lock(&vdev->igate);
858
859 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index,
860 hdr.start, hdr.count, data);
861
862 mutex_unlock(&vdev->igate);
863 kfree(data);
864
865 return ret;
866
867 } else if (cmd == VFIO_DEVICE_RESET) {
868 return vdev->reset_works ?
869 pci_try_reset_function(vdev->pdev) : -EINVAL;
870
871 } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) {
872 struct vfio_pci_hot_reset_info hdr;
873 struct vfio_pci_fill_info fill = { 0 };
874 struct vfio_pci_dependent_device *devices = NULL;
875 bool slot = false;
876 int ret = 0;
877
878 minsz = offsetofend(struct vfio_pci_hot_reset_info, count);
879
880 if (copy_from_user(&hdr, (void __user *)arg, minsz))
881 return -EFAULT;
882
883 if (hdr.argsz < minsz)
884 return -EINVAL;
885
886 hdr.flags = 0;
887
888 /* Can we do a slot or bus reset or neither? */
889 if (!pci_probe_reset_slot(vdev->pdev->slot))
890 slot = true;
891 else if (pci_probe_reset_bus(vdev->pdev->bus))
892 return -ENODEV;
893
894 /* How many devices are affected? */
895 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
896 vfio_pci_count_devs,
897 &fill.max, slot);
898 if (ret)
899 return ret;
900
901 WARN_ON(!fill.max); /* Should always be at least one */
902
903 /*
904 * If there's enough space, fill it now, otherwise return
905 * -ENOSPC and the number of devices affected.
906 */
907 if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) {
908 ret = -ENOSPC;
909 hdr.count = fill.max;
910 goto reset_info_exit;
911 }
912
913 devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL);
914 if (!devices)
915 return -ENOMEM;
916
917 fill.devices = devices;
918
919 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
920 vfio_pci_fill_devs,
921 &fill, slot);
922
923 /*
924 * If a device was removed between counting and filling,
925 * we may come up short of fill.max. If a device was
926 * added, we'll have a return of -EAGAIN above.
927 */
928 if (!ret)
929 hdr.count = fill.cur;
930
931 reset_info_exit:
932 if (copy_to_user((void __user *)arg, &hdr, minsz))
933 ret = -EFAULT;
934
935 if (!ret) {
936 if (copy_to_user((void __user *)(arg + minsz), devices,
937 hdr.count * sizeof(*devices)))
938 ret = -EFAULT;
939 }
940
941 kfree(devices);
942 return ret;
943
944 } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) {
945 struct vfio_pci_hot_reset hdr;
946 int32_t *group_fds;
947 struct vfio_pci_group_entry *groups;
948 struct vfio_pci_group_info info;
949 bool slot = false;
950 int i, count = 0, ret = 0;
951
952 minsz = offsetofend(struct vfio_pci_hot_reset, count);
953
954 if (copy_from_user(&hdr, (void __user *)arg, minsz))
955 return -EFAULT;
956
957 if (hdr.argsz < minsz || hdr.flags)
958 return -EINVAL;
959
960 /* Can we do a slot or bus reset or neither? */
961 if (!pci_probe_reset_slot(vdev->pdev->slot))
962 slot = true;
963 else if (pci_probe_reset_bus(vdev->pdev->bus))
964 return -ENODEV;
965
966 /*
967 * We can't let userspace give us an arbitrarily large
968 * buffer to copy, so verify how many we think there
969 * could be. Note groups can have multiple devices so
970 * one group per device is the max.
971 */
972 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
973 vfio_pci_count_devs,
974 &count, slot);
975 if (ret)
976 return ret;
977
978 /* Somewhere between 1 and count is OK */
979 if (!hdr.count || hdr.count > count)
980 return -EINVAL;
981
982 group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL);
983 groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL);
984 if (!group_fds || !groups) {
985 kfree(group_fds);
986 kfree(groups);
987 return -ENOMEM;
988 }
989
990 if (copy_from_user(group_fds, (void __user *)(arg + minsz),
991 hdr.count * sizeof(*group_fds))) {
992 kfree(group_fds);
993 kfree(groups);
994 return -EFAULT;
995 }
996
997 /*
998 * For each group_fd, get the group through the vfio external
999 * user interface and store the group and iommu ID. This
1000 * ensures the group is held across the reset.
1001 */
1002 for (i = 0; i < hdr.count; i++) {
1003 struct vfio_group *group;
1004 struct fd f = fdget(group_fds[i]);
1005 if (!f.file) {
1006 ret = -EBADF;
1007 break;
1008 }
1009
1010 group = vfio_group_get_external_user(f.file);
1011 fdput(f);
1012 if (IS_ERR(group)) {
1013 ret = PTR_ERR(group);
1014 break;
1015 }
1016
1017 groups[i].group = group;
1018 groups[i].id = vfio_external_user_iommu_id(group);
1019 }
1020
1021 kfree(group_fds);
1022
1023 /* release reference to groups on error */
1024 if (ret)
1025 goto hot_reset_release;
1026
1027 info.count = hdr.count;
1028 info.groups = groups;
1029
1030 /*
1031 * Test whether all the affected devices are contained
1032 * by the set of groups provided by the user.
1033 */
1034 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
1035 vfio_pci_validate_devs,
1036 &info, slot);
1037 if (!ret)
1038 /* User has access, do the reset */
1039 ret = pci_reset_bus(vdev->pdev);
1040
1041 hot_reset_release:
1042 for (i--; i >= 0; i--)
1043 vfio_group_put_external_user(groups[i].group);
1044
1045 kfree(groups);
1046 return ret;
1047 } else if (cmd == VFIO_DEVICE_IOEVENTFD) {
1048 struct vfio_device_ioeventfd ioeventfd;
1049 int count;
1050
1051 minsz = offsetofend(struct vfio_device_ioeventfd, fd);
1052
1053 if (copy_from_user(&ioeventfd, (void __user *)arg, minsz))
1054 return -EFAULT;
1055
1056 if (ioeventfd.argsz < minsz)
1057 return -EINVAL;
1058
1059 if (ioeventfd.flags & ~VFIO_DEVICE_IOEVENTFD_SIZE_MASK)
1060 return -EINVAL;
1061
1062 count = ioeventfd.flags & VFIO_DEVICE_IOEVENTFD_SIZE_MASK;
1063
1064 if (hweight8(count) != 1 || ioeventfd.fd < -1)
1065 return -EINVAL;
1066
1067 return vfio_pci_ioeventfd(vdev, ioeventfd.offset,
1068 ioeventfd.data, count, ioeventfd.fd);
1069 }
1070
1071 return -ENOTTY;
1072 }
1073
1074 static ssize_t vfio_pci_rw(void *device_data, char __user *buf,
1075 size_t count, loff_t *ppos, bool iswrite)
1076 {
1077 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
1078 struct vfio_pci_device *vdev = device_data;
1079
1080 if (index >= VFIO_PCI_NUM_REGIONS + vdev->num_regions)
1081 return -EINVAL;
1082
1083 switch (index) {
1084 case VFIO_PCI_CONFIG_REGION_INDEX:
1085 return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite);
1086
1087 case VFIO_PCI_ROM_REGION_INDEX:
1088 if (iswrite)
1089 return -EINVAL;
1090 return vfio_pci_bar_rw(vdev, buf, count, ppos, false);
1091
1092 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
1093 return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite);
1094
1095 case VFIO_PCI_VGA_REGION_INDEX:
1096 return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite);
1097 default:
1098 index -= VFIO_PCI_NUM_REGIONS;
1099 return vdev->region[index].ops->rw(vdev, buf,
1100 count, ppos, iswrite);
1101 }
1102
1103 return -EINVAL;
1104 }
1105
1106 static ssize_t vfio_pci_read(void *device_data, char __user *buf,
1107 size_t count, loff_t *ppos)
1108 {
1109 if (!count)
1110 return 0;
1111
1112 return vfio_pci_rw(device_data, buf, count, ppos, false);
1113 }
1114
1115 static ssize_t vfio_pci_write(void *device_data, const char __user *buf,
1116 size_t count, loff_t *ppos)
1117 {
1118 if (!count)
1119 return 0;
1120
1121 return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true);
1122 }
1123
1124 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
1125 {
1126 struct vfio_pci_device *vdev = device_data;
1127 struct pci_dev *pdev = vdev->pdev;
1128 unsigned int index;
1129 u64 phys_len, req_len, pgoff, req_start;
1130 int ret;
1131
1132 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
1133
1134 if (vma->vm_end < vma->vm_start)
1135 return -EINVAL;
1136 if ((vma->vm_flags & VM_SHARED) == 0)
1137 return -EINVAL;
1138 if (index >= VFIO_PCI_ROM_REGION_INDEX)
1139 return -EINVAL;
1140 if (!vdev->bar_mmap_supported[index])
1141 return -EINVAL;
1142
1143 phys_len = PAGE_ALIGN(pci_resource_len(pdev, index));
1144 req_len = vma->vm_end - vma->vm_start;
1145 pgoff = vma->vm_pgoff &
1146 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
1147 req_start = pgoff << PAGE_SHIFT;
1148
1149 if (req_start + req_len > phys_len)
1150 return -EINVAL;
1151
1152 /*
1153 * Even though we don't make use of the barmap for the mmap,
1154 * we need to request the region and the barmap tracks that.
1155 */
1156 if (!vdev->barmap[index]) {
1157 ret = pci_request_selected_regions(pdev,
1158 1 << index, "vfio-pci");
1159 if (ret)
1160 return ret;
1161
1162 vdev->barmap[index] = pci_iomap(pdev, index, 0);
1163 if (!vdev->barmap[index]) {
1164 pci_release_selected_regions(pdev, 1 << index);
1165 return -ENOMEM;
1166 }
1167 }
1168
1169 vma->vm_private_data = vdev;
1170 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1171 vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff;
1172
1173 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
1174 req_len, vma->vm_page_prot);
1175 }
1176
1177 static void vfio_pci_request(void *device_data, unsigned int count)
1178 {
1179 struct vfio_pci_device *vdev = device_data;
1180
1181 mutex_lock(&vdev->igate);
1182
1183 if (vdev->req_trigger) {
1184 if (!(count % 10))
1185 dev_notice_ratelimited(&vdev->pdev->dev,
1186 "Relaying device request to user (#%u)\n",
1187 count);
1188 eventfd_signal(vdev->req_trigger, 1);
1189 } else if (count == 0) {
1190 dev_warn(&vdev->pdev->dev,
1191 "No device request channel registered, blocked until released by user\n");
1192 }
1193
1194 mutex_unlock(&vdev->igate);
1195 }
1196
1197 static const struct vfio_device_ops vfio_pci_ops = {
1198 .name = "vfio-pci",
1199 .open = vfio_pci_open,
1200 .release = vfio_pci_release,
1201 .ioctl = vfio_pci_ioctl,
1202 .read = vfio_pci_read,
1203 .write = vfio_pci_write,
1204 .mmap = vfio_pci_mmap,
1205 .request = vfio_pci_request,
1206 };
1207
1208 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1209 {
1210 struct vfio_pci_device *vdev;
1211 struct iommu_group *group;
1212 int ret;
1213
1214 if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL)
1215 return -EINVAL;
1216
1217 /*
1218 * Prevent binding to PFs with VFs enabled, this too easily allows
1219 * userspace instance with VFs and PFs from the same device, which
1220 * cannot work. Disabling SR-IOV here would initiate removing the
1221 * VFs, which would unbind the driver, which is prone to blocking
1222 * if that VF is also in use by vfio-pci. Just reject these PFs
1223 * and let the user sort it out.
1224 */
1225 if (pci_num_vf(pdev)) {
1226 pci_warn(pdev, "Cannot bind to PF with SR-IOV enabled\n");
1227 return -EBUSY;
1228 }
1229
1230 group = vfio_iommu_group_get(&pdev->dev);
1231 if (!group)
1232 return -EINVAL;
1233
1234 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
1235 if (!vdev) {
1236 vfio_iommu_group_put(group, &pdev->dev);
1237 return -ENOMEM;
1238 }
1239
1240 vdev->pdev = pdev;
1241 vdev->irq_type = VFIO_PCI_NUM_IRQS;
1242 mutex_init(&vdev->igate);
1243 spin_lock_init(&vdev->irqlock);
1244 mutex_init(&vdev->ioeventfds_lock);
1245 INIT_LIST_HEAD(&vdev->ioeventfds_list);
1246
1247 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
1248 if (ret) {
1249 vfio_iommu_group_put(group, &pdev->dev);
1250 kfree(vdev);
1251 return ret;
1252 }
1253
1254 if (vfio_pci_is_vga(pdev)) {
1255 vga_client_register(pdev, vdev, NULL, vfio_pci_set_vga_decode);
1256 vga_set_legacy_decoding(pdev,
1257 vfio_pci_set_vga_decode(vdev, false));
1258 }
1259
1260 if (!disable_idle_d3) {
1261 /*
1262 * pci-core sets the device power state to an unknown value at
1263 * bootup and after being removed from a driver. The only
1264 * transition it allows from this unknown state is to D0, which
1265 * typically happens when a driver calls pci_enable_device().
1266 * We're not ready to enable the device yet, but we do want to
1267 * be able to get to D3. Therefore first do a D0 transition
1268 * before going to D3.
1269 */
1270 pci_set_power_state(pdev, PCI_D0);
1271 pci_set_power_state(pdev, PCI_D3hot);
1272 }
1273
1274 return ret;
1275 }
1276
1277 static void vfio_pci_remove(struct pci_dev *pdev)
1278 {
1279 struct vfio_pci_device *vdev;
1280
1281 vdev = vfio_del_group_dev(&pdev->dev);
1282 if (!vdev)
1283 return;
1284
1285 vfio_iommu_group_put(pdev->dev.iommu_group, &pdev->dev);
1286 kfree(vdev->region);
1287 mutex_destroy(&vdev->ioeventfds_lock);
1288 kfree(vdev);
1289
1290 if (vfio_pci_is_vga(pdev)) {
1291 vga_client_register(pdev, NULL, NULL, NULL);
1292 vga_set_legacy_decoding(pdev,
1293 VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
1294 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM);
1295 }
1296
1297 if (!disable_idle_d3)
1298 pci_set_power_state(pdev, PCI_D0);
1299 }
1300
1301 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev,
1302 pci_channel_state_t state)
1303 {
1304 struct vfio_pci_device *vdev;
1305 struct vfio_device *device;
1306
1307 device = vfio_device_get_from_dev(&pdev->dev);
1308 if (device == NULL)
1309 return PCI_ERS_RESULT_DISCONNECT;
1310
1311 vdev = vfio_device_data(device);
1312 if (vdev == NULL) {
1313 vfio_device_put(device);
1314 return PCI_ERS_RESULT_DISCONNECT;
1315 }
1316
1317 mutex_lock(&vdev->igate);
1318
1319 if (vdev->err_trigger)
1320 eventfd_signal(vdev->err_trigger, 1);
1321
1322 mutex_unlock(&vdev->igate);
1323
1324 vfio_device_put(device);
1325
1326 return PCI_ERS_RESULT_CAN_RECOVER;
1327 }
1328
1329 static const struct pci_error_handlers vfio_err_handlers = {
1330 .error_detected = vfio_pci_aer_err_detected,
1331 };
1332
1333 static struct pci_driver vfio_pci_driver = {
1334 .name = "vfio-pci",
1335 .id_table = NULL, /* only dynamic ids */
1336 .probe = vfio_pci_probe,
1337 .remove = vfio_pci_remove,
1338 .err_handler = &vfio_err_handlers,
1339 };
1340
1341 struct vfio_devices {
1342 struct vfio_device **devices;
1343 int cur_index;
1344 int max_index;
1345 };
1346
1347 static int vfio_pci_get_devs(struct pci_dev *pdev, void *data)
1348 {
1349 struct vfio_devices *devs = data;
1350 struct vfio_device *device;
1351
1352 if (devs->cur_index == devs->max_index)
1353 return -ENOSPC;
1354
1355 device = vfio_device_get_from_dev(&pdev->dev);
1356 if (!device)
1357 return -EINVAL;
1358
1359 if (pci_dev_driver(pdev) != &vfio_pci_driver) {
1360 vfio_device_put(device);
1361 return -EBUSY;
1362 }
1363
1364 devs->devices[devs->cur_index++] = device;
1365 return 0;
1366 }
1367
1368 /*
1369 * Attempt to do a bus/slot reset if there are devices affected by a reset for
1370 * this device that are needs_reset and all of the affected devices are unused
1371 * (!refcnt). Callers are required to hold driver_lock when calling this to
1372 * prevent device opens and concurrent bus reset attempts. We prevent device
1373 * unbinds by acquiring and holding a reference to the vfio_device.
1374 *
1375 * NB: vfio-core considers a group to be viable even if some devices are
1376 * bound to drivers like pci-stub or pcieport. Here we require all devices
1377 * to be bound to vfio_pci since that's the only way we can be sure they
1378 * stay put.
1379 */
1380 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
1381 {
1382 struct vfio_devices devs = { .cur_index = 0 };
1383 int i = 0, ret = -EINVAL;
1384 bool needs_reset = false, slot = false;
1385 struct vfio_pci_device *tmp;
1386
1387 if (!pci_probe_reset_slot(vdev->pdev->slot))
1388 slot = true;
1389 else if (pci_probe_reset_bus(vdev->pdev->bus))
1390 return;
1391
1392 if (vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs,
1393 &i, slot) || !i)
1394 return;
1395
1396 devs.max_index = i;
1397 devs.devices = kcalloc(i, sizeof(struct vfio_device *), GFP_KERNEL);
1398 if (!devs.devices)
1399 return;
1400
1401 if (vfio_pci_for_each_slot_or_bus(vdev->pdev,
1402 vfio_pci_get_devs, &devs, slot))
1403 goto put_devs;
1404
1405 for (i = 0; i < devs.cur_index; i++) {
1406 tmp = vfio_device_data(devs.devices[i]);
1407 if (tmp->needs_reset)
1408 needs_reset = true;
1409 if (tmp->refcnt)
1410 goto put_devs;
1411 }
1412
1413 if (needs_reset)
1414 ret = pci_reset_bus(vdev->pdev);
1415
1416 put_devs:
1417 for (i = 0; i < devs.cur_index; i++) {
1418 tmp = vfio_device_data(devs.devices[i]);
1419 if (!ret)
1420 tmp->needs_reset = false;
1421
1422 if (!tmp->refcnt && !disable_idle_d3)
1423 pci_set_power_state(tmp->pdev, PCI_D3hot);
1424
1425 vfio_device_put(devs.devices[i]);
1426 }
1427
1428 kfree(devs.devices);
1429 }
1430
1431 static void __exit vfio_pci_cleanup(void)
1432 {
1433 pci_unregister_driver(&vfio_pci_driver);
1434 vfio_pci_uninit_perm_bits();
1435 }
1436
1437 static void __init vfio_pci_fill_ids(void)
1438 {
1439 char *p, *id;
1440 int rc;
1441
1442 /* no ids passed actually */
1443 if (ids[0] == '\0')
1444 return;
1445
1446 /* add ids specified in the module parameter */
1447 p = ids;
1448 while ((id = strsep(&p, ","))) {
1449 unsigned int vendor, device, subvendor = PCI_ANY_ID,
1450 subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
1451 int fields;
1452
1453 if (!strlen(id))
1454 continue;
1455
1456 fields = sscanf(id, "%x:%x:%x:%x:%x:%x",
1457 &vendor, &device, &subvendor, &subdevice,
1458 &class, &class_mask);
1459
1460 if (fields < 2) {
1461 pr_warn("invalid id string \"%s\"\n", id);
1462 continue;
1463 }
1464
1465 rc = pci_add_dynid(&vfio_pci_driver, vendor, device,
1466 subvendor, subdevice, class, class_mask, 0);
1467 if (rc)
1468 pr_warn("failed to add dynamic id [%04x:%04x[%04x:%04x]] class %#08x/%08x (%d)\n",
1469 vendor, device, subvendor, subdevice,
1470 class, class_mask, rc);
1471 else
1472 pr_info("add [%04x:%04x[%04x:%04x]] class %#08x/%08x\n",
1473 vendor, device, subvendor, subdevice,
1474 class, class_mask);
1475 }
1476 }
1477
1478 static int __init vfio_pci_init(void)
1479 {
1480 int ret;
1481
1482 /* Allocate shared config space permision data used by all devices */
1483 ret = vfio_pci_init_perm_bits();
1484 if (ret)
1485 return ret;
1486
1487 /* Register and scan for devices */
1488 ret = pci_register_driver(&vfio_pci_driver);
1489 if (ret)
1490 goto out_driver;
1491
1492 vfio_pci_fill_ids();
1493
1494 return 0;
1495
1496 out_driver:
1497 vfio_pci_uninit_perm_bits();
1498 return ret;
1499 }
1500
1501 module_init(vfio_pci_init);
1502 module_exit(vfio_pci_cleanup);
1503
1504 MODULE_VERSION(DRIVER_VERSION);
1505 MODULE_LICENSE("GPL v2");
1506 MODULE_AUTHOR(DRIVER_AUTHOR);
1507 MODULE_DESCRIPTION(DRIVER_DESC);
Linux with added FPC-III support