?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm thin metadata: fix __udivdi3 undefined on 32-bit
[linux/fpc-iii.git] / drivers / vfio / pci / vfio_pci.c
blobb31b84f56e8fd5a68678f2e842b9f9115174b012
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
32 #include "vfio_pci_private.h"
33
34 #define DRIVER_VERSION "0.2"
35 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>"
36 #define DRIVER_DESC "VFIO PCI - User Level meta-driver"
37
38 static char ids[1024] __initdata;
39 module_param_string(ids, ids, sizeof(ids), 0);
40 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");
41
42 static bool nointxmask;
43 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
44 MODULE_PARM_DESC(nointxmask,
45 "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.");
46
47 #ifdef CONFIG_VFIO_PCI_VGA
48 static bool disable_vga;
49 module_param(disable_vga, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vga, "Disable VGA resource access through vfio-pci");
51 #endif
52
53 static bool disable_idle_d3;
54 module_param(disable_idle_d3, bool, S_IRUGO | S_IWUSR);
55 MODULE_PARM_DESC(disable_idle_d3,
56 "Disable using the PCI D3 low power state for idle, unused devices");
57
58 static DEFINE_MUTEX(driver_lock);
59
60 static inline bool vfio_vga_disabled(void)
61 {
62 #ifdef CONFIG_VFIO_PCI_VGA
63 return disable_vga;
64 #else
65 return true;
66 #endif
67 }
68
69 /*
70 * Our VGA arbiter participation is limited since we don't know anything
71 * about the device itself. However, if the device is the only VGA device
72 * downstream of a bridge and VFIO VGA support is disabled, then we can
73 * safely return legacy VGA IO and memory as not decoded since the user
74 * has no way to get to it and routing can be disabled externally at the
75 * bridge.
76 */
77 static unsigned int vfio_pci_set_vga_decode(void *opaque, bool single_vga)
78 {
79 struct vfio_pci_device *vdev = opaque;
80 struct pci_dev *tmp = NULL, *pdev = vdev->pdev;
81 unsigned char max_busnr;
82 unsigned int decodes;
83
84 if (single_vga || !vfio_vga_disabled() || pci_is_root_bus(pdev->bus))
85 return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
86 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
87
88 max_busnr = pci_bus_max_busnr(pdev->bus);
89 decodes = VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
90
91 while ((tmp = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, tmp)) != NULL) {
92 if (tmp == pdev ||
93 pci_domain_nr(tmp->bus) != pci_domain_nr(pdev->bus) ||
94 pci_is_root_bus(tmp->bus))
95 continue;
96
97 if (tmp->bus->number >= pdev->bus->number &&
98 tmp->bus->number <= max_busnr) {
99 pci_dev_put(tmp);
100 decodes |= VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
101 break;
102 }
103 }
104
105 return decodes;
106 }
107
108 static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
109 {
110 return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
111 }
112
113 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
114
115 static int vfio_pci_enable(struct vfio_pci_device *vdev)
116 {
117 struct pci_dev *pdev = vdev->pdev;
118 int ret;
119 u16 cmd;
120 u8 msix_pos;
121
122 pci_set_power_state(pdev, PCI_D0);
123
124 /* Don't allow our initial saved state to include busmaster */
125 pci_clear_master(pdev);
126
127 ret = pci_enable_device(pdev);
128 if (ret)
129 return ret;
130
131 vdev->reset_works = (pci_reset_function(pdev) == 0);
132 pci_save_state(pdev);
133 vdev->pci_saved_state = pci_store_saved_state(pdev);
134 if (!vdev->pci_saved_state)
135 pr_debug("%s: Couldn't store %s saved state\n",
136 __func__, dev_name(&pdev->dev));
137
138 ret = vfio_config_init(vdev);
139 if (ret) {
140 kfree(vdev->pci_saved_state);
141 vdev->pci_saved_state = NULL;
142 pci_disable_device(pdev);
143 return ret;
144 }
145
146 if (likely(!nointxmask))
147 vdev->pci_2_3 = pci_intx_mask_supported(pdev);
148
149 pci_read_config_word(pdev, PCI_COMMAND, &cmd);
150 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) {
151 cmd &= ~PCI_COMMAND_INTX_DISABLE;
152 pci_write_config_word(pdev, PCI_COMMAND, cmd);
153 }
154
155 msix_pos = pdev->msix_cap;
156 if (msix_pos) {
157 u16 flags;
158 u32 table;
159
160 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags);
161 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table);
162
163 vdev->msix_bar = table & PCI_MSIX_TABLE_BIR;
164 vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET;
165 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16;
166 } else
167 vdev->msix_bar = 0xFF;
168
169 if (!vfio_vga_disabled() && vfio_pci_is_vga(pdev))
170 vdev->has_vga = true;
171
172 return 0;
173 }
174
175 static void vfio_pci_disable(struct vfio_pci_device *vdev)
176 {
177 struct pci_dev *pdev = vdev->pdev;
178 int bar;
179
180 /* Stop the device from further DMA */
181 pci_clear_master(pdev);
182
183 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE |
184 VFIO_IRQ_SET_ACTION_TRIGGER,
185 vdev->irq_type, 0, 0, NULL);
186
187 vdev->virq_disabled = false;
188
189 vfio_config_free(vdev);
190
191 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
192 if (!vdev->barmap[bar])
193 continue;
194 pci_iounmap(pdev, vdev->barmap[bar]);
195 pci_release_selected_regions(pdev, 1 << bar);
196 vdev->barmap[bar] = NULL;
197 }
198
199 vdev->needs_reset = true;
200
201 /*
202 * If we have saved state, restore it. If we can reset the device,
203 * even better. Resetting with current state seems better than
204 * nothing, but saving and restoring current state without reset
205 * is just busy work.
206 */
207 if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) {
208 pr_info("%s: Couldn't reload %s saved state\n",
209 __func__, dev_name(&pdev->dev));
210
211 if (!vdev->reset_works)
212 goto out;
213
214 pci_save_state(pdev);
215 }
216
217 /*
218 * Disable INTx and MSI, presumably to avoid spurious interrupts
219 * during reset. Stolen from pci_reset_function()
220 */
221 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
222
223 /*
224 * Try to reset the device. The success of this is dependent on
225 * being able to lock the device, which is not always possible.
226 */
227 if (vdev->reset_works && !pci_try_reset_function(pdev))
228 vdev->needs_reset = false;
229
230 pci_restore_state(pdev);
231 out:
232 pci_disable_device(pdev);
233
234 vfio_pci_try_bus_reset(vdev);
235
236 if (!disable_idle_d3)
237 pci_set_power_state(pdev, PCI_D3hot);
238 }
239
240 static void vfio_pci_release(void *device_data)
241 {
242 struct vfio_pci_device *vdev = device_data;
243
244 mutex_lock(&driver_lock);
245
246 if (!(--vdev->refcnt)) {
247 vfio_spapr_pci_eeh_release(vdev->pdev);
248 vfio_pci_disable(vdev);
249 }
250
251 mutex_unlock(&driver_lock);
252
253 module_put(THIS_MODULE);
254 }
255
256 static int vfio_pci_open(void *device_data)
257 {
258 struct vfio_pci_device *vdev = device_data;
259 int ret = 0;
260
261 if (!try_module_get(THIS_MODULE))
262 return -ENODEV;
263
264 mutex_lock(&driver_lock);
265
266 if (!vdev->refcnt) {
267 ret = vfio_pci_enable(vdev);
268 if (ret)
269 goto error;
270
271 vfio_spapr_pci_eeh_open(vdev->pdev);
272 }
273 vdev->refcnt++;
274 error:
275 mutex_unlock(&driver_lock);
276 if (ret)
277 module_put(THIS_MODULE);
278 return ret;
279 }
280
281 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type)
282 {
283 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) {
284 u8 pin;
285 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin);
286 if (IS_ENABLED(CONFIG_VFIO_PCI_INTX) && pin)
287 return 1;
288
289 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) {
290 u8 pos;
291 u16 flags;
292
293 pos = vdev->pdev->msi_cap;
294 if (pos) {
295 pci_read_config_word(vdev->pdev,
296 pos + PCI_MSI_FLAGS, &flags);
297 return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1);
298 }
299 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) {
300 u8 pos;
301 u16 flags;
302
303 pos = vdev->pdev->msix_cap;
304 if (pos) {
305 pci_read_config_word(vdev->pdev,
306 pos + PCI_MSIX_FLAGS, &flags);
307
308 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1;
309 }
310 } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) {
311 if (pci_is_pcie(vdev->pdev))
312 return 1;
313 } else if (irq_type == VFIO_PCI_REQ_IRQ_INDEX) {
314 return 1;
315 }
316
317 return 0;
318 }
319
320 static int vfio_pci_count_devs(struct pci_dev *pdev, void *data)
321 {
322 (*(int *)data)++;
323 return 0;
324 }
325
326 struct vfio_pci_fill_info {
327 int max;
328 int cur;
329 struct vfio_pci_dependent_device *devices;
330 };
331
332 static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data)
333 {
334 struct vfio_pci_fill_info *fill = data;
335 struct iommu_group *iommu_group;
336
337 if (fill->cur == fill->max)
338 return -EAGAIN; /* Something changed, try again */
339
340 iommu_group = iommu_group_get(&pdev->dev);
341 if (!iommu_group)
342 return -EPERM; /* Cannot reset non-isolated devices */
343
344 fill->devices[fill->cur].group_id = iommu_group_id(iommu_group);
345 fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus);
346 fill->devices[fill->cur].bus = pdev->bus->number;
347 fill->devices[fill->cur].devfn = pdev->devfn;
348 fill->cur++;
349 iommu_group_put(iommu_group);
350 return 0;
351 }
352
353 struct vfio_pci_group_entry {
354 struct vfio_group *group;
355 int id;
356 };
357
358 struct vfio_pci_group_info {
359 int count;
360 struct vfio_pci_group_entry *groups;
361 };
362
363 static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data)
364 {
365 struct vfio_pci_group_info *info = data;
366 struct iommu_group *group;
367 int id, i;
368
369 group = iommu_group_get(&pdev->dev);
370 if (!group)
371 return -EPERM;
372
373 id = iommu_group_id(group);
374
375 for (i = 0; i < info->count; i++)
376 if (info->groups[i].id == id)
377 break;
378
379 iommu_group_put(group);
380
381 return (i == info->count) ? -EINVAL : 0;
382 }
383
384 static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot)
385 {
386 for (; pdev; pdev = pdev->bus->self)
387 if (pdev->bus == slot->bus)
388 return (pdev->slot == slot);
389 return false;
390 }
391
392 struct vfio_pci_walk_info {
393 int (*fn)(struct pci_dev *, void *data);
394 void *data;
395 struct pci_dev *pdev;
396 bool slot;
397 int ret;
398 };
399
400 static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data)
401 {
402 struct vfio_pci_walk_info *walk = data;
403
404 if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot))
405 walk->ret = walk->fn(pdev, walk->data);
406
407 return walk->ret;
408 }
409
410 static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev,
411 int (*fn)(struct pci_dev *,
412 void *data), void *data,
413 bool slot)
414 {
415 struct vfio_pci_walk_info walk = {
416 .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0,
417 };
418
419 pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk);
420
421 return walk.ret;
422 }
423
424 static long vfio_pci_ioctl(void *device_data,
425 unsigned int cmd, unsigned long arg)
426 {
427 struct vfio_pci_device *vdev = device_data;
428 unsigned long minsz;
429
430 if (cmd == VFIO_DEVICE_GET_INFO) {
431 struct vfio_device_info info;
432
433 minsz = offsetofend(struct vfio_device_info, num_irqs);
434
435 if (copy_from_user(&info, (void __user *)arg, minsz))
436 return -EFAULT;
437
438 if (info.argsz < minsz)
439 return -EINVAL;
440
441 info.flags = VFIO_DEVICE_FLAGS_PCI;
442
443 if (vdev->reset_works)
444 info.flags |= VFIO_DEVICE_FLAGS_RESET;
445
446 info.num_regions = VFIO_PCI_NUM_REGIONS;
447 info.num_irqs = VFIO_PCI_NUM_IRQS;
448
449 return copy_to_user((void __user *)arg, &info, minsz) ?
450 -EFAULT : 0;
451
452 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
453 struct pci_dev *pdev = vdev->pdev;
454 struct vfio_region_info info;
455
456 minsz = offsetofend(struct vfio_region_info, offset);
457
458 if (copy_from_user(&info, (void __user *)arg, minsz))
459 return -EFAULT;
460
461 if (info.argsz < minsz)
462 return -EINVAL;
463
464 switch (info.index) {
465 case VFIO_PCI_CONFIG_REGION_INDEX:
466 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
467 info.size = pdev->cfg_size;
468 info.flags = VFIO_REGION_INFO_FLAG_READ |
469 VFIO_REGION_INFO_FLAG_WRITE;
470 break;
471 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
472 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
473 info.size = pci_resource_len(pdev, info.index);
474 if (!info.size) {
475 info.flags = 0;
476 break;
477 }
478
479 info.flags = VFIO_REGION_INFO_FLAG_READ |
480 VFIO_REGION_INFO_FLAG_WRITE;
481 if (IS_ENABLED(CONFIG_VFIO_PCI_MMAP) &&
482 pci_resource_flags(pdev, info.index) &
483 IORESOURCE_MEM && info.size >= PAGE_SIZE)
484 info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
485 break;
486 case VFIO_PCI_ROM_REGION_INDEX:
487 {
488 void __iomem *io;
489 size_t size;
490
491 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
492 info.flags = 0;
493
494 /* Report the BAR size, not the ROM size */
495 info.size = pci_resource_len(pdev, info.index);
496 if (!info.size)
497 break;
498
499 /* Is it really there? */
500 io = pci_map_rom(pdev, &size);
501 if (!io || !size) {
502 info.size = 0;
503 break;
504 }
505 pci_unmap_rom(pdev, io);
506
507 info.flags = VFIO_REGION_INFO_FLAG_READ;
508 break;
509 }
510 case VFIO_PCI_VGA_REGION_INDEX:
511 if (!vdev->has_vga)
512 return -EINVAL;
513
514 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
515 info.size = 0xc0000;
516 info.flags = VFIO_REGION_INFO_FLAG_READ |
517 VFIO_REGION_INFO_FLAG_WRITE;
518
519 break;
520 default:
521 return -EINVAL;
522 }
523
524 return copy_to_user((void __user *)arg, &info, minsz) ?
525 -EFAULT : 0;
526
527 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
528 struct vfio_irq_info info;
529
530 minsz = offsetofend(struct vfio_irq_info, count);
531
532 if (copy_from_user(&info, (void __user *)arg, minsz))
533 return -EFAULT;
534
535 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
536 return -EINVAL;
537
538 switch (info.index) {
539 case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX:
540 case VFIO_PCI_REQ_IRQ_INDEX:
541 break;
542 case VFIO_PCI_ERR_IRQ_INDEX:
543 if (pci_is_pcie(vdev->pdev))
544 break;
545 /* pass thru to return error */
546 default:
547 return -EINVAL;
548 }
549
550 info.flags = VFIO_IRQ_INFO_EVENTFD;
551
552 info.count = vfio_pci_get_irq_count(vdev, info.index);
553
554 if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
555 info.flags |= (VFIO_IRQ_INFO_MASKABLE |
556 VFIO_IRQ_INFO_AUTOMASKED);
557 else
558 info.flags |= VFIO_IRQ_INFO_NORESIZE;
559
560 return copy_to_user((void __user *)arg, &info, minsz) ?
561 -EFAULT : 0;
562
563 } else if (cmd == VFIO_DEVICE_SET_IRQS) {
564 struct vfio_irq_set hdr;
565 size_t size;
566 u8 *data = NULL;
567 int max, ret = 0;
568
569 minsz = offsetofend(struct vfio_irq_set, count);
570
571 if (copy_from_user(&hdr, (void __user *)arg, minsz))
572 return -EFAULT;
573
574 if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS ||
575 hdr.count >= (U32_MAX - hdr.start) ||
576 hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
577 VFIO_IRQ_SET_ACTION_TYPE_MASK))
578 return -EINVAL;
579
580 max = vfio_pci_get_irq_count(vdev, hdr.index);
581 if (hdr.start >= max || hdr.start + hdr.count > max)
582 return -EINVAL;
583
584 switch (hdr.flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
585 case VFIO_IRQ_SET_DATA_NONE:
586 size = 0;
587 break;
588 case VFIO_IRQ_SET_DATA_BOOL:
589 size = sizeof(uint8_t);
590 break;
591 case VFIO_IRQ_SET_DATA_EVENTFD:
592 size = sizeof(int32_t);
593 break;
594 default:
595 return -EINVAL;
596 }
597
598 if (size) {
599 if (hdr.argsz - minsz < hdr.count * size)
600 return -EINVAL;
601
602 data = memdup_user((void __user *)(arg + minsz),
603 hdr.count * size);
604 if (IS_ERR(data))
605 return PTR_ERR(data);
606 }
607
608 mutex_lock(&vdev->igate);
609
610 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index,
611 hdr.start, hdr.count, data);
612
613 mutex_unlock(&vdev->igate);
614 kfree(data);
615
616 return ret;
617
618 } else if (cmd == VFIO_DEVICE_RESET) {
619 return vdev->reset_works ?
620 pci_try_reset_function(vdev->pdev) : -EINVAL;
621
622 } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) {
623 struct vfio_pci_hot_reset_info hdr;
624 struct vfio_pci_fill_info fill = { 0 };
625 struct vfio_pci_dependent_device *devices = NULL;
626 bool slot = false;
627 int ret = 0;
628
629 minsz = offsetofend(struct vfio_pci_hot_reset_info, count);
630
631 if (copy_from_user(&hdr, (void __user *)arg, minsz))
632 return -EFAULT;
633
634 if (hdr.argsz < minsz)
635 return -EINVAL;
636
637 hdr.flags = 0;
638
639 /* Can we do a slot or bus reset or neither? */
640 if (!pci_probe_reset_slot(vdev->pdev->slot))
641 slot = true;
642 else if (pci_probe_reset_bus(vdev->pdev->bus))
643 return -ENODEV;
644
645 /* How many devices are affected? */
646 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
647 vfio_pci_count_devs,
648 &fill.max, slot);
649 if (ret)
650 return ret;
651
652 WARN_ON(!fill.max); /* Should always be at least one */
653
654 /*
655 * If there's enough space, fill it now, otherwise return
656 * -ENOSPC and the number of devices affected.
657 */
658 if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) {
659 ret = -ENOSPC;
660 hdr.count = fill.max;
661 goto reset_info_exit;
662 }
663
664 devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL);
665 if (!devices)
666 return -ENOMEM;
667
668 fill.devices = devices;
669
670 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
671 vfio_pci_fill_devs,
672 &fill, slot);
673
674 /*
675 * If a device was removed between counting and filling,
676 * we may come up short of fill.max. If a device was
677 * added, we'll have a return of -EAGAIN above.
678 */
679 if (!ret)
680 hdr.count = fill.cur;
681
682 reset_info_exit:
683 if (copy_to_user((void __user *)arg, &hdr, minsz))
684 ret = -EFAULT;
685
686 if (!ret) {
687 if (copy_to_user((void __user *)(arg + minsz), devices,
688 hdr.count * sizeof(*devices)))
689 ret = -EFAULT;
690 }
691
692 kfree(devices);
693 return ret;
694
695 } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) {
696 struct vfio_pci_hot_reset hdr;
697 int32_t *group_fds;
698 struct vfio_pci_group_entry *groups;
699 struct vfio_pci_group_info info;
700 bool slot = false;
701 int i, count = 0, ret = 0;
702
703 minsz = offsetofend(struct vfio_pci_hot_reset, count);
704
705 if (copy_from_user(&hdr, (void __user *)arg, minsz))
706 return -EFAULT;
707
708 if (hdr.argsz < minsz || hdr.flags)
709 return -EINVAL;
710
711 /* Can we do a slot or bus reset or neither? */
712 if (!pci_probe_reset_slot(vdev->pdev->slot))
713 slot = true;
714 else if (pci_probe_reset_bus(vdev->pdev->bus))
715 return -ENODEV;
716
717 /*
718 * We can't let userspace give us an arbitrarily large
719 * buffer to copy, so verify how many we think there
720 * could be. Note groups can have multiple devices so
721 * one group per device is the max.
722 */
723 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
724 vfio_pci_count_devs,
725 &count, slot);
726 if (ret)
727 return ret;
728
729 /* Somewhere between 1 and count is OK */
730 if (!hdr.count || hdr.count > count)
731 return -EINVAL;
732
733 group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL);
734 groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL);
735 if (!group_fds || !groups) {
736 kfree(group_fds);
737 kfree(groups);
738 return -ENOMEM;
739 }
740
741 if (copy_from_user(group_fds, (void __user *)(arg + minsz),
742 hdr.count * sizeof(*group_fds))) {
743 kfree(group_fds);
744 kfree(groups);
745 return -EFAULT;
746 }
747
748 /*
749 * For each group_fd, get the group through the vfio external
750 * user interface and store the group and iommu ID. This
751 * ensures the group is held across the reset.
752 */
753 for (i = 0; i < hdr.count; i++) {
754 struct vfio_group *group;
755 struct fd f = fdget(group_fds[i]);
756 if (!f.file) {
757 ret = -EBADF;
758 break;
759 }
760
761 group = vfio_group_get_external_user(f.file);
762 fdput(f);
763 if (IS_ERR(group)) {
764 ret = PTR_ERR(group);
765 break;
766 }
767
768 groups[i].group = group;
769 groups[i].id = vfio_external_user_iommu_id(group);
770 }
771
772 kfree(group_fds);
773
774 /* release reference to groups on error */
775 if (ret)
776 goto hot_reset_release;
777
778 info.count = hdr.count;
779 info.groups = groups;
780
781 /*
782 * Test whether all the affected devices are contained
783 * by the set of groups provided by the user.
784 */
785 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
786 vfio_pci_validate_devs,
787 &info, slot);
788 if (!ret)
789 /* User has access, do the reset */
790 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
791 pci_try_reset_bus(vdev->pdev->bus);
792
793 hot_reset_release:
794 for (i--; i >= 0; i--)
795 vfio_group_put_external_user(groups[i].group);
796
797 kfree(groups);
798 return ret;
799 }
800
801 return -ENOTTY;
802 }
803
804 static ssize_t vfio_pci_rw(void *device_data, char __user *buf,
805 size_t count, loff_t *ppos, bool iswrite)
806 {
807 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
808 struct vfio_pci_device *vdev = device_data;
809
810 if (index >= VFIO_PCI_NUM_REGIONS)
811 return -EINVAL;
812
813 switch (index) {
814 case VFIO_PCI_CONFIG_REGION_INDEX:
815 return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite);
816
817 case VFIO_PCI_ROM_REGION_INDEX:
818 if (iswrite)
819 return -EINVAL;
820 return vfio_pci_bar_rw(vdev, buf, count, ppos, false);
821
822 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
823 return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite);
824
825 case VFIO_PCI_VGA_REGION_INDEX:
826 return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite);
827 }
828
829 return -EINVAL;
830 }
831
832 static ssize_t vfio_pci_read(void *device_data, char __user *buf,
833 size_t count, loff_t *ppos)
834 {
835 if (!count)
836 return 0;
837
838 return vfio_pci_rw(device_data, buf, count, ppos, false);
839 }
840
841 static ssize_t vfio_pci_write(void *device_data, const char __user *buf,
842 size_t count, loff_t *ppos)
843 {
844 if (!count)
845 return 0;
846
847 return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true);
848 }
849
850 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
851 {
852 struct vfio_pci_device *vdev = device_data;
853 struct pci_dev *pdev = vdev->pdev;
854 unsigned int index;
855 u64 phys_len, req_len, pgoff, req_start;
856 int ret;
857
858 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
859
860 if (vma->vm_end < vma->vm_start)
861 return -EINVAL;
862 if ((vma->vm_flags & VM_SHARED) == 0)
863 return -EINVAL;
864 if (index >= VFIO_PCI_ROM_REGION_INDEX)
865 return -EINVAL;
866 if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM))
867 return -EINVAL;
868
869 phys_len = pci_resource_len(pdev, index);
870 req_len = vma->vm_end - vma->vm_start;
871 pgoff = vma->vm_pgoff &
872 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
873 req_start = pgoff << PAGE_SHIFT;
874
875 if (phys_len < PAGE_SIZE || req_start + req_len > phys_len)
876 return -EINVAL;
877
878 if (index == vdev->msix_bar) {
879 /*
880 * Disallow mmaps overlapping the MSI-X table; users don't
881 * get to touch this directly. We could find somewhere
882 * else to map the overlap, but page granularity is only
883 * a recommendation, not a requirement, so the user needs
884 * to know which bits are real. Requiring them to mmap
885 * around the table makes that clear.
886 */
887
888 /* If neither entirely above nor below, then it overlaps */
889 if (!(req_start >= vdev->msix_offset + vdev->msix_size ||
890 req_start + req_len <= vdev->msix_offset))
891 return -EINVAL;
892 }
893
894 /*
895 * Even though we don't make use of the barmap for the mmap,
896 * we need to request the region and the barmap tracks that.
897 */
898 if (!vdev->barmap[index]) {
899 ret = pci_request_selected_regions(pdev,
900 1 << index, "vfio-pci");
901 if (ret)
902 return ret;
903
904 vdev->barmap[index] = pci_iomap(pdev, index, 0);
905 if (!vdev->barmap[index]) {
906 pci_release_selected_regions(pdev, 1 << index);
907 return -ENOMEM;
908 }
909 }
910
911 vma->vm_private_data = vdev;
912 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
913 vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff;
914
915 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
916 req_len, vma->vm_page_prot);
917 }
918
919 static void vfio_pci_request(void *device_data, unsigned int count)
920 {
921 struct vfio_pci_device *vdev = device_data;
922
923 mutex_lock(&vdev->igate);
924
925 if (vdev->req_trigger) {
926 if (!(count % 10))
927 dev_notice_ratelimited(&vdev->pdev->dev,
928 "Relaying device request to user (#%u)\n",
929 count);
930 eventfd_signal(vdev->req_trigger, 1);
931 } else if (count == 0) {
932 dev_warn(&vdev->pdev->dev,
933 "No device request channel registered, blocked until released by user\n");
934 }
935
936 mutex_unlock(&vdev->igate);
937 }
938
939 static const struct vfio_device_ops vfio_pci_ops = {
940 .name = "vfio-pci",
941 .open = vfio_pci_open,
942 .release = vfio_pci_release,
943 .ioctl = vfio_pci_ioctl,
944 .read = vfio_pci_read,
945 .write = vfio_pci_write,
946 .mmap = vfio_pci_mmap,
947 .request = vfio_pci_request,
948 };
949
950 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
951 {
952 struct vfio_pci_device *vdev;
953 struct iommu_group *group;
954 int ret;
955
956 if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL)
957 return -EINVAL;
958
959 group = iommu_group_get(&pdev->dev);
960 if (!group)
961 return -EINVAL;
962
963 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
964 if (!vdev) {
965 iommu_group_put(group);
966 return -ENOMEM;
967 }
968
969 vdev->pdev = pdev;
970 vdev->irq_type = VFIO_PCI_NUM_IRQS;
971 mutex_init(&vdev->igate);
972 spin_lock_init(&vdev->irqlock);
973
974 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
975 if (ret) {
976 iommu_group_put(group);
977 kfree(vdev);
978 return ret;
979 }
980
981 if (vfio_pci_is_vga(pdev)) {
982 vga_client_register(pdev, vdev, NULL, vfio_pci_set_vga_decode);
983 vga_set_legacy_decoding(pdev,
984 vfio_pci_set_vga_decode(vdev, false));
985 }
986
987 if (!disable_idle_d3) {
988 /*
989 * pci-core sets the device power state to an unknown value at
990 * bootup and after being removed from a driver. The only
991 * transition it allows from this unknown state is to D0, which
992 * typically happens when a driver calls pci_enable_device().
993 * We're not ready to enable the device yet, but we do want to
994 * be able to get to D3. Therefore first do a D0 transition
995 * before going to D3.
996 */
997 pci_set_power_state(pdev, PCI_D0);
998 pci_set_power_state(pdev, PCI_D3hot);
999 }
1000
1001 return ret;
1002 }
1003
1004 static void vfio_pci_remove(struct pci_dev *pdev)
1005 {
1006 struct vfio_pci_device *vdev;
1007
1008 vdev = vfio_del_group_dev(&pdev->dev);
1009 if (!vdev)
1010 return;
1011
1012 iommu_group_put(pdev->dev.iommu_group);
1013 kfree(vdev);
1014
1015 if (vfio_pci_is_vga(pdev)) {
1016 vga_client_register(pdev, NULL, NULL, NULL);
1017 vga_set_legacy_decoding(pdev,
1018 VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
1019 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM);
1020 }
1021
1022 if (!disable_idle_d3)
1023 pci_set_power_state(pdev, PCI_D0);
1024 }
1025
1026 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev,
1027 pci_channel_state_t state)
1028 {
1029 struct vfio_pci_device *vdev;
1030 struct vfio_device *device;
1031
1032 device = vfio_device_get_from_dev(&pdev->dev);
1033 if (device == NULL)
1034 return PCI_ERS_RESULT_DISCONNECT;
1035
1036 vdev = vfio_device_data(device);
1037 if (vdev == NULL) {
1038 vfio_device_put(device);
1039 return PCI_ERS_RESULT_DISCONNECT;
1040 }
1041
1042 mutex_lock(&vdev->igate);
1043
1044 if (vdev->err_trigger)
1045 eventfd_signal(vdev->err_trigger, 1);
1046
1047 mutex_unlock(&vdev->igate);
1048
1049 vfio_device_put(device);
1050
1051 return PCI_ERS_RESULT_CAN_RECOVER;
1052 }
1053
1054 static const struct pci_error_handlers vfio_err_handlers = {
1055 .error_detected = vfio_pci_aer_err_detected,
1056 };
1057
1058 static struct pci_driver vfio_pci_driver = {
1059 .name = "vfio-pci",
1060 .id_table = NULL, /* only dynamic ids */
1061 .probe = vfio_pci_probe,
1062 .remove = vfio_pci_remove,
1063 .err_handler = &vfio_err_handlers,
1064 };
1065
1066 struct vfio_devices {
1067 struct vfio_device **devices;
1068 int cur_index;
1069 int max_index;
1070 };
1071
1072 static int vfio_pci_get_devs(struct pci_dev *pdev, void *data)
1073 {
1074 struct vfio_devices *devs = data;
1075 struct vfio_device *device;
1076
1077 if (devs->cur_index == devs->max_index)
1078 return -ENOSPC;
1079
1080 device = vfio_device_get_from_dev(&pdev->dev);
1081 if (!device)
1082 return -EINVAL;
1083
1084 if (pci_dev_driver(pdev) != &vfio_pci_driver) {
1085 vfio_device_put(device);
1086 return -EBUSY;
1087 }
1088
1089 devs->devices[devs->cur_index++] = device;
1090 return 0;
1091 }
1092
1093 /*
1094 * Attempt to do a bus/slot reset if there are devices affected by a reset for
1095 * this device that are needs_reset and all of the affected devices are unused
1096 * (!refcnt). Callers are required to hold driver_lock when calling this to
1097 * prevent device opens and concurrent bus reset attempts. We prevent device
1098 * unbinds by acquiring and holding a reference to the vfio_device.
1099 *
1100 * NB: vfio-core considers a group to be viable even if some devices are
1101 * bound to drivers like pci-stub or pcieport. Here we require all devices
1102 * to be bound to vfio_pci since that's the only way we can be sure they
1103 * stay put.
1104 */
1105 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
1106 {
1107 struct vfio_devices devs = { .cur_index = 0 };
1108 int i = 0, ret = -EINVAL;
1109 bool needs_reset = false, slot = false;
1110 struct vfio_pci_device *tmp;
1111
1112 if (!pci_probe_reset_slot(vdev->pdev->slot))
1113 slot = true;
1114 else if (pci_probe_reset_bus(vdev->pdev->bus))
1115 return;
1116
1117 if (vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs,
1118 &i, slot) || !i)
1119 return;
1120
1121 devs.max_index = i;
1122 devs.devices = kcalloc(i, sizeof(struct vfio_device *), GFP_KERNEL);
1123 if (!devs.devices)
1124 return;
1125
1126 if (vfio_pci_for_each_slot_or_bus(vdev->pdev,
1127 vfio_pci_get_devs, &devs, slot))
1128 goto put_devs;
1129
1130 for (i = 0; i < devs.cur_index; i++) {
1131 tmp = vfio_device_data(devs.devices[i]);
1132 if (tmp->needs_reset)
1133 needs_reset = true;
1134 if (tmp->refcnt)
1135 goto put_devs;
1136 }
1137
1138 if (needs_reset)
1139 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
1140 pci_try_reset_bus(vdev->pdev->bus);
1141
1142 put_devs:
1143 for (i = 0; i < devs.cur_index; i++) {
1144 tmp = vfio_device_data(devs.devices[i]);
1145 if (!ret)
1146 tmp->needs_reset = false;
1147
1148 if (!tmp->refcnt && !disable_idle_d3)
1149 pci_set_power_state(tmp->pdev, PCI_D3hot);
1150
1151 vfio_device_put(devs.devices[i]);
1152 }
1153
1154 kfree(devs.devices);
1155 }
1156
1157 static void __exit vfio_pci_cleanup(void)
1158 {
1159 pci_unregister_driver(&vfio_pci_driver);
1160 vfio_pci_uninit_perm_bits();
1161 }
1162
1163 static void __init vfio_pci_fill_ids(void)
1164 {
1165 char *p, *id;
1166 int rc;
1167
1168 /* no ids passed actually */
1169 if (ids[0] == '\0')
1170 return;
1171
1172 /* add ids specified in the module parameter */
1173 p = ids;
1174 while ((id = strsep(&p, ","))) {
1175 unsigned int vendor, device, subvendor = PCI_ANY_ID,
1176 subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
1177 int fields;
1178
1179 if (!strlen(id))
1180 continue;
1181
1182 fields = sscanf(id, "%x:%x:%x:%x:%x:%x",
1183 &vendor, &device, &subvendor, &subdevice,
1184 &class, &class_mask);
1185
1186 if (fields < 2) {
1187 pr_warn("invalid id string \"%s\"\n", id);
1188 continue;
1189 }
1190
1191 rc = pci_add_dynid(&vfio_pci_driver, vendor, device,
1192 subvendor, subdevice, class, class_mask, 0);
1193 if (rc)
1194 pr_warn("failed to add dynamic id [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x (%d)\n",
1195 vendor, device, subvendor, subdevice,
1196 class, class_mask, rc);
1197 else
1198 pr_info("add [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x\n",
1199 vendor, device, subvendor, subdevice,
1200 class, class_mask);
1201 }
1202 }
1203
1204 static int __init vfio_pci_init(void)
1205 {
1206 int ret;
1207
1208 /* Allocate shared config space permision data used by all devices */
1209 ret = vfio_pci_init_perm_bits();
1210 if (ret)
1211 return ret;
1212
1213 /* Register and scan for devices */
1214 ret = pci_register_driver(&vfio_pci_driver);
1215 if (ret)
1216 goto out_driver;
1217
1218 vfio_pci_fill_ids();
1219
1220 return 0;
1221
1222 out_driver:
1223 vfio_pci_uninit_perm_bits();
1224 return ret;
1225 }
1226
1227 module_init(vfio_pci_init);
1228 module_exit(vfio_pci_cleanup);
1229
1230 MODULE_VERSION(DRIVER_VERSION);
1231 MODULE_LICENSE("GPL v2");
1232 MODULE_AUTHOR(DRIVER_AUTHOR);
1233 MODULE_DESCRIPTION(DRIVER_DESC);
Linux with added FPC-III support