Commit | Line | Data |
---|---|---|
89e1f7d4 AW |
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 | #include <linux/device.h> | |
15 | #include <linux/eventfd.h> | |
8b27ee60 | 16 | #include <linux/file.h> |
89e1f7d4 AW |
17 | #include <linux/interrupt.h> |
18 | #include <linux/iommu.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/mutex.h> | |
21 | #include <linux/notifier.h> | |
22 | #include <linux/pci.h> | |
23 | #include <linux/pm_runtime.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/types.h> | |
26 | #include <linux/uaccess.h> | |
27 | #include <linux/vfio.h> | |
28 | ||
29 | #include "vfio_pci_private.h" | |
30 | ||
31 | #define DRIVER_VERSION "0.2" | |
32 | #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>" | |
33 | #define DRIVER_DESC "VFIO PCI - User Level meta-driver" | |
34 | ||
35 | static bool nointxmask; | |
36 | module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR); | |
37 | MODULE_PARM_DESC(nointxmask, | |
38 | "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."); | |
39 | ||
40 | static int vfio_pci_enable(struct vfio_pci_device *vdev) | |
41 | { | |
42 | struct pci_dev *pdev = vdev->pdev; | |
43 | int ret; | |
44 | u16 cmd; | |
45 | u8 msix_pos; | |
46 | ||
9c22e660 AW |
47 | /* Don't allow our initial saved state to include busmaster */ |
48 | pci_clear_master(pdev); | |
49 | ||
9a92c509 AW |
50 | ret = pci_enable_device(pdev); |
51 | if (ret) | |
52 | return ret; | |
53 | ||
89e1f7d4 AW |
54 | vdev->reset_works = (pci_reset_function(pdev) == 0); |
55 | pci_save_state(pdev); | |
56 | vdev->pci_saved_state = pci_store_saved_state(pdev); | |
57 | if (!vdev->pci_saved_state) | |
58 | pr_debug("%s: Couldn't store %s saved state\n", | |
59 | __func__, dev_name(&pdev->dev)); | |
60 | ||
61 | ret = vfio_config_init(vdev); | |
9a92c509 | 62 | if (ret) { |
eb5685f0 AW |
63 | kfree(vdev->pci_saved_state); |
64 | vdev->pci_saved_state = NULL; | |
9a92c509 AW |
65 | pci_disable_device(pdev); |
66 | return ret; | |
67 | } | |
89e1f7d4 AW |
68 | |
69 | if (likely(!nointxmask)) | |
70 | vdev->pci_2_3 = pci_intx_mask_supported(pdev); | |
71 | ||
72 | pci_read_config_word(pdev, PCI_COMMAND, &cmd); | |
73 | if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) { | |
74 | cmd &= ~PCI_COMMAND_INTX_DISABLE; | |
75 | pci_write_config_word(pdev, PCI_COMMAND, cmd); | |
76 | } | |
77 | ||
a9047f24 | 78 | msix_pos = pdev->msix_cap; |
89e1f7d4 AW |
79 | if (msix_pos) { |
80 | u16 flags; | |
81 | u32 table; | |
82 | ||
83 | pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags); | |
84 | pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table); | |
85 | ||
508d1aa6 BH |
86 | vdev->msix_bar = table & PCI_MSIX_TABLE_BIR; |
87 | vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET; | |
89e1f7d4 AW |
88 | vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16; |
89 | } else | |
90 | vdev->msix_bar = 0xFF; | |
91 | ||
84237a82 AW |
92 | #ifdef CONFIG_VFIO_PCI_VGA |
93 | if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) | |
94 | vdev->has_vga = true; | |
95 | #endif | |
96 | ||
9a92c509 | 97 | return 0; |
89e1f7d4 AW |
98 | } |
99 | ||
100 | static void vfio_pci_disable(struct vfio_pci_device *vdev) | |
101 | { | |
2007722a | 102 | struct pci_dev *pdev = vdev->pdev; |
89e1f7d4 AW |
103 | int bar; |
104 | ||
9c22e660 AW |
105 | /* Stop the device from further DMA */ |
106 | pci_clear_master(pdev); | |
89e1f7d4 AW |
107 | |
108 | vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE | | |
109 | VFIO_IRQ_SET_ACTION_TRIGGER, | |
110 | vdev->irq_type, 0, 0, NULL); | |
111 | ||
112 | vdev->virq_disabled = false; | |
113 | ||
114 | vfio_config_free(vdev); | |
115 | ||
89e1f7d4 AW |
116 | for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) { |
117 | if (!vdev->barmap[bar]) | |
118 | continue; | |
2007722a AW |
119 | pci_iounmap(pdev, vdev->barmap[bar]); |
120 | pci_release_selected_regions(pdev, 1 << bar); | |
89e1f7d4 AW |
121 | vdev->barmap[bar] = NULL; |
122 | } | |
2007722a AW |
123 | |
124 | /* | |
125 | * If we have saved state, restore it. If we can reset the device, | |
126 | * even better. Resetting with current state seems better than | |
127 | * nothing, but saving and restoring current state without reset | |
128 | * is just busy work. | |
129 | */ | |
130 | if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) { | |
131 | pr_info("%s: Couldn't reload %s saved state\n", | |
132 | __func__, dev_name(&pdev->dev)); | |
133 | ||
134 | if (!vdev->reset_works) | |
9c22e660 | 135 | goto out; |
2007722a AW |
136 | |
137 | pci_save_state(pdev); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Disable INTx and MSI, presumably to avoid spurious interrupts | |
142 | * during reset. Stolen from pci_reset_function() | |
143 | */ | |
144 | pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE); | |
145 | ||
d24cdbfd | 146 | /* |
890ed578 AW |
147 | * Try to reset the device. The success of this is dependent on |
148 | * being able to lock the device, which is not always possible. | |
d24cdbfd AW |
149 | */ |
150 | if (vdev->reset_works) { | |
890ed578 AW |
151 | int ret = pci_try_reset_function(pdev); |
152 | if (ret) | |
153 | pr_warn("%s: Failed to reset device %s (%d)\n", | |
154 | __func__, dev_name(&pdev->dev), ret); | |
d24cdbfd | 155 | } |
2007722a AW |
156 | |
157 | pci_restore_state(pdev); | |
9c22e660 AW |
158 | out: |
159 | pci_disable_device(pdev); | |
89e1f7d4 AW |
160 | } |
161 | ||
162 | static void vfio_pci_release(void *device_data) | |
163 | { | |
164 | struct vfio_pci_device *vdev = device_data; | |
165 | ||
1b69be5e GS |
166 | if (atomic_dec_and_test(&vdev->refcnt)) { |
167 | vfio_spapr_pci_eeh_release(vdev->pdev); | |
89e1f7d4 | 168 | vfio_pci_disable(vdev); |
1b69be5e | 169 | } |
89e1f7d4 AW |
170 | |
171 | module_put(THIS_MODULE); | |
172 | } | |
173 | ||
174 | static int vfio_pci_open(void *device_data) | |
175 | { | |
176 | struct vfio_pci_device *vdev = device_data; | |
1b69be5e | 177 | int ret; |
89e1f7d4 AW |
178 | |
179 | if (!try_module_get(THIS_MODULE)) | |
180 | return -ENODEV; | |
181 | ||
182 | if (atomic_inc_return(&vdev->refcnt) == 1) { | |
1b69be5e GS |
183 | ret = vfio_pci_enable(vdev); |
184 | if (ret) | |
185 | goto error; | |
186 | ||
187 | ret = vfio_spapr_pci_eeh_open(vdev->pdev); | |
89e1f7d4 | 188 | if (ret) { |
1b69be5e GS |
189 | vfio_pci_disable(vdev); |
190 | goto error; | |
89e1f7d4 AW |
191 | } |
192 | } | |
193 | ||
194 | return 0; | |
1b69be5e GS |
195 | error: |
196 | module_put(THIS_MODULE); | |
197 | return ret; | |
89e1f7d4 AW |
198 | } |
199 | ||
200 | static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type) | |
201 | { | |
202 | if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) { | |
203 | u8 pin; | |
204 | pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin); | |
205 | if (pin) | |
206 | return 1; | |
207 | ||
208 | } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) { | |
209 | u8 pos; | |
210 | u16 flags; | |
211 | ||
a9047f24 | 212 | pos = vdev->pdev->msi_cap; |
89e1f7d4 AW |
213 | if (pos) { |
214 | pci_read_config_word(vdev->pdev, | |
215 | pos + PCI_MSI_FLAGS, &flags); | |
fd49c81f | 216 | return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1); |
89e1f7d4 AW |
217 | } |
218 | } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) { | |
219 | u8 pos; | |
220 | u16 flags; | |
221 | ||
a9047f24 | 222 | pos = vdev->pdev->msix_cap; |
89e1f7d4 AW |
223 | if (pos) { |
224 | pci_read_config_word(vdev->pdev, | |
225 | pos + PCI_MSIX_FLAGS, &flags); | |
226 | ||
227 | return (flags & PCI_MSIX_FLAGS_QSIZE) + 1; | |
228 | } | |
dad9f897 VMP |
229 | } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) |
230 | if (pci_is_pcie(vdev->pdev)) | |
231 | return 1; | |
89e1f7d4 AW |
232 | |
233 | return 0; | |
234 | } | |
235 | ||
8b27ee60 AW |
236 | static int vfio_pci_count_devs(struct pci_dev *pdev, void *data) |
237 | { | |
238 | (*(int *)data)++; | |
239 | return 0; | |
240 | } | |
241 | ||
242 | struct vfio_pci_fill_info { | |
243 | int max; | |
244 | int cur; | |
245 | struct vfio_pci_dependent_device *devices; | |
246 | }; | |
247 | ||
248 | static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data) | |
249 | { | |
250 | struct vfio_pci_fill_info *fill = data; | |
251 | struct iommu_group *iommu_group; | |
252 | ||
253 | if (fill->cur == fill->max) | |
254 | return -EAGAIN; /* Something changed, try again */ | |
255 | ||
256 | iommu_group = iommu_group_get(&pdev->dev); | |
257 | if (!iommu_group) | |
258 | return -EPERM; /* Cannot reset non-isolated devices */ | |
259 | ||
260 | fill->devices[fill->cur].group_id = iommu_group_id(iommu_group); | |
261 | fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus); | |
262 | fill->devices[fill->cur].bus = pdev->bus->number; | |
263 | fill->devices[fill->cur].devfn = pdev->devfn; | |
264 | fill->cur++; | |
265 | iommu_group_put(iommu_group); | |
266 | return 0; | |
267 | } | |
268 | ||
269 | struct vfio_pci_group_entry { | |
270 | struct vfio_group *group; | |
271 | int id; | |
272 | }; | |
273 | ||
274 | struct vfio_pci_group_info { | |
275 | int count; | |
276 | struct vfio_pci_group_entry *groups; | |
277 | }; | |
278 | ||
279 | static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data) | |
280 | { | |
281 | struct vfio_pci_group_info *info = data; | |
282 | struct iommu_group *group; | |
283 | int id, i; | |
284 | ||
285 | group = iommu_group_get(&pdev->dev); | |
286 | if (!group) | |
287 | return -EPERM; | |
288 | ||
289 | id = iommu_group_id(group); | |
290 | ||
291 | for (i = 0; i < info->count; i++) | |
292 | if (info->groups[i].id == id) | |
293 | break; | |
294 | ||
295 | iommu_group_put(group); | |
296 | ||
297 | return (i == info->count) ? -EINVAL : 0; | |
298 | } | |
299 | ||
300 | static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot) | |
301 | { | |
302 | for (; pdev; pdev = pdev->bus->self) | |
303 | if (pdev->bus == slot->bus) | |
304 | return (pdev->slot == slot); | |
305 | return false; | |
306 | } | |
307 | ||
308 | struct vfio_pci_walk_info { | |
309 | int (*fn)(struct pci_dev *, void *data); | |
310 | void *data; | |
311 | struct pci_dev *pdev; | |
312 | bool slot; | |
313 | int ret; | |
314 | }; | |
315 | ||
316 | static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data) | |
317 | { | |
318 | struct vfio_pci_walk_info *walk = data; | |
319 | ||
320 | if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot)) | |
321 | walk->ret = walk->fn(pdev, walk->data); | |
322 | ||
323 | return walk->ret; | |
324 | } | |
325 | ||
326 | static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev, | |
327 | int (*fn)(struct pci_dev *, | |
328 | void *data), void *data, | |
329 | bool slot) | |
330 | { | |
331 | struct vfio_pci_walk_info walk = { | |
332 | .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0, | |
333 | }; | |
334 | ||
335 | pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk); | |
336 | ||
337 | return walk.ret; | |
338 | } | |
339 | ||
89e1f7d4 AW |
340 | static long vfio_pci_ioctl(void *device_data, |
341 | unsigned int cmd, unsigned long arg) | |
342 | { | |
343 | struct vfio_pci_device *vdev = device_data; | |
344 | unsigned long minsz; | |
345 | ||
346 | if (cmd == VFIO_DEVICE_GET_INFO) { | |
347 | struct vfio_device_info info; | |
348 | ||
349 | minsz = offsetofend(struct vfio_device_info, num_irqs); | |
350 | ||
351 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
352 | return -EFAULT; | |
353 | ||
354 | if (info.argsz < minsz) | |
355 | return -EINVAL; | |
356 | ||
357 | info.flags = VFIO_DEVICE_FLAGS_PCI; | |
358 | ||
359 | if (vdev->reset_works) | |
360 | info.flags |= VFIO_DEVICE_FLAGS_RESET; | |
361 | ||
362 | info.num_regions = VFIO_PCI_NUM_REGIONS; | |
363 | info.num_irqs = VFIO_PCI_NUM_IRQS; | |
364 | ||
365 | return copy_to_user((void __user *)arg, &info, minsz); | |
366 | ||
367 | } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) { | |
368 | struct pci_dev *pdev = vdev->pdev; | |
369 | struct vfio_region_info info; | |
370 | ||
371 | minsz = offsetofend(struct vfio_region_info, offset); | |
372 | ||
373 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
374 | return -EFAULT; | |
375 | ||
376 | if (info.argsz < minsz) | |
377 | return -EINVAL; | |
378 | ||
379 | switch (info.index) { | |
380 | case VFIO_PCI_CONFIG_REGION_INDEX: | |
381 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
382 | info.size = pdev->cfg_size; | |
383 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
384 | VFIO_REGION_INFO_FLAG_WRITE; | |
385 | break; | |
386 | case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: | |
387 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
388 | info.size = pci_resource_len(pdev, info.index); | |
389 | if (!info.size) { | |
390 | info.flags = 0; | |
391 | break; | |
392 | } | |
393 | ||
394 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
395 | VFIO_REGION_INFO_FLAG_WRITE; | |
396 | if (pci_resource_flags(pdev, info.index) & | |
397 | IORESOURCE_MEM && info.size >= PAGE_SIZE) | |
398 | info.flags |= VFIO_REGION_INFO_FLAG_MMAP; | |
399 | break; | |
400 | case VFIO_PCI_ROM_REGION_INDEX: | |
401 | { | |
402 | void __iomem *io; | |
403 | size_t size; | |
404 | ||
405 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
406 | info.flags = 0; | |
407 | ||
408 | /* Report the BAR size, not the ROM size */ | |
409 | info.size = pci_resource_len(pdev, info.index); | |
410 | if (!info.size) | |
411 | break; | |
412 | ||
413 | /* Is it really there? */ | |
414 | io = pci_map_rom(pdev, &size); | |
415 | if (!io || !size) { | |
416 | info.size = 0; | |
417 | break; | |
418 | } | |
419 | pci_unmap_rom(pdev, io); | |
420 | ||
421 | info.flags = VFIO_REGION_INFO_FLAG_READ; | |
422 | break; | |
423 | } | |
84237a82 AW |
424 | case VFIO_PCI_VGA_REGION_INDEX: |
425 | if (!vdev->has_vga) | |
426 | return -EINVAL; | |
427 | ||
428 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
429 | info.size = 0xc0000; | |
430 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
431 | VFIO_REGION_INFO_FLAG_WRITE; | |
432 | ||
433 | break; | |
89e1f7d4 AW |
434 | default: |
435 | return -EINVAL; | |
436 | } | |
437 | ||
438 | return copy_to_user((void __user *)arg, &info, minsz); | |
439 | ||
440 | } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) { | |
441 | struct vfio_irq_info info; | |
442 | ||
443 | minsz = offsetofend(struct vfio_irq_info, count); | |
444 | ||
445 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
446 | return -EFAULT; | |
447 | ||
448 | if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS) | |
449 | return -EINVAL; | |
450 | ||
dad9f897 VMP |
451 | switch (info.index) { |
452 | case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX: | |
453 | break; | |
454 | case VFIO_PCI_ERR_IRQ_INDEX: | |
455 | if (pci_is_pcie(vdev->pdev)) | |
456 | break; | |
457 | /* pass thru to return error */ | |
458 | default: | |
459 | return -EINVAL; | |
460 | } | |
461 | ||
89e1f7d4 AW |
462 | info.flags = VFIO_IRQ_INFO_EVENTFD; |
463 | ||
464 | info.count = vfio_pci_get_irq_count(vdev, info.index); | |
465 | ||
466 | if (info.index == VFIO_PCI_INTX_IRQ_INDEX) | |
467 | info.flags |= (VFIO_IRQ_INFO_MASKABLE | | |
468 | VFIO_IRQ_INFO_AUTOMASKED); | |
469 | else | |
470 | info.flags |= VFIO_IRQ_INFO_NORESIZE; | |
471 | ||
472 | return copy_to_user((void __user *)arg, &info, minsz); | |
473 | ||
474 | } else if (cmd == VFIO_DEVICE_SET_IRQS) { | |
475 | struct vfio_irq_set hdr; | |
476 | u8 *data = NULL; | |
477 | int ret = 0; | |
478 | ||
479 | minsz = offsetofend(struct vfio_irq_set, count); | |
480 | ||
481 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
482 | return -EFAULT; | |
483 | ||
484 | if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS || | |
485 | hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK | | |
486 | VFIO_IRQ_SET_ACTION_TYPE_MASK)) | |
487 | return -EINVAL; | |
488 | ||
489 | if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) { | |
490 | size_t size; | |
904c680c | 491 | int max = vfio_pci_get_irq_count(vdev, hdr.index); |
89e1f7d4 AW |
492 | |
493 | if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL) | |
494 | size = sizeof(uint8_t); | |
495 | else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD) | |
496 | size = sizeof(int32_t); | |
497 | else | |
498 | return -EINVAL; | |
499 | ||
500 | if (hdr.argsz - minsz < hdr.count * size || | |
904c680c | 501 | hdr.start >= max || hdr.start + hdr.count > max) |
89e1f7d4 AW |
502 | return -EINVAL; |
503 | ||
3a1f7041 FW |
504 | data = memdup_user((void __user *)(arg + minsz), |
505 | hdr.count * size); | |
506 | if (IS_ERR(data)) | |
507 | return PTR_ERR(data); | |
89e1f7d4 AW |
508 | } |
509 | ||
510 | mutex_lock(&vdev->igate); | |
511 | ||
512 | ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index, | |
513 | hdr.start, hdr.count, data); | |
514 | ||
515 | mutex_unlock(&vdev->igate); | |
516 | kfree(data); | |
517 | ||
518 | return ret; | |
519 | ||
8b27ee60 | 520 | } else if (cmd == VFIO_DEVICE_RESET) { |
89e1f7d4 | 521 | return vdev->reset_works ? |
890ed578 | 522 | pci_try_reset_function(vdev->pdev) : -EINVAL; |
89e1f7d4 | 523 | |
8b27ee60 AW |
524 | } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) { |
525 | struct vfio_pci_hot_reset_info hdr; | |
526 | struct vfio_pci_fill_info fill = { 0 }; | |
527 | struct vfio_pci_dependent_device *devices = NULL; | |
528 | bool slot = false; | |
529 | int ret = 0; | |
530 | ||
531 | minsz = offsetofend(struct vfio_pci_hot_reset_info, count); | |
532 | ||
533 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
534 | return -EFAULT; | |
535 | ||
536 | if (hdr.argsz < minsz) | |
537 | return -EINVAL; | |
538 | ||
539 | hdr.flags = 0; | |
540 | ||
541 | /* Can we do a slot or bus reset or neither? */ | |
542 | if (!pci_probe_reset_slot(vdev->pdev->slot)) | |
543 | slot = true; | |
544 | else if (pci_probe_reset_bus(vdev->pdev->bus)) | |
545 | return -ENODEV; | |
546 | ||
547 | /* How many devices are affected? */ | |
548 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
549 | vfio_pci_count_devs, | |
550 | &fill.max, slot); | |
551 | if (ret) | |
552 | return ret; | |
553 | ||
554 | WARN_ON(!fill.max); /* Should always be at least one */ | |
555 | ||
556 | /* | |
557 | * If there's enough space, fill it now, otherwise return | |
558 | * -ENOSPC and the number of devices affected. | |
559 | */ | |
560 | if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) { | |
561 | ret = -ENOSPC; | |
562 | hdr.count = fill.max; | |
563 | goto reset_info_exit; | |
564 | } | |
565 | ||
566 | devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL); | |
567 | if (!devices) | |
568 | return -ENOMEM; | |
569 | ||
570 | fill.devices = devices; | |
571 | ||
572 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
573 | vfio_pci_fill_devs, | |
574 | &fill, slot); | |
575 | ||
576 | /* | |
577 | * If a device was removed between counting and filling, | |
578 | * we may come up short of fill.max. If a device was | |
579 | * added, we'll have a return of -EAGAIN above. | |
580 | */ | |
581 | if (!ret) | |
582 | hdr.count = fill.cur; | |
583 | ||
584 | reset_info_exit: | |
585 | if (copy_to_user((void __user *)arg, &hdr, minsz)) | |
586 | ret = -EFAULT; | |
587 | ||
588 | if (!ret) { | |
589 | if (copy_to_user((void __user *)(arg + minsz), devices, | |
590 | hdr.count * sizeof(*devices))) | |
591 | ret = -EFAULT; | |
592 | } | |
593 | ||
594 | kfree(devices); | |
595 | return ret; | |
596 | ||
597 | } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) { | |
598 | struct vfio_pci_hot_reset hdr; | |
599 | int32_t *group_fds; | |
600 | struct vfio_pci_group_entry *groups; | |
601 | struct vfio_pci_group_info info; | |
602 | bool slot = false; | |
603 | int i, count = 0, ret = 0; | |
604 | ||
605 | minsz = offsetofend(struct vfio_pci_hot_reset, count); | |
606 | ||
607 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
608 | return -EFAULT; | |
609 | ||
610 | if (hdr.argsz < minsz || hdr.flags) | |
611 | return -EINVAL; | |
612 | ||
613 | /* Can we do a slot or bus reset or neither? */ | |
614 | if (!pci_probe_reset_slot(vdev->pdev->slot)) | |
615 | slot = true; | |
616 | else if (pci_probe_reset_bus(vdev->pdev->bus)) | |
617 | return -ENODEV; | |
618 | ||
619 | /* | |
620 | * We can't let userspace give us an arbitrarily large | |
621 | * buffer to copy, so verify how many we think there | |
622 | * could be. Note groups can have multiple devices so | |
623 | * one group per device is the max. | |
624 | */ | |
625 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
626 | vfio_pci_count_devs, | |
627 | &count, slot); | |
628 | if (ret) | |
629 | return ret; | |
630 | ||
631 | /* Somewhere between 1 and count is OK */ | |
632 | if (!hdr.count || hdr.count > count) | |
633 | return -EINVAL; | |
634 | ||
635 | group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL); | |
636 | groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL); | |
637 | if (!group_fds || !groups) { | |
638 | kfree(group_fds); | |
639 | kfree(groups); | |
640 | return -ENOMEM; | |
641 | } | |
642 | ||
643 | if (copy_from_user(group_fds, (void __user *)(arg + minsz), | |
644 | hdr.count * sizeof(*group_fds))) { | |
645 | kfree(group_fds); | |
646 | kfree(groups); | |
647 | return -EFAULT; | |
648 | } | |
649 | ||
650 | /* | |
651 | * For each group_fd, get the group through the vfio external | |
652 | * user interface and store the group and iommu ID. This | |
653 | * ensures the group is held across the reset. | |
654 | */ | |
655 | for (i = 0; i < hdr.count; i++) { | |
656 | struct vfio_group *group; | |
657 | struct fd f = fdget(group_fds[i]); | |
658 | if (!f.file) { | |
659 | ret = -EBADF; | |
660 | break; | |
661 | } | |
662 | ||
663 | group = vfio_group_get_external_user(f.file); | |
664 | fdput(f); | |
665 | if (IS_ERR(group)) { | |
666 | ret = PTR_ERR(group); | |
667 | break; | |
668 | } | |
669 | ||
670 | groups[i].group = group; | |
671 | groups[i].id = vfio_external_user_iommu_id(group); | |
672 | } | |
673 | ||
674 | kfree(group_fds); | |
675 | ||
676 | /* release reference to groups on error */ | |
677 | if (ret) | |
678 | goto hot_reset_release; | |
679 | ||
680 | info.count = hdr.count; | |
681 | info.groups = groups; | |
682 | ||
683 | /* | |
684 | * Test whether all the affected devices are contained | |
685 | * by the set of groups provided by the user. | |
686 | */ | |
687 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
688 | vfio_pci_validate_devs, | |
689 | &info, slot); | |
690 | if (!ret) | |
691 | /* User has access, do the reset */ | |
890ed578 AW |
692 | ret = slot ? pci_try_reset_slot(vdev->pdev->slot) : |
693 | pci_try_reset_bus(vdev->pdev->bus); | |
8b27ee60 AW |
694 | |
695 | hot_reset_release: | |
696 | for (i--; i >= 0; i--) | |
697 | vfio_group_put_external_user(groups[i].group); | |
698 | ||
699 | kfree(groups); | |
700 | return ret; | |
701 | } | |
702 | ||
89e1f7d4 AW |
703 | return -ENOTTY; |
704 | } | |
705 | ||
5b279a11 AW |
706 | static ssize_t vfio_pci_rw(void *device_data, char __user *buf, |
707 | size_t count, loff_t *ppos, bool iswrite) | |
89e1f7d4 AW |
708 | { |
709 | unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); | |
710 | struct vfio_pci_device *vdev = device_data; | |
89e1f7d4 AW |
711 | |
712 | if (index >= VFIO_PCI_NUM_REGIONS) | |
713 | return -EINVAL; | |
714 | ||
5b279a11 AW |
715 | switch (index) { |
716 | case VFIO_PCI_CONFIG_REGION_INDEX: | |
906ee99d AW |
717 | return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite); |
718 | ||
5b279a11 AW |
719 | case VFIO_PCI_ROM_REGION_INDEX: |
720 | if (iswrite) | |
721 | return -EINVAL; | |
906ee99d | 722 | return vfio_pci_bar_rw(vdev, buf, count, ppos, false); |
89e1f7d4 | 723 | |
5b279a11 | 724 | case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: |
906ee99d | 725 | return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite); |
84237a82 AW |
726 | |
727 | case VFIO_PCI_VGA_REGION_INDEX: | |
728 | return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite); | |
5b279a11 AW |
729 | } |
730 | ||
89e1f7d4 AW |
731 | return -EINVAL; |
732 | } | |
733 | ||
5b279a11 AW |
734 | static ssize_t vfio_pci_read(void *device_data, char __user *buf, |
735 | size_t count, loff_t *ppos) | |
736 | { | |
906ee99d AW |
737 | if (!count) |
738 | return 0; | |
739 | ||
5b279a11 AW |
740 | return vfio_pci_rw(device_data, buf, count, ppos, false); |
741 | } | |
742 | ||
89e1f7d4 AW |
743 | static ssize_t vfio_pci_write(void *device_data, const char __user *buf, |
744 | size_t count, loff_t *ppos) | |
745 | { | |
906ee99d AW |
746 | if (!count) |
747 | return 0; | |
748 | ||
749 | return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true); | |
89e1f7d4 AW |
750 | } |
751 | ||
752 | static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma) | |
753 | { | |
754 | struct vfio_pci_device *vdev = device_data; | |
755 | struct pci_dev *pdev = vdev->pdev; | |
756 | unsigned int index; | |
34002f54 | 757 | u64 phys_len, req_len, pgoff, req_start; |
89e1f7d4 AW |
758 | int ret; |
759 | ||
760 | index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); | |
761 | ||
762 | if (vma->vm_end < vma->vm_start) | |
763 | return -EINVAL; | |
764 | if ((vma->vm_flags & VM_SHARED) == 0) | |
765 | return -EINVAL; | |
766 | if (index >= VFIO_PCI_ROM_REGION_INDEX) | |
767 | return -EINVAL; | |
768 | if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM)) | |
769 | return -EINVAL; | |
770 | ||
771 | phys_len = pci_resource_len(pdev, index); | |
772 | req_len = vma->vm_end - vma->vm_start; | |
773 | pgoff = vma->vm_pgoff & | |
774 | ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); | |
775 | req_start = pgoff << PAGE_SHIFT; | |
776 | ||
777 | if (phys_len < PAGE_SIZE || req_start + req_len > phys_len) | |
778 | return -EINVAL; | |
779 | ||
780 | if (index == vdev->msix_bar) { | |
781 | /* | |
782 | * Disallow mmaps overlapping the MSI-X table; users don't | |
783 | * get to touch this directly. We could find somewhere | |
784 | * else to map the overlap, but page granularity is only | |
785 | * a recommendation, not a requirement, so the user needs | |
786 | * to know which bits are real. Requiring them to mmap | |
787 | * around the table makes that clear. | |
788 | */ | |
789 | ||
790 | /* If neither entirely above nor below, then it overlaps */ | |
791 | if (!(req_start >= vdev->msix_offset + vdev->msix_size || | |
792 | req_start + req_len <= vdev->msix_offset)) | |
793 | return -EINVAL; | |
794 | } | |
795 | ||
796 | /* | |
797 | * Even though we don't make use of the barmap for the mmap, | |
798 | * we need to request the region and the barmap tracks that. | |
799 | */ | |
800 | if (!vdev->barmap[index]) { | |
801 | ret = pci_request_selected_regions(pdev, | |
802 | 1 << index, "vfio-pci"); | |
803 | if (ret) | |
804 | return ret; | |
805 | ||
806 | vdev->barmap[index] = pci_iomap(pdev, index, 0); | |
807 | } | |
808 | ||
809 | vma->vm_private_data = vdev; | |
89e1f7d4 | 810 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); |
34002f54 | 811 | vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff; |
89e1f7d4 | 812 | |
34002f54 | 813 | return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff, |
89e1f7d4 AW |
814 | req_len, vma->vm_page_prot); |
815 | } | |
816 | ||
817 | static const struct vfio_device_ops vfio_pci_ops = { | |
818 | .name = "vfio-pci", | |
819 | .open = vfio_pci_open, | |
820 | .release = vfio_pci_release, | |
821 | .ioctl = vfio_pci_ioctl, | |
822 | .read = vfio_pci_read, | |
823 | .write = vfio_pci_write, | |
824 | .mmap = vfio_pci_mmap, | |
825 | }; | |
826 | ||
827 | static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) | |
828 | { | |
829 | u8 type; | |
830 | struct vfio_pci_device *vdev; | |
831 | struct iommu_group *group; | |
832 | int ret; | |
833 | ||
834 | pci_read_config_byte(pdev, PCI_HEADER_TYPE, &type); | |
835 | if ((type & PCI_HEADER_TYPE) != PCI_HEADER_TYPE_NORMAL) | |
836 | return -EINVAL; | |
837 | ||
838 | group = iommu_group_get(&pdev->dev); | |
839 | if (!group) | |
840 | return -EINVAL; | |
841 | ||
842 | vdev = kzalloc(sizeof(*vdev), GFP_KERNEL); | |
843 | if (!vdev) { | |
844 | iommu_group_put(group); | |
845 | return -ENOMEM; | |
846 | } | |
847 | ||
848 | vdev->pdev = pdev; | |
849 | vdev->irq_type = VFIO_PCI_NUM_IRQS; | |
850 | mutex_init(&vdev->igate); | |
851 | spin_lock_init(&vdev->irqlock); | |
852 | atomic_set(&vdev->refcnt, 0); | |
853 | ||
854 | ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev); | |
855 | if (ret) { | |
856 | iommu_group_put(group); | |
857 | kfree(vdev); | |
858 | } | |
859 | ||
860 | return ret; | |
861 | } | |
862 | ||
863 | static void vfio_pci_remove(struct pci_dev *pdev) | |
864 | { | |
865 | struct vfio_pci_device *vdev; | |
866 | ||
867 | vdev = vfio_del_group_dev(&pdev->dev); | |
868 | if (!vdev) | |
869 | return; | |
870 | ||
871 | iommu_group_put(pdev->dev.iommu_group); | |
872 | kfree(vdev); | |
873 | } | |
874 | ||
dad9f897 VMP |
875 | static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev, |
876 | pci_channel_state_t state) | |
877 | { | |
878 | struct vfio_pci_device *vdev; | |
879 | struct vfio_device *device; | |
880 | ||
881 | device = vfio_device_get_from_dev(&pdev->dev); | |
882 | if (device == NULL) | |
883 | return PCI_ERS_RESULT_DISCONNECT; | |
884 | ||
885 | vdev = vfio_device_data(device); | |
886 | if (vdev == NULL) { | |
887 | vfio_device_put(device); | |
888 | return PCI_ERS_RESULT_DISCONNECT; | |
889 | } | |
890 | ||
3be3a074 AW |
891 | mutex_lock(&vdev->igate); |
892 | ||
dad9f897 VMP |
893 | if (vdev->err_trigger) |
894 | eventfd_signal(vdev->err_trigger, 1); | |
895 | ||
3be3a074 AW |
896 | mutex_unlock(&vdev->igate); |
897 | ||
dad9f897 VMP |
898 | vfio_device_put(device); |
899 | ||
900 | return PCI_ERS_RESULT_CAN_RECOVER; | |
901 | } | |
902 | ||
903 | static struct pci_error_handlers vfio_err_handlers = { | |
904 | .error_detected = vfio_pci_aer_err_detected, | |
905 | }; | |
906 | ||
89e1f7d4 AW |
907 | static struct pci_driver vfio_pci_driver = { |
908 | .name = "vfio-pci", | |
909 | .id_table = NULL, /* only dynamic ids */ | |
910 | .probe = vfio_pci_probe, | |
911 | .remove = vfio_pci_remove, | |
dad9f897 | 912 | .err_handler = &vfio_err_handlers, |
89e1f7d4 AW |
913 | }; |
914 | ||
915 | static void __exit vfio_pci_cleanup(void) | |
916 | { | |
917 | pci_unregister_driver(&vfio_pci_driver); | |
918 | vfio_pci_virqfd_exit(); | |
919 | vfio_pci_uninit_perm_bits(); | |
920 | } | |
921 | ||
922 | static int __init vfio_pci_init(void) | |
923 | { | |
924 | int ret; | |
925 | ||
926 | /* Allocate shared config space permision data used by all devices */ | |
927 | ret = vfio_pci_init_perm_bits(); | |
928 | if (ret) | |
929 | return ret; | |
930 | ||
931 | /* Start the virqfd cleanup handler */ | |
932 | ret = vfio_pci_virqfd_init(); | |
933 | if (ret) | |
934 | goto out_virqfd; | |
935 | ||
936 | /* Register and scan for devices */ | |
937 | ret = pci_register_driver(&vfio_pci_driver); | |
938 | if (ret) | |
939 | goto out_driver; | |
940 | ||
941 | return 0; | |
942 | ||
89e1f7d4 | 943 | out_driver: |
05bf3aac JL |
944 | vfio_pci_virqfd_exit(); |
945 | out_virqfd: | |
89e1f7d4 AW |
946 | vfio_pci_uninit_perm_bits(); |
947 | return ret; | |
948 | } | |
949 | ||
950 | module_init(vfio_pci_init); | |
951 | module_exit(vfio_pci_cleanup); | |
952 | ||
953 | MODULE_VERSION(DRIVER_VERSION); | |
954 | MODULE_LICENSE("GPL v2"); | |
955 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
956 | MODULE_DESCRIPTION(DRIVER_DESC); |