]>
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 | ||
61d79256 AW |
40 | static DEFINE_MUTEX(driver_lock); |
41 | ||
89e1f7d4 AW |
42 | static int vfio_pci_enable(struct vfio_pci_device *vdev) |
43 | { | |
44 | struct pci_dev *pdev = vdev->pdev; | |
45 | int ret; | |
46 | u16 cmd; | |
47 | u8 msix_pos; | |
48 | ||
9c22e660 AW |
49 | /* Don't allow our initial saved state to include busmaster */ |
50 | pci_clear_master(pdev); | |
51 | ||
9a92c509 AW |
52 | ret = pci_enable_device(pdev); |
53 | if (ret) | |
54 | return ret; | |
55 | ||
89e1f7d4 AW |
56 | vdev->reset_works = (pci_reset_function(pdev) == 0); |
57 | pci_save_state(pdev); | |
58 | vdev->pci_saved_state = pci_store_saved_state(pdev); | |
59 | if (!vdev->pci_saved_state) | |
60 | pr_debug("%s: Couldn't store %s saved state\n", | |
61 | __func__, dev_name(&pdev->dev)); | |
62 | ||
63 | ret = vfio_config_init(vdev); | |
9a92c509 | 64 | if (ret) { |
eb5685f0 AW |
65 | kfree(vdev->pci_saved_state); |
66 | vdev->pci_saved_state = NULL; | |
9a92c509 AW |
67 | pci_disable_device(pdev); |
68 | return ret; | |
69 | } | |
89e1f7d4 AW |
70 | |
71 | if (likely(!nointxmask)) | |
72 | vdev->pci_2_3 = pci_intx_mask_supported(pdev); | |
73 | ||
74 | pci_read_config_word(pdev, PCI_COMMAND, &cmd); | |
75 | if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) { | |
76 | cmd &= ~PCI_COMMAND_INTX_DISABLE; | |
77 | pci_write_config_word(pdev, PCI_COMMAND, cmd); | |
78 | } | |
79 | ||
a9047f24 | 80 | msix_pos = pdev->msix_cap; |
89e1f7d4 AW |
81 | if (msix_pos) { |
82 | u16 flags; | |
83 | u32 table; | |
84 | ||
85 | pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags); | |
86 | pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table); | |
87 | ||
508d1aa6 BH |
88 | vdev->msix_bar = table & PCI_MSIX_TABLE_BIR; |
89 | vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET; | |
89e1f7d4 AW |
90 | vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16; |
91 | } else | |
92 | vdev->msix_bar = 0xFF; | |
93 | ||
84237a82 AW |
94 | #ifdef CONFIG_VFIO_PCI_VGA |
95 | if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) | |
96 | vdev->has_vga = true; | |
97 | #endif | |
98 | ||
9a92c509 | 99 | return 0; |
89e1f7d4 AW |
100 | } |
101 | ||
102 | static void vfio_pci_disable(struct vfio_pci_device *vdev) | |
103 | { | |
2007722a | 104 | struct pci_dev *pdev = vdev->pdev; |
89e1f7d4 AW |
105 | int bar; |
106 | ||
9c22e660 AW |
107 | /* Stop the device from further DMA */ |
108 | pci_clear_master(pdev); | |
89e1f7d4 AW |
109 | |
110 | vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE | | |
111 | VFIO_IRQ_SET_ACTION_TRIGGER, | |
112 | vdev->irq_type, 0, 0, NULL); | |
113 | ||
114 | vdev->virq_disabled = false; | |
115 | ||
116 | vfio_config_free(vdev); | |
117 | ||
89e1f7d4 AW |
118 | for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) { |
119 | if (!vdev->barmap[bar]) | |
120 | continue; | |
2007722a AW |
121 | pci_iounmap(pdev, vdev->barmap[bar]); |
122 | pci_release_selected_regions(pdev, 1 << bar); | |
89e1f7d4 AW |
123 | vdev->barmap[bar] = NULL; |
124 | } | |
2007722a AW |
125 | |
126 | /* | |
127 | * If we have saved state, restore it. If we can reset the device, | |
128 | * even better. Resetting with current state seems better than | |
129 | * nothing, but saving and restoring current state without reset | |
130 | * is just busy work. | |
131 | */ | |
132 | if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) { | |
133 | pr_info("%s: Couldn't reload %s saved state\n", | |
134 | __func__, dev_name(&pdev->dev)); | |
135 | ||
136 | if (!vdev->reset_works) | |
9c22e660 | 137 | goto out; |
2007722a AW |
138 | |
139 | pci_save_state(pdev); | |
140 | } | |
141 | ||
142 | /* | |
143 | * Disable INTx and MSI, presumably to avoid spurious interrupts | |
144 | * during reset. Stolen from pci_reset_function() | |
145 | */ | |
146 | pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE); | |
147 | ||
d24cdbfd | 148 | /* |
890ed578 AW |
149 | * Try to reset the device. The success of this is dependent on |
150 | * being able to lock the device, which is not always possible. | |
d24cdbfd AW |
151 | */ |
152 | if (vdev->reset_works) { | |
890ed578 AW |
153 | int ret = pci_try_reset_function(pdev); |
154 | if (ret) | |
155 | pr_warn("%s: Failed to reset device %s (%d)\n", | |
156 | __func__, dev_name(&pdev->dev), ret); | |
d24cdbfd | 157 | } |
2007722a AW |
158 | |
159 | pci_restore_state(pdev); | |
9c22e660 AW |
160 | out: |
161 | pci_disable_device(pdev); | |
89e1f7d4 AW |
162 | } |
163 | ||
164 | static void vfio_pci_release(void *device_data) | |
165 | { | |
166 | struct vfio_pci_device *vdev = device_data; | |
167 | ||
61d79256 AW |
168 | mutex_lock(&driver_lock); |
169 | ||
170 | if (!(--vdev->refcnt)) { | |
1b69be5e | 171 | vfio_spapr_pci_eeh_release(vdev->pdev); |
89e1f7d4 | 172 | vfio_pci_disable(vdev); |
1b69be5e | 173 | } |
89e1f7d4 | 174 | |
61d79256 AW |
175 | mutex_unlock(&driver_lock); |
176 | ||
89e1f7d4 AW |
177 | module_put(THIS_MODULE); |
178 | } | |
179 | ||
180 | static int vfio_pci_open(void *device_data) | |
181 | { | |
182 | struct vfio_pci_device *vdev = device_data; | |
61d79256 | 183 | int ret = 0; |
89e1f7d4 AW |
184 | |
185 | if (!try_module_get(THIS_MODULE)) | |
186 | return -ENODEV; | |
187 | ||
61d79256 AW |
188 | mutex_lock(&driver_lock); |
189 | ||
190 | if (!vdev->refcnt) { | |
1b69be5e GS |
191 | ret = vfio_pci_enable(vdev); |
192 | if (ret) | |
193 | goto error; | |
194 | ||
195 | ret = vfio_spapr_pci_eeh_open(vdev->pdev); | |
89e1f7d4 | 196 | if (ret) { |
1b69be5e GS |
197 | vfio_pci_disable(vdev); |
198 | goto error; | |
89e1f7d4 AW |
199 | } |
200 | } | |
61d79256 | 201 | vdev->refcnt++; |
1b69be5e | 202 | error: |
61d79256 AW |
203 | mutex_unlock(&driver_lock); |
204 | if (ret) | |
205 | module_put(THIS_MODULE); | |
1b69be5e | 206 | return ret; |
89e1f7d4 AW |
207 | } |
208 | ||
209 | static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type) | |
210 | { | |
211 | if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) { | |
212 | u8 pin; | |
213 | pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin); | |
214 | if (pin) | |
215 | return 1; | |
216 | ||
217 | } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) { | |
218 | u8 pos; | |
219 | u16 flags; | |
220 | ||
a9047f24 | 221 | pos = vdev->pdev->msi_cap; |
89e1f7d4 AW |
222 | if (pos) { |
223 | pci_read_config_word(vdev->pdev, | |
224 | pos + PCI_MSI_FLAGS, &flags); | |
fd49c81f | 225 | return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1); |
89e1f7d4 AW |
226 | } |
227 | } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) { | |
228 | u8 pos; | |
229 | u16 flags; | |
230 | ||
a9047f24 | 231 | pos = vdev->pdev->msix_cap; |
89e1f7d4 AW |
232 | if (pos) { |
233 | pci_read_config_word(vdev->pdev, | |
234 | pos + PCI_MSIX_FLAGS, &flags); | |
235 | ||
236 | return (flags & PCI_MSIX_FLAGS_QSIZE) + 1; | |
237 | } | |
dad9f897 VMP |
238 | } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) |
239 | if (pci_is_pcie(vdev->pdev)) | |
240 | return 1; | |
89e1f7d4 AW |
241 | |
242 | return 0; | |
243 | } | |
244 | ||
8b27ee60 AW |
245 | static int vfio_pci_count_devs(struct pci_dev *pdev, void *data) |
246 | { | |
247 | (*(int *)data)++; | |
248 | return 0; | |
249 | } | |
250 | ||
251 | struct vfio_pci_fill_info { | |
252 | int max; | |
253 | int cur; | |
254 | struct vfio_pci_dependent_device *devices; | |
255 | }; | |
256 | ||
257 | static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data) | |
258 | { | |
259 | struct vfio_pci_fill_info *fill = data; | |
260 | struct iommu_group *iommu_group; | |
261 | ||
262 | if (fill->cur == fill->max) | |
263 | return -EAGAIN; /* Something changed, try again */ | |
264 | ||
265 | iommu_group = iommu_group_get(&pdev->dev); | |
266 | if (!iommu_group) | |
267 | return -EPERM; /* Cannot reset non-isolated devices */ | |
268 | ||
269 | fill->devices[fill->cur].group_id = iommu_group_id(iommu_group); | |
270 | fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus); | |
271 | fill->devices[fill->cur].bus = pdev->bus->number; | |
272 | fill->devices[fill->cur].devfn = pdev->devfn; | |
273 | fill->cur++; | |
274 | iommu_group_put(iommu_group); | |
275 | return 0; | |
276 | } | |
277 | ||
278 | struct vfio_pci_group_entry { | |
279 | struct vfio_group *group; | |
280 | int id; | |
281 | }; | |
282 | ||
283 | struct vfio_pci_group_info { | |
284 | int count; | |
285 | struct vfio_pci_group_entry *groups; | |
286 | }; | |
287 | ||
288 | static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data) | |
289 | { | |
290 | struct vfio_pci_group_info *info = data; | |
291 | struct iommu_group *group; | |
292 | int id, i; | |
293 | ||
294 | group = iommu_group_get(&pdev->dev); | |
295 | if (!group) | |
296 | return -EPERM; | |
297 | ||
298 | id = iommu_group_id(group); | |
299 | ||
300 | for (i = 0; i < info->count; i++) | |
301 | if (info->groups[i].id == id) | |
302 | break; | |
303 | ||
304 | iommu_group_put(group); | |
305 | ||
306 | return (i == info->count) ? -EINVAL : 0; | |
307 | } | |
308 | ||
309 | static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot) | |
310 | { | |
311 | for (; pdev; pdev = pdev->bus->self) | |
312 | if (pdev->bus == slot->bus) | |
313 | return (pdev->slot == slot); | |
314 | return false; | |
315 | } | |
316 | ||
317 | struct vfio_pci_walk_info { | |
318 | int (*fn)(struct pci_dev *, void *data); | |
319 | void *data; | |
320 | struct pci_dev *pdev; | |
321 | bool slot; | |
322 | int ret; | |
323 | }; | |
324 | ||
325 | static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data) | |
326 | { | |
327 | struct vfio_pci_walk_info *walk = data; | |
328 | ||
329 | if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot)) | |
330 | walk->ret = walk->fn(pdev, walk->data); | |
331 | ||
332 | return walk->ret; | |
333 | } | |
334 | ||
335 | static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev, | |
336 | int (*fn)(struct pci_dev *, | |
337 | void *data), void *data, | |
338 | bool slot) | |
339 | { | |
340 | struct vfio_pci_walk_info walk = { | |
341 | .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0, | |
342 | }; | |
343 | ||
344 | pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk); | |
345 | ||
346 | return walk.ret; | |
347 | } | |
348 | ||
89e1f7d4 AW |
349 | static long vfio_pci_ioctl(void *device_data, |
350 | unsigned int cmd, unsigned long arg) | |
351 | { | |
352 | struct vfio_pci_device *vdev = device_data; | |
353 | unsigned long minsz; | |
354 | ||
355 | if (cmd == VFIO_DEVICE_GET_INFO) { | |
356 | struct vfio_device_info info; | |
357 | ||
358 | minsz = offsetofend(struct vfio_device_info, num_irqs); | |
359 | ||
360 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
361 | return -EFAULT; | |
362 | ||
363 | if (info.argsz < minsz) | |
364 | return -EINVAL; | |
365 | ||
366 | info.flags = VFIO_DEVICE_FLAGS_PCI; | |
367 | ||
368 | if (vdev->reset_works) | |
369 | info.flags |= VFIO_DEVICE_FLAGS_RESET; | |
370 | ||
371 | info.num_regions = VFIO_PCI_NUM_REGIONS; | |
372 | info.num_irqs = VFIO_PCI_NUM_IRQS; | |
373 | ||
374 | return copy_to_user((void __user *)arg, &info, minsz); | |
375 | ||
376 | } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) { | |
377 | struct pci_dev *pdev = vdev->pdev; | |
378 | struct vfio_region_info info; | |
379 | ||
380 | minsz = offsetofend(struct vfio_region_info, offset); | |
381 | ||
382 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
383 | return -EFAULT; | |
384 | ||
385 | if (info.argsz < minsz) | |
386 | return -EINVAL; | |
387 | ||
388 | switch (info.index) { | |
389 | case VFIO_PCI_CONFIG_REGION_INDEX: | |
390 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
391 | info.size = pdev->cfg_size; | |
392 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
393 | VFIO_REGION_INFO_FLAG_WRITE; | |
394 | break; | |
395 | case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: | |
396 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
397 | info.size = pci_resource_len(pdev, info.index); | |
398 | if (!info.size) { | |
399 | info.flags = 0; | |
400 | break; | |
401 | } | |
402 | ||
403 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
404 | VFIO_REGION_INFO_FLAG_WRITE; | |
405 | if (pci_resource_flags(pdev, info.index) & | |
406 | IORESOURCE_MEM && info.size >= PAGE_SIZE) | |
407 | info.flags |= VFIO_REGION_INFO_FLAG_MMAP; | |
408 | break; | |
409 | case VFIO_PCI_ROM_REGION_INDEX: | |
410 | { | |
411 | void __iomem *io; | |
412 | size_t size; | |
413 | ||
414 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
415 | info.flags = 0; | |
416 | ||
417 | /* Report the BAR size, not the ROM size */ | |
418 | info.size = pci_resource_len(pdev, info.index); | |
419 | if (!info.size) | |
420 | break; | |
421 | ||
422 | /* Is it really there? */ | |
423 | io = pci_map_rom(pdev, &size); | |
424 | if (!io || !size) { | |
425 | info.size = 0; | |
426 | break; | |
427 | } | |
428 | pci_unmap_rom(pdev, io); | |
429 | ||
430 | info.flags = VFIO_REGION_INFO_FLAG_READ; | |
431 | break; | |
432 | } | |
84237a82 AW |
433 | case VFIO_PCI_VGA_REGION_INDEX: |
434 | if (!vdev->has_vga) | |
435 | return -EINVAL; | |
436 | ||
437 | info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); | |
438 | info.size = 0xc0000; | |
439 | info.flags = VFIO_REGION_INFO_FLAG_READ | | |
440 | VFIO_REGION_INFO_FLAG_WRITE; | |
441 | ||
442 | break; | |
89e1f7d4 AW |
443 | default: |
444 | return -EINVAL; | |
445 | } | |
446 | ||
447 | return copy_to_user((void __user *)arg, &info, minsz); | |
448 | ||
449 | } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) { | |
450 | struct vfio_irq_info info; | |
451 | ||
452 | minsz = offsetofend(struct vfio_irq_info, count); | |
453 | ||
454 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
455 | return -EFAULT; | |
456 | ||
457 | if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS) | |
458 | return -EINVAL; | |
459 | ||
dad9f897 VMP |
460 | switch (info.index) { |
461 | case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX: | |
462 | break; | |
463 | case VFIO_PCI_ERR_IRQ_INDEX: | |
464 | if (pci_is_pcie(vdev->pdev)) | |
465 | break; | |
466 | /* pass thru to return error */ | |
467 | default: | |
468 | return -EINVAL; | |
469 | } | |
470 | ||
89e1f7d4 AW |
471 | info.flags = VFIO_IRQ_INFO_EVENTFD; |
472 | ||
473 | info.count = vfio_pci_get_irq_count(vdev, info.index); | |
474 | ||
475 | if (info.index == VFIO_PCI_INTX_IRQ_INDEX) | |
476 | info.flags |= (VFIO_IRQ_INFO_MASKABLE | | |
477 | VFIO_IRQ_INFO_AUTOMASKED); | |
478 | else | |
479 | info.flags |= VFIO_IRQ_INFO_NORESIZE; | |
480 | ||
481 | return copy_to_user((void __user *)arg, &info, minsz); | |
482 | ||
483 | } else if (cmd == VFIO_DEVICE_SET_IRQS) { | |
484 | struct vfio_irq_set hdr; | |
485 | u8 *data = NULL; | |
486 | int ret = 0; | |
487 | ||
488 | minsz = offsetofend(struct vfio_irq_set, count); | |
489 | ||
490 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
491 | return -EFAULT; | |
492 | ||
493 | if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS || | |
494 | hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK | | |
495 | VFIO_IRQ_SET_ACTION_TYPE_MASK)) | |
496 | return -EINVAL; | |
497 | ||
498 | if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) { | |
499 | size_t size; | |
904c680c | 500 | int max = vfio_pci_get_irq_count(vdev, hdr.index); |
89e1f7d4 AW |
501 | |
502 | if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL) | |
503 | size = sizeof(uint8_t); | |
504 | else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD) | |
505 | size = sizeof(int32_t); | |
506 | else | |
507 | return -EINVAL; | |
508 | ||
509 | if (hdr.argsz - minsz < hdr.count * size || | |
904c680c | 510 | hdr.start >= max || hdr.start + hdr.count > max) |
89e1f7d4 AW |
511 | return -EINVAL; |
512 | ||
3a1f7041 FW |
513 | data = memdup_user((void __user *)(arg + minsz), |
514 | hdr.count * size); | |
515 | if (IS_ERR(data)) | |
516 | return PTR_ERR(data); | |
89e1f7d4 AW |
517 | } |
518 | ||
519 | mutex_lock(&vdev->igate); | |
520 | ||
521 | ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index, | |
522 | hdr.start, hdr.count, data); | |
523 | ||
524 | mutex_unlock(&vdev->igate); | |
525 | kfree(data); | |
526 | ||
527 | return ret; | |
528 | ||
8b27ee60 | 529 | } else if (cmd == VFIO_DEVICE_RESET) { |
89e1f7d4 | 530 | return vdev->reset_works ? |
890ed578 | 531 | pci_try_reset_function(vdev->pdev) : -EINVAL; |
89e1f7d4 | 532 | |
8b27ee60 AW |
533 | } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) { |
534 | struct vfio_pci_hot_reset_info hdr; | |
535 | struct vfio_pci_fill_info fill = { 0 }; | |
536 | struct vfio_pci_dependent_device *devices = NULL; | |
537 | bool slot = false; | |
538 | int ret = 0; | |
539 | ||
540 | minsz = offsetofend(struct vfio_pci_hot_reset_info, count); | |
541 | ||
542 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
543 | return -EFAULT; | |
544 | ||
545 | if (hdr.argsz < minsz) | |
546 | return -EINVAL; | |
547 | ||
548 | hdr.flags = 0; | |
549 | ||
550 | /* Can we do a slot or bus reset or neither? */ | |
551 | if (!pci_probe_reset_slot(vdev->pdev->slot)) | |
552 | slot = true; | |
553 | else if (pci_probe_reset_bus(vdev->pdev->bus)) | |
554 | return -ENODEV; | |
555 | ||
556 | /* How many devices are affected? */ | |
557 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
558 | vfio_pci_count_devs, | |
559 | &fill.max, slot); | |
560 | if (ret) | |
561 | return ret; | |
562 | ||
563 | WARN_ON(!fill.max); /* Should always be at least one */ | |
564 | ||
565 | /* | |
566 | * If there's enough space, fill it now, otherwise return | |
567 | * -ENOSPC and the number of devices affected. | |
568 | */ | |
569 | if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) { | |
570 | ret = -ENOSPC; | |
571 | hdr.count = fill.max; | |
572 | goto reset_info_exit; | |
573 | } | |
574 | ||
575 | devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL); | |
576 | if (!devices) | |
577 | return -ENOMEM; | |
578 | ||
579 | fill.devices = devices; | |
580 | ||
581 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
582 | vfio_pci_fill_devs, | |
583 | &fill, slot); | |
584 | ||
585 | /* | |
586 | * If a device was removed between counting and filling, | |
587 | * we may come up short of fill.max. If a device was | |
588 | * added, we'll have a return of -EAGAIN above. | |
589 | */ | |
590 | if (!ret) | |
591 | hdr.count = fill.cur; | |
592 | ||
593 | reset_info_exit: | |
594 | if (copy_to_user((void __user *)arg, &hdr, minsz)) | |
595 | ret = -EFAULT; | |
596 | ||
597 | if (!ret) { | |
598 | if (copy_to_user((void __user *)(arg + minsz), devices, | |
599 | hdr.count * sizeof(*devices))) | |
600 | ret = -EFAULT; | |
601 | } | |
602 | ||
603 | kfree(devices); | |
604 | return ret; | |
605 | ||
606 | } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) { | |
607 | struct vfio_pci_hot_reset hdr; | |
608 | int32_t *group_fds; | |
609 | struct vfio_pci_group_entry *groups; | |
610 | struct vfio_pci_group_info info; | |
611 | bool slot = false; | |
612 | int i, count = 0, ret = 0; | |
613 | ||
614 | minsz = offsetofend(struct vfio_pci_hot_reset, count); | |
615 | ||
616 | if (copy_from_user(&hdr, (void __user *)arg, minsz)) | |
617 | return -EFAULT; | |
618 | ||
619 | if (hdr.argsz < minsz || hdr.flags) | |
620 | return -EINVAL; | |
621 | ||
622 | /* Can we do a slot or bus reset or neither? */ | |
623 | if (!pci_probe_reset_slot(vdev->pdev->slot)) | |
624 | slot = true; | |
625 | else if (pci_probe_reset_bus(vdev->pdev->bus)) | |
626 | return -ENODEV; | |
627 | ||
628 | /* | |
629 | * We can't let userspace give us an arbitrarily large | |
630 | * buffer to copy, so verify how many we think there | |
631 | * could be. Note groups can have multiple devices so | |
632 | * one group per device is the max. | |
633 | */ | |
634 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
635 | vfio_pci_count_devs, | |
636 | &count, slot); | |
637 | if (ret) | |
638 | return ret; | |
639 | ||
640 | /* Somewhere between 1 and count is OK */ | |
641 | if (!hdr.count || hdr.count > count) | |
642 | return -EINVAL; | |
643 | ||
644 | group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL); | |
645 | groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL); | |
646 | if (!group_fds || !groups) { | |
647 | kfree(group_fds); | |
648 | kfree(groups); | |
649 | return -ENOMEM; | |
650 | } | |
651 | ||
652 | if (copy_from_user(group_fds, (void __user *)(arg + minsz), | |
653 | hdr.count * sizeof(*group_fds))) { | |
654 | kfree(group_fds); | |
655 | kfree(groups); | |
656 | return -EFAULT; | |
657 | } | |
658 | ||
659 | /* | |
660 | * For each group_fd, get the group through the vfio external | |
661 | * user interface and store the group and iommu ID. This | |
662 | * ensures the group is held across the reset. | |
663 | */ | |
664 | for (i = 0; i < hdr.count; i++) { | |
665 | struct vfio_group *group; | |
666 | struct fd f = fdget(group_fds[i]); | |
667 | if (!f.file) { | |
668 | ret = -EBADF; | |
669 | break; | |
670 | } | |
671 | ||
672 | group = vfio_group_get_external_user(f.file); | |
673 | fdput(f); | |
674 | if (IS_ERR(group)) { | |
675 | ret = PTR_ERR(group); | |
676 | break; | |
677 | } | |
678 | ||
679 | groups[i].group = group; | |
680 | groups[i].id = vfio_external_user_iommu_id(group); | |
681 | } | |
682 | ||
683 | kfree(group_fds); | |
684 | ||
685 | /* release reference to groups on error */ | |
686 | if (ret) | |
687 | goto hot_reset_release; | |
688 | ||
689 | info.count = hdr.count; | |
690 | info.groups = groups; | |
691 | ||
692 | /* | |
693 | * Test whether all the affected devices are contained | |
694 | * by the set of groups provided by the user. | |
695 | */ | |
696 | ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, | |
697 | vfio_pci_validate_devs, | |
698 | &info, slot); | |
699 | if (!ret) | |
700 | /* User has access, do the reset */ | |
890ed578 AW |
701 | ret = slot ? pci_try_reset_slot(vdev->pdev->slot) : |
702 | pci_try_reset_bus(vdev->pdev->bus); | |
8b27ee60 AW |
703 | |
704 | hot_reset_release: | |
705 | for (i--; i >= 0; i--) | |
706 | vfio_group_put_external_user(groups[i].group); | |
707 | ||
708 | kfree(groups); | |
709 | return ret; | |
710 | } | |
711 | ||
89e1f7d4 AW |
712 | return -ENOTTY; |
713 | } | |
714 | ||
5b279a11 AW |
715 | static ssize_t vfio_pci_rw(void *device_data, char __user *buf, |
716 | size_t count, loff_t *ppos, bool iswrite) | |
89e1f7d4 AW |
717 | { |
718 | unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); | |
719 | struct vfio_pci_device *vdev = device_data; | |
89e1f7d4 AW |
720 | |
721 | if (index >= VFIO_PCI_NUM_REGIONS) | |
722 | return -EINVAL; | |
723 | ||
5b279a11 AW |
724 | switch (index) { |
725 | case VFIO_PCI_CONFIG_REGION_INDEX: | |
906ee99d AW |
726 | return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite); |
727 | ||
5b279a11 AW |
728 | case VFIO_PCI_ROM_REGION_INDEX: |
729 | if (iswrite) | |
730 | return -EINVAL; | |
906ee99d | 731 | return vfio_pci_bar_rw(vdev, buf, count, ppos, false); |
89e1f7d4 | 732 | |
5b279a11 | 733 | case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: |
906ee99d | 734 | return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite); |
84237a82 AW |
735 | |
736 | case VFIO_PCI_VGA_REGION_INDEX: | |
737 | return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite); | |
5b279a11 AW |
738 | } |
739 | ||
89e1f7d4 AW |
740 | return -EINVAL; |
741 | } | |
742 | ||
5b279a11 AW |
743 | static ssize_t vfio_pci_read(void *device_data, char __user *buf, |
744 | size_t count, loff_t *ppos) | |
745 | { | |
906ee99d AW |
746 | if (!count) |
747 | return 0; | |
748 | ||
5b279a11 AW |
749 | return vfio_pci_rw(device_data, buf, count, ppos, false); |
750 | } | |
751 | ||
89e1f7d4 AW |
752 | static ssize_t vfio_pci_write(void *device_data, const char __user *buf, |
753 | size_t count, loff_t *ppos) | |
754 | { | |
906ee99d AW |
755 | if (!count) |
756 | return 0; | |
757 | ||
758 | return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true); | |
89e1f7d4 AW |
759 | } |
760 | ||
761 | static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma) | |
762 | { | |
763 | struct vfio_pci_device *vdev = device_data; | |
764 | struct pci_dev *pdev = vdev->pdev; | |
765 | unsigned int index; | |
34002f54 | 766 | u64 phys_len, req_len, pgoff, req_start; |
89e1f7d4 AW |
767 | int ret; |
768 | ||
769 | index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); | |
770 | ||
771 | if (vma->vm_end < vma->vm_start) | |
772 | return -EINVAL; | |
773 | if ((vma->vm_flags & VM_SHARED) == 0) | |
774 | return -EINVAL; | |
775 | if (index >= VFIO_PCI_ROM_REGION_INDEX) | |
776 | return -EINVAL; | |
777 | if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM)) | |
778 | return -EINVAL; | |
779 | ||
780 | phys_len = pci_resource_len(pdev, index); | |
781 | req_len = vma->vm_end - vma->vm_start; | |
782 | pgoff = vma->vm_pgoff & | |
783 | ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); | |
784 | req_start = pgoff << PAGE_SHIFT; | |
785 | ||
786 | if (phys_len < PAGE_SIZE || req_start + req_len > phys_len) | |
787 | return -EINVAL; | |
788 | ||
789 | if (index == vdev->msix_bar) { | |
790 | /* | |
791 | * Disallow mmaps overlapping the MSI-X table; users don't | |
792 | * get to touch this directly. We could find somewhere | |
793 | * else to map the overlap, but page granularity is only | |
794 | * a recommendation, not a requirement, so the user needs | |
795 | * to know which bits are real. Requiring them to mmap | |
796 | * around the table makes that clear. | |
797 | */ | |
798 | ||
799 | /* If neither entirely above nor below, then it overlaps */ | |
800 | if (!(req_start >= vdev->msix_offset + vdev->msix_size || | |
801 | req_start + req_len <= vdev->msix_offset)) | |
802 | return -EINVAL; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Even though we don't make use of the barmap for the mmap, | |
807 | * we need to request the region and the barmap tracks that. | |
808 | */ | |
809 | if (!vdev->barmap[index]) { | |
810 | ret = pci_request_selected_regions(pdev, | |
811 | 1 << index, "vfio-pci"); | |
812 | if (ret) | |
813 | return ret; | |
814 | ||
815 | vdev->barmap[index] = pci_iomap(pdev, index, 0); | |
816 | } | |
817 | ||
818 | vma->vm_private_data = vdev; | |
89e1f7d4 | 819 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); |
34002f54 | 820 | vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff; |
89e1f7d4 | 821 | |
34002f54 | 822 | return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff, |
89e1f7d4 AW |
823 | req_len, vma->vm_page_prot); |
824 | } | |
825 | ||
826 | static const struct vfio_device_ops vfio_pci_ops = { | |
827 | .name = "vfio-pci", | |
828 | .open = vfio_pci_open, | |
829 | .release = vfio_pci_release, | |
830 | .ioctl = vfio_pci_ioctl, | |
831 | .read = vfio_pci_read, | |
832 | .write = vfio_pci_write, | |
833 | .mmap = vfio_pci_mmap, | |
834 | }; | |
835 | ||
836 | static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) | |
837 | { | |
838 | u8 type; | |
839 | struct vfio_pci_device *vdev; | |
840 | struct iommu_group *group; | |
841 | int ret; | |
842 | ||
843 | pci_read_config_byte(pdev, PCI_HEADER_TYPE, &type); | |
844 | if ((type & PCI_HEADER_TYPE) != PCI_HEADER_TYPE_NORMAL) | |
845 | return -EINVAL; | |
846 | ||
847 | group = iommu_group_get(&pdev->dev); | |
848 | if (!group) | |
849 | return -EINVAL; | |
850 | ||
851 | vdev = kzalloc(sizeof(*vdev), GFP_KERNEL); | |
852 | if (!vdev) { | |
853 | iommu_group_put(group); | |
854 | return -ENOMEM; | |
855 | } | |
856 | ||
857 | vdev->pdev = pdev; | |
858 | vdev->irq_type = VFIO_PCI_NUM_IRQS; | |
859 | mutex_init(&vdev->igate); | |
860 | spin_lock_init(&vdev->irqlock); | |
89e1f7d4 AW |
861 | |
862 | ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev); | |
863 | if (ret) { | |
864 | iommu_group_put(group); | |
865 | kfree(vdev); | |
866 | } | |
867 | ||
868 | return ret; | |
869 | } | |
870 | ||
871 | static void vfio_pci_remove(struct pci_dev *pdev) | |
872 | { | |
873 | struct vfio_pci_device *vdev; | |
874 | ||
61d79256 AW |
875 | mutex_lock(&driver_lock); |
876 | ||
89e1f7d4 | 877 | vdev = vfio_del_group_dev(&pdev->dev); |
61d79256 AW |
878 | if (vdev) { |
879 | iommu_group_put(pdev->dev.iommu_group); | |
880 | kfree(vdev); | |
881 | } | |
89e1f7d4 | 882 | |
61d79256 | 883 | mutex_unlock(&driver_lock); |
89e1f7d4 AW |
884 | } |
885 | ||
dad9f897 VMP |
886 | static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev, |
887 | pci_channel_state_t state) | |
888 | { | |
889 | struct vfio_pci_device *vdev; | |
890 | struct vfio_device *device; | |
891 | ||
892 | device = vfio_device_get_from_dev(&pdev->dev); | |
893 | if (device == NULL) | |
894 | return PCI_ERS_RESULT_DISCONNECT; | |
895 | ||
896 | vdev = vfio_device_data(device); | |
897 | if (vdev == NULL) { | |
898 | vfio_device_put(device); | |
899 | return PCI_ERS_RESULT_DISCONNECT; | |
900 | } | |
901 | ||
3be3a074 AW |
902 | mutex_lock(&vdev->igate); |
903 | ||
dad9f897 VMP |
904 | if (vdev->err_trigger) |
905 | eventfd_signal(vdev->err_trigger, 1); | |
906 | ||
3be3a074 AW |
907 | mutex_unlock(&vdev->igate); |
908 | ||
dad9f897 VMP |
909 | vfio_device_put(device); |
910 | ||
911 | return PCI_ERS_RESULT_CAN_RECOVER; | |
912 | } | |
913 | ||
914 | static struct pci_error_handlers vfio_err_handlers = { | |
915 | .error_detected = vfio_pci_aer_err_detected, | |
916 | }; | |
917 | ||
89e1f7d4 AW |
918 | static struct pci_driver vfio_pci_driver = { |
919 | .name = "vfio-pci", | |
920 | .id_table = NULL, /* only dynamic ids */ | |
921 | .probe = vfio_pci_probe, | |
922 | .remove = vfio_pci_remove, | |
dad9f897 | 923 | .err_handler = &vfio_err_handlers, |
89e1f7d4 AW |
924 | }; |
925 | ||
926 | static void __exit vfio_pci_cleanup(void) | |
927 | { | |
928 | pci_unregister_driver(&vfio_pci_driver); | |
929 | vfio_pci_virqfd_exit(); | |
930 | vfio_pci_uninit_perm_bits(); | |
931 | } | |
932 | ||
933 | static int __init vfio_pci_init(void) | |
934 | { | |
935 | int ret; | |
936 | ||
937 | /* Allocate shared config space permision data used by all devices */ | |
938 | ret = vfio_pci_init_perm_bits(); | |
939 | if (ret) | |
940 | return ret; | |
941 | ||
942 | /* Start the virqfd cleanup handler */ | |
943 | ret = vfio_pci_virqfd_init(); | |
944 | if (ret) | |
945 | goto out_virqfd; | |
946 | ||
947 | /* Register and scan for devices */ | |
948 | ret = pci_register_driver(&vfio_pci_driver); | |
949 | if (ret) | |
950 | goto out_driver; | |
951 | ||
952 | return 0; | |
953 | ||
89e1f7d4 | 954 | out_driver: |
05bf3aac JL |
955 | vfio_pci_virqfd_exit(); |
956 | out_virqfd: | |
89e1f7d4 AW |
957 | vfio_pci_uninit_perm_bits(); |
958 | return ret; | |
959 | } | |
960 | ||
961 | module_init(vfio_pci_init); | |
962 | module_exit(vfio_pci_cleanup); | |
963 | ||
964 | MODULE_VERSION(DRIVER_VERSION); | |
965 | MODULE_LICENSE("GPL v2"); | |
966 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
967 | MODULE_DESCRIPTION(DRIVER_DESC); |