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