1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * VDPA device simulator core. |
4 | * |
5 | * Copyright (c) 2020, Red Hat Inc. All rights reserved. |
6 | * Author: Jason Wang <jasowang@redhat.com> |
7 | * |
8 | */ |
9 | |
10 | #include <linux/init.h> |
11 | #include <linux/module.h> |
12 | #include <linux/device.h> |
13 | #include <linux/kernel.h> |
14 | #include <linux/kthread.h> |
15 | #include <linux/slab.h> |
16 | #include <linux/dma-map-ops.h> |
17 | #include <linux/vringh.h> |
18 | #include <linux/vdpa.h> |
19 | #include <linux/vhost_iotlb.h> |
20 | #include <uapi/linux/vdpa.h> |
21 | #include <uapi/linux/vhost_types.h> |
22 | |
23 | #include "vdpa_sim.h" |
24 | |
25 | #define DRV_VERSION "0.1" |
26 | #define DRV_AUTHOR "Jason Wang <jasowang@redhat.com>" |
27 | #define DRV_DESC "vDPA Device Simulator core" |
28 | #define DRV_LICENSE "GPL v2" |
29 | |
30 | static int batch_mapping = 1; |
31 | module_param(batch_mapping, int, 0444); |
32 | MODULE_PARM_DESC(batch_mapping, "Batched mapping 1 -Enable; 0 - Disable" ); |
33 | |
34 | static int max_iotlb_entries = 2048; |
35 | module_param(max_iotlb_entries, int, 0444); |
36 | MODULE_PARM_DESC(max_iotlb_entries, |
37 | "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)" ); |
38 | |
39 | static bool use_va = true; |
40 | module_param(use_va, bool, 0444); |
41 | MODULE_PARM_DESC(use_va, "Enable/disable the device's ability to use VA" ); |
42 | |
43 | #define VDPASIM_QUEUE_ALIGN PAGE_SIZE |
44 | #define VDPASIM_QUEUE_MAX 256 |
45 | #define VDPASIM_VENDOR_ID 0 |
46 | |
47 | struct vdpasim_mm_work { |
48 | struct kthread_work work; |
49 | struct vdpasim *vdpasim; |
50 | struct mm_struct *mm_to_bind; |
51 | int ret; |
52 | }; |
53 | |
54 | static void vdpasim_mm_work_fn(struct kthread_work *work) |
55 | { |
56 | struct vdpasim_mm_work *mm_work = |
57 | container_of(work, struct vdpasim_mm_work, work); |
58 | struct vdpasim *vdpasim = mm_work->vdpasim; |
59 | |
60 | mm_work->ret = 0; |
61 | |
62 | //TODO: should we attach the cgroup of the mm owner? |
63 | vdpasim->mm_bound = mm_work->mm_to_bind; |
64 | } |
65 | |
66 | static void vdpasim_worker_change_mm_sync(struct vdpasim *vdpasim, |
67 | struct vdpasim_mm_work *mm_work) |
68 | { |
69 | struct kthread_work *work = &mm_work->work; |
70 | |
71 | kthread_init_work(work, vdpasim_mm_work_fn); |
72 | kthread_queue_work(worker: vdpasim->worker, work); |
73 | |
74 | kthread_flush_work(work); |
75 | } |
76 | |
77 | static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa) |
78 | { |
79 | return container_of(vdpa, struct vdpasim, vdpa); |
80 | } |
81 | |
82 | static void vdpasim_vq_notify(struct vringh *vring) |
83 | { |
84 | struct vdpasim_virtqueue *vq = |
85 | container_of(vring, struct vdpasim_virtqueue, vring); |
86 | |
87 | if (!vq->cb) |
88 | return; |
89 | |
90 | vq->cb(vq->private); |
91 | } |
92 | |
93 | static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx) |
94 | { |
95 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
96 | uint16_t last_avail_idx = vq->vring.last_avail_idx; |
97 | struct vring_desc *desc = (struct vring_desc *) |
98 | (uintptr_t)vq->desc_addr; |
99 | struct vring_avail *avail = (struct vring_avail *) |
100 | (uintptr_t)vq->driver_addr; |
101 | struct vring_used *used = (struct vring_used *) |
102 | (uintptr_t)vq->device_addr; |
103 | |
104 | if (use_va && vdpasim->mm_bound) { |
105 | vringh_init_iotlb_va(vrh: &vq->vring, features: vdpasim->features, num: vq->num, |
106 | weak_barriers: true, desc, avail, used); |
107 | } else { |
108 | vringh_init_iotlb(vrh: &vq->vring, features: vdpasim->features, num: vq->num, |
109 | weak_barriers: true, desc, avail, used); |
110 | } |
111 | |
112 | vq->vring.last_avail_idx = last_avail_idx; |
113 | |
114 | /* |
115 | * Since vdpa_sim does not support receive inflight descriptors as a |
116 | * destination of a migration, let's set both avail_idx and used_idx |
117 | * the same at vq start. This is how vhost-user works in a |
118 | * VHOST_SET_VRING_BASE call. |
119 | * |
120 | * Although the simple fix is to set last_used_idx at |
121 | * vdpasim_set_vq_state, it would be reset at vdpasim_queue_ready. |
122 | */ |
123 | vq->vring.last_used_idx = last_avail_idx; |
124 | vq->vring.notify = vdpasim_vq_notify; |
125 | } |
126 | |
127 | static void vdpasim_vq_reset(struct vdpasim *vdpasim, |
128 | struct vdpasim_virtqueue *vq) |
129 | { |
130 | vq->ready = false; |
131 | vq->desc_addr = 0; |
132 | vq->driver_addr = 0; |
133 | vq->device_addr = 0; |
134 | vq->cb = NULL; |
135 | vq->private = NULL; |
136 | vringh_init_iotlb(vrh: &vq->vring, features: vdpasim->dev_attr.supported_features, |
137 | VDPASIM_QUEUE_MAX, weak_barriers: false, NULL, NULL, NULL); |
138 | |
139 | vq->vring.notify = NULL; |
140 | } |
141 | |
142 | static void vdpasim_do_reset(struct vdpasim *vdpasim, u32 flags) |
143 | { |
144 | int i; |
145 | |
146 | spin_lock(lock: &vdpasim->iommu_lock); |
147 | |
148 | for (i = 0; i < vdpasim->dev_attr.nvqs; i++) { |
149 | vdpasim_vq_reset(vdpasim, vq: &vdpasim->vqs[i]); |
150 | vringh_set_iotlb(vrh: &vdpasim->vqs[i].vring, iotlb: &vdpasim->iommu[0], |
151 | iotlb_lock: &vdpasim->iommu_lock); |
152 | } |
153 | |
154 | if (flags & VDPA_RESET_F_CLEAN_MAP) { |
155 | for (i = 0; i < vdpasim->dev_attr.nas; i++) { |
156 | vhost_iotlb_reset(iotlb: &vdpasim->iommu[i]); |
157 | vhost_iotlb_add_range(iotlb: &vdpasim->iommu[i], start: 0, ULONG_MAX, |
158 | addr: 0, VHOST_MAP_RW); |
159 | vdpasim->iommu_pt[i] = true; |
160 | } |
161 | } |
162 | |
163 | vdpasim->running = false; |
164 | spin_unlock(lock: &vdpasim->iommu_lock); |
165 | |
166 | vdpasim->features = 0; |
167 | vdpasim->status = 0; |
168 | ++vdpasim->generation; |
169 | } |
170 | |
171 | static const struct vdpa_config_ops vdpasim_config_ops; |
172 | static const struct vdpa_config_ops vdpasim_batch_config_ops; |
173 | |
174 | static void vdpasim_work_fn(struct kthread_work *work) |
175 | { |
176 | struct vdpasim *vdpasim = container_of(work, struct vdpasim, work); |
177 | struct mm_struct *mm = vdpasim->mm_bound; |
178 | |
179 | if (use_va && mm) { |
180 | if (!mmget_not_zero(mm)) |
181 | return; |
182 | kthread_use_mm(mm); |
183 | } |
184 | |
185 | vdpasim->dev_attr.work_fn(vdpasim); |
186 | |
187 | if (use_va && mm) { |
188 | kthread_unuse_mm(mm); |
189 | mmput(mm); |
190 | } |
191 | } |
192 | |
193 | struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr, |
194 | const struct vdpa_dev_set_config *config) |
195 | { |
196 | const struct vdpa_config_ops *ops; |
197 | struct vdpa_device *vdpa; |
198 | struct vdpasim *vdpasim; |
199 | struct device *dev; |
200 | int i, ret = -ENOMEM; |
201 | |
202 | if (!dev_attr->alloc_size) |
203 | return ERR_PTR(error: -EINVAL); |
204 | |
205 | if (config->mask & BIT_ULL(VDPA_ATTR_DEV_FEATURES)) { |
206 | if (config->device_features & |
207 | ~dev_attr->supported_features) |
208 | return ERR_PTR(error: -EINVAL); |
209 | dev_attr->supported_features = |
210 | config->device_features; |
211 | } |
212 | |
213 | if (batch_mapping) |
214 | ops = &vdpasim_batch_config_ops; |
215 | else |
216 | ops = &vdpasim_config_ops; |
217 | |
218 | vdpa = __vdpa_alloc_device(NULL, config: ops, |
219 | ngroups: dev_attr->ngroups, nas: dev_attr->nas, |
220 | size: dev_attr->alloc_size, |
221 | name: dev_attr->name, use_va); |
222 | if (IS_ERR(ptr: vdpa)) { |
223 | ret = PTR_ERR(ptr: vdpa); |
224 | goto err_alloc; |
225 | } |
226 | |
227 | vdpasim = vdpa_to_sim(vdpa); |
228 | vdpasim->dev_attr = *dev_attr; |
229 | dev = &vdpasim->vdpa.dev; |
230 | |
231 | kthread_init_work(&vdpasim->work, vdpasim_work_fn); |
232 | vdpasim->worker = kthread_create_worker(flags: 0, namefmt: "vDPA sim worker: %s" , |
233 | dev_attr->name); |
234 | if (IS_ERR(ptr: vdpasim->worker)) |
235 | goto err_iommu; |
236 | |
237 | mutex_init(&vdpasim->mutex); |
238 | spin_lock_init(&vdpasim->iommu_lock); |
239 | |
240 | dev->dma_mask = &dev->coherent_dma_mask; |
241 | if (dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64))) |
242 | goto err_iommu; |
243 | vdpasim->vdpa.mdev = dev_attr->mgmt_dev; |
244 | |
245 | vdpasim->config = kzalloc(size: dev_attr->config_size, GFP_KERNEL); |
246 | if (!vdpasim->config) |
247 | goto err_iommu; |
248 | |
249 | vdpasim->vqs = kcalloc(n: dev_attr->nvqs, size: sizeof(struct vdpasim_virtqueue), |
250 | GFP_KERNEL); |
251 | if (!vdpasim->vqs) |
252 | goto err_iommu; |
253 | |
254 | vdpasim->iommu = kmalloc_array(n: vdpasim->dev_attr.nas, |
255 | size: sizeof(*vdpasim->iommu), GFP_KERNEL); |
256 | if (!vdpasim->iommu) |
257 | goto err_iommu; |
258 | |
259 | vdpasim->iommu_pt = kmalloc_array(n: vdpasim->dev_attr.nas, |
260 | size: sizeof(*vdpasim->iommu_pt), GFP_KERNEL); |
261 | if (!vdpasim->iommu_pt) |
262 | goto err_iommu; |
263 | |
264 | for (i = 0; i < vdpasim->dev_attr.nas; i++) { |
265 | vhost_iotlb_init(iotlb: &vdpasim->iommu[i], limit: max_iotlb_entries, flags: 0); |
266 | vhost_iotlb_add_range(iotlb: &vdpasim->iommu[i], start: 0, ULONG_MAX, addr: 0, |
267 | VHOST_MAP_RW); |
268 | vdpasim->iommu_pt[i] = true; |
269 | } |
270 | |
271 | for (i = 0; i < dev_attr->nvqs; i++) |
272 | vringh_set_iotlb(vrh: &vdpasim->vqs[i].vring, iotlb: &vdpasim->iommu[0], |
273 | iotlb_lock: &vdpasim->iommu_lock); |
274 | |
275 | vdpasim->vdpa.dma_dev = dev; |
276 | |
277 | return vdpasim; |
278 | |
279 | err_iommu: |
280 | put_device(dev); |
281 | err_alloc: |
282 | return ERR_PTR(error: ret); |
283 | } |
284 | EXPORT_SYMBOL_GPL(vdpasim_create); |
285 | |
286 | void vdpasim_schedule_work(struct vdpasim *vdpasim) |
287 | { |
288 | kthread_queue_work(worker: vdpasim->worker, work: &vdpasim->work); |
289 | } |
290 | EXPORT_SYMBOL_GPL(vdpasim_schedule_work); |
291 | |
292 | static int vdpasim_set_vq_address(struct vdpa_device *vdpa, u16 idx, |
293 | u64 desc_area, u64 driver_area, |
294 | u64 device_area) |
295 | { |
296 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
297 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
298 | |
299 | vq->desc_addr = desc_area; |
300 | vq->driver_addr = driver_area; |
301 | vq->device_addr = device_area; |
302 | |
303 | return 0; |
304 | } |
305 | |
306 | static void vdpasim_set_vq_num(struct vdpa_device *vdpa, u16 idx, u32 num) |
307 | { |
308 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
309 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
310 | |
311 | vq->num = num; |
312 | } |
313 | |
314 | static u16 vdpasim_get_vq_size(struct vdpa_device *vdpa, u16 idx) |
315 | { |
316 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
317 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
318 | |
319 | if (vdpasim->status & VIRTIO_CONFIG_S_DRIVER_OK) |
320 | return vq->num; |
321 | else |
322 | return VDPASIM_QUEUE_MAX; |
323 | } |
324 | |
325 | static void vdpasim_kick_vq(struct vdpa_device *vdpa, u16 idx) |
326 | { |
327 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
328 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
329 | |
330 | if (!vdpasim->running && |
331 | (vdpasim->status & VIRTIO_CONFIG_S_DRIVER_OK)) { |
332 | vdpasim->pending_kick = true; |
333 | return; |
334 | } |
335 | |
336 | if (vq->ready) |
337 | vdpasim_schedule_work(vdpasim); |
338 | } |
339 | |
340 | static void vdpasim_set_vq_cb(struct vdpa_device *vdpa, u16 idx, |
341 | struct vdpa_callback *cb) |
342 | { |
343 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
344 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
345 | |
346 | vq->cb = cb->callback; |
347 | vq->private = cb->private; |
348 | } |
349 | |
350 | static void vdpasim_set_vq_ready(struct vdpa_device *vdpa, u16 idx, bool ready) |
351 | { |
352 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
353 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
354 | bool old_ready; |
355 | |
356 | mutex_lock(&vdpasim->mutex); |
357 | old_ready = vq->ready; |
358 | vq->ready = ready; |
359 | if (vq->ready && !old_ready) { |
360 | vdpasim_queue_ready(vdpasim, idx); |
361 | } |
362 | mutex_unlock(lock: &vdpasim->mutex); |
363 | } |
364 | |
365 | static bool vdpasim_get_vq_ready(struct vdpa_device *vdpa, u16 idx) |
366 | { |
367 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
368 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
369 | |
370 | return vq->ready; |
371 | } |
372 | |
373 | static int vdpasim_set_vq_state(struct vdpa_device *vdpa, u16 idx, |
374 | const struct vdpa_vq_state *state) |
375 | { |
376 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
377 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
378 | struct vringh *vrh = &vq->vring; |
379 | |
380 | mutex_lock(&vdpasim->mutex); |
381 | vrh->last_avail_idx = state->split.avail_index; |
382 | mutex_unlock(lock: &vdpasim->mutex); |
383 | |
384 | return 0; |
385 | } |
386 | |
387 | static int vdpasim_get_vq_state(struct vdpa_device *vdpa, u16 idx, |
388 | struct vdpa_vq_state *state) |
389 | { |
390 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
391 | struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx]; |
392 | struct vringh *vrh = &vq->vring; |
393 | |
394 | state->split.avail_index = vrh->last_avail_idx; |
395 | return 0; |
396 | } |
397 | |
398 | static int vdpasim_get_vq_stats(struct vdpa_device *vdpa, u16 idx, |
399 | struct sk_buff *msg, |
400 | struct netlink_ext_ack *extack) |
401 | { |
402 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
403 | |
404 | if (vdpasim->dev_attr.get_stats) |
405 | return vdpasim->dev_attr.get_stats(vdpasim, idx, |
406 | msg, extack); |
407 | return -EOPNOTSUPP; |
408 | } |
409 | |
410 | static u32 vdpasim_get_vq_align(struct vdpa_device *vdpa) |
411 | { |
412 | return VDPASIM_QUEUE_ALIGN; |
413 | } |
414 | |
415 | static u32 vdpasim_get_vq_group(struct vdpa_device *vdpa, u16 idx) |
416 | { |
417 | /* RX and TX belongs to group 0, CVQ belongs to group 1 */ |
418 | if (idx == 2) |
419 | return 1; |
420 | else |
421 | return 0; |
422 | } |
423 | |
424 | static u64 vdpasim_get_device_features(struct vdpa_device *vdpa) |
425 | { |
426 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
427 | |
428 | return vdpasim->dev_attr.supported_features; |
429 | } |
430 | |
431 | static u64 vdpasim_get_backend_features(const struct vdpa_device *vdpa) |
432 | { |
433 | return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK); |
434 | } |
435 | |
436 | static int vdpasim_set_driver_features(struct vdpa_device *vdpa, u64 features) |
437 | { |
438 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
439 | |
440 | /* DMA mapping must be done by driver */ |
441 | if (!(features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) |
442 | return -EINVAL; |
443 | |
444 | vdpasim->features = features & vdpasim->dev_attr.supported_features; |
445 | |
446 | return 0; |
447 | } |
448 | |
449 | static u64 vdpasim_get_driver_features(struct vdpa_device *vdpa) |
450 | { |
451 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
452 | |
453 | return vdpasim->features; |
454 | } |
455 | |
456 | static void vdpasim_set_config_cb(struct vdpa_device *vdpa, |
457 | struct vdpa_callback *cb) |
458 | { |
459 | /* We don't support config interrupt */ |
460 | } |
461 | |
462 | static u16 vdpasim_get_vq_num_max(struct vdpa_device *vdpa) |
463 | { |
464 | return VDPASIM_QUEUE_MAX; |
465 | } |
466 | |
467 | static u32 vdpasim_get_device_id(struct vdpa_device *vdpa) |
468 | { |
469 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
470 | |
471 | return vdpasim->dev_attr.id; |
472 | } |
473 | |
474 | static u32 vdpasim_get_vendor_id(struct vdpa_device *vdpa) |
475 | { |
476 | return VDPASIM_VENDOR_ID; |
477 | } |
478 | |
479 | static u8 vdpasim_get_status(struct vdpa_device *vdpa) |
480 | { |
481 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
482 | u8 status; |
483 | |
484 | mutex_lock(&vdpasim->mutex); |
485 | status = vdpasim->status; |
486 | mutex_unlock(lock: &vdpasim->mutex); |
487 | |
488 | return status; |
489 | } |
490 | |
491 | static void vdpasim_set_status(struct vdpa_device *vdpa, u8 status) |
492 | { |
493 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
494 | |
495 | mutex_lock(&vdpasim->mutex); |
496 | vdpasim->status = status; |
497 | vdpasim->running = (status & VIRTIO_CONFIG_S_DRIVER_OK) != 0; |
498 | mutex_unlock(lock: &vdpasim->mutex); |
499 | } |
500 | |
501 | static int vdpasim_compat_reset(struct vdpa_device *vdpa, u32 flags) |
502 | { |
503 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
504 | |
505 | mutex_lock(&vdpasim->mutex); |
506 | vdpasim->status = 0; |
507 | vdpasim_do_reset(vdpasim, flags); |
508 | mutex_unlock(lock: &vdpasim->mutex); |
509 | |
510 | return 0; |
511 | } |
512 | |
513 | static int vdpasim_reset(struct vdpa_device *vdpa) |
514 | { |
515 | return vdpasim_compat_reset(vdpa, flags: 0); |
516 | } |
517 | |
518 | static int vdpasim_suspend(struct vdpa_device *vdpa) |
519 | { |
520 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
521 | |
522 | mutex_lock(&vdpasim->mutex); |
523 | vdpasim->running = false; |
524 | mutex_unlock(lock: &vdpasim->mutex); |
525 | |
526 | return 0; |
527 | } |
528 | |
529 | static int vdpasim_resume(struct vdpa_device *vdpa) |
530 | { |
531 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
532 | int i; |
533 | |
534 | mutex_lock(&vdpasim->mutex); |
535 | vdpasim->running = true; |
536 | |
537 | if (vdpasim->pending_kick) { |
538 | /* Process pending descriptors */ |
539 | for (i = 0; i < vdpasim->dev_attr.nvqs; ++i) |
540 | vdpasim_kick_vq(vdpa, idx: i); |
541 | |
542 | vdpasim->pending_kick = false; |
543 | } |
544 | |
545 | mutex_unlock(lock: &vdpasim->mutex); |
546 | |
547 | return 0; |
548 | } |
549 | |
550 | static size_t vdpasim_get_config_size(struct vdpa_device *vdpa) |
551 | { |
552 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
553 | |
554 | return vdpasim->dev_attr.config_size; |
555 | } |
556 | |
557 | static void vdpasim_get_config(struct vdpa_device *vdpa, unsigned int offset, |
558 | void *buf, unsigned int len) |
559 | { |
560 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
561 | |
562 | if (offset + len > vdpasim->dev_attr.config_size) |
563 | return; |
564 | |
565 | if (vdpasim->dev_attr.get_config) |
566 | vdpasim->dev_attr.get_config(vdpasim, vdpasim->config); |
567 | |
568 | memcpy(buf, vdpasim->config + offset, len); |
569 | } |
570 | |
571 | static void vdpasim_set_config(struct vdpa_device *vdpa, unsigned int offset, |
572 | const void *buf, unsigned int len) |
573 | { |
574 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
575 | |
576 | if (offset + len > vdpasim->dev_attr.config_size) |
577 | return; |
578 | |
579 | memcpy(vdpasim->config + offset, buf, len); |
580 | |
581 | if (vdpasim->dev_attr.set_config) |
582 | vdpasim->dev_attr.set_config(vdpasim, vdpasim->config); |
583 | } |
584 | |
585 | static u32 vdpasim_get_generation(struct vdpa_device *vdpa) |
586 | { |
587 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
588 | |
589 | return vdpasim->generation; |
590 | } |
591 | |
592 | static struct vdpa_iova_range vdpasim_get_iova_range(struct vdpa_device *vdpa) |
593 | { |
594 | struct vdpa_iova_range range = { |
595 | .first = 0ULL, |
596 | .last = ULLONG_MAX, |
597 | }; |
598 | |
599 | return range; |
600 | } |
601 | |
602 | static int vdpasim_set_group_asid(struct vdpa_device *vdpa, unsigned int group, |
603 | unsigned int asid) |
604 | { |
605 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
606 | struct vhost_iotlb *iommu; |
607 | int i; |
608 | |
609 | if (group > vdpasim->dev_attr.ngroups) |
610 | return -EINVAL; |
611 | |
612 | if (asid >= vdpasim->dev_attr.nas) |
613 | return -EINVAL; |
614 | |
615 | iommu = &vdpasim->iommu[asid]; |
616 | |
617 | mutex_lock(&vdpasim->mutex); |
618 | |
619 | for (i = 0; i < vdpasim->dev_attr.nvqs; i++) |
620 | if (vdpasim_get_vq_group(vdpa, idx: i) == group) |
621 | vringh_set_iotlb(vrh: &vdpasim->vqs[i].vring, iotlb: iommu, |
622 | iotlb_lock: &vdpasim->iommu_lock); |
623 | |
624 | mutex_unlock(lock: &vdpasim->mutex); |
625 | |
626 | return 0; |
627 | } |
628 | |
629 | static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid, |
630 | struct vhost_iotlb *iotlb) |
631 | { |
632 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
633 | struct vhost_iotlb_map *map; |
634 | struct vhost_iotlb *iommu; |
635 | u64 start = 0ULL, last = 0ULL - 1; |
636 | int ret; |
637 | |
638 | if (asid >= vdpasim->dev_attr.nas) |
639 | return -EINVAL; |
640 | |
641 | spin_lock(lock: &vdpasim->iommu_lock); |
642 | |
643 | iommu = &vdpasim->iommu[asid]; |
644 | vhost_iotlb_reset(iotlb: iommu); |
645 | vdpasim->iommu_pt[asid] = false; |
646 | |
647 | for (map = vhost_iotlb_itree_first(iotlb, start, last); map; |
648 | map = vhost_iotlb_itree_next(map, start, last)) { |
649 | ret = vhost_iotlb_add_range(iotlb: iommu, start: map->start, |
650 | last: map->last, addr: map->addr, perm: map->perm); |
651 | if (ret) |
652 | goto err; |
653 | } |
654 | spin_unlock(lock: &vdpasim->iommu_lock); |
655 | return 0; |
656 | |
657 | err: |
658 | vhost_iotlb_reset(iotlb: iommu); |
659 | spin_unlock(lock: &vdpasim->iommu_lock); |
660 | return ret; |
661 | } |
662 | |
663 | static int vdpasim_reset_map(struct vdpa_device *vdpa, unsigned int asid) |
664 | { |
665 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
666 | |
667 | if (asid >= vdpasim->dev_attr.nas) |
668 | return -EINVAL; |
669 | |
670 | spin_lock(lock: &vdpasim->iommu_lock); |
671 | if (vdpasim->iommu_pt[asid]) |
672 | goto out; |
673 | vhost_iotlb_reset(iotlb: &vdpasim->iommu[asid]); |
674 | vhost_iotlb_add_range(iotlb: &vdpasim->iommu[asid], start: 0, ULONG_MAX, |
675 | addr: 0, VHOST_MAP_RW); |
676 | vdpasim->iommu_pt[asid] = true; |
677 | out: |
678 | spin_unlock(lock: &vdpasim->iommu_lock); |
679 | return 0; |
680 | } |
681 | |
682 | static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm) |
683 | { |
684 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
685 | struct vdpasim_mm_work mm_work; |
686 | |
687 | mm_work.vdpasim = vdpasim; |
688 | mm_work.mm_to_bind = mm; |
689 | |
690 | vdpasim_worker_change_mm_sync(vdpasim, mm_work: &mm_work); |
691 | |
692 | return mm_work.ret; |
693 | } |
694 | |
695 | static void vdpasim_unbind_mm(struct vdpa_device *vdpa) |
696 | { |
697 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
698 | struct vdpasim_mm_work mm_work; |
699 | |
700 | mm_work.vdpasim = vdpasim; |
701 | mm_work.mm_to_bind = NULL; |
702 | |
703 | vdpasim_worker_change_mm_sync(vdpasim, mm_work: &mm_work); |
704 | } |
705 | |
706 | static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid, |
707 | u64 iova, u64 size, |
708 | u64 pa, u32 perm, void *opaque) |
709 | { |
710 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
711 | int ret; |
712 | |
713 | if (asid >= vdpasim->dev_attr.nas) |
714 | return -EINVAL; |
715 | |
716 | spin_lock(lock: &vdpasim->iommu_lock); |
717 | if (vdpasim->iommu_pt[asid]) { |
718 | vhost_iotlb_reset(iotlb: &vdpasim->iommu[asid]); |
719 | vdpasim->iommu_pt[asid] = false; |
720 | } |
721 | ret = vhost_iotlb_add_range_ctx(iotlb: &vdpasim->iommu[asid], start: iova, |
722 | last: iova + size - 1, addr: pa, perm, opaque); |
723 | spin_unlock(lock: &vdpasim->iommu_lock); |
724 | |
725 | return ret; |
726 | } |
727 | |
728 | static int vdpasim_dma_unmap(struct vdpa_device *vdpa, unsigned int asid, |
729 | u64 iova, u64 size) |
730 | { |
731 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
732 | |
733 | if (asid >= vdpasim->dev_attr.nas) |
734 | return -EINVAL; |
735 | |
736 | if (vdpasim->iommu_pt[asid]) { |
737 | vhost_iotlb_reset(iotlb: &vdpasim->iommu[asid]); |
738 | vdpasim->iommu_pt[asid] = false; |
739 | } |
740 | |
741 | spin_lock(lock: &vdpasim->iommu_lock); |
742 | vhost_iotlb_del_range(iotlb: &vdpasim->iommu[asid], start: iova, last: iova + size - 1); |
743 | spin_unlock(lock: &vdpasim->iommu_lock); |
744 | |
745 | return 0; |
746 | } |
747 | |
748 | static void vdpasim_free(struct vdpa_device *vdpa) |
749 | { |
750 | struct vdpasim *vdpasim = vdpa_to_sim(vdpa); |
751 | int i; |
752 | |
753 | kthread_cancel_work_sync(work: &vdpasim->work); |
754 | kthread_destroy_worker(worker: vdpasim->worker); |
755 | |
756 | for (i = 0; i < vdpasim->dev_attr.nvqs; i++) { |
757 | vringh_kiov_cleanup(kiov: &vdpasim->vqs[i].out_iov); |
758 | vringh_kiov_cleanup(kiov: &vdpasim->vqs[i].in_iov); |
759 | } |
760 | |
761 | vdpasim->dev_attr.free(vdpasim); |
762 | |
763 | for (i = 0; i < vdpasim->dev_attr.nas; i++) |
764 | vhost_iotlb_reset(iotlb: &vdpasim->iommu[i]); |
765 | kfree(objp: vdpasim->iommu); |
766 | kfree(objp: vdpasim->iommu_pt); |
767 | kfree(objp: vdpasim->vqs); |
768 | kfree(objp: vdpasim->config); |
769 | } |
770 | |
771 | static const struct vdpa_config_ops vdpasim_config_ops = { |
772 | .set_vq_address = vdpasim_set_vq_address, |
773 | .set_vq_num = vdpasim_set_vq_num, |
774 | .kick_vq = vdpasim_kick_vq, |
775 | .set_vq_cb = vdpasim_set_vq_cb, |
776 | .set_vq_ready = vdpasim_set_vq_ready, |
777 | .get_vq_ready = vdpasim_get_vq_ready, |
778 | .set_vq_state = vdpasim_set_vq_state, |
779 | .get_vendor_vq_stats = vdpasim_get_vq_stats, |
780 | .get_vq_state = vdpasim_get_vq_state, |
781 | .get_vq_align = vdpasim_get_vq_align, |
782 | .get_vq_group = vdpasim_get_vq_group, |
783 | .get_device_features = vdpasim_get_device_features, |
784 | .get_backend_features = vdpasim_get_backend_features, |
785 | .set_driver_features = vdpasim_set_driver_features, |
786 | .get_driver_features = vdpasim_get_driver_features, |
787 | .set_config_cb = vdpasim_set_config_cb, |
788 | .get_vq_num_max = vdpasim_get_vq_num_max, |
789 | .get_vq_size = vdpasim_get_vq_size, |
790 | .get_device_id = vdpasim_get_device_id, |
791 | .get_vendor_id = vdpasim_get_vendor_id, |
792 | .get_status = vdpasim_get_status, |
793 | .set_status = vdpasim_set_status, |
794 | .reset = vdpasim_reset, |
795 | .compat_reset = vdpasim_compat_reset, |
796 | .suspend = vdpasim_suspend, |
797 | .resume = vdpasim_resume, |
798 | .get_config_size = vdpasim_get_config_size, |
799 | .get_config = vdpasim_get_config, |
800 | .set_config = vdpasim_set_config, |
801 | .get_generation = vdpasim_get_generation, |
802 | .get_iova_range = vdpasim_get_iova_range, |
803 | .set_group_asid = vdpasim_set_group_asid, |
804 | .dma_map = vdpasim_dma_map, |
805 | .dma_unmap = vdpasim_dma_unmap, |
806 | .reset_map = vdpasim_reset_map, |
807 | .bind_mm = vdpasim_bind_mm, |
808 | .unbind_mm = vdpasim_unbind_mm, |
809 | .free = vdpasim_free, |
810 | }; |
811 | |
812 | static const struct vdpa_config_ops vdpasim_batch_config_ops = { |
813 | .set_vq_address = vdpasim_set_vq_address, |
814 | .set_vq_num = vdpasim_set_vq_num, |
815 | .kick_vq = vdpasim_kick_vq, |
816 | .set_vq_cb = vdpasim_set_vq_cb, |
817 | .set_vq_ready = vdpasim_set_vq_ready, |
818 | .get_vq_ready = vdpasim_get_vq_ready, |
819 | .set_vq_state = vdpasim_set_vq_state, |
820 | .get_vendor_vq_stats = vdpasim_get_vq_stats, |
821 | .get_vq_state = vdpasim_get_vq_state, |
822 | .get_vq_align = vdpasim_get_vq_align, |
823 | .get_vq_group = vdpasim_get_vq_group, |
824 | .get_device_features = vdpasim_get_device_features, |
825 | .get_backend_features = vdpasim_get_backend_features, |
826 | .set_driver_features = vdpasim_set_driver_features, |
827 | .get_driver_features = vdpasim_get_driver_features, |
828 | .set_config_cb = vdpasim_set_config_cb, |
829 | .get_vq_num_max = vdpasim_get_vq_num_max, |
830 | .get_device_id = vdpasim_get_device_id, |
831 | .get_vendor_id = vdpasim_get_vendor_id, |
832 | .get_status = vdpasim_get_status, |
833 | .set_status = vdpasim_set_status, |
834 | .reset = vdpasim_reset, |
835 | .compat_reset = vdpasim_compat_reset, |
836 | .suspend = vdpasim_suspend, |
837 | .resume = vdpasim_resume, |
838 | .get_config_size = vdpasim_get_config_size, |
839 | .get_config = vdpasim_get_config, |
840 | .set_config = vdpasim_set_config, |
841 | .get_generation = vdpasim_get_generation, |
842 | .get_iova_range = vdpasim_get_iova_range, |
843 | .set_group_asid = vdpasim_set_group_asid, |
844 | .set_map = vdpasim_set_map, |
845 | .reset_map = vdpasim_reset_map, |
846 | .bind_mm = vdpasim_bind_mm, |
847 | .unbind_mm = vdpasim_unbind_mm, |
848 | .free = vdpasim_free, |
849 | }; |
850 | |
851 | MODULE_VERSION(DRV_VERSION); |
852 | MODULE_LICENSE(DRV_LICENSE); |
853 | MODULE_AUTHOR(DRV_AUTHOR); |
854 | MODULE_DESCRIPTION(DRV_DESC); |
855 | |