?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.1.15
[linux/fpc-iii.git] / drivers / misc / mic / vop / vop_vringh.c
blob3632fce405908f70270e20c7ae657fcff0d61018
1 /*
2 * Intel MIC Platform Software Stack (MPSS)
3 *
4 * Copyright(c) 2016 Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2, as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
14 *
15 * The full GNU General Public License is included in this distribution in
16 * the file called "COPYING".
17 *
18 * Intel Virtio Over PCIe (VOP) driver.
19 *
20 */
21 #include <linux/sched.h>
22 #include <linux/poll.h>
23 #include <linux/dma-mapping.h>
24
25 #include <linux/mic_common.h>
26 #include "../common/mic_dev.h"
27
28 #include <linux/mic_ioctl.h>
29 #include "vop_main.h"
30
31 /* Helper API to obtain the VOP PCIe device */
32 static inline struct device *vop_dev(struct vop_vdev *vdev)
33 {
34 return vdev->vpdev->dev.parent;
35 }
36
37 /* Helper API to check if a virtio device is initialized */
38 static inline int vop_vdev_inited(struct vop_vdev *vdev)
39 {
40 if (!vdev)
41 return -EINVAL;
42 /* Device has not been created yet */
43 if (!vdev->dd || !vdev->dd->type) {
44 dev_err(vop_dev(vdev), "%s %d err %d\n",
45 __func__, __LINE__, -EINVAL);
46 return -EINVAL;
47 }
48 /* Device has been removed/deleted */
49 if (vdev->dd->type == -1) {
50 dev_dbg(vop_dev(vdev), "%s %d err %d\n",
51 __func__, __LINE__, -ENODEV);
52 return -ENODEV;
53 }
54 return 0;
55 }
56
57 static void _vop_notify(struct vringh *vrh)
58 {
59 struct vop_vringh *vvrh = container_of(vrh, struct vop_vringh, vrh);
60 struct vop_vdev *vdev = vvrh->vdev;
61 struct vop_device *vpdev = vdev->vpdev;
62 s8 db = vdev->dc->h2c_vdev_db;
63
64 if (db != -1)
65 vpdev->hw_ops->send_intr(vpdev, db);
66 }
67
68 static void vop_virtio_init_post(struct vop_vdev *vdev)
69 {
70 struct mic_vqconfig *vqconfig = mic_vq_config(vdev->dd);
71 struct vop_device *vpdev = vdev->vpdev;
72 int i, used_size;
73
74 for (i = 0; i < vdev->dd->num_vq; i++) {
75 used_size = PAGE_ALIGN(sizeof(u16) * 3 +
76 sizeof(struct vring_used_elem) *
77 le16_to_cpu(vqconfig->num));
78 if (!le64_to_cpu(vqconfig[i].used_address)) {
79 dev_warn(vop_dev(vdev), "used_address zero??\n");
80 continue;
81 }
82 vdev->vvr[i].vrh.vring.used =
83 (void __force *)vpdev->hw_ops->remap(
84 vpdev,
85 le64_to_cpu(vqconfig[i].used_address),
86 used_size);
87 }
88
89 vdev->dc->used_address_updated = 0;
90
91 dev_info(vop_dev(vdev), "%s: device type %d LINKUP\n",
92 __func__, vdev->virtio_id);
93 }
94
95 static inline void vop_virtio_device_reset(struct vop_vdev *vdev)
96 {
97 int i;
98
99 dev_dbg(vop_dev(vdev), "%s: status %d device type %d RESET\n",
100 __func__, vdev->dd->status, vdev->virtio_id);
101
102 for (i = 0; i < vdev->dd->num_vq; i++)
103 /*
104 * Avoid lockdep false positive. The + 1 is for the vop
105 * mutex which is held in the reset devices code path.
106 */
107 mutex_lock_nested(&vdev->vvr[i].vr_mutex, i + 1);
108
109 /* 0 status means "reset" */
110 vdev->dd->status = 0;
111 vdev->dc->vdev_reset = 0;
112 vdev->dc->host_ack = 1;
113
114 for (i = 0; i < vdev->dd->num_vq; i++) {
115 struct vringh *vrh = &vdev->vvr[i].vrh;
116
117 vdev->vvr[i].vring.info->avail_idx = 0;
118 vrh->completed = 0;
119 vrh->last_avail_idx = 0;
120 vrh->last_used_idx = 0;
121 }
122
123 for (i = 0; i < vdev->dd->num_vq; i++)
124 mutex_unlock(&vdev->vvr[i].vr_mutex);
125 }
126
127 static void vop_virtio_reset_devices(struct vop_info *vi)
128 {
129 struct list_head *pos, *tmp;
130 struct vop_vdev *vdev;
131
132 list_for_each_safe(pos, tmp, &vi->vdev_list) {
133 vdev = list_entry(pos, struct vop_vdev, list);
134 vop_virtio_device_reset(vdev);
135 vdev->poll_wake = 1;
136 wake_up(&vdev->waitq);
137 }
138 }
139
140 static void vop_bh_handler(struct work_struct *work)
141 {
142 struct vop_vdev *vdev = container_of(work, struct vop_vdev,
143 virtio_bh_work);
144
145 if (vdev->dc->used_address_updated)
146 vop_virtio_init_post(vdev);
147
148 if (vdev->dc->vdev_reset)
149 vop_virtio_device_reset(vdev);
150
151 vdev->poll_wake = 1;
152 wake_up(&vdev->waitq);
153 }
154
155 static irqreturn_t _vop_virtio_intr_handler(int irq, void *data)
156 {
157 struct vop_vdev *vdev = data;
158 struct vop_device *vpdev = vdev->vpdev;
159
160 vpdev->hw_ops->ack_interrupt(vpdev, vdev->virtio_db);
161 schedule_work(&vdev->virtio_bh_work);
162 return IRQ_HANDLED;
163 }
164
165 static int vop_virtio_config_change(struct vop_vdev *vdev, void *argp)
166 {
167 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wake);
168 int ret = 0, retry, i;
169 struct vop_device *vpdev = vdev->vpdev;
170 struct vop_info *vi = dev_get_drvdata(&vpdev->dev);
171 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
172 s8 db = bootparam->h2c_config_db;
173
174 mutex_lock(&vi->vop_mutex);
175 for (i = 0; i < vdev->dd->num_vq; i++)
176 mutex_lock_nested(&vdev->vvr[i].vr_mutex, i + 1);
177
178 if (db == -1 || vdev->dd->type == -1) {
179 ret = -EIO;
180 goto exit;
181 }
182
183 memcpy(mic_vq_configspace(vdev->dd), argp, vdev->dd->config_len);
184 vdev->dc->config_change = MIC_VIRTIO_PARAM_CONFIG_CHANGED;
185 vpdev->hw_ops->send_intr(vpdev, db);
186
187 for (retry = 100; retry--;) {
188 ret = wait_event_timeout(wake, vdev->dc->guest_ack,
189 msecs_to_jiffies(100));
190 if (ret)
191 break;
192 }
193
194 dev_dbg(vop_dev(vdev),
195 "%s %d retry: %d\n", __func__, __LINE__, retry);
196 vdev->dc->config_change = 0;
197 vdev->dc->guest_ack = 0;
198 exit:
199 for (i = 0; i < vdev->dd->num_vq; i++)
200 mutex_unlock(&vdev->vvr[i].vr_mutex);
201 mutex_unlock(&vi->vop_mutex);
202 return ret;
203 }
204
205 static int vop_copy_dp_entry(struct vop_vdev *vdev,
206 struct mic_device_desc *argp, __u8 *type,
207 struct mic_device_desc **devpage)
208 {
209 struct vop_device *vpdev = vdev->vpdev;
210 struct mic_device_desc *devp;
211 struct mic_vqconfig *vqconfig;
212 int ret = 0, i;
213 bool slot_found = false;
214
215 vqconfig = mic_vq_config(argp);
216 for (i = 0; i < argp->num_vq; i++) {
217 if (le16_to_cpu(vqconfig[i].num) > MIC_MAX_VRING_ENTRIES) {
218 ret = -EINVAL;
219 dev_err(vop_dev(vdev), "%s %d err %d\n",
220 __func__, __LINE__, ret);
221 goto exit;
222 }
223 }
224
225 /* Find the first free device page entry */
226 for (i = sizeof(struct mic_bootparam);
227 i < MIC_DP_SIZE - mic_total_desc_size(argp);
228 i += mic_total_desc_size(devp)) {
229 devp = vpdev->hw_ops->get_dp(vpdev) + i;
230 if (devp->type == 0 || devp->type == -1) {
231 slot_found = true;
232 break;
233 }
234 }
235 if (!slot_found) {
236 ret = -EINVAL;
237 dev_err(vop_dev(vdev), "%s %d err %d\n",
238 __func__, __LINE__, ret);
239 goto exit;
240 }
241 /*
242 * Save off the type before doing the memcpy. Type will be set in the
243 * end after completing all initialization for the new device.
244 */
245 *type = argp->type;
246 argp->type = 0;
247 memcpy(devp, argp, mic_desc_size(argp));
248
249 *devpage = devp;
250 exit:
251 return ret;
252 }
253
254 static void vop_init_device_ctrl(struct vop_vdev *vdev,
255 struct mic_device_desc *devpage)
256 {
257 struct mic_device_ctrl *dc;
258
259 dc = (void *)devpage + mic_aligned_desc_size(devpage);
260
261 dc->config_change = 0;
262 dc->guest_ack = 0;
263 dc->vdev_reset = 0;
264 dc->host_ack = 0;
265 dc->used_address_updated = 0;
266 dc->c2h_vdev_db = -1;
267 dc->h2c_vdev_db = -1;
268 vdev->dc = dc;
269 }
270
271 static int vop_virtio_add_device(struct vop_vdev *vdev,
272 struct mic_device_desc *argp)
273 {
274 struct vop_info *vi = vdev->vi;
275 struct vop_device *vpdev = vi->vpdev;
276 struct mic_device_desc *dd = NULL;
277 struct mic_vqconfig *vqconfig;
278 int vr_size, i, j, ret;
279 u8 type = 0;
280 s8 db = -1;
281 char irqname[16];
282 struct mic_bootparam *bootparam;
283 u16 num;
284 dma_addr_t vr_addr;
285
286 bootparam = vpdev->hw_ops->get_dp(vpdev);
287 init_waitqueue_head(&vdev->waitq);
288 INIT_LIST_HEAD(&vdev->list);
289 vdev->vpdev = vpdev;
290
291 ret = vop_copy_dp_entry(vdev, argp, &type, &dd);
292 if (ret) {
293 dev_err(vop_dev(vdev), "%s %d err %d\n",
294 __func__, __LINE__, ret);
295 return ret;
296 }
297
298 vop_init_device_ctrl(vdev, dd);
299
300 vdev->dd = dd;
301 vdev->virtio_id = type;
302 vqconfig = mic_vq_config(dd);
303 INIT_WORK(&vdev->virtio_bh_work, vop_bh_handler);
304
305 for (i = 0; i < dd->num_vq; i++) {
306 struct vop_vringh *vvr = &vdev->vvr[i];
307 struct mic_vring *vr = &vdev->vvr[i].vring;
308
309 num = le16_to_cpu(vqconfig[i].num);
310 mutex_init(&vvr->vr_mutex);
311 vr_size = PAGE_ALIGN(vring_size(num, MIC_VIRTIO_RING_ALIGN) +
312 sizeof(struct _mic_vring_info));
313 vr->va = (void *)
314 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
315 get_order(vr_size));
316 if (!vr->va) {
317 ret = -ENOMEM;
318 dev_err(vop_dev(vdev), "%s %d err %d\n",
319 __func__, __LINE__, ret);
320 goto err;
321 }
322 vr->len = vr_size;
323 vr->info = vr->va + vring_size(num, MIC_VIRTIO_RING_ALIGN);
324 vr->info->magic = cpu_to_le32(MIC_MAGIC + vdev->virtio_id + i);
325 vr_addr = dma_map_single(&vpdev->dev, vr->va, vr_size,
326 DMA_BIDIRECTIONAL);
327 if (dma_mapping_error(&vpdev->dev, vr_addr)) {
328 free_pages((unsigned long)vr->va, get_order(vr_size));
329 ret = -ENOMEM;
330 dev_err(vop_dev(vdev), "%s %d err %d\n",
331 __func__, __LINE__, ret);
332 goto err;
333 }
334 vqconfig[i].address = cpu_to_le64(vr_addr);
335
336 vring_init(&vr->vr, num, vr->va, MIC_VIRTIO_RING_ALIGN);
337 ret = vringh_init_kern(&vvr->vrh,
338 *(u32 *)mic_vq_features(vdev->dd),
339 num, false, vr->vr.desc, vr->vr.avail,
340 vr->vr.used);
341 if (ret) {
342 dev_err(vop_dev(vdev), "%s %d err %d\n",
343 __func__, __LINE__, ret);
344 goto err;
345 }
346 vringh_kiov_init(&vvr->riov, NULL, 0);
347 vringh_kiov_init(&vvr->wiov, NULL, 0);
348 vvr->head = USHRT_MAX;
349 vvr->vdev = vdev;
350 vvr->vrh.notify = _vop_notify;
351 dev_dbg(&vpdev->dev,
352 "%s %d index %d va %p info %p vr_size 0x%x\n",
353 __func__, __LINE__, i, vr->va, vr->info, vr_size);
354 vvr->buf = (void *)__get_free_pages(GFP_KERNEL,
355 get_order(VOP_INT_DMA_BUF_SIZE));
356 vvr->buf_da = dma_map_single(&vpdev->dev,
357 vvr->buf, VOP_INT_DMA_BUF_SIZE,
358 DMA_BIDIRECTIONAL);
359 }
360
361 snprintf(irqname, sizeof(irqname), "vop%dvirtio%d", vpdev->index,
362 vdev->virtio_id);
363 vdev->virtio_db = vpdev->hw_ops->next_db(vpdev);
364 vdev->virtio_cookie = vpdev->hw_ops->request_irq(vpdev,
365 _vop_virtio_intr_handler, irqname, vdev,
366 vdev->virtio_db);
367 if (IS_ERR(vdev->virtio_cookie)) {
368 ret = PTR_ERR(vdev->virtio_cookie);
369 dev_dbg(&vpdev->dev, "request irq failed\n");
370 goto err;
371 }
372
373 vdev->dc->c2h_vdev_db = vdev->virtio_db;
374
375 /*
376 * Order the type update with previous stores. This write barrier
377 * is paired with the corresponding read barrier before the uncached
378 * system memory read of the type, on the card while scanning the
379 * device page.
380 */
381 smp_wmb();
382 dd->type = type;
383 argp->type = type;
384
385 if (bootparam) {
386 db = bootparam->h2c_config_db;
387 if (db != -1)
388 vpdev->hw_ops->send_intr(vpdev, db);
389 }
390 dev_dbg(&vpdev->dev, "Added virtio id %d db %d\n", dd->type, db);
391 return 0;
392 err:
393 vqconfig = mic_vq_config(dd);
394 for (j = 0; j < i; j++) {
395 struct vop_vringh *vvr = &vdev->vvr[j];
396
397 dma_unmap_single(&vpdev->dev, le64_to_cpu(vqconfig[j].address),
398 vvr->vring.len, DMA_BIDIRECTIONAL);
399 free_pages((unsigned long)vvr->vring.va,
400 get_order(vvr->vring.len));
401 }
402 return ret;
403 }
404
405 static void vop_dev_remove(struct vop_info *pvi, struct mic_device_ctrl *devp,
406 struct vop_device *vpdev)
407 {
408 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
409 s8 db;
410 int ret, retry;
411 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wake);
412
413 devp->config_change = MIC_VIRTIO_PARAM_DEV_REMOVE;
414 db = bootparam->h2c_config_db;
415 if (db != -1)
416 vpdev->hw_ops->send_intr(vpdev, db);
417 else
418 goto done;
419 for (retry = 15; retry--;) {
420 ret = wait_event_timeout(wake, devp->guest_ack,
421 msecs_to_jiffies(1000));
422 if (ret)
423 break;
424 }
425 done:
426 devp->config_change = 0;
427 devp->guest_ack = 0;
428 }
429
430 static void vop_virtio_del_device(struct vop_vdev *vdev)
431 {
432 struct vop_info *vi = vdev->vi;
433 struct vop_device *vpdev = vdev->vpdev;
434 int i;
435 struct mic_vqconfig *vqconfig;
436 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
437
438 if (!bootparam)
439 goto skip_hot_remove;
440 vop_dev_remove(vi, vdev->dc, vpdev);
441 skip_hot_remove:
442 vpdev->hw_ops->free_irq(vpdev, vdev->virtio_cookie, vdev);
443 flush_work(&vdev->virtio_bh_work);
444 vqconfig = mic_vq_config(vdev->dd);
445 for (i = 0; i < vdev->dd->num_vq; i++) {
446 struct vop_vringh *vvr = &vdev->vvr[i];
447
448 dma_unmap_single(&vpdev->dev,
449 vvr->buf_da, VOP_INT_DMA_BUF_SIZE,
450 DMA_BIDIRECTIONAL);
451 free_pages((unsigned long)vvr->buf,
452 get_order(VOP_INT_DMA_BUF_SIZE));
453 vringh_kiov_cleanup(&vvr->riov);
454 vringh_kiov_cleanup(&vvr->wiov);
455 dma_unmap_single(&vpdev->dev, le64_to_cpu(vqconfig[i].address),
456 vvr->vring.len, DMA_BIDIRECTIONAL);
457 free_pages((unsigned long)vvr->vring.va,
458 get_order(vvr->vring.len));
459 }
460 /*
461 * Order the type update with previous stores. This write barrier
462 * is paired with the corresponding read barrier before the uncached
463 * system memory read of the type, on the card while scanning the
464 * device page.
465 */
466 smp_wmb();
467 vdev->dd->type = -1;
468 }
469
470 /*
471 * vop_sync_dma - Wrapper for synchronous DMAs.
472 *
473 * @dev - The address of the pointer to the device instance used
474 * for DMA registration.
475 * @dst - destination DMA address.
476 * @src - source DMA address.
477 * @len - size of the transfer.
478 *
479 * Return DMA_SUCCESS on success
480 */
481 static int vop_sync_dma(struct vop_vdev *vdev, dma_addr_t dst, dma_addr_t src,
482 size_t len)
483 {
484 int err = 0;
485 struct dma_device *ddev;
486 struct dma_async_tx_descriptor *tx;
487 struct vop_info *vi = dev_get_drvdata(&vdev->vpdev->dev);
488 struct dma_chan *vop_ch = vi->dma_ch;
489
490 if (!vop_ch) {
491 err = -EBUSY;
492 goto error;
493 }
494 ddev = vop_ch->device;
495 tx = ddev->device_prep_dma_memcpy(vop_ch, dst, src, len,
496 DMA_PREP_FENCE);
497 if (!tx) {
498 err = -ENOMEM;
499 goto error;
500 } else {
501 dma_cookie_t cookie;
502
503 cookie = tx->tx_submit(tx);
504 if (dma_submit_error(cookie)) {
505 err = -ENOMEM;
506 goto error;
507 }
508 dma_async_issue_pending(vop_ch);
509 err = dma_sync_wait(vop_ch, cookie);
510 }
511 error:
512 if (err)
513 dev_err(&vi->vpdev->dev, "%s %d err %d\n",
514 __func__, __LINE__, err);
515 return err;
516 }
517
518 #define VOP_USE_DMA true
519
520 /*
521 * Initiates the copies across the PCIe bus from card memory to a user
522 * space buffer. When transfers are done using DMA, source/destination
523 * addresses and transfer length must follow the alignment requirements of
524 * the MIC DMA engine.
525 */
526 static int vop_virtio_copy_to_user(struct vop_vdev *vdev, void __user *ubuf,
527 size_t len, u64 daddr, size_t dlen,
528 int vr_idx)
529 {
530 struct vop_device *vpdev = vdev->vpdev;
531 void __iomem *dbuf = vpdev->hw_ops->remap(vpdev, daddr, len);
532 struct vop_vringh *vvr = &vdev->vvr[vr_idx];
533 struct vop_info *vi = dev_get_drvdata(&vpdev->dev);
534 size_t dma_alignment;
535 bool x200;
536 size_t dma_offset, partlen;
537 int err;
538
539 if (!VOP_USE_DMA || !vi->dma_ch) {
540 if (copy_to_user(ubuf, (void __force *)dbuf, len)) {
541 err = -EFAULT;
542 dev_err(vop_dev(vdev), "%s %d err %d\n",
543 __func__, __LINE__, err);
544 goto err;
545 }
546 vdev->in_bytes += len;
547 err = 0;
548 goto err;
549 }
550
551 dma_alignment = 1 << vi->dma_ch->device->copy_align;
552 x200 = is_dma_copy_aligned(vi->dma_ch->device, 1, 1, 1);
553
554 dma_offset = daddr - round_down(daddr, dma_alignment);
555 daddr -= dma_offset;
556 len += dma_offset;
557 /*
558 * X100 uses DMA addresses as seen by the card so adding
559 * the aperture base is not required for DMA. However x200
560 * requires DMA addresses to be an offset into the bar so
561 * add the aperture base for x200.
562 */
563 if (x200)
564 daddr += vpdev->aper->pa;
565 while (len) {
566 partlen = min_t(size_t, len, VOP_INT_DMA_BUF_SIZE);
567 err = vop_sync_dma(vdev, vvr->buf_da, daddr,
568 ALIGN(partlen, dma_alignment));
569 if (err) {
570 dev_err(vop_dev(vdev), "%s %d err %d\n",
571 __func__, __LINE__, err);
572 goto err;
573 }
574 if (copy_to_user(ubuf, vvr->buf + dma_offset,
575 partlen - dma_offset)) {
576 err = -EFAULT;
577 dev_err(vop_dev(vdev), "%s %d err %d\n",
578 __func__, __LINE__, err);
579 goto err;
580 }
581 daddr += partlen;
582 ubuf += partlen;
583 dbuf += partlen;
584 vdev->in_bytes_dma += partlen;
585 vdev->in_bytes += partlen;
586 len -= partlen;
587 dma_offset = 0;
588 }
589 err = 0;
590 err:
591 vpdev->hw_ops->unmap(vpdev, dbuf);
592 dev_dbg(vop_dev(vdev),
593 "%s: ubuf %p dbuf %p len 0x%zx vr_idx 0x%x\n",
594 __func__, ubuf, dbuf, len, vr_idx);
595 return err;
596 }
597
598 /*
599 * Initiates copies across the PCIe bus from a user space buffer to card
600 * memory. When transfers are done using DMA, source/destination addresses
601 * and transfer length must follow the alignment requirements of the MIC
602 * DMA engine.
603 */
604 static int vop_virtio_copy_from_user(struct vop_vdev *vdev, void __user *ubuf,
605 size_t len, u64 daddr, size_t dlen,
606 int vr_idx)
607 {
608 struct vop_device *vpdev = vdev->vpdev;
609 void __iomem *dbuf = vpdev->hw_ops->remap(vpdev, daddr, len);
610 struct vop_vringh *vvr = &vdev->vvr[vr_idx];
611 struct vop_info *vi = dev_get_drvdata(&vdev->vpdev->dev);
612 size_t dma_alignment;
613 bool x200;
614 size_t partlen;
615 bool dma = VOP_USE_DMA && vi->dma_ch;
616 int err = 0;
617
618 if (dma) {
619 dma_alignment = 1 << vi->dma_ch->device->copy_align;
620 x200 = is_dma_copy_aligned(vi->dma_ch->device, 1, 1, 1);
621
622 if (daddr & (dma_alignment - 1)) {
623 vdev->tx_dst_unaligned += len;
624 dma = false;
625 } else if (ALIGN(len, dma_alignment) > dlen) {
626 vdev->tx_len_unaligned += len;
627 dma = false;
628 }
629 }
630
631 if (!dma)
632 goto memcpy;
633
634 /*
635 * X100 uses DMA addresses as seen by the card so adding
636 * the aperture base is not required for DMA. However x200
637 * requires DMA addresses to be an offset into the bar so
638 * add the aperture base for x200.
639 */
640 if (x200)
641 daddr += vpdev->aper->pa;
642 while (len) {
643 partlen = min_t(size_t, len, VOP_INT_DMA_BUF_SIZE);
644
645 if (copy_from_user(vvr->buf, ubuf, partlen)) {
646 err = -EFAULT;
647 dev_err(vop_dev(vdev), "%s %d err %d\n",
648 __func__, __LINE__, err);
649 goto err;
650 }
651 err = vop_sync_dma(vdev, daddr, vvr->buf_da,
652 ALIGN(partlen, dma_alignment));
653 if (err) {
654 dev_err(vop_dev(vdev), "%s %d err %d\n",
655 __func__, __LINE__, err);
656 goto err;
657 }
658 daddr += partlen;
659 ubuf += partlen;
660 dbuf += partlen;
661 vdev->out_bytes_dma += partlen;
662 vdev->out_bytes += partlen;
663 len -= partlen;
664 }
665 memcpy:
666 /*
667 * We are copying to IO below and should ideally use something
668 * like copy_from_user_toio(..) if it existed.
669 */
670 if (copy_from_user((void __force *)dbuf, ubuf, len)) {
671 err = -EFAULT;
672 dev_err(vop_dev(vdev), "%s %d err %d\n",
673 __func__, __LINE__, err);
674 goto err;
675 }
676 vdev->out_bytes += len;
677 err = 0;
678 err:
679 vpdev->hw_ops->unmap(vpdev, dbuf);
680 dev_dbg(vop_dev(vdev),
681 "%s: ubuf %p dbuf %p len 0x%zx vr_idx 0x%x\n",
682 __func__, ubuf, dbuf, len, vr_idx);
683 return err;
684 }
685
686 #define MIC_VRINGH_READ true
687
688 /* Determine the total number of bytes consumed in a VRINGH KIOV */
689 static inline u32 vop_vringh_iov_consumed(struct vringh_kiov *iov)
690 {
691 int i;
692 u32 total = iov->consumed;
693
694 for (i = 0; i < iov->i; i++)
695 total += iov->iov[i].iov_len;
696 return total;
697 }
698
699 /*
700 * Traverse the VRINGH KIOV and issue the APIs to trigger the copies.
701 * This API is heavily based on the vringh_iov_xfer(..) implementation
702 * in vringh.c. The reason we cannot reuse vringh_iov_pull_kern(..)
703 * and vringh_iov_push_kern(..) directly is because there is no
704 * way to override the VRINGH xfer(..) routines as of v3.10.
705 */
706 static int vop_vringh_copy(struct vop_vdev *vdev, struct vringh_kiov *iov,
707 void __user *ubuf, size_t len, bool read, int vr_idx,
708 size_t *out_len)
709 {
710 int ret = 0;
711 size_t partlen, tot_len = 0;
712
713 while (len && iov->i < iov->used) {
714 struct kvec *kiov = &iov->iov[iov->i];
715 unsigned long daddr = (unsigned long)kiov->iov_base;
716
717 partlen = min(kiov->iov_len, len);
718 if (read)
719 ret = vop_virtio_copy_to_user(vdev, ubuf, partlen,
720 daddr,
721 kiov->iov_len,
722 vr_idx);
723 else
724 ret = vop_virtio_copy_from_user(vdev, ubuf, partlen,
725 daddr,
726 kiov->iov_len,
727 vr_idx);
728 if (ret) {
729 dev_err(vop_dev(vdev), "%s %d err %d\n",
730 __func__, __LINE__, ret);
731 break;
732 }
733 len -= partlen;
734 ubuf += partlen;
735 tot_len += partlen;
736 iov->consumed += partlen;
737 kiov->iov_len -= partlen;
738 kiov->iov_base += partlen;
739 if (!kiov->iov_len) {
740 /* Fix up old iov element then increment. */
741 kiov->iov_len = iov->consumed;
742 kiov->iov_base -= iov->consumed;
743
744 iov->consumed = 0;
745 iov->i++;
746 }
747 }
748 *out_len = tot_len;
749 return ret;
750 }
751
752 /*
753 * Use the standard VRINGH infrastructure in the kernel to fetch new
754 * descriptors, initiate the copies and update the used ring.
755 */
756 static int _vop_virtio_copy(struct vop_vdev *vdev, struct mic_copy_desc *copy)
757 {
758 int ret = 0;
759 u32 iovcnt = copy->iovcnt;
760 struct iovec iov;
761 struct iovec __user *u_iov = copy->iov;
762 void __user *ubuf = NULL;
763 struct vop_vringh *vvr = &vdev->vvr[copy->vr_idx];
764 struct vringh_kiov *riov = &vvr->riov;
765 struct vringh_kiov *wiov = &vvr->wiov;
766 struct vringh *vrh = &vvr->vrh;
767 u16 *head = &vvr->head;
768 struct mic_vring *vr = &vvr->vring;
769 size_t len = 0, out_len;
770
771 copy->out_len = 0;
772 /* Fetch a new IOVEC if all previous elements have been processed */
773 if (riov->i == riov->used && wiov->i == wiov->used) {
774 ret = vringh_getdesc_kern(vrh, riov, wiov,
775 head, GFP_KERNEL);
776 /* Check if there are available descriptors */
777 if (ret <= 0)
778 return ret;
779 }
780 while (iovcnt) {
781 if (!len) {
782 /* Copy over a new iovec from user space. */
783 ret = copy_from_user(&iov, u_iov, sizeof(*u_iov));
784 if (ret) {
785 ret = -EINVAL;
786 dev_err(vop_dev(vdev), "%s %d err %d\n",
787 __func__, __LINE__, ret);
788 break;
789 }
790 len = iov.iov_len;
791 ubuf = iov.iov_base;
792 }
793 /* Issue all the read descriptors first */
794 ret = vop_vringh_copy(vdev, riov, ubuf, len,
795 MIC_VRINGH_READ, copy->vr_idx, &out_len);
796 if (ret) {
797 dev_err(vop_dev(vdev), "%s %d err %d\n",
798 __func__, __LINE__, ret);
799 break;
800 }
801 len -= out_len;
802 ubuf += out_len;
803 copy->out_len += out_len;
804 /* Issue the write descriptors next */
805 ret = vop_vringh_copy(vdev, wiov, ubuf, len,
806 !MIC_VRINGH_READ, copy->vr_idx, &out_len);
807 if (ret) {
808 dev_err(vop_dev(vdev), "%s %d err %d\n",
809 __func__, __LINE__, ret);
810 break;
811 }
812 len -= out_len;
813 ubuf += out_len;
814 copy->out_len += out_len;
815 if (!len) {
816 /* One user space iovec is now completed */
817 iovcnt--;
818 u_iov++;
819 }
820 /* Exit loop if all elements in KIOVs have been processed. */
821 if (riov->i == riov->used && wiov->i == wiov->used)
822 break;
823 }
824 /*
825 * Update the used ring if a descriptor was available and some data was
826 * copied in/out and the user asked for a used ring update.
827 */
828 if (*head != USHRT_MAX && copy->out_len && copy->update_used) {
829 u32 total = 0;
830
831 /* Determine the total data consumed */
832 total += vop_vringh_iov_consumed(riov);
833 total += vop_vringh_iov_consumed(wiov);
834 vringh_complete_kern(vrh, *head, total);
835 *head = USHRT_MAX;
836 if (vringh_need_notify_kern(vrh) > 0)
837 vringh_notify(vrh);
838 vringh_kiov_cleanup(riov);
839 vringh_kiov_cleanup(wiov);
840 /* Update avail idx for user space */
841 vr->info->avail_idx = vrh->last_avail_idx;
842 }
843 return ret;
844 }
845
846 static inline int vop_verify_copy_args(struct vop_vdev *vdev,
847 struct mic_copy_desc *copy)
848 {
849 if (!vdev || copy->vr_idx >= vdev->dd->num_vq)
850 return -EINVAL;
851 return 0;
852 }
853
854 /* Copy a specified number of virtio descriptors in a chain */
855 static int vop_virtio_copy_desc(struct vop_vdev *vdev,
856 struct mic_copy_desc *copy)
857 {
858 int err;
859 struct vop_vringh *vvr;
860
861 err = vop_verify_copy_args(vdev, copy);
862 if (err)
863 return err;
864
865 vvr = &vdev->vvr[copy->vr_idx];
866 mutex_lock(&vvr->vr_mutex);
867 if (!vop_vdevup(vdev)) {
868 err = -ENODEV;
869 dev_err(vop_dev(vdev), "%s %d err %d\n",
870 __func__, __LINE__, err);
871 goto err;
872 }
873 err = _vop_virtio_copy(vdev, copy);
874 if (err) {
875 dev_err(vop_dev(vdev), "%s %d err %d\n",
876 __func__, __LINE__, err);
877 }
878 err:
879 mutex_unlock(&vvr->vr_mutex);
880 return err;
881 }
882
883 static int vop_open(struct inode *inode, struct file *f)
884 {
885 struct vop_vdev *vdev;
886 struct vop_info *vi = container_of(f->private_data,
887 struct vop_info, miscdev);
888
889 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
890 if (!vdev)
891 return -ENOMEM;
892 vdev->vi = vi;
893 mutex_init(&vdev->vdev_mutex);
894 f->private_data = vdev;
895 init_completion(&vdev->destroy);
896 complete(&vdev->destroy);
897 return 0;
898 }
899
900 static int vop_release(struct inode *inode, struct file *f)
901 {
902 struct vop_vdev *vdev = f->private_data, *vdev_tmp;
903 struct vop_info *vi = vdev->vi;
904 struct list_head *pos, *tmp;
905 bool found = false;
906
907 mutex_lock(&vdev->vdev_mutex);
908 if (vdev->deleted)
909 goto unlock;
910 mutex_lock(&vi->vop_mutex);
911 list_for_each_safe(pos, tmp, &vi->vdev_list) {
912 vdev_tmp = list_entry(pos, struct vop_vdev, list);
913 if (vdev == vdev_tmp) {
914 vop_virtio_del_device(vdev);
915 list_del(pos);
916 found = true;
917 break;
918 }
919 }
920 mutex_unlock(&vi->vop_mutex);
921 unlock:
922 mutex_unlock(&vdev->vdev_mutex);
923 if (!found)
924 wait_for_completion(&vdev->destroy);
925 f->private_data = NULL;
926 kfree(vdev);
927 return 0;
928 }
929
930 static long vop_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
931 {
932 struct vop_vdev *vdev = f->private_data;
933 struct vop_info *vi = vdev->vi;
934 void __user *argp = (void __user *)arg;
935 int ret;
936
937 switch (cmd) {
938 case MIC_VIRTIO_ADD_DEVICE:
939 {
940 struct mic_device_desc dd, *dd_config;
941
942 if (copy_from_user(&dd, argp, sizeof(dd)))
943 return -EFAULT;
944
945 if (mic_aligned_desc_size(&dd) > MIC_MAX_DESC_BLK_SIZE ||
946 dd.num_vq > MIC_MAX_VRINGS)
947 return -EINVAL;
948
949 dd_config = memdup_user(argp, mic_desc_size(&dd));
950 if (IS_ERR(dd_config))
951 return PTR_ERR(dd_config);
952
953 /* Ensure desc has not changed between the two reads */
954 if (memcmp(&dd, dd_config, sizeof(dd))) {
955 ret = -EINVAL;
956 goto free_ret;
957 }
958 mutex_lock(&vdev->vdev_mutex);
959 mutex_lock(&vi->vop_mutex);
960 ret = vop_virtio_add_device(vdev, dd_config);
961 if (ret)
962 goto unlock_ret;
963 list_add_tail(&vdev->list, &vi->vdev_list);
964 unlock_ret:
965 mutex_unlock(&vi->vop_mutex);
966 mutex_unlock(&vdev->vdev_mutex);
967 free_ret:
968 kfree(dd_config);
969 return ret;
970 }
971 case MIC_VIRTIO_COPY_DESC:
972 {
973 struct mic_copy_desc copy;
974
975 mutex_lock(&vdev->vdev_mutex);
976 ret = vop_vdev_inited(vdev);
977 if (ret)
978 goto _unlock_ret;
979
980 if (copy_from_user(&copy, argp, sizeof(copy))) {
981 ret = -EFAULT;
982 goto _unlock_ret;
983 }
984
985 ret = vop_virtio_copy_desc(vdev, &copy);
986 if (ret < 0)
987 goto _unlock_ret;
988 if (copy_to_user(
989 &((struct mic_copy_desc __user *)argp)->out_len,
990 &copy.out_len, sizeof(copy.out_len)))
991 ret = -EFAULT;
992 _unlock_ret:
993 mutex_unlock(&vdev->vdev_mutex);
994 return ret;
995 }
996 case MIC_VIRTIO_CONFIG_CHANGE:
997 {
998 void *buf;
999
1000 mutex_lock(&vdev->vdev_mutex);
1001 ret = vop_vdev_inited(vdev);
1002 if (ret)
1003 goto __unlock_ret;
1004 buf = memdup_user(argp, vdev->dd->config_len);
1005 if (IS_ERR(buf)) {
1006 ret = PTR_ERR(buf);
1007 goto __unlock_ret;
1008 }
1009 ret = vop_virtio_config_change(vdev, buf);
1010 kfree(buf);
1011 __unlock_ret:
1012 mutex_unlock(&vdev->vdev_mutex);
1013 return ret;
1014 }
1015 default:
1016 return -ENOIOCTLCMD;
1017 };
1018 return 0;
1019 }
1020
1021 /*
1022 * We return EPOLLIN | EPOLLOUT from poll when new buffers are enqueued, and
1023 * not when previously enqueued buffers may be available. This means that
1024 * in the card->host (TX) path, when userspace is unblocked by poll it
1025 * must drain all available descriptors or it can stall.
1026 */
1027 static __poll_t vop_poll(struct file *f, poll_table *wait)
1028 {
1029 struct vop_vdev *vdev = f->private_data;
1030 __poll_t mask = 0;
1031
1032 mutex_lock(&vdev->vdev_mutex);
1033 if (vop_vdev_inited(vdev)) {
1034 mask = EPOLLERR;
1035 goto done;
1036 }
1037 poll_wait(f, &vdev->waitq, wait);
1038 if (vop_vdev_inited(vdev)) {
1039 mask = EPOLLERR;
1040 } else if (vdev->poll_wake) {
1041 vdev->poll_wake = 0;
1042 mask = EPOLLIN | EPOLLOUT;
1043 }
1044 done:
1045 mutex_unlock(&vdev->vdev_mutex);
1046 return mask;
1047 }
1048
1049 static inline int
1050 vop_query_offset(struct vop_vdev *vdev, unsigned long offset,
1051 unsigned long *size, unsigned long *pa)
1052 {
1053 struct vop_device *vpdev = vdev->vpdev;
1054 unsigned long start = MIC_DP_SIZE;
1055 int i;
1056
1057 /*
1058 * MMAP interface is as follows:
1059 * offset region
1060 * 0x0 virtio device_page
1061 * 0x1000 first vring
1062 * 0x1000 + size of 1st vring second vring
1063 * ....
1064 */
1065 if (!offset) {
1066 *pa = virt_to_phys(vpdev->hw_ops->get_dp(vpdev));
1067 *size = MIC_DP_SIZE;
1068 return 0;
1069 }
1070
1071 for (i = 0; i < vdev->dd->num_vq; i++) {
1072 struct vop_vringh *vvr = &vdev->vvr[i];
1073
1074 if (offset == start) {
1075 *pa = virt_to_phys(vvr->vring.va);
1076 *size = vvr->vring.len;
1077 return 0;
1078 }
1079 start += vvr->vring.len;
1080 }
1081 return -1;
1082 }
1083
1084 /*
1085 * Maps the device page and virtio rings to user space for readonly access.
1086 */
1087 static int vop_mmap(struct file *f, struct vm_area_struct *vma)
1088 {
1089 struct vop_vdev *vdev = f->private_data;
1090 unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
1091 unsigned long pa, size = vma->vm_end - vma->vm_start, size_rem = size;
1092 int i, err;
1093
1094 err = vop_vdev_inited(vdev);
1095 if (err)
1096 goto ret;
1097 if (vma->vm_flags & VM_WRITE) {
1098 err = -EACCES;
1099 goto ret;
1100 }
1101 while (size_rem) {
1102 i = vop_query_offset(vdev, offset, &size, &pa);
1103 if (i < 0) {
1104 err = -EINVAL;
1105 goto ret;
1106 }
1107 err = remap_pfn_range(vma, vma->vm_start + offset,
1108 pa >> PAGE_SHIFT, size,
1109 vma->vm_page_prot);
1110 if (err)
1111 goto ret;
1112 size_rem -= size;
1113 offset += size;
1114 }
1115 ret:
1116 return err;
1117 }
1118
1119 static const struct file_operations vop_fops = {
1120 .open = vop_open,
1121 .release = vop_release,
1122 .unlocked_ioctl = vop_ioctl,
1123 .poll = vop_poll,
1124 .mmap = vop_mmap,
1125 .owner = THIS_MODULE,
1126 };
1127
1128 int vop_host_init(struct vop_info *vi)
1129 {
1130 int rc;
1131 struct miscdevice *mdev;
1132 struct vop_device *vpdev = vi->vpdev;
1133
1134 INIT_LIST_HEAD(&vi->vdev_list);
1135 vi->dma_ch = vpdev->dma_ch;
1136 mdev = &vi->miscdev;
1137 mdev->minor = MISC_DYNAMIC_MINOR;
1138 snprintf(vi->name, sizeof(vi->name), "vop_virtio%d", vpdev->index);
1139 mdev->name = vi->name;
1140 mdev->fops = &vop_fops;
1141 mdev->parent = &vpdev->dev;
1142
1143 rc = misc_register(mdev);
1144 if (rc)
1145 dev_err(&vpdev->dev, "%s failed rc %d\n", __func__, rc);
1146 return rc;
1147 }
1148
1149 void vop_host_uninit(struct vop_info *vi)
1150 {
1151 struct list_head *pos, *tmp;
1152 struct vop_vdev *vdev;
1153
1154 mutex_lock(&vi->vop_mutex);
1155 vop_virtio_reset_devices(vi);
1156 list_for_each_safe(pos, tmp, &vi->vdev_list) {
1157 vdev = list_entry(pos, struct vop_vdev, list);
1158 list_del(pos);
1159 reinit_completion(&vdev->destroy);
1160 mutex_unlock(&vi->vop_mutex);
1161 mutex_lock(&vdev->vdev_mutex);
1162 vop_virtio_del_device(vdev);
1163 vdev->deleted = true;
1164 mutex_unlock(&vdev->vdev_mutex);
1165 complete(&vdev->destroy);
1166 mutex_lock(&vi->vop_mutex);
1167 }
1168 mutex_unlock(&vi->vop_mutex);
1169 misc_deregister(&vi->miscdev);
1170 }
Linux with added FPC-III support