Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
[linux-2.6-microblaze.git] / drivers / s390 / cio / vfio_ccw_ops.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Physical device callbacks for vfio_ccw
4  *
5  * Copyright IBM Corp. 2017
6  * Copyright Red Hat, Inc. 2019
7  *
8  * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com>
9  *            Xiao Feng Ren <renxiaof@linux.vnet.ibm.com>
10  *            Cornelia Huck <cohuck@redhat.com>
11  */
12
13 #include <linux/vfio.h>
14 #include <linux/mdev.h>
15 #include <linux/nospec.h>
16 #include <linux/slab.h>
17
18 #include "vfio_ccw_private.h"
19
20 static const struct vfio_device_ops vfio_ccw_dev_ops;
21
22 static int vfio_ccw_mdev_reset(struct vfio_ccw_private *private)
23 {
24         struct subchannel *sch;
25         int ret;
26
27         sch = private->sch;
28         /*
29          * TODO:
30          * In the cureent stage, some things like "no I/O running" and "no
31          * interrupt pending" are clear, but we are not sure what other state
32          * we need to care about.
33          * There are still a lot more instructions need to be handled. We
34          * should come back here later.
35          */
36         ret = vfio_ccw_sch_quiesce(sch);
37         if (ret)
38                 return ret;
39
40         ret = cio_enable_subchannel(sch, (u32)(unsigned long)sch);
41         if (!ret)
42                 private->state = VFIO_CCW_STATE_IDLE;
43
44         return ret;
45 }
46
47 static int vfio_ccw_mdev_notifier(struct notifier_block *nb,
48                                   unsigned long action,
49                                   void *data)
50 {
51         struct vfio_ccw_private *private =
52                 container_of(nb, struct vfio_ccw_private, nb);
53
54         /*
55          * Vendor drivers MUST unpin pages in response to an
56          * invalidation.
57          */
58         if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
59                 struct vfio_iommu_type1_dma_unmap *unmap = data;
60
61                 if (!cp_iova_pinned(&private->cp, unmap->iova))
62                         return NOTIFY_OK;
63
64                 if (vfio_ccw_mdev_reset(private))
65                         return NOTIFY_BAD;
66
67                 cp_free(&private->cp);
68                 return NOTIFY_OK;
69         }
70
71         return NOTIFY_DONE;
72 }
73
74 static ssize_t name_show(struct mdev_type *mtype,
75                          struct mdev_type_attribute *attr, char *buf)
76 {
77         return sprintf(buf, "I/O subchannel (Non-QDIO)\n");
78 }
79 static MDEV_TYPE_ATTR_RO(name);
80
81 static ssize_t device_api_show(struct mdev_type *mtype,
82                                struct mdev_type_attribute *attr, char *buf)
83 {
84         return sprintf(buf, "%s\n", VFIO_DEVICE_API_CCW_STRING);
85 }
86 static MDEV_TYPE_ATTR_RO(device_api);
87
88 static ssize_t available_instances_show(struct mdev_type *mtype,
89                                         struct mdev_type_attribute *attr,
90                                         char *buf)
91 {
92         struct vfio_ccw_private *private =
93                 dev_get_drvdata(mtype_get_parent_dev(mtype));
94
95         return sprintf(buf, "%d\n", atomic_read(&private->avail));
96 }
97 static MDEV_TYPE_ATTR_RO(available_instances);
98
99 static struct attribute *mdev_types_attrs[] = {
100         &mdev_type_attr_name.attr,
101         &mdev_type_attr_device_api.attr,
102         &mdev_type_attr_available_instances.attr,
103         NULL,
104 };
105
106 static struct attribute_group mdev_type_group = {
107         .name  = "io",
108         .attrs = mdev_types_attrs,
109 };
110
111 static struct attribute_group *mdev_type_groups[] = {
112         &mdev_type_group,
113         NULL,
114 };
115
116 static int vfio_ccw_mdev_probe(struct mdev_device *mdev)
117 {
118         struct vfio_ccw_private *private = dev_get_drvdata(mdev->dev.parent);
119         int ret;
120
121         if (private->state == VFIO_CCW_STATE_NOT_OPER)
122                 return -ENODEV;
123
124         if (atomic_dec_if_positive(&private->avail) < 0)
125                 return -EPERM;
126
127         memset(&private->vdev, 0, sizeof(private->vdev));
128         vfio_init_group_dev(&private->vdev, &mdev->dev,
129                             &vfio_ccw_dev_ops);
130
131         private->mdev = mdev;
132         private->state = VFIO_CCW_STATE_IDLE;
133
134         VFIO_CCW_MSG_EVENT(2, "mdev %pUl, sch %x.%x.%04x: create\n",
135                            mdev_uuid(mdev), private->sch->schid.cssid,
136                            private->sch->schid.ssid,
137                            private->sch->schid.sch_no);
138
139         ret = vfio_register_emulated_iommu_dev(&private->vdev);
140         if (ret)
141                 goto err_atomic;
142         dev_set_drvdata(&mdev->dev, private);
143         return 0;
144
145 err_atomic:
146         vfio_uninit_group_dev(&private->vdev);
147         atomic_inc(&private->avail);
148         private->mdev = NULL;
149         private->state = VFIO_CCW_STATE_IDLE;
150         return ret;
151 }
152
153 static void vfio_ccw_mdev_remove(struct mdev_device *mdev)
154 {
155         struct vfio_ccw_private *private = dev_get_drvdata(mdev->dev.parent);
156
157         VFIO_CCW_MSG_EVENT(2, "mdev %pUl, sch %x.%x.%04x: remove\n",
158                            mdev_uuid(mdev), private->sch->schid.cssid,
159                            private->sch->schid.ssid,
160                            private->sch->schid.sch_no);
161
162         vfio_unregister_group_dev(&private->vdev);
163
164         if ((private->state != VFIO_CCW_STATE_NOT_OPER) &&
165             (private->state != VFIO_CCW_STATE_STANDBY)) {
166                 if (!vfio_ccw_sch_quiesce(private->sch))
167                         private->state = VFIO_CCW_STATE_STANDBY;
168                 /* The state will be NOT_OPER on error. */
169         }
170
171         vfio_uninit_group_dev(&private->vdev);
172         cp_free(&private->cp);
173         private->mdev = NULL;
174         atomic_inc(&private->avail);
175 }
176
177 static int vfio_ccw_mdev_open_device(struct vfio_device *vdev)
178 {
179         struct vfio_ccw_private *private =
180                 container_of(vdev, struct vfio_ccw_private, vdev);
181         unsigned long events = VFIO_IOMMU_NOTIFY_DMA_UNMAP;
182         int ret;
183
184         private->nb.notifier_call = vfio_ccw_mdev_notifier;
185
186         ret = vfio_register_notifier(vdev->dev, VFIO_IOMMU_NOTIFY,
187                                      &events, &private->nb);
188         if (ret)
189                 return ret;
190
191         ret = vfio_ccw_register_async_dev_regions(private);
192         if (ret)
193                 goto out_unregister;
194
195         ret = vfio_ccw_register_schib_dev_regions(private);
196         if (ret)
197                 goto out_unregister;
198
199         ret = vfio_ccw_register_crw_dev_regions(private);
200         if (ret)
201                 goto out_unregister;
202
203         return ret;
204
205 out_unregister:
206         vfio_ccw_unregister_dev_regions(private);
207         vfio_unregister_notifier(vdev->dev, VFIO_IOMMU_NOTIFY,
208                                  &private->nb);
209         return ret;
210 }
211
212 static void vfio_ccw_mdev_close_device(struct vfio_device *vdev)
213 {
214         struct vfio_ccw_private *private =
215                 container_of(vdev, struct vfio_ccw_private, vdev);
216
217         if ((private->state != VFIO_CCW_STATE_NOT_OPER) &&
218             (private->state != VFIO_CCW_STATE_STANDBY)) {
219                 if (!vfio_ccw_mdev_reset(private))
220                         private->state = VFIO_CCW_STATE_STANDBY;
221                 /* The state will be NOT_OPER on error. */
222         }
223
224         cp_free(&private->cp);
225         vfio_ccw_unregister_dev_regions(private);
226         vfio_unregister_notifier(vdev->dev, VFIO_IOMMU_NOTIFY, &private->nb);
227 }
228
229 static ssize_t vfio_ccw_mdev_read_io_region(struct vfio_ccw_private *private,
230                                             char __user *buf, size_t count,
231                                             loff_t *ppos)
232 {
233         loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
234         struct ccw_io_region *region;
235         int ret;
236
237         if (pos + count > sizeof(*region))
238                 return -EINVAL;
239
240         mutex_lock(&private->io_mutex);
241         region = private->io_region;
242         if (copy_to_user(buf, (void *)region + pos, count))
243                 ret = -EFAULT;
244         else
245                 ret = count;
246         mutex_unlock(&private->io_mutex);
247         return ret;
248 }
249
250 static ssize_t vfio_ccw_mdev_read(struct vfio_device *vdev,
251                                   char __user *buf,
252                                   size_t count,
253                                   loff_t *ppos)
254 {
255         struct vfio_ccw_private *private =
256                 container_of(vdev, struct vfio_ccw_private, vdev);
257         unsigned int index = VFIO_CCW_OFFSET_TO_INDEX(*ppos);
258
259         if (index >= VFIO_CCW_NUM_REGIONS + private->num_regions)
260                 return -EINVAL;
261
262         switch (index) {
263         case VFIO_CCW_CONFIG_REGION_INDEX:
264                 return vfio_ccw_mdev_read_io_region(private, buf, count, ppos);
265         default:
266                 index -= VFIO_CCW_NUM_REGIONS;
267                 return private->region[index].ops->read(private, buf, count,
268                                                         ppos);
269         }
270
271         return -EINVAL;
272 }
273
274 static ssize_t vfio_ccw_mdev_write_io_region(struct vfio_ccw_private *private,
275                                              const char __user *buf,
276                                              size_t count, loff_t *ppos)
277 {
278         loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
279         struct ccw_io_region *region;
280         int ret;
281
282         if (pos + count > sizeof(*region))
283                 return -EINVAL;
284
285         if (!mutex_trylock(&private->io_mutex))
286                 return -EAGAIN;
287
288         region = private->io_region;
289         if (copy_from_user((void *)region + pos, buf, count)) {
290                 ret = -EFAULT;
291                 goto out_unlock;
292         }
293
294         vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_IO_REQ);
295         ret = (region->ret_code != 0) ? region->ret_code : count;
296
297 out_unlock:
298         mutex_unlock(&private->io_mutex);
299         return ret;
300 }
301
302 static ssize_t vfio_ccw_mdev_write(struct vfio_device *vdev,
303                                    const char __user *buf,
304                                    size_t count,
305                                    loff_t *ppos)
306 {
307         struct vfio_ccw_private *private =
308                 container_of(vdev, struct vfio_ccw_private, vdev);
309         unsigned int index = VFIO_CCW_OFFSET_TO_INDEX(*ppos);
310
311         if (index >= VFIO_CCW_NUM_REGIONS + private->num_regions)
312                 return -EINVAL;
313
314         switch (index) {
315         case VFIO_CCW_CONFIG_REGION_INDEX:
316                 return vfio_ccw_mdev_write_io_region(private, buf, count, ppos);
317         default:
318                 index -= VFIO_CCW_NUM_REGIONS;
319                 return private->region[index].ops->write(private, buf, count,
320                                                          ppos);
321         }
322
323         return -EINVAL;
324 }
325
326 static int vfio_ccw_mdev_get_device_info(struct vfio_ccw_private *private,
327                                          struct vfio_device_info *info)
328 {
329         info->flags = VFIO_DEVICE_FLAGS_CCW | VFIO_DEVICE_FLAGS_RESET;
330         info->num_regions = VFIO_CCW_NUM_REGIONS + private->num_regions;
331         info->num_irqs = VFIO_CCW_NUM_IRQS;
332
333         return 0;
334 }
335
336 static int vfio_ccw_mdev_get_region_info(struct vfio_ccw_private *private,
337                                          struct vfio_region_info *info,
338                                          unsigned long arg)
339 {
340         int i;
341
342         switch (info->index) {
343         case VFIO_CCW_CONFIG_REGION_INDEX:
344                 info->offset = 0;
345                 info->size = sizeof(struct ccw_io_region);
346                 info->flags = VFIO_REGION_INFO_FLAG_READ
347                               | VFIO_REGION_INFO_FLAG_WRITE;
348                 return 0;
349         default: /* all other regions are handled via capability chain */
350         {
351                 struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
352                 struct vfio_region_info_cap_type cap_type = {
353                         .header.id = VFIO_REGION_INFO_CAP_TYPE,
354                         .header.version = 1 };
355                 int ret;
356
357                 if (info->index >=
358                     VFIO_CCW_NUM_REGIONS + private->num_regions)
359                         return -EINVAL;
360
361                 info->index = array_index_nospec(info->index,
362                                                  VFIO_CCW_NUM_REGIONS +
363                                                  private->num_regions);
364
365                 i = info->index - VFIO_CCW_NUM_REGIONS;
366
367                 info->offset = VFIO_CCW_INDEX_TO_OFFSET(info->index);
368                 info->size = private->region[i].size;
369                 info->flags = private->region[i].flags;
370
371                 cap_type.type = private->region[i].type;
372                 cap_type.subtype = private->region[i].subtype;
373
374                 ret = vfio_info_add_capability(&caps, &cap_type.header,
375                                                sizeof(cap_type));
376                 if (ret)
377                         return ret;
378
379                 info->flags |= VFIO_REGION_INFO_FLAG_CAPS;
380                 if (info->argsz < sizeof(*info) + caps.size) {
381                         info->argsz = sizeof(*info) + caps.size;
382                         info->cap_offset = 0;
383                 } else {
384                         vfio_info_cap_shift(&caps, sizeof(*info));
385                         if (copy_to_user((void __user *)arg + sizeof(*info),
386                                          caps.buf, caps.size)) {
387                                 kfree(caps.buf);
388                                 return -EFAULT;
389                         }
390                         info->cap_offset = sizeof(*info);
391                 }
392
393                 kfree(caps.buf);
394
395         }
396         }
397         return 0;
398 }
399
400 static int vfio_ccw_mdev_get_irq_info(struct vfio_irq_info *info)
401 {
402         switch (info->index) {
403         case VFIO_CCW_IO_IRQ_INDEX:
404         case VFIO_CCW_CRW_IRQ_INDEX:
405         case VFIO_CCW_REQ_IRQ_INDEX:
406                 info->count = 1;
407                 info->flags = VFIO_IRQ_INFO_EVENTFD;
408                 break;
409         default:
410                 return -EINVAL;
411         }
412
413         return 0;
414 }
415
416 static int vfio_ccw_mdev_set_irqs(struct vfio_ccw_private *private,
417                                   uint32_t flags,
418                                   uint32_t index,
419                                   void __user *data)
420 {
421         struct eventfd_ctx **ctx;
422
423         if (!(flags & VFIO_IRQ_SET_ACTION_TRIGGER))
424                 return -EINVAL;
425
426         switch (index) {
427         case VFIO_CCW_IO_IRQ_INDEX:
428                 ctx = &private->io_trigger;
429                 break;
430         case VFIO_CCW_CRW_IRQ_INDEX:
431                 ctx = &private->crw_trigger;
432                 break;
433         case VFIO_CCW_REQ_IRQ_INDEX:
434                 ctx = &private->req_trigger;
435                 break;
436         default:
437                 return -EINVAL;
438         }
439
440         switch (flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
441         case VFIO_IRQ_SET_DATA_NONE:
442         {
443                 if (*ctx)
444                         eventfd_signal(*ctx, 1);
445                 return 0;
446         }
447         case VFIO_IRQ_SET_DATA_BOOL:
448         {
449                 uint8_t trigger;
450
451                 if (get_user(trigger, (uint8_t __user *)data))
452                         return -EFAULT;
453
454                 if (trigger && *ctx)
455                         eventfd_signal(*ctx, 1);
456                 return 0;
457         }
458         case VFIO_IRQ_SET_DATA_EVENTFD:
459         {
460                 int32_t fd;
461
462                 if (get_user(fd, (int32_t __user *)data))
463                         return -EFAULT;
464
465                 if (fd == -1) {
466                         if (*ctx)
467                                 eventfd_ctx_put(*ctx);
468                         *ctx = NULL;
469                 } else if (fd >= 0) {
470                         struct eventfd_ctx *efdctx;
471
472                         efdctx = eventfd_ctx_fdget(fd);
473                         if (IS_ERR(efdctx))
474                                 return PTR_ERR(efdctx);
475
476                         if (*ctx)
477                                 eventfd_ctx_put(*ctx);
478
479                         *ctx = efdctx;
480                 } else
481                         return -EINVAL;
482
483                 return 0;
484         }
485         default:
486                 return -EINVAL;
487         }
488 }
489
490 int vfio_ccw_register_dev_region(struct vfio_ccw_private *private,
491                                  unsigned int subtype,
492                                  const struct vfio_ccw_regops *ops,
493                                  size_t size, u32 flags, void *data)
494 {
495         struct vfio_ccw_region *region;
496
497         region = krealloc(private->region,
498                           (private->num_regions + 1) * sizeof(*region),
499                           GFP_KERNEL);
500         if (!region)
501                 return -ENOMEM;
502
503         private->region = region;
504         private->region[private->num_regions].type = VFIO_REGION_TYPE_CCW;
505         private->region[private->num_regions].subtype = subtype;
506         private->region[private->num_regions].ops = ops;
507         private->region[private->num_regions].size = size;
508         private->region[private->num_regions].flags = flags;
509         private->region[private->num_regions].data = data;
510
511         private->num_regions++;
512
513         return 0;
514 }
515
516 void vfio_ccw_unregister_dev_regions(struct vfio_ccw_private *private)
517 {
518         int i;
519
520         for (i = 0; i < private->num_regions; i++)
521                 private->region[i].ops->release(private, &private->region[i]);
522         private->num_regions = 0;
523         kfree(private->region);
524         private->region = NULL;
525 }
526
527 static ssize_t vfio_ccw_mdev_ioctl(struct vfio_device *vdev,
528                                    unsigned int cmd,
529                                    unsigned long arg)
530 {
531         struct vfio_ccw_private *private =
532                 container_of(vdev, struct vfio_ccw_private, vdev);
533         int ret = 0;
534         unsigned long minsz;
535
536         switch (cmd) {
537         case VFIO_DEVICE_GET_INFO:
538         {
539                 struct vfio_device_info info;
540
541                 minsz = offsetofend(struct vfio_device_info, num_irqs);
542
543                 if (copy_from_user(&info, (void __user *)arg, minsz))
544                         return -EFAULT;
545
546                 if (info.argsz < minsz)
547                         return -EINVAL;
548
549                 ret = vfio_ccw_mdev_get_device_info(private, &info);
550                 if (ret)
551                         return ret;
552
553                 return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
554         }
555         case VFIO_DEVICE_GET_REGION_INFO:
556         {
557                 struct vfio_region_info info;
558
559                 minsz = offsetofend(struct vfio_region_info, offset);
560
561                 if (copy_from_user(&info, (void __user *)arg, minsz))
562                         return -EFAULT;
563
564                 if (info.argsz < minsz)
565                         return -EINVAL;
566
567                 ret = vfio_ccw_mdev_get_region_info(private, &info, arg);
568                 if (ret)
569                         return ret;
570
571                 return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
572         }
573         case VFIO_DEVICE_GET_IRQ_INFO:
574         {
575                 struct vfio_irq_info info;
576
577                 minsz = offsetofend(struct vfio_irq_info, count);
578
579                 if (copy_from_user(&info, (void __user *)arg, minsz))
580                         return -EFAULT;
581
582                 if (info.argsz < minsz || info.index >= VFIO_CCW_NUM_IRQS)
583                         return -EINVAL;
584
585                 ret = vfio_ccw_mdev_get_irq_info(&info);
586                 if (ret)
587                         return ret;
588
589                 if (info.count == -1)
590                         return -EINVAL;
591
592                 return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
593         }
594         case VFIO_DEVICE_SET_IRQS:
595         {
596                 struct vfio_irq_set hdr;
597                 size_t data_size;
598                 void __user *data;
599
600                 minsz = offsetofend(struct vfio_irq_set, count);
601
602                 if (copy_from_user(&hdr, (void __user *)arg, minsz))
603                         return -EFAULT;
604
605                 ret = vfio_set_irqs_validate_and_prepare(&hdr, 1,
606                                                          VFIO_CCW_NUM_IRQS,
607                                                          &data_size);
608                 if (ret)
609                         return ret;
610
611                 data = (void __user *)(arg + minsz);
612                 return vfio_ccw_mdev_set_irqs(private, hdr.flags, hdr.index,
613                                               data);
614         }
615         case VFIO_DEVICE_RESET:
616                 return vfio_ccw_mdev_reset(private);
617         default:
618                 return -ENOTTY;
619         }
620 }
621
622 /* Request removal of the device*/
623 static void vfio_ccw_mdev_request(struct vfio_device *vdev, unsigned int count)
624 {
625         struct vfio_ccw_private *private =
626                 container_of(vdev, struct vfio_ccw_private, vdev);
627         struct device *dev = vdev->dev;
628
629         if (private->req_trigger) {
630                 if (!(count % 10))
631                         dev_notice_ratelimited(dev,
632                                                "Relaying device request to user (#%u)\n",
633                                                count);
634
635                 eventfd_signal(private->req_trigger, 1);
636         } else if (count == 0) {
637                 dev_notice(dev,
638                            "No device request channel registered, blocked until released by user\n");
639         }
640 }
641
642 static const struct vfio_device_ops vfio_ccw_dev_ops = {
643         .open_device = vfio_ccw_mdev_open_device,
644         .close_device = vfio_ccw_mdev_close_device,
645         .read = vfio_ccw_mdev_read,
646         .write = vfio_ccw_mdev_write,
647         .ioctl = vfio_ccw_mdev_ioctl,
648         .request = vfio_ccw_mdev_request,
649 };
650
651 struct mdev_driver vfio_ccw_mdev_driver = {
652         .driver = {
653                 .name = "vfio_ccw_mdev",
654                 .owner = THIS_MODULE,
655                 .mod_name = KBUILD_MODNAME,
656         },
657         .probe = vfio_ccw_mdev_probe,
658         .remove = vfio_ccw_mdev_remove,
659 };
660
661 static const struct mdev_parent_ops vfio_ccw_mdev_ops = {
662         .owner                  = THIS_MODULE,
663         .device_driver          = &vfio_ccw_mdev_driver,
664         .supported_type_groups  = mdev_type_groups,
665 };
666
667 int vfio_ccw_mdev_reg(struct subchannel *sch)
668 {
669         return mdev_register_device(&sch->dev, &vfio_ccw_mdev_ops);
670 }
671
672 void vfio_ccw_mdev_unreg(struct subchannel *sch)
673 {
674         mdev_unregister_device(&sch->dev);
675 }