1 // SPDX-License-Identifier: GPL-2.0
3 * Physical device callbacks for vfio_ccw
5 * Copyright IBM Corp. 2017
6 * Copyright Red Hat, Inc. 2019
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>
13 #include <linux/vfio.h>
14 #include <linux/mdev.h>
15 #include <linux/nospec.h>
16 #include <linux/slab.h>
18 #include "vfio_ccw_private.h"
20 static int vfio_ccw_mdev_reset(struct mdev_device *mdev)
22 struct vfio_ccw_private *private;
23 struct subchannel *sch;
26 private = dev_get_drvdata(mdev_parent_dev(mdev));
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.
36 ret = vfio_ccw_sch_quiesce(sch);
40 ret = cio_enable_subchannel(sch, (u32)(unsigned long)sch);
42 private->state = VFIO_CCW_STATE_IDLE;
47 static int vfio_ccw_mdev_notifier(struct notifier_block *nb,
51 struct vfio_ccw_private *private =
52 container_of(nb, struct vfio_ccw_private, nb);
55 * Vendor drivers MUST unpin pages in response to an
58 if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
59 struct vfio_iommu_type1_dma_unmap *unmap = data;
61 if (!cp_iova_pinned(&private->cp, unmap->iova))
64 if (vfio_ccw_mdev_reset(private->mdev))
67 cp_free(&private->cp);
74 static ssize_t name_show(struct kobject *kobj, struct device *dev, char *buf)
76 return sprintf(buf, "I/O subchannel (Non-QDIO)\n");
78 static MDEV_TYPE_ATTR_RO(name);
80 static ssize_t device_api_show(struct kobject *kobj, struct device *dev,
83 return sprintf(buf, "%s\n", VFIO_DEVICE_API_CCW_STRING);
85 static MDEV_TYPE_ATTR_RO(device_api);
87 static ssize_t available_instances_show(struct kobject *kobj,
88 struct device *dev, char *buf)
90 struct vfio_ccw_private *private = dev_get_drvdata(dev);
92 return sprintf(buf, "%d\n", atomic_read(&private->avail));
94 static MDEV_TYPE_ATTR_RO(available_instances);
96 static struct attribute *mdev_types_attrs[] = {
97 &mdev_type_attr_name.attr,
98 &mdev_type_attr_device_api.attr,
99 &mdev_type_attr_available_instances.attr,
103 static struct attribute_group mdev_type_group = {
105 .attrs = mdev_types_attrs,
108 static struct attribute_group *mdev_type_groups[] = {
113 static int vfio_ccw_mdev_create(struct kobject *kobj, struct mdev_device *mdev)
115 struct vfio_ccw_private *private =
116 dev_get_drvdata(mdev_parent_dev(mdev));
118 if (private->state == VFIO_CCW_STATE_NOT_OPER)
121 if (atomic_dec_if_positive(&private->avail) < 0)
124 private->mdev = mdev;
125 private->state = VFIO_CCW_STATE_IDLE;
127 VFIO_CCW_MSG_EVENT(2, "mdev %pUl, sch %x.%x.%04x: create\n",
128 mdev_uuid(mdev), private->sch->schid.cssid,
129 private->sch->schid.ssid,
130 private->sch->schid.sch_no);
135 static int vfio_ccw_mdev_remove(struct mdev_device *mdev)
137 struct vfio_ccw_private *private =
138 dev_get_drvdata(mdev_parent_dev(mdev));
140 VFIO_CCW_MSG_EVENT(2, "mdev %pUl, sch %x.%x.%04x: remove\n",
141 mdev_uuid(mdev), private->sch->schid.cssid,
142 private->sch->schid.ssid,
143 private->sch->schid.sch_no);
145 if ((private->state != VFIO_CCW_STATE_NOT_OPER) &&
146 (private->state != VFIO_CCW_STATE_STANDBY)) {
147 if (!vfio_ccw_sch_quiesce(private->sch))
148 private->state = VFIO_CCW_STATE_STANDBY;
149 /* The state will be NOT_OPER on error. */
152 cp_free(&private->cp);
153 private->mdev = NULL;
154 atomic_inc(&private->avail);
159 static int vfio_ccw_mdev_open(struct mdev_device *mdev)
161 struct vfio_ccw_private *private =
162 dev_get_drvdata(mdev_parent_dev(mdev));
163 unsigned long events = VFIO_IOMMU_NOTIFY_DMA_UNMAP;
166 private->nb.notifier_call = vfio_ccw_mdev_notifier;
168 ret = vfio_register_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
169 &events, &private->nb);
173 ret = vfio_ccw_register_async_dev_regions(private);
175 vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
180 static void vfio_ccw_mdev_release(struct mdev_device *mdev)
182 struct vfio_ccw_private *private =
183 dev_get_drvdata(mdev_parent_dev(mdev));
186 if ((private->state != VFIO_CCW_STATE_NOT_OPER) &&
187 (private->state != VFIO_CCW_STATE_STANDBY)) {
188 if (!vfio_ccw_mdev_reset(mdev))
189 private->state = VFIO_CCW_STATE_STANDBY;
190 /* The state will be NOT_OPER on error. */
193 cp_free(&private->cp);
194 vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
197 for (i = 0; i < private->num_regions; i++)
198 private->region[i].ops->release(private, &private->region[i]);
200 private->num_regions = 0;
201 kfree(private->region);
202 private->region = NULL;
205 static ssize_t vfio_ccw_mdev_read_io_region(struct vfio_ccw_private *private,
206 char __user *buf, size_t count,
209 loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
210 struct ccw_io_region *region;
213 if (pos + count > sizeof(*region))
216 mutex_lock(&private->io_mutex);
217 region = private->io_region;
218 if (copy_to_user(buf, (void *)region + pos, count))
222 mutex_unlock(&private->io_mutex);
226 static ssize_t vfio_ccw_mdev_read(struct mdev_device *mdev,
231 unsigned int index = VFIO_CCW_OFFSET_TO_INDEX(*ppos);
232 struct vfio_ccw_private *private;
234 private = dev_get_drvdata(mdev_parent_dev(mdev));
236 if (index >= VFIO_CCW_NUM_REGIONS + private->num_regions)
240 case VFIO_CCW_CONFIG_REGION_INDEX:
241 return vfio_ccw_mdev_read_io_region(private, buf, count, ppos);
243 index -= VFIO_CCW_NUM_REGIONS;
244 return private->region[index].ops->read(private, buf, count,
251 static ssize_t vfio_ccw_mdev_write_io_region(struct vfio_ccw_private *private,
252 const char __user *buf,
253 size_t count, loff_t *ppos)
255 loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
256 struct ccw_io_region *region;
259 if (pos + count > sizeof(*region))
262 if (!mutex_trylock(&private->io_mutex))
265 region = private->io_region;
266 if (copy_from_user((void *)region + pos, buf, count)) {
271 vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_IO_REQ);
272 if (region->ret_code != 0)
273 private->state = VFIO_CCW_STATE_IDLE;
274 ret = (region->ret_code != 0) ? region->ret_code : count;
277 mutex_unlock(&private->io_mutex);
281 static ssize_t vfio_ccw_mdev_write(struct mdev_device *mdev,
282 const char __user *buf,
286 unsigned int index = VFIO_CCW_OFFSET_TO_INDEX(*ppos);
287 struct vfio_ccw_private *private;
289 private = dev_get_drvdata(mdev_parent_dev(mdev));
291 if (index >= VFIO_CCW_NUM_REGIONS + private->num_regions)
295 case VFIO_CCW_CONFIG_REGION_INDEX:
296 return vfio_ccw_mdev_write_io_region(private, buf, count, ppos);
298 index -= VFIO_CCW_NUM_REGIONS;
299 return private->region[index].ops->write(private, buf, count,
306 static int vfio_ccw_mdev_get_device_info(struct vfio_device_info *info,
307 struct mdev_device *mdev)
309 struct vfio_ccw_private *private;
311 private = dev_get_drvdata(mdev_parent_dev(mdev));
312 info->flags = VFIO_DEVICE_FLAGS_CCW | VFIO_DEVICE_FLAGS_RESET;
313 info->num_regions = VFIO_CCW_NUM_REGIONS + private->num_regions;
314 info->num_irqs = VFIO_CCW_NUM_IRQS;
319 static int vfio_ccw_mdev_get_region_info(struct vfio_region_info *info,
320 struct mdev_device *mdev,
323 struct vfio_ccw_private *private;
326 private = dev_get_drvdata(mdev_parent_dev(mdev));
327 switch (info->index) {
328 case VFIO_CCW_CONFIG_REGION_INDEX:
330 info->size = sizeof(struct ccw_io_region);
331 info->flags = VFIO_REGION_INFO_FLAG_READ
332 | VFIO_REGION_INFO_FLAG_WRITE;
334 default: /* all other regions are handled via capability chain */
336 struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
337 struct vfio_region_info_cap_type cap_type = {
338 .header.id = VFIO_REGION_INFO_CAP_TYPE,
339 .header.version = 1 };
343 VFIO_CCW_NUM_REGIONS + private->num_regions)
346 info->index = array_index_nospec(info->index,
347 VFIO_CCW_NUM_REGIONS +
348 private->num_regions);
350 i = info->index - VFIO_CCW_NUM_REGIONS;
352 info->offset = VFIO_CCW_INDEX_TO_OFFSET(info->index);
353 info->size = private->region[i].size;
354 info->flags = private->region[i].flags;
356 cap_type.type = private->region[i].type;
357 cap_type.subtype = private->region[i].subtype;
359 ret = vfio_info_add_capability(&caps, &cap_type.header,
364 info->flags |= VFIO_REGION_INFO_FLAG_CAPS;
365 if (info->argsz < sizeof(*info) + caps.size) {
366 info->argsz = sizeof(*info) + caps.size;
367 info->cap_offset = 0;
369 vfio_info_cap_shift(&caps, sizeof(*info));
370 if (copy_to_user((void __user *)arg + sizeof(*info),
371 caps.buf, caps.size)) {
375 info->cap_offset = sizeof(*info);
385 static int vfio_ccw_mdev_get_irq_info(struct vfio_irq_info *info)
387 if (info->index != VFIO_CCW_IO_IRQ_INDEX)
391 info->flags = VFIO_IRQ_INFO_EVENTFD;
396 static int vfio_ccw_mdev_set_irqs(struct mdev_device *mdev,
400 struct vfio_ccw_private *private;
401 struct eventfd_ctx **ctx;
403 if (!(flags & VFIO_IRQ_SET_ACTION_TRIGGER))
406 private = dev_get_drvdata(mdev_parent_dev(mdev));
407 ctx = &private->io_trigger;
409 switch (flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
410 case VFIO_IRQ_SET_DATA_NONE:
413 eventfd_signal(*ctx, 1);
416 case VFIO_IRQ_SET_DATA_BOOL:
420 if (get_user(trigger, (uint8_t __user *)data))
424 eventfd_signal(*ctx, 1);
427 case VFIO_IRQ_SET_DATA_EVENTFD:
431 if (get_user(fd, (int32_t __user *)data))
436 eventfd_ctx_put(*ctx);
438 } else if (fd >= 0) {
439 struct eventfd_ctx *efdctx;
441 efdctx = eventfd_ctx_fdget(fd);
443 return PTR_ERR(efdctx);
446 eventfd_ctx_put(*ctx);
459 int vfio_ccw_register_dev_region(struct vfio_ccw_private *private,
460 unsigned int subtype,
461 const struct vfio_ccw_regops *ops,
462 size_t size, u32 flags, void *data)
464 struct vfio_ccw_region *region;
466 region = krealloc(private->region,
467 (private->num_regions + 1) * sizeof(*region),
472 private->region = region;
473 private->region[private->num_regions].type = VFIO_REGION_TYPE_CCW;
474 private->region[private->num_regions].subtype = subtype;
475 private->region[private->num_regions].ops = ops;
476 private->region[private->num_regions].size = size;
477 private->region[private->num_regions].flags = flags;
478 private->region[private->num_regions].data = data;
480 private->num_regions++;
485 static ssize_t vfio_ccw_mdev_ioctl(struct mdev_device *mdev,
493 case VFIO_DEVICE_GET_INFO:
495 struct vfio_device_info info;
497 minsz = offsetofend(struct vfio_device_info, num_irqs);
499 if (copy_from_user(&info, (void __user *)arg, minsz))
502 if (info.argsz < minsz)
505 ret = vfio_ccw_mdev_get_device_info(&info, mdev);
509 return copy_to_user((void __user *)arg, &info, minsz);
511 case VFIO_DEVICE_GET_REGION_INFO:
513 struct vfio_region_info info;
515 minsz = offsetofend(struct vfio_region_info, offset);
517 if (copy_from_user(&info, (void __user *)arg, minsz))
520 if (info.argsz < minsz)
523 ret = vfio_ccw_mdev_get_region_info(&info, mdev, arg);
527 return copy_to_user((void __user *)arg, &info, minsz);
529 case VFIO_DEVICE_GET_IRQ_INFO:
531 struct vfio_irq_info info;
533 minsz = offsetofend(struct vfio_irq_info, count);
535 if (copy_from_user(&info, (void __user *)arg, minsz))
538 if (info.argsz < minsz || info.index >= VFIO_CCW_NUM_IRQS)
541 ret = vfio_ccw_mdev_get_irq_info(&info);
545 if (info.count == -1)
548 return copy_to_user((void __user *)arg, &info, minsz);
550 case VFIO_DEVICE_SET_IRQS:
552 struct vfio_irq_set hdr;
556 minsz = offsetofend(struct vfio_irq_set, count);
558 if (copy_from_user(&hdr, (void __user *)arg, minsz))
561 ret = vfio_set_irqs_validate_and_prepare(&hdr, 1,
567 data = (void __user *)(arg + minsz);
568 return vfio_ccw_mdev_set_irqs(mdev, hdr.flags, data);
570 case VFIO_DEVICE_RESET:
571 return vfio_ccw_mdev_reset(mdev);
577 static const struct mdev_parent_ops vfio_ccw_mdev_ops = {
578 .owner = THIS_MODULE,
579 .supported_type_groups = mdev_type_groups,
580 .create = vfio_ccw_mdev_create,
581 .remove = vfio_ccw_mdev_remove,
582 .open = vfio_ccw_mdev_open,
583 .release = vfio_ccw_mdev_release,
584 .read = vfio_ccw_mdev_read,
585 .write = vfio_ccw_mdev_write,
586 .ioctl = vfio_ccw_mdev_ioctl,
589 int vfio_ccw_mdev_reg(struct subchannel *sch)
591 return mdev_register_device(&sch->dev, &vfio_ccw_mdev_ops);
594 void vfio_ccw_mdev_unreg(struct subchannel *sch)
596 mdev_unregister_device(&sch->dev);