1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/libnvdimm.h>
7 #include <linux/sched/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/uaccess.h>
10 #include <linux/module.h>
11 #include <linux/blkdev.h>
12 #include <linux/fcntl.h>
13 #include <linux/async.h>
14 #include <linux/genhd.h>
15 #include <linux/ndctl.h>
16 #include <linux/sched.h>
17 #include <linux/slab.h>
18 #include <linux/cpu.h>
28 static int nvdimm_bus_major;
29 struct class *nd_class;
30 static DEFINE_IDA(nd_ida);
32 static int to_nd_device_type(struct device *dev)
35 return ND_DEVICE_DIMM;
36 else if (is_memory(dev))
37 return ND_DEVICE_REGION_PMEM;
38 else if (is_nd_blk(dev))
39 return ND_DEVICE_REGION_BLK;
40 else if (is_nd_dax(dev))
41 return ND_DEVICE_DAX_PMEM;
42 else if (is_nd_region(dev->parent))
43 return nd_region_to_nstype(to_nd_region(dev->parent));
48 static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
50 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
51 to_nd_device_type(dev));
54 static struct module *to_bus_provider(struct device *dev)
56 /* pin bus providers while regions are enabled */
57 if (is_nd_region(dev)) {
58 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
60 return nvdimm_bus->nd_desc->module;
65 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
67 nvdimm_bus_lock(&nvdimm_bus->dev);
68 nvdimm_bus->probe_active++;
69 nvdimm_bus_unlock(&nvdimm_bus->dev);
72 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
74 nvdimm_bus_lock(&nvdimm_bus->dev);
75 if (--nvdimm_bus->probe_active == 0)
76 wake_up(&nvdimm_bus->wait);
77 nvdimm_bus_unlock(&nvdimm_bus->dev);
80 static int nvdimm_bus_probe(struct device *dev)
82 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
83 struct module *provider = to_bus_provider(dev);
84 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
87 if (!try_module_get(provider))
90 dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n",
91 dev->driver->name, dev_name(dev));
93 nvdimm_bus_probe_start(nvdimm_bus);
94 debug_nvdimm_lock(dev);
95 rc = nd_drv->probe(dev);
96 debug_nvdimm_unlock(dev);
98 if ((rc == 0 || rc == -EOPNOTSUPP) &&
99 dev->parent && is_nd_region(dev->parent))
100 nd_region_advance_seeds(to_nd_region(dev->parent), dev);
101 nvdimm_bus_probe_end(nvdimm_bus);
103 dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name,
107 module_put(provider);
111 static int nvdimm_bus_remove(struct device *dev)
113 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
114 struct module *provider = to_bus_provider(dev);
115 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
118 if (nd_drv->remove) {
119 debug_nvdimm_lock(dev);
120 rc = nd_drv->remove(dev);
121 debug_nvdimm_unlock(dev);
124 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
126 module_put(provider);
130 static void nvdimm_bus_shutdown(struct device *dev)
132 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
133 struct nd_device_driver *nd_drv = NULL;
136 nd_drv = to_nd_device_driver(dev->driver);
138 if (nd_drv && nd_drv->shutdown) {
139 nd_drv->shutdown(dev);
140 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
141 dev->driver->name, dev_name(dev));
145 void nd_device_notify(struct device *dev, enum nvdimm_event event)
149 struct nd_device_driver *nd_drv;
151 nd_drv = to_nd_device_driver(dev->driver);
153 nd_drv->notify(dev, event);
155 nd_device_unlock(dev);
157 EXPORT_SYMBOL(nd_device_notify);
159 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
161 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
166 /* caller is responsible for holding a reference on the device */
167 nd_device_notify(&nd_region->dev, event);
169 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
171 struct clear_badblocks_context {
172 resource_size_t phys, cleared;
175 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
177 struct clear_badblocks_context *ctx = data;
178 struct nd_region *nd_region;
179 resource_size_t ndr_end;
182 /* make sure device is a region */
186 nd_region = to_nd_region(dev);
187 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
189 /* make sure we are in the region */
190 if (ctx->phys < nd_region->ndr_start
191 || (ctx->phys + ctx->cleared) > ndr_end)
194 sector = (ctx->phys - nd_region->ndr_start) / 512;
195 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
197 if (nd_region->bb_state)
198 sysfs_notify_dirent(nd_region->bb_state);
203 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
204 phys_addr_t phys, u64 cleared)
206 struct clear_badblocks_context ctx = {
211 device_for_each_child(&nvdimm_bus->dev, &ctx,
212 nvdimm_clear_badblocks_region);
215 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
216 phys_addr_t phys, u64 cleared)
219 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
221 if (cleared > 0 && cleared / 512)
222 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
225 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
228 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
229 struct nvdimm_bus_descriptor *nd_desc;
230 struct nd_cmd_clear_error clear_err;
231 struct nd_cmd_ars_cap ars_cap;
232 u32 clear_err_unit, mask;
233 unsigned int noio_flag;
239 nd_desc = nvdimm_bus->nd_desc;
241 * if ndctl does not exist, it's PMEM_LEGACY and
242 * we want to just pretend everything is handled.
247 memset(&ars_cap, 0, sizeof(ars_cap));
248 ars_cap.address = phys;
249 ars_cap.length = len;
250 noio_flag = memalloc_noio_save();
251 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
252 sizeof(ars_cap), &cmd_rc);
253 memalloc_noio_restore(noio_flag);
258 clear_err_unit = ars_cap.clear_err_unit;
259 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
262 mask = clear_err_unit - 1;
263 if ((phys | len) & mask)
265 memset(&clear_err, 0, sizeof(clear_err));
266 clear_err.address = phys;
267 clear_err.length = len;
268 noio_flag = memalloc_noio_save();
269 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
270 sizeof(clear_err), &cmd_rc);
271 memalloc_noio_restore(noio_flag);
277 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
279 return clear_err.cleared;
281 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
283 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
285 static struct bus_type nvdimm_bus_type = {
287 .uevent = nvdimm_bus_uevent,
288 .match = nvdimm_bus_match,
289 .probe = nvdimm_bus_probe,
290 .remove = nvdimm_bus_remove,
291 .shutdown = nvdimm_bus_shutdown,
294 static void nvdimm_bus_release(struct device *dev)
296 struct nvdimm_bus *nvdimm_bus;
298 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
299 ida_simple_remove(&nd_ida, nvdimm_bus->id);
303 bool is_nvdimm_bus(struct device *dev)
305 return dev->release == nvdimm_bus_release;
308 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
312 for (dev = nd_dev; dev; dev = dev->parent)
313 if (is_nvdimm_bus(dev))
315 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
317 return to_nvdimm_bus(dev);
321 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
323 struct nvdimm_bus *nvdimm_bus;
325 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
326 WARN_ON(!is_nvdimm_bus(dev));
329 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
331 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
333 return to_nvdimm_bus(nvdimm->dev.parent);
335 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
337 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
338 struct nvdimm_bus_descriptor *nd_desc)
340 struct nvdimm_bus *nvdimm_bus;
343 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
346 INIT_LIST_HEAD(&nvdimm_bus->list);
347 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
348 init_waitqueue_head(&nvdimm_bus->wait);
349 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
350 if (nvdimm_bus->id < 0) {
354 mutex_init(&nvdimm_bus->reconfig_mutex);
355 badrange_init(&nvdimm_bus->badrange);
356 nvdimm_bus->nd_desc = nd_desc;
357 nvdimm_bus->dev.parent = parent;
358 nvdimm_bus->dev.release = nvdimm_bus_release;
359 nvdimm_bus->dev.groups = nd_desc->attr_groups;
360 nvdimm_bus->dev.bus = &nvdimm_bus_type;
361 nvdimm_bus->dev.of_node = nd_desc->of_node;
362 dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
363 rc = device_register(&nvdimm_bus->dev);
365 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
371 put_device(&nvdimm_bus->dev);
374 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
376 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
380 device_unregister(&nvdimm_bus->dev);
382 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
384 static int child_unregister(struct device *dev, void *data)
387 * the singular ndctl class device per bus needs to be
388 * "device_destroy"ed, so skip it here
390 * i.e. remove classless children
395 if (is_nvdimm(dev)) {
396 struct nvdimm *nvdimm = to_nvdimm(dev);
397 bool dev_put = false;
399 /* We are shutting down. Make state frozen artificially. */
400 nvdimm_bus_lock(dev);
401 set_bit(NVDIMM_SECURITY_FROZEN, &nvdimm->sec.flags);
402 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
404 nvdimm_bus_unlock(dev);
405 cancel_delayed_work_sync(&nvdimm->dwork);
409 nd_device_unregister(dev, ND_SYNC);
414 static void free_badrange_list(struct list_head *badrange_list)
416 struct badrange_entry *bre, *next;
418 list_for_each_entry_safe(bre, next, badrange_list, list) {
419 list_del(&bre->list);
422 list_del_init(badrange_list);
425 static int nd_bus_remove(struct device *dev)
427 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
429 mutex_lock(&nvdimm_bus_list_mutex);
430 list_del_init(&nvdimm_bus->list);
431 mutex_unlock(&nvdimm_bus_list_mutex);
433 wait_event(nvdimm_bus->wait,
434 atomic_read(&nvdimm_bus->ioctl_active) == 0);
437 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
439 spin_lock(&nvdimm_bus->badrange.lock);
440 free_badrange_list(&nvdimm_bus->badrange.list);
441 spin_unlock(&nvdimm_bus->badrange.lock);
443 nvdimm_bus_destroy_ndctl(nvdimm_bus);
448 static int nd_bus_probe(struct device *dev)
450 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
453 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
457 mutex_lock(&nvdimm_bus_list_mutex);
458 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
459 mutex_unlock(&nvdimm_bus_list_mutex);
461 /* enable bus provider attributes to look up their local context */
462 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
467 static struct nd_device_driver nd_bus_driver = {
468 .probe = nd_bus_probe,
469 .remove = nd_bus_remove,
472 .suppress_bind_attrs = true,
473 .bus = &nvdimm_bus_type,
474 .owner = THIS_MODULE,
475 .mod_name = KBUILD_MODNAME,
479 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
481 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
483 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
486 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
489 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
491 void nd_synchronize(void)
493 async_synchronize_full_domain(&nd_async_domain);
495 EXPORT_SYMBOL_GPL(nd_synchronize);
497 static void nd_async_device_register(void *d, async_cookie_t cookie)
499 struct device *dev = d;
501 if (device_add(dev) != 0) {
502 dev_err(dev, "%s: failed\n", __func__);
507 put_device(dev->parent);
510 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
512 struct device *dev = d;
514 /* flush bus operations before delete */
515 nvdimm_bus_lock(dev);
516 nvdimm_bus_unlock(dev);
518 device_unregister(dev);
522 void __nd_device_register(struct device *dev)
528 * Ensure that region devices always have their NUMA node set as
529 * early as possible. This way we are able to make certain that
530 * any memory associated with the creation and the creation
531 * itself of the region is associated with the correct node.
533 if (is_nd_region(dev))
534 set_dev_node(dev, to_nd_region(dev)->numa_node);
536 dev->bus = &nvdimm_bus_type;
538 get_device(dev->parent);
539 if (dev_to_node(dev) == NUMA_NO_NODE)
540 set_dev_node(dev, dev_to_node(dev->parent));
544 async_schedule_dev_domain(nd_async_device_register, dev,
548 void nd_device_register(struct device *dev)
550 device_initialize(dev);
551 __nd_device_register(dev);
553 EXPORT_SYMBOL(nd_device_register);
555 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
562 * In the async case this is being triggered with the
563 * device lock held and the unregistration work needs to
564 * be moved out of line iff this is thread has won the
565 * race to schedule the deletion.
567 if (!kill_device(dev))
571 async_schedule_domain(nd_async_device_unregister, dev,
576 * In the sync case the device is being unregistered due
577 * to a state change of the parent. Claim the kill state
578 * to synchronize against other unregistration requests,
579 * or otherwise let the async path handle it if the
580 * unregistration was already queued.
583 killed = kill_device(dev);
584 nd_device_unlock(dev);
590 device_unregister(dev);
594 EXPORT_SYMBOL(nd_device_unregister);
597 * __nd_driver_register() - register a region or a namespace driver
598 * @nd_drv: driver to register
599 * @owner: automatically set by nd_driver_register() macro
600 * @mod_name: automatically set by nd_driver_register() macro
602 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
603 const char *mod_name)
605 struct device_driver *drv = &nd_drv->drv;
608 pr_debug("driver type bitmask not set (%ps)\n",
609 __builtin_return_address(0));
613 if (!nd_drv->probe) {
614 pr_debug("%s ->probe() must be specified\n", mod_name);
618 drv->bus = &nvdimm_bus_type;
620 drv->mod_name = mod_name;
622 return driver_register(drv);
624 EXPORT_SYMBOL(__nd_driver_register);
626 int nvdimm_revalidate_disk(struct gendisk *disk)
628 struct device *dev = disk_to_dev(disk)->parent;
629 struct nd_region *nd_region = to_nd_region(dev->parent);
630 int disk_ro = get_disk_ro(disk);
633 * Upgrade to read-only if the region is read-only preserve as
634 * read-only if the disk is already read-only.
636 if (disk_ro || nd_region->ro == disk_ro)
639 dev_info(dev, "%s read-only, marking %s read-only\n",
640 dev_name(&nd_region->dev), disk->disk_name);
641 set_disk_ro(disk, 1);
646 EXPORT_SYMBOL(nvdimm_revalidate_disk);
648 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
651 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
652 to_nd_device_type(dev));
654 static DEVICE_ATTR_RO(modalias);
656 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
659 return sprintf(buf, "%s\n", dev->type->name);
661 static DEVICE_ATTR_RO(devtype);
663 static struct attribute *nd_device_attributes[] = {
664 &dev_attr_modalias.attr,
665 &dev_attr_devtype.attr,
670 * nd_device_attribute_group - generic attributes for all devices on an nd bus
672 const struct attribute_group nd_device_attribute_group = {
673 .attrs = nd_device_attributes,
676 static ssize_t numa_node_show(struct device *dev,
677 struct device_attribute *attr, char *buf)
679 return sprintf(buf, "%d\n", dev_to_node(dev));
681 static DEVICE_ATTR_RO(numa_node);
683 static struct attribute *nd_numa_attributes[] = {
684 &dev_attr_numa_node.attr,
688 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
691 if (!IS_ENABLED(CONFIG_NUMA))
698 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
700 struct attribute_group nd_numa_attribute_group = {
701 .attrs = nd_numa_attributes,
702 .is_visible = nd_numa_attr_visible,
704 EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
706 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
708 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
711 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
712 "ndctl%d", nvdimm_bus->id);
715 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
716 nvdimm_bus->id, PTR_ERR(dev));
717 return PTR_ERR_OR_ZERO(dev);
720 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
722 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
725 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
726 [ND_CMD_IMPLEMENTED] = { },
729 .out_sizes = { 4, 128, },
731 [ND_CMD_SMART_THRESHOLD] = {
733 .out_sizes = { 4, 8, },
735 [ND_CMD_DIMM_FLAGS] = {
737 .out_sizes = { 4, 4 },
739 [ND_CMD_GET_CONFIG_SIZE] = {
741 .out_sizes = { 4, 4, 4, },
743 [ND_CMD_GET_CONFIG_DATA] = {
745 .in_sizes = { 4, 4, },
747 .out_sizes = { 4, UINT_MAX, },
749 [ND_CMD_SET_CONFIG_DATA] = {
751 .in_sizes = { 4, 4, UINT_MAX, },
757 .in_sizes = { 4, 4, UINT_MAX, },
759 .out_sizes = { 4, 4, UINT_MAX, },
763 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
765 .out_sizes = { UINT_MAX, },
769 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
771 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
772 return &__nd_cmd_dimm_descs[cmd];
775 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
777 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
778 [ND_CMD_IMPLEMENTED] = { },
781 .in_sizes = { 8, 8, },
783 .out_sizes = { 4, 4, 4, 4, },
785 [ND_CMD_ARS_START] = {
787 .in_sizes = { 8, 8, 2, 1, 5, },
789 .out_sizes = { 4, 4, },
791 [ND_CMD_ARS_STATUS] = {
793 .out_sizes = { 4, 4, UINT_MAX, },
795 [ND_CMD_CLEAR_ERROR] = {
797 .in_sizes = { 8, 8, },
799 .out_sizes = { 4, 4, 8, },
803 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
805 .out_sizes = { UINT_MAX, },
809 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
811 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
812 return &__nd_cmd_bus_descs[cmd];
815 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
817 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
818 const struct nd_cmd_desc *desc, int idx, void *buf)
820 if (idx >= desc->in_num)
823 if (desc->in_sizes[idx] < UINT_MAX)
824 return desc->in_sizes[idx];
826 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
827 struct nd_cmd_set_config_hdr *hdr = buf;
829 return hdr->in_length;
830 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
831 struct nd_cmd_vendor_hdr *hdr = buf;
833 return hdr->in_length;
834 } else if (cmd == ND_CMD_CALL) {
835 struct nd_cmd_pkg *pkg = buf;
837 return pkg->nd_size_in;
842 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
844 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
845 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
846 const u32 *out_field, unsigned long remainder)
848 if (idx >= desc->out_num)
851 if (desc->out_sizes[idx] < UINT_MAX)
852 return desc->out_sizes[idx];
854 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
856 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
858 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
860 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
861 * "Size of Output Buffer in bytes, including this
864 if (out_field[1] < 4)
867 * ACPI 6.1 is ambiguous if 'status' is included in the
868 * output size. If we encounter an output size that
869 * overshoots the remainder by 4 bytes, assume it was
870 * including 'status'.
872 if (out_field[1] - 4 == remainder)
874 return out_field[1] - 8;
875 } else if (cmd == ND_CMD_CALL) {
876 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
878 return pkg->nd_size_out;
884 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
886 void wait_nvdimm_bus_probe_idle(struct device *dev)
888 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
891 if (nvdimm_bus->probe_active == 0)
893 nvdimm_bus_unlock(dev);
894 nd_device_unlock(dev);
895 wait_event(nvdimm_bus->wait,
896 nvdimm_bus->probe_active == 0);
898 nvdimm_bus_lock(dev);
902 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
904 struct nd_cmd_clear_error *clear_err =
905 (struct nd_cmd_clear_error *)data;
906 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
907 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
908 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
909 struct nd_namespace_common *ndns = NULL;
910 struct nd_namespace_io *nsio;
911 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
913 if (nd_dax || !dev->driver)
916 start = clear_err->address;
917 end = clear_err->address + clear_err->cleared - 1;
919 if (nd_btt || nd_pfn || nd_dax) {
925 ndns = nd_dax->nd_pfn.ndns;
932 nsio = to_nd_namespace_io(&ndns->dev);
933 pstart = nsio->res.start + offset;
934 pend = nsio->res.end - end_trunc;
936 if ((pstart >= start) && (pend <= end))
943 static int nd_ns_forget_poison_check(struct device *dev, void *data)
945 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
948 /* set_config requires an idle interleave set */
949 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
950 struct nvdimm *nvdimm, unsigned int cmd, void *data)
952 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
954 /* ask the bus provider if it would like to block this request */
955 if (nd_desc->clear_to_send) {
956 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
962 /* require clear error to go through the pmem driver */
963 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
964 return device_for_each_child(&nvdimm_bus->dev, data,
965 nd_ns_forget_poison_check);
967 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
970 /* prevent label manipulation while the kernel owns label updates */
971 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
972 if (atomic_read(&nvdimm->busy))
977 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
978 int read_only, unsigned int ioctl_cmd, unsigned long arg)
980 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
981 const struct nd_cmd_desc *desc = NULL;
982 unsigned int cmd = _IOC_NR(ioctl_cmd);
983 struct device *dev = &nvdimm_bus->dev;
984 void __user *p = (void __user *) arg;
985 char *out_env = NULL, *in_env = NULL;
986 const char *cmd_name, *dimm_name;
987 u32 in_len = 0, out_len = 0;
988 unsigned int func = cmd;
989 unsigned long cmd_mask;
990 struct nd_cmd_pkg pkg;
996 desc = nd_cmd_dimm_desc(cmd);
997 cmd_name = nvdimm_cmd_name(cmd);
998 cmd_mask = nvdimm->cmd_mask;
999 dimm_name = dev_name(&nvdimm->dev);
1001 desc = nd_cmd_bus_desc(cmd);
1002 cmd_name = nvdimm_bus_cmd_name(cmd);
1003 cmd_mask = nd_desc->cmd_mask;
1007 if (cmd == ND_CMD_CALL) {
1008 if (copy_from_user(&pkg, p, sizeof(pkg)))
1012 if (!desc || (desc->out_num + desc->in_num == 0) ||
1013 !test_bit(cmd, &cmd_mask))
1016 /* fail write commands (when read-only) */
1020 case ND_CMD_SET_CONFIG_DATA:
1021 case ND_CMD_ARS_START:
1022 case ND_CMD_CLEAR_ERROR:
1024 dev_dbg(dev, "'%s' command while read-only.\n",
1025 nvdimm ? nvdimm_cmd_name(cmd)
1026 : nvdimm_bus_cmd_name(cmd));
1032 /* process an input envelope */
1033 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1036 for (i = 0; i < desc->in_num; i++) {
1039 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1040 if (in_size == UINT_MAX) {
1041 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1042 __func__, dimm_name, cmd_name, i);
1046 if (in_len < ND_CMD_MAX_ENVELOPE)
1047 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1050 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1057 if (cmd == ND_CMD_CALL) {
1058 func = pkg.nd_command;
1059 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1060 dimm_name, pkg.nd_command,
1061 in_len, out_len, buf_len);
1064 /* process an output envelope */
1065 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1071 for (i = 0; i < desc->out_num; i++) {
1072 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1073 (u32 *) in_env, (u32 *) out_env, 0);
1076 if (out_size == UINT_MAX) {
1077 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1078 dimm_name, cmd_name, i);
1082 if (out_len < ND_CMD_MAX_ENVELOPE)
1083 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1086 if (copy && copy_from_user(&out_env[out_len],
1087 p + in_len + out_len, copy)) {
1091 out_len += out_size;
1094 buf_len = (u64) out_len + (u64) in_len;
1095 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1096 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1097 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1102 buf = vmalloc(buf_len);
1108 if (copy_from_user(buf, p, buf_len)) {
1113 nd_device_lock(dev);
1114 nvdimm_bus_lock(dev);
1115 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1119 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1123 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1124 struct nd_cmd_clear_error *clear_err = buf;
1126 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1127 clear_err->cleared);
1130 if (copy_to_user(p, buf, buf_len))
1134 nvdimm_bus_unlock(dev);
1135 nd_device_unlock(dev);
1143 enum nd_ioctl_mode {
1148 static int match_dimm(struct device *dev, void *data)
1150 long id = (long) data;
1152 if (is_nvdimm(dev)) {
1153 struct nvdimm *nvdimm = to_nvdimm(dev);
1155 return nvdimm->id == id;
1161 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1162 enum nd_ioctl_mode mode)
1165 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1166 long id = (long) file->private_data;
1167 struct nvdimm *nvdimm = NULL;
1170 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1171 mutex_lock(&nvdimm_bus_list_mutex);
1172 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1173 if (mode == DIMM_IOCTL) {
1176 dev = device_find_child(&nvdimm_bus->dev,
1177 file->private_data, match_dimm);
1180 nvdimm = to_nvdimm(dev);
1182 } else if (nvdimm_bus->id == id) {
1187 atomic_inc(&nvdimm_bus->ioctl_active);
1191 mutex_unlock(&nvdimm_bus_list_mutex);
1197 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1200 put_device(&nvdimm->dev);
1201 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1202 wake_up(&nvdimm_bus->wait);
1207 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1209 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1212 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1214 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1217 static int nd_open(struct inode *inode, struct file *file)
1219 long minor = iminor(inode);
1221 file->private_data = (void *) minor;
1225 static const struct file_operations nvdimm_bus_fops = {
1226 .owner = THIS_MODULE,
1228 .unlocked_ioctl = bus_ioctl,
1229 .compat_ioctl = bus_ioctl,
1230 .llseek = noop_llseek,
1233 static const struct file_operations nvdimm_fops = {
1234 .owner = THIS_MODULE,
1236 .unlocked_ioctl = dimm_ioctl,
1237 .compat_ioctl = dimm_ioctl,
1238 .llseek = noop_llseek,
1241 int __init nvdimm_bus_init(void)
1245 rc = bus_register(&nvdimm_bus_type);
1249 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1251 goto err_bus_chrdev;
1252 nvdimm_bus_major = rc;
1254 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1256 goto err_dimm_chrdev;
1259 nd_class = class_create(THIS_MODULE, "nd");
1260 if (IS_ERR(nd_class)) {
1261 rc = PTR_ERR(nd_class);
1265 rc = driver_register(&nd_bus_driver.drv);
1272 class_destroy(nd_class);
1274 unregister_chrdev(nvdimm_major, "dimmctl");
1276 unregister_chrdev(nvdimm_bus_major, "ndctl");
1278 bus_unregister(&nvdimm_bus_type);
1283 void nvdimm_bus_exit(void)
1285 driver_unregister(&nd_bus_driver.drv);
1286 class_destroy(nd_class);
1287 unregister_chrdev(nvdimm_bus_major, "ndctl");
1288 unregister_chrdev(nvdimm_major, "dimmctl");
1289 bus_unregister(&nvdimm_bus_type);
1290 ida_destroy(&nd_ida);