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 static const struct device_type nvdimm_bus_dev_type = {
304 .release = nvdimm_bus_release,
305 .groups = nvdimm_bus_attribute_groups,
308 bool is_nvdimm_bus(struct device *dev)
310 return dev->type == &nvdimm_bus_dev_type;
313 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
317 for (dev = nd_dev; dev; dev = dev->parent)
318 if (is_nvdimm_bus(dev))
320 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
322 return to_nvdimm_bus(dev);
326 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
328 struct nvdimm_bus *nvdimm_bus;
330 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
331 WARN_ON(!is_nvdimm_bus(dev));
334 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
336 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
338 return to_nvdimm_bus(nvdimm->dev.parent);
340 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
342 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
343 struct nvdimm_bus_descriptor *nd_desc)
345 struct nvdimm_bus *nvdimm_bus;
348 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
351 INIT_LIST_HEAD(&nvdimm_bus->list);
352 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
353 init_waitqueue_head(&nvdimm_bus->wait);
354 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
355 if (nvdimm_bus->id < 0) {
359 mutex_init(&nvdimm_bus->reconfig_mutex);
360 badrange_init(&nvdimm_bus->badrange);
361 nvdimm_bus->nd_desc = nd_desc;
362 nvdimm_bus->dev.parent = parent;
363 nvdimm_bus->dev.type = &nvdimm_bus_dev_type;
364 nvdimm_bus->dev.groups = nd_desc->attr_groups;
365 nvdimm_bus->dev.bus = &nvdimm_bus_type;
366 nvdimm_bus->dev.of_node = nd_desc->of_node;
367 dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
368 rc = device_register(&nvdimm_bus->dev);
370 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
376 put_device(&nvdimm_bus->dev);
379 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
381 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
385 device_unregister(&nvdimm_bus->dev);
387 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
389 static int child_unregister(struct device *dev, void *data)
392 * the singular ndctl class device per bus needs to be
393 * "device_destroy"ed, so skip it here
395 * i.e. remove classless children
400 if (is_nvdimm(dev)) {
401 struct nvdimm *nvdimm = to_nvdimm(dev);
402 bool dev_put = false;
404 /* We are shutting down. Make state frozen artificially. */
405 nvdimm_bus_lock(dev);
406 set_bit(NVDIMM_SECURITY_FROZEN, &nvdimm->sec.flags);
407 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
409 nvdimm_bus_unlock(dev);
410 cancel_delayed_work_sync(&nvdimm->dwork);
414 nd_device_unregister(dev, ND_SYNC);
419 static void free_badrange_list(struct list_head *badrange_list)
421 struct badrange_entry *bre, *next;
423 list_for_each_entry_safe(bre, next, badrange_list, list) {
424 list_del(&bre->list);
427 list_del_init(badrange_list);
430 static int nd_bus_remove(struct device *dev)
432 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
434 mutex_lock(&nvdimm_bus_list_mutex);
435 list_del_init(&nvdimm_bus->list);
436 mutex_unlock(&nvdimm_bus_list_mutex);
438 wait_event(nvdimm_bus->wait,
439 atomic_read(&nvdimm_bus->ioctl_active) == 0);
442 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
444 spin_lock(&nvdimm_bus->badrange.lock);
445 free_badrange_list(&nvdimm_bus->badrange.list);
446 spin_unlock(&nvdimm_bus->badrange.lock);
448 nvdimm_bus_destroy_ndctl(nvdimm_bus);
453 static int nd_bus_probe(struct device *dev)
455 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
458 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
462 mutex_lock(&nvdimm_bus_list_mutex);
463 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
464 mutex_unlock(&nvdimm_bus_list_mutex);
466 /* enable bus provider attributes to look up their local context */
467 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
472 static struct nd_device_driver nd_bus_driver = {
473 .probe = nd_bus_probe,
474 .remove = nd_bus_remove,
477 .suppress_bind_attrs = true,
478 .bus = &nvdimm_bus_type,
479 .owner = THIS_MODULE,
480 .mod_name = KBUILD_MODNAME,
484 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
486 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
488 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
491 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
494 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
496 void nd_synchronize(void)
498 async_synchronize_full_domain(&nd_async_domain);
500 EXPORT_SYMBOL_GPL(nd_synchronize);
502 static void nd_async_device_register(void *d, async_cookie_t cookie)
504 struct device *dev = d;
506 if (device_add(dev) != 0) {
507 dev_err(dev, "%s: failed\n", __func__);
512 put_device(dev->parent);
515 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
517 struct device *dev = d;
519 /* flush bus operations before delete */
520 nvdimm_bus_lock(dev);
521 nvdimm_bus_unlock(dev);
523 device_unregister(dev);
527 void __nd_device_register(struct device *dev)
533 * Ensure that region devices always have their NUMA node set as
534 * early as possible. This way we are able to make certain that
535 * any memory associated with the creation and the creation
536 * itself of the region is associated with the correct node.
538 if (is_nd_region(dev))
539 set_dev_node(dev, to_nd_region(dev)->numa_node);
541 dev->bus = &nvdimm_bus_type;
543 get_device(dev->parent);
544 if (dev_to_node(dev) == NUMA_NO_NODE)
545 set_dev_node(dev, dev_to_node(dev->parent));
549 async_schedule_dev_domain(nd_async_device_register, dev,
553 void nd_device_register(struct device *dev)
555 device_initialize(dev);
556 __nd_device_register(dev);
558 EXPORT_SYMBOL(nd_device_register);
560 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
567 * In the async case this is being triggered with the
568 * device lock held and the unregistration work needs to
569 * be moved out of line iff this is thread has won the
570 * race to schedule the deletion.
572 if (!kill_device(dev))
576 async_schedule_domain(nd_async_device_unregister, dev,
581 * In the sync case the device is being unregistered due
582 * to a state change of the parent. Claim the kill state
583 * to synchronize against other unregistration requests,
584 * or otherwise let the async path handle it if the
585 * unregistration was already queued.
588 killed = kill_device(dev);
589 nd_device_unlock(dev);
595 device_unregister(dev);
599 EXPORT_SYMBOL(nd_device_unregister);
602 * __nd_driver_register() - register a region or a namespace driver
603 * @nd_drv: driver to register
604 * @owner: automatically set by nd_driver_register() macro
605 * @mod_name: automatically set by nd_driver_register() macro
607 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
608 const char *mod_name)
610 struct device_driver *drv = &nd_drv->drv;
613 pr_debug("driver type bitmask not set (%ps)\n",
614 __builtin_return_address(0));
618 if (!nd_drv->probe) {
619 pr_debug("%s ->probe() must be specified\n", mod_name);
623 drv->bus = &nvdimm_bus_type;
625 drv->mod_name = mod_name;
627 return driver_register(drv);
629 EXPORT_SYMBOL(__nd_driver_register);
631 void nvdimm_check_and_set_ro(struct gendisk *disk)
633 struct device *dev = disk_to_dev(disk)->parent;
634 struct nd_region *nd_region = to_nd_region(dev->parent);
635 int disk_ro = get_disk_ro(disk);
638 * Upgrade to read-only if the region is read-only preserve as
639 * read-only if the disk is already read-only.
641 if (disk_ro || nd_region->ro == disk_ro)
644 dev_info(dev, "%s read-only, marking %s read-only\n",
645 dev_name(&nd_region->dev), disk->disk_name);
646 set_disk_ro(disk, 1);
648 EXPORT_SYMBOL(nvdimm_check_and_set_ro);
650 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
653 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
654 to_nd_device_type(dev));
656 static DEVICE_ATTR_RO(modalias);
658 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
661 return sprintf(buf, "%s\n", dev->type->name);
663 static DEVICE_ATTR_RO(devtype);
665 static struct attribute *nd_device_attributes[] = {
666 &dev_attr_modalias.attr,
667 &dev_attr_devtype.attr,
672 * nd_device_attribute_group - generic attributes for all devices on an nd bus
674 const struct attribute_group nd_device_attribute_group = {
675 .attrs = nd_device_attributes,
678 static ssize_t numa_node_show(struct device *dev,
679 struct device_attribute *attr, char *buf)
681 return sprintf(buf, "%d\n", dev_to_node(dev));
683 static DEVICE_ATTR_RO(numa_node);
685 static int nvdimm_dev_to_target_node(struct device *dev)
687 struct device *parent = dev->parent;
688 struct nd_region *nd_region = NULL;
690 if (is_nd_region(dev))
691 nd_region = to_nd_region(dev);
692 else if (parent && is_nd_region(parent))
693 nd_region = to_nd_region(parent);
697 return nd_region->target_node;
700 static ssize_t target_node_show(struct device *dev,
701 struct device_attribute *attr, char *buf)
703 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev));
705 static DEVICE_ATTR_RO(target_node);
707 static struct attribute *nd_numa_attributes[] = {
708 &dev_attr_numa_node.attr,
709 &dev_attr_target_node.attr,
713 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
716 struct device *dev = container_of(kobj, typeof(*dev), kobj);
718 if (!IS_ENABLED(CONFIG_NUMA))
721 if (a == &dev_attr_target_node.attr &&
722 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE)
729 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
731 const struct attribute_group nd_numa_attribute_group = {
732 .attrs = nd_numa_attributes,
733 .is_visible = nd_numa_attr_visible,
736 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
738 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
741 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
742 "ndctl%d", nvdimm_bus->id);
745 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
746 nvdimm_bus->id, PTR_ERR(dev));
747 return PTR_ERR_OR_ZERO(dev);
750 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
752 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
755 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
756 [ND_CMD_IMPLEMENTED] = { },
759 .out_sizes = { 4, 128, },
761 [ND_CMD_SMART_THRESHOLD] = {
763 .out_sizes = { 4, 8, },
765 [ND_CMD_DIMM_FLAGS] = {
767 .out_sizes = { 4, 4 },
769 [ND_CMD_GET_CONFIG_SIZE] = {
771 .out_sizes = { 4, 4, 4, },
773 [ND_CMD_GET_CONFIG_DATA] = {
775 .in_sizes = { 4, 4, },
777 .out_sizes = { 4, UINT_MAX, },
779 [ND_CMD_SET_CONFIG_DATA] = {
781 .in_sizes = { 4, 4, UINT_MAX, },
787 .in_sizes = { 4, 4, UINT_MAX, },
789 .out_sizes = { 4, 4, UINT_MAX, },
793 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
795 .out_sizes = { UINT_MAX, },
799 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
801 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
802 return &__nd_cmd_dimm_descs[cmd];
805 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
807 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
808 [ND_CMD_IMPLEMENTED] = { },
811 .in_sizes = { 8, 8, },
813 .out_sizes = { 4, 4, 4, 4, },
815 [ND_CMD_ARS_START] = {
817 .in_sizes = { 8, 8, 2, 1, 5, },
819 .out_sizes = { 4, 4, },
821 [ND_CMD_ARS_STATUS] = {
823 .out_sizes = { 4, 4, UINT_MAX, },
825 [ND_CMD_CLEAR_ERROR] = {
827 .in_sizes = { 8, 8, },
829 .out_sizes = { 4, 4, 8, },
833 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
835 .out_sizes = { UINT_MAX, },
839 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
841 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
842 return &__nd_cmd_bus_descs[cmd];
845 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
847 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
848 const struct nd_cmd_desc *desc, int idx, void *buf)
850 if (idx >= desc->in_num)
853 if (desc->in_sizes[idx] < UINT_MAX)
854 return desc->in_sizes[idx];
856 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
857 struct nd_cmd_set_config_hdr *hdr = buf;
859 return hdr->in_length;
860 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
861 struct nd_cmd_vendor_hdr *hdr = buf;
863 return hdr->in_length;
864 } else if (cmd == ND_CMD_CALL) {
865 struct nd_cmd_pkg *pkg = buf;
867 return pkg->nd_size_in;
872 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
874 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
875 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
876 const u32 *out_field, unsigned long remainder)
878 if (idx >= desc->out_num)
881 if (desc->out_sizes[idx] < UINT_MAX)
882 return desc->out_sizes[idx];
884 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
886 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
888 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
890 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
891 * "Size of Output Buffer in bytes, including this
894 if (out_field[1] < 4)
897 * ACPI 6.1 is ambiguous if 'status' is included in the
898 * output size. If we encounter an output size that
899 * overshoots the remainder by 4 bytes, assume it was
900 * including 'status'.
902 if (out_field[1] - 4 == remainder)
904 return out_field[1] - 8;
905 } else if (cmd == ND_CMD_CALL) {
906 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
908 return pkg->nd_size_out;
914 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
916 void wait_nvdimm_bus_probe_idle(struct device *dev)
918 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
921 if (nvdimm_bus->probe_active == 0)
923 nvdimm_bus_unlock(dev);
924 nd_device_unlock(dev);
925 wait_event(nvdimm_bus->wait,
926 nvdimm_bus->probe_active == 0);
928 nvdimm_bus_lock(dev);
932 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
934 struct nd_cmd_clear_error *clear_err =
935 (struct nd_cmd_clear_error *)data;
936 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
937 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
938 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
939 struct nd_namespace_common *ndns = NULL;
940 struct nd_namespace_io *nsio;
941 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
943 if (nd_dax || !dev->driver)
946 start = clear_err->address;
947 end = clear_err->address + clear_err->cleared - 1;
949 if (nd_btt || nd_pfn || nd_dax) {
955 ndns = nd_dax->nd_pfn.ndns;
962 nsio = to_nd_namespace_io(&ndns->dev);
963 pstart = nsio->res.start + offset;
964 pend = nsio->res.end - end_trunc;
966 if ((pstart >= start) && (pend <= end))
973 static int nd_ns_forget_poison_check(struct device *dev, void *data)
975 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
978 /* set_config requires an idle interleave set */
979 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
980 struct nvdimm *nvdimm, unsigned int cmd, void *data)
982 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
984 /* ask the bus provider if it would like to block this request */
985 if (nd_desc->clear_to_send) {
986 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
992 /* require clear error to go through the pmem driver */
993 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
994 return device_for_each_child(&nvdimm_bus->dev, data,
995 nd_ns_forget_poison_check);
997 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
1000 /* prevent label manipulation while the kernel owns label updates */
1001 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
1002 if (atomic_read(&nvdimm->busy))
1007 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
1008 int read_only, unsigned int ioctl_cmd, unsigned long arg)
1010 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
1011 const struct nd_cmd_desc *desc = NULL;
1012 unsigned int cmd = _IOC_NR(ioctl_cmd);
1013 struct device *dev = &nvdimm_bus->dev;
1014 void __user *p = (void __user *) arg;
1015 char *out_env = NULL, *in_env = NULL;
1016 const char *cmd_name, *dimm_name;
1017 u32 in_len = 0, out_len = 0;
1018 unsigned int func = cmd;
1019 unsigned long cmd_mask;
1020 struct nd_cmd_pkg pkg;
1026 desc = nd_cmd_dimm_desc(cmd);
1027 cmd_name = nvdimm_cmd_name(cmd);
1028 cmd_mask = nvdimm->cmd_mask;
1029 dimm_name = dev_name(&nvdimm->dev);
1031 desc = nd_cmd_bus_desc(cmd);
1032 cmd_name = nvdimm_bus_cmd_name(cmd);
1033 cmd_mask = nd_desc->cmd_mask;
1037 /* Validate command family support against bus declared support */
1038 if (cmd == ND_CMD_CALL) {
1039 unsigned long *mask;
1041 if (copy_from_user(&pkg, p, sizeof(pkg)))
1045 if (pkg.nd_family > NVDIMM_FAMILY_MAX)
1047 mask = &nd_desc->dimm_family_mask;
1049 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX)
1051 mask = &nd_desc->bus_family_mask;
1054 if (!test_bit(pkg.nd_family, mask))
1059 (desc->out_num + desc->in_num == 0) ||
1060 cmd > ND_CMD_CALL ||
1061 !test_bit(cmd, &cmd_mask))
1064 /* fail write commands (when read-only) */
1068 case ND_CMD_SET_CONFIG_DATA:
1069 case ND_CMD_ARS_START:
1070 case ND_CMD_CLEAR_ERROR:
1072 dev_dbg(dev, "'%s' command while read-only.\n",
1073 nvdimm ? nvdimm_cmd_name(cmd)
1074 : nvdimm_bus_cmd_name(cmd));
1080 /* process an input envelope */
1081 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1084 for (i = 0; i < desc->in_num; i++) {
1087 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1088 if (in_size == UINT_MAX) {
1089 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1090 __func__, dimm_name, cmd_name, i);
1094 if (in_len < ND_CMD_MAX_ENVELOPE)
1095 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1098 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1105 if (cmd == ND_CMD_CALL) {
1106 func = pkg.nd_command;
1107 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1108 dimm_name, pkg.nd_command,
1109 in_len, out_len, buf_len);
1112 /* process an output envelope */
1113 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1119 for (i = 0; i < desc->out_num; i++) {
1120 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1121 (u32 *) in_env, (u32 *) out_env, 0);
1124 if (out_size == UINT_MAX) {
1125 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1126 dimm_name, cmd_name, i);
1130 if (out_len < ND_CMD_MAX_ENVELOPE)
1131 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1134 if (copy && copy_from_user(&out_env[out_len],
1135 p + in_len + out_len, copy)) {
1139 out_len += out_size;
1142 buf_len = (u64) out_len + (u64) in_len;
1143 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1144 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1145 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1150 buf = vmalloc(buf_len);
1156 if (copy_from_user(buf, p, buf_len)) {
1161 nd_device_lock(dev);
1162 nvdimm_bus_lock(dev);
1163 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1167 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1171 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1172 struct nd_cmd_clear_error *clear_err = buf;
1174 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1175 clear_err->cleared);
1178 if (copy_to_user(p, buf, buf_len))
1182 nvdimm_bus_unlock(dev);
1183 nd_device_unlock(dev);
1191 enum nd_ioctl_mode {
1196 static int match_dimm(struct device *dev, void *data)
1198 long id = (long) data;
1200 if (is_nvdimm(dev)) {
1201 struct nvdimm *nvdimm = to_nvdimm(dev);
1203 return nvdimm->id == id;
1209 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1210 enum nd_ioctl_mode mode)
1213 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1214 long id = (long) file->private_data;
1215 struct nvdimm *nvdimm = NULL;
1218 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1219 mutex_lock(&nvdimm_bus_list_mutex);
1220 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1221 if (mode == DIMM_IOCTL) {
1224 dev = device_find_child(&nvdimm_bus->dev,
1225 file->private_data, match_dimm);
1228 nvdimm = to_nvdimm(dev);
1230 } else if (nvdimm_bus->id == id) {
1235 atomic_inc(&nvdimm_bus->ioctl_active);
1239 mutex_unlock(&nvdimm_bus_list_mutex);
1245 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1248 put_device(&nvdimm->dev);
1249 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1250 wake_up(&nvdimm_bus->wait);
1255 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1257 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1260 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1262 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1265 static int nd_open(struct inode *inode, struct file *file)
1267 long minor = iminor(inode);
1269 file->private_data = (void *) minor;
1273 static const struct file_operations nvdimm_bus_fops = {
1274 .owner = THIS_MODULE,
1276 .unlocked_ioctl = bus_ioctl,
1277 .compat_ioctl = compat_ptr_ioctl,
1278 .llseek = noop_llseek,
1281 static const struct file_operations nvdimm_fops = {
1282 .owner = THIS_MODULE,
1284 .unlocked_ioctl = dimm_ioctl,
1285 .compat_ioctl = compat_ptr_ioctl,
1286 .llseek = noop_llseek,
1289 int __init nvdimm_bus_init(void)
1293 rc = bus_register(&nvdimm_bus_type);
1297 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1299 goto err_bus_chrdev;
1300 nvdimm_bus_major = rc;
1302 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1304 goto err_dimm_chrdev;
1307 nd_class = class_create(THIS_MODULE, "nd");
1308 if (IS_ERR(nd_class)) {
1309 rc = PTR_ERR(nd_class);
1313 rc = driver_register(&nd_bus_driver.drv);
1320 class_destroy(nd_class);
1322 unregister_chrdev(nvdimm_major, "dimmctl");
1324 unregister_chrdev(nvdimm_bus_major, "ndctl");
1326 bus_unregister(&nvdimm_bus_type);
1331 void nvdimm_bus_exit(void)
1333 driver_unregister(&nd_bus_driver.drv);
1334 class_destroy(nd_class);
1335 unregister_chrdev(nvdimm_bus_major, "ndctl");
1336 unregister_chrdev(nvdimm_major, "dimmctl");
1337 bus_unregister(&nvdimm_bus_type);
1338 ida_destroy(&nd_ida);