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/ndctl.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/cpu.h>
27 static int nvdimm_bus_major;
28 static struct class *nd_class;
29 static DEFINE_IDA(nd_ida);
31 static int to_nd_device_type(const struct device *dev)
34 return ND_DEVICE_DIMM;
35 else if (is_memory(dev))
36 return ND_DEVICE_REGION_PMEM;
37 else if (is_nd_dax(dev))
38 return ND_DEVICE_DAX_PMEM;
39 else if (is_nd_region(dev->parent))
40 return nd_region_to_nstype(to_nd_region(dev->parent));
45 static int nvdimm_bus_uevent(const struct device *dev, struct kobj_uevent_env *env)
47 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
48 to_nd_device_type(dev));
51 static struct module *to_bus_provider(struct device *dev)
53 /* pin bus providers while regions are enabled */
54 if (is_nd_region(dev)) {
55 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
57 return nvdimm_bus->nd_desc->module;
62 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
64 nvdimm_bus_lock(&nvdimm_bus->dev);
65 nvdimm_bus->probe_active++;
66 nvdimm_bus_unlock(&nvdimm_bus->dev);
69 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
71 nvdimm_bus_lock(&nvdimm_bus->dev);
72 if (--nvdimm_bus->probe_active == 0)
73 wake_up(&nvdimm_bus->wait);
74 nvdimm_bus_unlock(&nvdimm_bus->dev);
77 static int nvdimm_bus_probe(struct device *dev)
79 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
80 struct module *provider = to_bus_provider(dev);
81 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
84 if (!try_module_get(provider))
87 dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n",
88 dev->driver->name, dev_name(dev));
90 nvdimm_bus_probe_start(nvdimm_bus);
91 rc = nd_drv->probe(dev);
92 if ((rc == 0 || rc == -EOPNOTSUPP) &&
93 dev->parent && is_nd_region(dev->parent))
94 nd_region_advance_seeds(to_nd_region(dev->parent), dev);
95 nvdimm_bus_probe_end(nvdimm_bus);
97 dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name,
101 module_put(provider);
105 static void nvdimm_bus_remove(struct device *dev)
107 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
108 struct module *provider = to_bus_provider(dev);
109 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
114 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s)\n", dev->driver->name,
116 module_put(provider);
119 static void nvdimm_bus_shutdown(struct device *dev)
121 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
122 struct nd_device_driver *nd_drv = NULL;
125 nd_drv = to_nd_device_driver(dev->driver);
127 if (nd_drv && nd_drv->shutdown) {
128 nd_drv->shutdown(dev);
129 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
130 dev->driver->name, dev_name(dev));
134 void nd_device_notify(struct device *dev, enum nvdimm_event event)
138 struct nd_device_driver *nd_drv;
140 nd_drv = to_nd_device_driver(dev->driver);
142 nd_drv->notify(dev, event);
146 EXPORT_SYMBOL(nd_device_notify);
148 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
150 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
155 /* caller is responsible for holding a reference on the device */
156 nd_device_notify(&nd_region->dev, event);
158 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
160 struct clear_badblocks_context {
161 resource_size_t phys, cleared;
164 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
166 struct clear_badblocks_context *ctx = data;
167 struct nd_region *nd_region;
168 resource_size_t ndr_end;
171 /* make sure device is a region */
175 nd_region = to_nd_region(dev);
176 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
178 /* make sure we are in the region */
179 if (ctx->phys < nd_region->ndr_start ||
180 (ctx->phys + ctx->cleared - 1) > ndr_end)
183 sector = (ctx->phys - nd_region->ndr_start) / 512;
184 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
186 if (nd_region->bb_state)
187 sysfs_notify_dirent(nd_region->bb_state);
192 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
193 phys_addr_t phys, u64 cleared)
195 struct clear_badblocks_context ctx = {
200 device_for_each_child(&nvdimm_bus->dev, &ctx,
201 nvdimm_clear_badblocks_region);
204 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
205 phys_addr_t phys, u64 cleared)
208 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
210 if (cleared > 0 && cleared / 512)
211 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
214 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
217 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
218 struct nvdimm_bus_descriptor *nd_desc;
219 struct nd_cmd_clear_error clear_err;
220 struct nd_cmd_ars_cap ars_cap;
221 u32 clear_err_unit, mask;
222 unsigned int noio_flag;
228 nd_desc = nvdimm_bus->nd_desc;
230 * if ndctl does not exist, it's PMEM_LEGACY and
231 * we want to just pretend everything is handled.
236 memset(&ars_cap, 0, sizeof(ars_cap));
237 ars_cap.address = phys;
238 ars_cap.length = len;
239 noio_flag = memalloc_noio_save();
240 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
241 sizeof(ars_cap), &cmd_rc);
242 memalloc_noio_restore(noio_flag);
247 clear_err_unit = ars_cap.clear_err_unit;
248 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
251 mask = clear_err_unit - 1;
252 if ((phys | len) & mask)
254 memset(&clear_err, 0, sizeof(clear_err));
255 clear_err.address = phys;
256 clear_err.length = len;
257 noio_flag = memalloc_noio_save();
258 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
259 sizeof(clear_err), &cmd_rc);
260 memalloc_noio_restore(noio_flag);
266 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
268 return clear_err.cleared;
270 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
272 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
274 static struct bus_type nvdimm_bus_type = {
276 .uevent = nvdimm_bus_uevent,
277 .match = nvdimm_bus_match,
278 .probe = nvdimm_bus_probe,
279 .remove = nvdimm_bus_remove,
280 .shutdown = nvdimm_bus_shutdown,
283 static void nvdimm_bus_release(struct device *dev)
285 struct nvdimm_bus *nvdimm_bus;
287 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
288 ida_simple_remove(&nd_ida, nvdimm_bus->id);
292 static const struct device_type nvdimm_bus_dev_type = {
293 .release = nvdimm_bus_release,
294 .groups = nvdimm_bus_attribute_groups,
297 bool is_nvdimm_bus(struct device *dev)
299 return dev->type == &nvdimm_bus_dev_type;
302 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
306 for (dev = nd_dev; dev; dev = dev->parent)
307 if (is_nvdimm_bus(dev))
309 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
311 return to_nvdimm_bus(dev);
315 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
317 struct nvdimm_bus *nvdimm_bus;
319 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
320 WARN_ON(!is_nvdimm_bus(dev));
323 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
325 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
327 return to_nvdimm_bus(nvdimm->dev.parent);
329 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
331 static struct lock_class_key nvdimm_bus_key;
333 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
334 struct nvdimm_bus_descriptor *nd_desc)
336 struct nvdimm_bus *nvdimm_bus;
339 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
342 INIT_LIST_HEAD(&nvdimm_bus->list);
343 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
344 init_waitqueue_head(&nvdimm_bus->wait);
345 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
346 if (nvdimm_bus->id < 0) {
350 mutex_init(&nvdimm_bus->reconfig_mutex);
351 badrange_init(&nvdimm_bus->badrange);
352 nvdimm_bus->nd_desc = nd_desc;
353 nvdimm_bus->dev.parent = parent;
354 nvdimm_bus->dev.type = &nvdimm_bus_dev_type;
355 nvdimm_bus->dev.groups = nd_desc->attr_groups;
356 nvdimm_bus->dev.bus = &nvdimm_bus_type;
357 nvdimm_bus->dev.of_node = nd_desc->of_node;
358 device_initialize(&nvdimm_bus->dev);
359 lockdep_set_class(&nvdimm_bus->dev.mutex, &nvdimm_bus_key);
360 device_set_pm_not_required(&nvdimm_bus->dev);
361 rc = dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
365 rc = device_add(&nvdimm_bus->dev);
367 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
373 put_device(&nvdimm_bus->dev);
376 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
378 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
382 device_unregister(&nvdimm_bus->dev);
384 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
386 static int child_unregister(struct device *dev, void *data)
389 * the singular ndctl class device per bus needs to be
390 * "device_destroy"ed, so skip it here
392 * i.e. remove classless children
398 nvdimm_delete(to_nvdimm(dev));
400 nd_device_unregister(dev, ND_SYNC);
405 static void free_badrange_list(struct list_head *badrange_list)
407 struct badrange_entry *bre, *next;
409 list_for_each_entry_safe(bre, next, badrange_list, list) {
410 list_del(&bre->list);
413 list_del_init(badrange_list);
416 static void nd_bus_remove(struct device *dev)
418 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
420 mutex_lock(&nvdimm_bus_list_mutex);
421 list_del_init(&nvdimm_bus->list);
422 mutex_unlock(&nvdimm_bus_list_mutex);
424 wait_event(nvdimm_bus->wait,
425 atomic_read(&nvdimm_bus->ioctl_active) == 0);
428 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
430 spin_lock(&nvdimm_bus->badrange.lock);
431 free_badrange_list(&nvdimm_bus->badrange.list);
432 spin_unlock(&nvdimm_bus->badrange.lock);
434 nvdimm_bus_destroy_ndctl(nvdimm_bus);
437 static int nd_bus_probe(struct device *dev)
439 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
442 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
446 mutex_lock(&nvdimm_bus_list_mutex);
447 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
448 mutex_unlock(&nvdimm_bus_list_mutex);
450 /* enable bus provider attributes to look up their local context */
451 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
456 static struct nd_device_driver nd_bus_driver = {
457 .probe = nd_bus_probe,
458 .remove = nd_bus_remove,
461 .suppress_bind_attrs = true,
462 .bus = &nvdimm_bus_type,
463 .owner = THIS_MODULE,
464 .mod_name = KBUILD_MODNAME,
468 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
470 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
472 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
475 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
478 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
480 void nd_synchronize(void)
482 async_synchronize_full_domain(&nd_async_domain);
484 EXPORT_SYMBOL_GPL(nd_synchronize);
486 static void nd_async_device_register(void *d, async_cookie_t cookie)
488 struct device *dev = d;
490 if (device_add(dev) != 0) {
491 dev_err(dev, "%s: failed\n", __func__);
496 put_device(dev->parent);
499 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
501 struct device *dev = d;
503 /* flush bus operations before delete */
504 nvdimm_bus_lock(dev);
505 nvdimm_bus_unlock(dev);
507 device_unregister(dev);
511 static void __nd_device_register(struct device *dev, bool sync)
517 * Ensure that region devices always have their NUMA node set as
518 * early as possible. This way we are able to make certain that
519 * any memory associated with the creation and the creation
520 * itself of the region is associated with the correct node.
522 if (is_nd_region(dev))
523 set_dev_node(dev, to_nd_region(dev)->numa_node);
525 dev->bus = &nvdimm_bus_type;
526 device_set_pm_not_required(dev);
528 get_device(dev->parent);
529 if (dev_to_node(dev) == NUMA_NO_NODE)
530 set_dev_node(dev, dev_to_node(dev->parent));
535 nd_async_device_register(dev, 0);
537 async_schedule_dev_domain(nd_async_device_register, dev,
541 void nd_device_register(struct device *dev)
543 __nd_device_register(dev, false);
545 EXPORT_SYMBOL(nd_device_register);
547 void nd_device_register_sync(struct device *dev)
549 __nd_device_register(dev, true);
552 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
559 * In the async case this is being triggered with the
560 * device lock held and the unregistration work needs to
561 * be moved out of line iff this is thread has won the
562 * race to schedule the deletion.
564 if (!kill_device(dev))
568 async_schedule_domain(nd_async_device_unregister, dev,
573 * In the sync case the device is being unregistered due
574 * to a state change of the parent. Claim the kill state
575 * to synchronize against other unregistration requests,
576 * or otherwise let the async path handle it if the
577 * unregistration was already queued.
580 killed = kill_device(dev);
587 device_unregister(dev);
591 EXPORT_SYMBOL(nd_device_unregister);
594 * __nd_driver_register() - register a region or a namespace driver
595 * @nd_drv: driver to register
596 * @owner: automatically set by nd_driver_register() macro
597 * @mod_name: automatically set by nd_driver_register() macro
599 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
600 const char *mod_name)
602 struct device_driver *drv = &nd_drv->drv;
605 pr_debug("driver type bitmask not set (%ps)\n",
606 __builtin_return_address(0));
610 if (!nd_drv->probe) {
611 pr_debug("%s ->probe() must be specified\n", mod_name);
615 drv->bus = &nvdimm_bus_type;
617 drv->mod_name = mod_name;
619 return driver_register(drv);
621 EXPORT_SYMBOL(__nd_driver_register);
623 void nvdimm_check_and_set_ro(struct gendisk *disk)
625 struct device *dev = disk_to_dev(disk)->parent;
626 struct nd_region *nd_region = to_nd_region(dev->parent);
627 int disk_ro = get_disk_ro(disk);
629 /* catch the disk up with the region ro state */
630 if (disk_ro == nd_region->ro)
633 dev_info(dev, "%s read-%s, marking %s read-%s\n",
634 dev_name(&nd_region->dev), nd_region->ro ? "only" : "write",
635 disk->disk_name, nd_region->ro ? "only" : "write");
636 set_disk_ro(disk, nd_region->ro);
638 EXPORT_SYMBOL(nvdimm_check_and_set_ro);
640 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
643 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
644 to_nd_device_type(dev));
646 static DEVICE_ATTR_RO(modalias);
648 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
651 return sprintf(buf, "%s\n", dev->type->name);
653 static DEVICE_ATTR_RO(devtype);
655 static struct attribute *nd_device_attributes[] = {
656 &dev_attr_modalias.attr,
657 &dev_attr_devtype.attr,
662 * nd_device_attribute_group - generic attributes for all devices on an nd bus
664 const struct attribute_group nd_device_attribute_group = {
665 .attrs = nd_device_attributes,
668 static ssize_t numa_node_show(struct device *dev,
669 struct device_attribute *attr, char *buf)
671 return sprintf(buf, "%d\n", dev_to_node(dev));
673 static DEVICE_ATTR_RO(numa_node);
675 static int nvdimm_dev_to_target_node(struct device *dev)
677 struct device *parent = dev->parent;
678 struct nd_region *nd_region = NULL;
680 if (is_nd_region(dev))
681 nd_region = to_nd_region(dev);
682 else if (parent && is_nd_region(parent))
683 nd_region = to_nd_region(parent);
687 return nd_region->target_node;
690 static ssize_t target_node_show(struct device *dev,
691 struct device_attribute *attr, char *buf)
693 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev));
695 static DEVICE_ATTR_RO(target_node);
697 static struct attribute *nd_numa_attributes[] = {
698 &dev_attr_numa_node.attr,
699 &dev_attr_target_node.attr,
703 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
706 struct device *dev = container_of(kobj, typeof(*dev), kobj);
708 if (!IS_ENABLED(CONFIG_NUMA))
711 if (a == &dev_attr_target_node.attr &&
712 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE)
719 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
721 const struct attribute_group nd_numa_attribute_group = {
722 .attrs = nd_numa_attributes,
723 .is_visible = nd_numa_attr_visible,
726 static void ndctl_release(struct device *dev)
731 static struct lock_class_key nvdimm_ndctl_key;
733 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
735 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
739 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
742 device_initialize(dev);
743 lockdep_set_class(&dev->mutex, &nvdimm_ndctl_key);
744 device_set_pm_not_required(dev);
745 dev->class = nd_class;
746 dev->parent = &nvdimm_bus->dev;
748 dev->release = ndctl_release;
749 rc = dev_set_name(dev, "ndctl%d", nvdimm_bus->id);
753 rc = device_add(dev);
755 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %d\n",
766 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
768 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
771 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
772 [ND_CMD_IMPLEMENTED] = { },
775 .out_sizes = { 4, 128, },
777 [ND_CMD_SMART_THRESHOLD] = {
779 .out_sizes = { 4, 8, },
781 [ND_CMD_DIMM_FLAGS] = {
783 .out_sizes = { 4, 4 },
785 [ND_CMD_GET_CONFIG_SIZE] = {
787 .out_sizes = { 4, 4, 4, },
789 [ND_CMD_GET_CONFIG_DATA] = {
791 .in_sizes = { 4, 4, },
793 .out_sizes = { 4, UINT_MAX, },
795 [ND_CMD_SET_CONFIG_DATA] = {
797 .in_sizes = { 4, 4, UINT_MAX, },
803 .in_sizes = { 4, 4, UINT_MAX, },
805 .out_sizes = { 4, 4, UINT_MAX, },
809 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
811 .out_sizes = { UINT_MAX, },
815 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
817 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
818 return &__nd_cmd_dimm_descs[cmd];
821 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
823 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
824 [ND_CMD_IMPLEMENTED] = { },
827 .in_sizes = { 8, 8, },
829 .out_sizes = { 4, 4, 4, 4, },
831 [ND_CMD_ARS_START] = {
833 .in_sizes = { 8, 8, 2, 1, 5, },
835 .out_sizes = { 4, 4, },
837 [ND_CMD_ARS_STATUS] = {
839 .out_sizes = { 4, 4, UINT_MAX, },
841 [ND_CMD_CLEAR_ERROR] = {
843 .in_sizes = { 8, 8, },
845 .out_sizes = { 4, 4, 8, },
849 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
851 .out_sizes = { UINT_MAX, },
855 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
857 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
858 return &__nd_cmd_bus_descs[cmd];
861 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
863 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
864 const struct nd_cmd_desc *desc, int idx, void *buf)
866 if (idx >= desc->in_num)
869 if (desc->in_sizes[idx] < UINT_MAX)
870 return desc->in_sizes[idx];
872 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
873 struct nd_cmd_set_config_hdr *hdr = buf;
875 return hdr->in_length;
876 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
877 struct nd_cmd_vendor_hdr *hdr = buf;
879 return hdr->in_length;
880 } else if (cmd == ND_CMD_CALL) {
881 struct nd_cmd_pkg *pkg = buf;
883 return pkg->nd_size_in;
888 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
890 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
891 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
892 const u32 *out_field, unsigned long remainder)
894 if (idx >= desc->out_num)
897 if (desc->out_sizes[idx] < UINT_MAX)
898 return desc->out_sizes[idx];
900 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
902 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
904 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
906 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
907 * "Size of Output Buffer in bytes, including this
910 if (out_field[1] < 4)
913 * ACPI 6.1 is ambiguous if 'status' is included in the
914 * output size. If we encounter an output size that
915 * overshoots the remainder by 4 bytes, assume it was
916 * including 'status'.
918 if (out_field[1] - 4 == remainder)
920 return out_field[1] - 8;
921 } else if (cmd == ND_CMD_CALL) {
922 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
924 return pkg->nd_size_out;
930 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
932 void wait_nvdimm_bus_probe_idle(struct device *dev)
934 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
937 if (nvdimm_bus->probe_active == 0)
939 nvdimm_bus_unlock(dev);
941 wait_event(nvdimm_bus->wait,
942 nvdimm_bus->probe_active == 0);
944 nvdimm_bus_lock(dev);
948 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
950 struct nd_cmd_clear_error *clear_err =
951 (struct nd_cmd_clear_error *)data;
952 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
953 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
954 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
955 struct nd_namespace_common *ndns = NULL;
956 struct nd_namespace_io *nsio;
957 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
959 if (nd_dax || !dev->driver)
962 start = clear_err->address;
963 end = clear_err->address + clear_err->cleared - 1;
965 if (nd_btt || nd_pfn || nd_dax) {
971 ndns = nd_dax->nd_pfn.ndns;
978 nsio = to_nd_namespace_io(&ndns->dev);
979 pstart = nsio->res.start + offset;
980 pend = nsio->res.end - end_trunc;
982 if ((pstart >= start) && (pend <= end))
989 static int nd_ns_forget_poison_check(struct device *dev, void *data)
991 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
994 /* set_config requires an idle interleave set */
995 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
996 struct nvdimm *nvdimm, unsigned int cmd, void *data)
998 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
1000 /* ask the bus provider if it would like to block this request */
1001 if (nd_desc->clear_to_send) {
1002 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
1008 /* require clear error to go through the pmem driver */
1009 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
1010 return device_for_each_child(&nvdimm_bus->dev, data,
1011 nd_ns_forget_poison_check);
1013 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
1016 /* prevent label manipulation while the kernel owns label updates */
1017 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
1018 if (atomic_read(&nvdimm->busy))
1023 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
1024 int read_only, unsigned int ioctl_cmd, unsigned long arg)
1026 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
1027 const struct nd_cmd_desc *desc = NULL;
1028 unsigned int cmd = _IOC_NR(ioctl_cmd);
1029 struct device *dev = &nvdimm_bus->dev;
1030 void __user *p = (void __user *) arg;
1031 char *out_env = NULL, *in_env = NULL;
1032 const char *cmd_name, *dimm_name;
1033 u32 in_len = 0, out_len = 0;
1034 unsigned int func = cmd;
1035 unsigned long cmd_mask;
1036 struct nd_cmd_pkg pkg;
1042 desc = nd_cmd_dimm_desc(cmd);
1043 cmd_name = nvdimm_cmd_name(cmd);
1044 cmd_mask = nvdimm->cmd_mask;
1045 dimm_name = dev_name(&nvdimm->dev);
1047 desc = nd_cmd_bus_desc(cmd);
1048 cmd_name = nvdimm_bus_cmd_name(cmd);
1049 cmd_mask = nd_desc->cmd_mask;
1053 /* Validate command family support against bus declared support */
1054 if (cmd == ND_CMD_CALL) {
1055 unsigned long *mask;
1057 if (copy_from_user(&pkg, p, sizeof(pkg)))
1061 if (pkg.nd_family > NVDIMM_FAMILY_MAX)
1063 mask = &nd_desc->dimm_family_mask;
1065 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX)
1067 mask = &nd_desc->bus_family_mask;
1070 if (!test_bit(pkg.nd_family, mask))
1075 (desc->out_num + desc->in_num == 0) ||
1076 cmd > ND_CMD_CALL ||
1077 !test_bit(cmd, &cmd_mask))
1080 /* fail write commands (when read-only) */
1084 case ND_CMD_SET_CONFIG_DATA:
1085 case ND_CMD_ARS_START:
1086 case ND_CMD_CLEAR_ERROR:
1088 dev_dbg(dev, "'%s' command while read-only.\n",
1089 nvdimm ? nvdimm_cmd_name(cmd)
1090 : nvdimm_bus_cmd_name(cmd));
1096 /* process an input envelope */
1097 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1100 for (i = 0; i < desc->in_num; i++) {
1103 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1104 if (in_size == UINT_MAX) {
1105 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1106 __func__, dimm_name, cmd_name, i);
1110 if (in_len < ND_CMD_MAX_ENVELOPE)
1111 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1114 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1121 if (cmd == ND_CMD_CALL) {
1122 func = pkg.nd_command;
1123 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1124 dimm_name, pkg.nd_command,
1125 in_len, out_len, buf_len);
1128 /* process an output envelope */
1129 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1135 for (i = 0; i < desc->out_num; i++) {
1136 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1137 (u32 *) in_env, (u32 *) out_env, 0);
1140 if (out_size == UINT_MAX) {
1141 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1142 dimm_name, cmd_name, i);
1146 if (out_len < ND_CMD_MAX_ENVELOPE)
1147 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1150 if (copy && copy_from_user(&out_env[out_len],
1151 p + in_len + out_len, copy)) {
1155 out_len += out_size;
1158 buf_len = (u64) out_len + (u64) in_len;
1159 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1160 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1161 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1166 buf = vmalloc(buf_len);
1172 if (copy_from_user(buf, p, buf_len)) {
1178 nvdimm_bus_lock(dev);
1179 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1183 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1187 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1188 struct nd_cmd_clear_error *clear_err = buf;
1190 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1191 clear_err->cleared);
1194 if (copy_to_user(p, buf, buf_len))
1198 nvdimm_bus_unlock(dev);
1207 enum nd_ioctl_mode {
1212 static int match_dimm(struct device *dev, void *data)
1214 long id = (long) data;
1216 if (is_nvdimm(dev)) {
1217 struct nvdimm *nvdimm = to_nvdimm(dev);
1219 return nvdimm->id == id;
1225 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1226 enum nd_ioctl_mode mode)
1229 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1230 long id = (long) file->private_data;
1231 struct nvdimm *nvdimm = NULL;
1234 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1235 mutex_lock(&nvdimm_bus_list_mutex);
1236 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1237 if (mode == DIMM_IOCTL) {
1240 dev = device_find_child(&nvdimm_bus->dev,
1241 file->private_data, match_dimm);
1244 nvdimm = to_nvdimm(dev);
1246 } else if (nvdimm_bus->id == id) {
1251 atomic_inc(&nvdimm_bus->ioctl_active);
1255 mutex_unlock(&nvdimm_bus_list_mutex);
1261 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1264 put_device(&nvdimm->dev);
1265 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1266 wake_up(&nvdimm_bus->wait);
1271 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1273 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1276 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1278 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1281 static int nd_open(struct inode *inode, struct file *file)
1283 long minor = iminor(inode);
1285 file->private_data = (void *) minor;
1289 static const struct file_operations nvdimm_bus_fops = {
1290 .owner = THIS_MODULE,
1292 .unlocked_ioctl = bus_ioctl,
1293 .compat_ioctl = compat_ptr_ioctl,
1294 .llseek = noop_llseek,
1297 static const struct file_operations nvdimm_fops = {
1298 .owner = THIS_MODULE,
1300 .unlocked_ioctl = dimm_ioctl,
1301 .compat_ioctl = compat_ptr_ioctl,
1302 .llseek = noop_llseek,
1305 int __init nvdimm_bus_init(void)
1309 rc = bus_register(&nvdimm_bus_type);
1313 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1315 goto err_bus_chrdev;
1316 nvdimm_bus_major = rc;
1318 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1320 goto err_dimm_chrdev;
1323 nd_class = class_create("nd");
1324 if (IS_ERR(nd_class)) {
1325 rc = PTR_ERR(nd_class);
1329 rc = driver_register(&nd_bus_driver.drv);
1336 class_destroy(nd_class);
1338 unregister_chrdev(nvdimm_major, "dimmctl");
1340 unregister_chrdev(nvdimm_bus_major, "ndctl");
1342 bus_unregister(&nvdimm_bus_type);
1347 void nvdimm_bus_exit(void)
1349 driver_unregister(&nd_bus_driver.drv);
1350 class_destroy(nd_class);
1351 unregister_chrdev(nvdimm_bus_major, "ndctl");
1352 unregister_chrdev(nvdimm_major, "dimmctl");
1353 bus_unregister(&nvdimm_bus_type);
1354 ida_destroy(&nd_ida);