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 void 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);
117 if (nd_drv->remove) {
118 debug_nvdimm_lock(dev);
120 debug_nvdimm_unlock(dev);
123 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s)\n", dev->driver->name,
125 module_put(provider);
128 static void nvdimm_bus_shutdown(struct device *dev)
130 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
131 struct nd_device_driver *nd_drv = NULL;
134 nd_drv = to_nd_device_driver(dev->driver);
136 if (nd_drv && nd_drv->shutdown) {
137 nd_drv->shutdown(dev);
138 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
139 dev->driver->name, dev_name(dev));
143 void nd_device_notify(struct device *dev, enum nvdimm_event event)
147 struct nd_device_driver *nd_drv;
149 nd_drv = to_nd_device_driver(dev->driver);
151 nd_drv->notify(dev, event);
153 nd_device_unlock(dev);
155 EXPORT_SYMBOL(nd_device_notify);
157 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
159 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
164 /* caller is responsible for holding a reference on the device */
165 nd_device_notify(&nd_region->dev, event);
167 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
169 struct clear_badblocks_context {
170 resource_size_t phys, cleared;
173 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
175 struct clear_badblocks_context *ctx = data;
176 struct nd_region *nd_region;
177 resource_size_t ndr_end;
180 /* make sure device is a region */
184 nd_region = to_nd_region(dev);
185 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
187 /* make sure we are in the region */
188 if (ctx->phys < nd_region->ndr_start
189 || (ctx->phys + ctx->cleared) > ndr_end)
192 sector = (ctx->phys - nd_region->ndr_start) / 512;
193 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
195 if (nd_region->bb_state)
196 sysfs_notify_dirent(nd_region->bb_state);
201 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
202 phys_addr_t phys, u64 cleared)
204 struct clear_badblocks_context ctx = {
209 device_for_each_child(&nvdimm_bus->dev, &ctx,
210 nvdimm_clear_badblocks_region);
213 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
214 phys_addr_t phys, u64 cleared)
217 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
219 if (cleared > 0 && cleared / 512)
220 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
223 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
226 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
227 struct nvdimm_bus_descriptor *nd_desc;
228 struct nd_cmd_clear_error clear_err;
229 struct nd_cmd_ars_cap ars_cap;
230 u32 clear_err_unit, mask;
231 unsigned int noio_flag;
237 nd_desc = nvdimm_bus->nd_desc;
239 * if ndctl does not exist, it's PMEM_LEGACY and
240 * we want to just pretend everything is handled.
245 memset(&ars_cap, 0, sizeof(ars_cap));
246 ars_cap.address = phys;
247 ars_cap.length = len;
248 noio_flag = memalloc_noio_save();
249 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
250 sizeof(ars_cap), &cmd_rc);
251 memalloc_noio_restore(noio_flag);
256 clear_err_unit = ars_cap.clear_err_unit;
257 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
260 mask = clear_err_unit - 1;
261 if ((phys | len) & mask)
263 memset(&clear_err, 0, sizeof(clear_err));
264 clear_err.address = phys;
265 clear_err.length = len;
266 noio_flag = memalloc_noio_save();
267 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
268 sizeof(clear_err), &cmd_rc);
269 memalloc_noio_restore(noio_flag);
275 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
277 return clear_err.cleared;
279 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
281 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
283 static struct bus_type nvdimm_bus_type = {
285 .uevent = nvdimm_bus_uevent,
286 .match = nvdimm_bus_match,
287 .probe = nvdimm_bus_probe,
288 .remove = nvdimm_bus_remove,
289 .shutdown = nvdimm_bus_shutdown,
292 static void nvdimm_bus_release(struct device *dev)
294 struct nvdimm_bus *nvdimm_bus;
296 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
297 ida_simple_remove(&nd_ida, nvdimm_bus->id);
301 static const struct device_type nvdimm_bus_dev_type = {
302 .release = nvdimm_bus_release,
303 .groups = nvdimm_bus_attribute_groups,
306 bool is_nvdimm_bus(struct device *dev)
308 return dev->type == &nvdimm_bus_dev_type;
311 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
315 for (dev = nd_dev; dev; dev = dev->parent)
316 if (is_nvdimm_bus(dev))
318 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
320 return to_nvdimm_bus(dev);
324 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
326 struct nvdimm_bus *nvdimm_bus;
328 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
329 WARN_ON(!is_nvdimm_bus(dev));
332 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
334 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
336 return to_nvdimm_bus(nvdimm->dev.parent);
338 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
340 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
341 struct nvdimm_bus_descriptor *nd_desc)
343 struct nvdimm_bus *nvdimm_bus;
346 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
349 INIT_LIST_HEAD(&nvdimm_bus->list);
350 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
351 init_waitqueue_head(&nvdimm_bus->wait);
352 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
353 if (nvdimm_bus->id < 0) {
357 mutex_init(&nvdimm_bus->reconfig_mutex);
358 badrange_init(&nvdimm_bus->badrange);
359 nvdimm_bus->nd_desc = nd_desc;
360 nvdimm_bus->dev.parent = parent;
361 nvdimm_bus->dev.type = &nvdimm_bus_dev_type;
362 nvdimm_bus->dev.groups = nd_desc->attr_groups;
363 nvdimm_bus->dev.bus = &nvdimm_bus_type;
364 nvdimm_bus->dev.of_node = nd_desc->of_node;
365 device_initialize(&nvdimm_bus->dev);
366 device_set_pm_not_required(&nvdimm_bus->dev);
367 rc = dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
371 rc = device_add(&nvdimm_bus->dev);
373 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
379 put_device(&nvdimm_bus->dev);
382 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
384 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
388 device_unregister(&nvdimm_bus->dev);
390 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
392 static int child_unregister(struct device *dev, void *data)
395 * the singular ndctl class device per bus needs to be
396 * "device_destroy"ed, so skip it here
398 * i.e. remove classless children
404 nvdimm_delete(to_nvdimm(dev));
406 nd_device_unregister(dev, ND_SYNC);
411 static void free_badrange_list(struct list_head *badrange_list)
413 struct badrange_entry *bre, *next;
415 list_for_each_entry_safe(bre, next, badrange_list, list) {
416 list_del(&bre->list);
419 list_del_init(badrange_list);
422 static void nd_bus_remove(struct device *dev)
424 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
426 mutex_lock(&nvdimm_bus_list_mutex);
427 list_del_init(&nvdimm_bus->list);
428 mutex_unlock(&nvdimm_bus_list_mutex);
430 wait_event(nvdimm_bus->wait,
431 atomic_read(&nvdimm_bus->ioctl_active) == 0);
434 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
436 spin_lock(&nvdimm_bus->badrange.lock);
437 free_badrange_list(&nvdimm_bus->badrange.list);
438 spin_unlock(&nvdimm_bus->badrange.lock);
440 nvdimm_bus_destroy_ndctl(nvdimm_bus);
443 static int nd_bus_probe(struct device *dev)
445 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
448 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
452 mutex_lock(&nvdimm_bus_list_mutex);
453 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
454 mutex_unlock(&nvdimm_bus_list_mutex);
456 /* enable bus provider attributes to look up their local context */
457 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
462 static struct nd_device_driver nd_bus_driver = {
463 .probe = nd_bus_probe,
464 .remove = nd_bus_remove,
467 .suppress_bind_attrs = true,
468 .bus = &nvdimm_bus_type,
469 .owner = THIS_MODULE,
470 .mod_name = KBUILD_MODNAME,
474 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
476 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
478 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
481 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
484 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
486 void nd_synchronize(void)
488 async_synchronize_full_domain(&nd_async_domain);
490 EXPORT_SYMBOL_GPL(nd_synchronize);
492 static void nd_async_device_register(void *d, async_cookie_t cookie)
494 struct device *dev = d;
496 if (device_add(dev) != 0) {
497 dev_err(dev, "%s: failed\n", __func__);
502 put_device(dev->parent);
505 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
507 struct device *dev = d;
509 /* flush bus operations before delete */
510 nvdimm_bus_lock(dev);
511 nvdimm_bus_unlock(dev);
513 device_unregister(dev);
517 void __nd_device_register(struct device *dev)
523 * Ensure that region devices always have their NUMA node set as
524 * early as possible. This way we are able to make certain that
525 * any memory associated with the creation and the creation
526 * itself of the region is associated with the correct node.
528 if (is_nd_region(dev))
529 set_dev_node(dev, to_nd_region(dev)->numa_node);
531 dev->bus = &nvdimm_bus_type;
532 device_set_pm_not_required(dev);
534 get_device(dev->parent);
535 if (dev_to_node(dev) == NUMA_NO_NODE)
536 set_dev_node(dev, dev_to_node(dev->parent));
540 async_schedule_dev_domain(nd_async_device_register, dev,
544 void nd_device_register(struct device *dev)
546 device_initialize(dev);
547 __nd_device_register(dev);
549 EXPORT_SYMBOL(nd_device_register);
551 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
558 * In the async case this is being triggered with the
559 * device lock held and the unregistration work needs to
560 * be moved out of line iff this is thread has won the
561 * race to schedule the deletion.
563 if (!kill_device(dev))
567 async_schedule_domain(nd_async_device_unregister, dev,
572 * In the sync case the device is being unregistered due
573 * to a state change of the parent. Claim the kill state
574 * to synchronize against other unregistration requests,
575 * or otherwise let the async path handle it if the
576 * unregistration was already queued.
579 killed = kill_device(dev);
580 nd_device_unlock(dev);
586 device_unregister(dev);
590 EXPORT_SYMBOL(nd_device_unregister);
593 * __nd_driver_register() - register a region or a namespace driver
594 * @nd_drv: driver to register
595 * @owner: automatically set by nd_driver_register() macro
596 * @mod_name: automatically set by nd_driver_register() macro
598 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
599 const char *mod_name)
601 struct device_driver *drv = &nd_drv->drv;
604 pr_debug("driver type bitmask not set (%ps)\n",
605 __builtin_return_address(0));
609 if (!nd_drv->probe) {
610 pr_debug("%s ->probe() must be specified\n", mod_name);
614 drv->bus = &nvdimm_bus_type;
616 drv->mod_name = mod_name;
618 return driver_register(drv);
620 EXPORT_SYMBOL(__nd_driver_register);
622 void nvdimm_check_and_set_ro(struct gendisk *disk)
624 struct device *dev = disk_to_dev(disk)->parent;
625 struct nd_region *nd_region = to_nd_region(dev->parent);
626 int disk_ro = get_disk_ro(disk);
628 /* catch the disk up with the region ro state */
629 if (disk_ro == nd_region->ro)
632 dev_info(dev, "%s read-%s, marking %s read-%s\n",
633 dev_name(&nd_region->dev), nd_region->ro ? "only" : "write",
634 disk->disk_name, nd_region->ro ? "only" : "write");
635 set_disk_ro(disk, nd_region->ro);
637 EXPORT_SYMBOL(nvdimm_check_and_set_ro);
639 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
642 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
643 to_nd_device_type(dev));
645 static DEVICE_ATTR_RO(modalias);
647 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
650 return sprintf(buf, "%s\n", dev->type->name);
652 static DEVICE_ATTR_RO(devtype);
654 static struct attribute *nd_device_attributes[] = {
655 &dev_attr_modalias.attr,
656 &dev_attr_devtype.attr,
661 * nd_device_attribute_group - generic attributes for all devices on an nd bus
663 const struct attribute_group nd_device_attribute_group = {
664 .attrs = nd_device_attributes,
667 static ssize_t numa_node_show(struct device *dev,
668 struct device_attribute *attr, char *buf)
670 return sprintf(buf, "%d\n", dev_to_node(dev));
672 static DEVICE_ATTR_RO(numa_node);
674 static int nvdimm_dev_to_target_node(struct device *dev)
676 struct device *parent = dev->parent;
677 struct nd_region *nd_region = NULL;
679 if (is_nd_region(dev))
680 nd_region = to_nd_region(dev);
681 else if (parent && is_nd_region(parent))
682 nd_region = to_nd_region(parent);
686 return nd_region->target_node;
689 static ssize_t target_node_show(struct device *dev,
690 struct device_attribute *attr, char *buf)
692 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev));
694 static DEVICE_ATTR_RO(target_node);
696 static struct attribute *nd_numa_attributes[] = {
697 &dev_attr_numa_node.attr,
698 &dev_attr_target_node.attr,
702 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
705 struct device *dev = container_of(kobj, typeof(*dev), kobj);
707 if (!IS_ENABLED(CONFIG_NUMA))
710 if (a == &dev_attr_target_node.attr &&
711 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE)
718 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
720 const struct attribute_group nd_numa_attribute_group = {
721 .attrs = nd_numa_attributes,
722 .is_visible = nd_numa_attr_visible,
725 static void ndctl_release(struct device *dev)
730 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
732 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
736 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
739 device_initialize(dev);
740 device_set_pm_not_required(dev);
741 dev->class = nd_class;
742 dev->parent = &nvdimm_bus->dev;
744 dev->release = ndctl_release;
745 rc = dev_set_name(dev, "ndctl%d", nvdimm_bus->id);
749 rc = device_add(dev);
751 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %d\n",
762 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
764 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
767 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
768 [ND_CMD_IMPLEMENTED] = { },
771 .out_sizes = { 4, 128, },
773 [ND_CMD_SMART_THRESHOLD] = {
775 .out_sizes = { 4, 8, },
777 [ND_CMD_DIMM_FLAGS] = {
779 .out_sizes = { 4, 4 },
781 [ND_CMD_GET_CONFIG_SIZE] = {
783 .out_sizes = { 4, 4, 4, },
785 [ND_CMD_GET_CONFIG_DATA] = {
787 .in_sizes = { 4, 4, },
789 .out_sizes = { 4, UINT_MAX, },
791 [ND_CMD_SET_CONFIG_DATA] = {
793 .in_sizes = { 4, 4, UINT_MAX, },
799 .in_sizes = { 4, 4, UINT_MAX, },
801 .out_sizes = { 4, 4, UINT_MAX, },
805 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
807 .out_sizes = { UINT_MAX, },
811 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
813 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
814 return &__nd_cmd_dimm_descs[cmd];
817 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
819 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
820 [ND_CMD_IMPLEMENTED] = { },
823 .in_sizes = { 8, 8, },
825 .out_sizes = { 4, 4, 4, 4, },
827 [ND_CMD_ARS_START] = {
829 .in_sizes = { 8, 8, 2, 1, 5, },
831 .out_sizes = { 4, 4, },
833 [ND_CMD_ARS_STATUS] = {
835 .out_sizes = { 4, 4, UINT_MAX, },
837 [ND_CMD_CLEAR_ERROR] = {
839 .in_sizes = { 8, 8, },
841 .out_sizes = { 4, 4, 8, },
845 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
847 .out_sizes = { UINT_MAX, },
851 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
853 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
854 return &__nd_cmd_bus_descs[cmd];
857 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
859 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
860 const struct nd_cmd_desc *desc, int idx, void *buf)
862 if (idx >= desc->in_num)
865 if (desc->in_sizes[idx] < UINT_MAX)
866 return desc->in_sizes[idx];
868 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
869 struct nd_cmd_set_config_hdr *hdr = buf;
871 return hdr->in_length;
872 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
873 struct nd_cmd_vendor_hdr *hdr = buf;
875 return hdr->in_length;
876 } else if (cmd == ND_CMD_CALL) {
877 struct nd_cmd_pkg *pkg = buf;
879 return pkg->nd_size_in;
884 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
886 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
887 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
888 const u32 *out_field, unsigned long remainder)
890 if (idx >= desc->out_num)
893 if (desc->out_sizes[idx] < UINT_MAX)
894 return desc->out_sizes[idx];
896 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
898 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
900 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
902 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
903 * "Size of Output Buffer in bytes, including this
906 if (out_field[1] < 4)
909 * ACPI 6.1 is ambiguous if 'status' is included in the
910 * output size. If we encounter an output size that
911 * overshoots the remainder by 4 bytes, assume it was
912 * including 'status'.
914 if (out_field[1] - 4 == remainder)
916 return out_field[1] - 8;
917 } else if (cmd == ND_CMD_CALL) {
918 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
920 return pkg->nd_size_out;
926 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
928 void wait_nvdimm_bus_probe_idle(struct device *dev)
930 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
933 if (nvdimm_bus->probe_active == 0)
935 nvdimm_bus_unlock(dev);
936 nd_device_unlock(dev);
937 wait_event(nvdimm_bus->wait,
938 nvdimm_bus->probe_active == 0);
940 nvdimm_bus_lock(dev);
944 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
946 struct nd_cmd_clear_error *clear_err =
947 (struct nd_cmd_clear_error *)data;
948 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
949 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
950 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
951 struct nd_namespace_common *ndns = NULL;
952 struct nd_namespace_io *nsio;
953 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
955 if (nd_dax || !dev->driver)
958 start = clear_err->address;
959 end = clear_err->address + clear_err->cleared - 1;
961 if (nd_btt || nd_pfn || nd_dax) {
967 ndns = nd_dax->nd_pfn.ndns;
974 nsio = to_nd_namespace_io(&ndns->dev);
975 pstart = nsio->res.start + offset;
976 pend = nsio->res.end - end_trunc;
978 if ((pstart >= start) && (pend <= end))
985 static int nd_ns_forget_poison_check(struct device *dev, void *data)
987 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
990 /* set_config requires an idle interleave set */
991 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
992 struct nvdimm *nvdimm, unsigned int cmd, void *data)
994 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
996 /* ask the bus provider if it would like to block this request */
997 if (nd_desc->clear_to_send) {
998 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
1004 /* require clear error to go through the pmem driver */
1005 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
1006 return device_for_each_child(&nvdimm_bus->dev, data,
1007 nd_ns_forget_poison_check);
1009 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
1012 /* prevent label manipulation while the kernel owns label updates */
1013 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
1014 if (atomic_read(&nvdimm->busy))
1019 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
1020 int read_only, unsigned int ioctl_cmd, unsigned long arg)
1022 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
1023 const struct nd_cmd_desc *desc = NULL;
1024 unsigned int cmd = _IOC_NR(ioctl_cmd);
1025 struct device *dev = &nvdimm_bus->dev;
1026 void __user *p = (void __user *) arg;
1027 char *out_env = NULL, *in_env = NULL;
1028 const char *cmd_name, *dimm_name;
1029 u32 in_len = 0, out_len = 0;
1030 unsigned int func = cmd;
1031 unsigned long cmd_mask;
1032 struct nd_cmd_pkg pkg;
1038 desc = nd_cmd_dimm_desc(cmd);
1039 cmd_name = nvdimm_cmd_name(cmd);
1040 cmd_mask = nvdimm->cmd_mask;
1041 dimm_name = dev_name(&nvdimm->dev);
1043 desc = nd_cmd_bus_desc(cmd);
1044 cmd_name = nvdimm_bus_cmd_name(cmd);
1045 cmd_mask = nd_desc->cmd_mask;
1049 /* Validate command family support against bus declared support */
1050 if (cmd == ND_CMD_CALL) {
1051 unsigned long *mask;
1053 if (copy_from_user(&pkg, p, sizeof(pkg)))
1057 if (pkg.nd_family > NVDIMM_FAMILY_MAX)
1059 mask = &nd_desc->dimm_family_mask;
1061 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX)
1063 mask = &nd_desc->bus_family_mask;
1066 if (!test_bit(pkg.nd_family, mask))
1071 (desc->out_num + desc->in_num == 0) ||
1072 cmd > ND_CMD_CALL ||
1073 !test_bit(cmd, &cmd_mask))
1076 /* fail write commands (when read-only) */
1080 case ND_CMD_SET_CONFIG_DATA:
1081 case ND_CMD_ARS_START:
1082 case ND_CMD_CLEAR_ERROR:
1084 dev_dbg(dev, "'%s' command while read-only.\n",
1085 nvdimm ? nvdimm_cmd_name(cmd)
1086 : nvdimm_bus_cmd_name(cmd));
1092 /* process an input envelope */
1093 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1096 for (i = 0; i < desc->in_num; i++) {
1099 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1100 if (in_size == UINT_MAX) {
1101 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1102 __func__, dimm_name, cmd_name, i);
1106 if (in_len < ND_CMD_MAX_ENVELOPE)
1107 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1110 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1117 if (cmd == ND_CMD_CALL) {
1118 func = pkg.nd_command;
1119 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1120 dimm_name, pkg.nd_command,
1121 in_len, out_len, buf_len);
1124 /* process an output envelope */
1125 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1131 for (i = 0; i < desc->out_num; i++) {
1132 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1133 (u32 *) in_env, (u32 *) out_env, 0);
1136 if (out_size == UINT_MAX) {
1137 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1138 dimm_name, cmd_name, i);
1142 if (out_len < ND_CMD_MAX_ENVELOPE)
1143 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1146 if (copy && copy_from_user(&out_env[out_len],
1147 p + in_len + out_len, copy)) {
1151 out_len += out_size;
1154 buf_len = (u64) out_len + (u64) in_len;
1155 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1156 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1157 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1162 buf = vmalloc(buf_len);
1168 if (copy_from_user(buf, p, buf_len)) {
1173 nd_device_lock(dev);
1174 nvdimm_bus_lock(dev);
1175 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1179 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1183 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1184 struct nd_cmd_clear_error *clear_err = buf;
1186 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1187 clear_err->cleared);
1190 if (copy_to_user(p, buf, buf_len))
1194 nvdimm_bus_unlock(dev);
1195 nd_device_unlock(dev);
1203 enum nd_ioctl_mode {
1208 static int match_dimm(struct device *dev, void *data)
1210 long id = (long) data;
1212 if (is_nvdimm(dev)) {
1213 struct nvdimm *nvdimm = to_nvdimm(dev);
1215 return nvdimm->id == id;
1221 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1222 enum nd_ioctl_mode mode)
1225 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1226 long id = (long) file->private_data;
1227 struct nvdimm *nvdimm = NULL;
1230 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1231 mutex_lock(&nvdimm_bus_list_mutex);
1232 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1233 if (mode == DIMM_IOCTL) {
1236 dev = device_find_child(&nvdimm_bus->dev,
1237 file->private_data, match_dimm);
1240 nvdimm = to_nvdimm(dev);
1242 } else if (nvdimm_bus->id == id) {
1247 atomic_inc(&nvdimm_bus->ioctl_active);
1251 mutex_unlock(&nvdimm_bus_list_mutex);
1257 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1260 put_device(&nvdimm->dev);
1261 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1262 wake_up(&nvdimm_bus->wait);
1267 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1269 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1272 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1274 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1277 static int nd_open(struct inode *inode, struct file *file)
1279 long minor = iminor(inode);
1281 file->private_data = (void *) minor;
1285 static const struct file_operations nvdimm_bus_fops = {
1286 .owner = THIS_MODULE,
1288 .unlocked_ioctl = bus_ioctl,
1289 .compat_ioctl = compat_ptr_ioctl,
1290 .llseek = noop_llseek,
1293 static const struct file_operations nvdimm_fops = {
1294 .owner = THIS_MODULE,
1296 .unlocked_ioctl = dimm_ioctl,
1297 .compat_ioctl = compat_ptr_ioctl,
1298 .llseek = noop_llseek,
1301 int __init nvdimm_bus_init(void)
1305 rc = bus_register(&nvdimm_bus_type);
1309 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1311 goto err_bus_chrdev;
1312 nvdimm_bus_major = rc;
1314 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1316 goto err_dimm_chrdev;
1319 nd_class = class_create(THIS_MODULE, "nd");
1320 if (IS_ERR(nd_class)) {
1321 rc = PTR_ERR(nd_class);
1325 rc = driver_register(&nd_bus_driver.drv);
1332 class_destroy(nd_class);
1334 unregister_chrdev(nvdimm_major, "dimmctl");
1336 unregister_chrdev(nvdimm_bus_major, "ndctl");
1338 bus_unregister(&nvdimm_bus_type);
1343 void nvdimm_bus_exit(void)
1345 driver_unregister(&nd_bus_driver.drv);
1346 class_destroy(nd_class);
1347 unregister_chrdev(nvdimm_bus_major, "ndctl");
1348 unregister_chrdev(nvdimm_major, "dimmctl");
1349 bus_unregister(&nvdimm_bus_type);
1350 ida_destroy(&nd_ida);