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);
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);
129 static void nvdimm_bus_shutdown(struct device *dev)
131 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
132 struct nd_device_driver *nd_drv = NULL;
135 nd_drv = to_nd_device_driver(dev->driver);
137 if (nd_drv && nd_drv->shutdown) {
138 nd_drv->shutdown(dev);
139 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
140 dev->driver->name, dev_name(dev));
144 void nd_device_notify(struct device *dev, enum nvdimm_event event)
148 struct nd_device_driver *nd_drv;
150 nd_drv = to_nd_device_driver(dev->driver);
152 nd_drv->notify(dev, event);
154 nd_device_unlock(dev);
156 EXPORT_SYMBOL(nd_device_notify);
158 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
160 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
165 /* caller is responsible for holding a reference on the device */
166 nd_device_notify(&nd_region->dev, event);
168 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
170 struct clear_badblocks_context {
171 resource_size_t phys, cleared;
174 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
176 struct clear_badblocks_context *ctx = data;
177 struct nd_region *nd_region;
178 resource_size_t ndr_end;
181 /* make sure device is a region */
185 nd_region = to_nd_region(dev);
186 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
188 /* make sure we are in the region */
189 if (ctx->phys < nd_region->ndr_start
190 || (ctx->phys + ctx->cleared) > ndr_end)
193 sector = (ctx->phys - nd_region->ndr_start) / 512;
194 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
196 if (nd_region->bb_state)
197 sysfs_notify_dirent(nd_region->bb_state);
202 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
203 phys_addr_t phys, u64 cleared)
205 struct clear_badblocks_context ctx = {
210 device_for_each_child(&nvdimm_bus->dev, &ctx,
211 nvdimm_clear_badblocks_region);
214 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
215 phys_addr_t phys, u64 cleared)
218 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
220 if (cleared > 0 && cleared / 512)
221 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
224 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
227 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
228 struct nvdimm_bus_descriptor *nd_desc;
229 struct nd_cmd_clear_error clear_err;
230 struct nd_cmd_ars_cap ars_cap;
231 u32 clear_err_unit, mask;
232 unsigned int noio_flag;
238 nd_desc = nvdimm_bus->nd_desc;
240 * if ndctl does not exist, it's PMEM_LEGACY and
241 * we want to just pretend everything is handled.
246 memset(&ars_cap, 0, sizeof(ars_cap));
247 ars_cap.address = phys;
248 ars_cap.length = len;
249 noio_flag = memalloc_noio_save();
250 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
251 sizeof(ars_cap), &cmd_rc);
252 memalloc_noio_restore(noio_flag);
257 clear_err_unit = ars_cap.clear_err_unit;
258 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
261 mask = clear_err_unit - 1;
262 if ((phys | len) & mask)
264 memset(&clear_err, 0, sizeof(clear_err));
265 clear_err.address = phys;
266 clear_err.length = len;
267 noio_flag = memalloc_noio_save();
268 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
269 sizeof(clear_err), &cmd_rc);
270 memalloc_noio_restore(noio_flag);
276 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
278 return clear_err.cleared;
280 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
282 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
284 static struct bus_type nvdimm_bus_type = {
286 .uevent = nvdimm_bus_uevent,
287 .match = nvdimm_bus_match,
288 .probe = nvdimm_bus_probe,
289 .remove = nvdimm_bus_remove,
290 .shutdown = nvdimm_bus_shutdown,
293 static void nvdimm_bus_release(struct device *dev)
295 struct nvdimm_bus *nvdimm_bus;
297 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
298 ida_simple_remove(&nd_ida, nvdimm_bus->id);
302 static const struct device_type nvdimm_bus_dev_type = {
303 .release = nvdimm_bus_release,
304 .groups = nvdimm_bus_attribute_groups,
307 bool is_nvdimm_bus(struct device *dev)
309 return dev->type == &nvdimm_bus_dev_type;
312 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
316 for (dev = nd_dev; dev; dev = dev->parent)
317 if (is_nvdimm_bus(dev))
319 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
321 return to_nvdimm_bus(dev);
325 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
327 struct nvdimm_bus *nvdimm_bus;
329 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
330 WARN_ON(!is_nvdimm_bus(dev));
333 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
335 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
337 return to_nvdimm_bus(nvdimm->dev.parent);
339 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
341 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
342 struct nvdimm_bus_descriptor *nd_desc)
344 struct nvdimm_bus *nvdimm_bus;
347 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
350 INIT_LIST_HEAD(&nvdimm_bus->list);
351 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
352 init_waitqueue_head(&nvdimm_bus->wait);
353 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
354 if (nvdimm_bus->id < 0) {
358 mutex_init(&nvdimm_bus->reconfig_mutex);
359 badrange_init(&nvdimm_bus->badrange);
360 nvdimm_bus->nd_desc = nd_desc;
361 nvdimm_bus->dev.parent = parent;
362 nvdimm_bus->dev.type = &nvdimm_bus_dev_type;
363 nvdimm_bus->dev.groups = nd_desc->attr_groups;
364 nvdimm_bus->dev.bus = &nvdimm_bus_type;
365 nvdimm_bus->dev.of_node = nd_desc->of_node;
366 dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
367 rc = device_register(&nvdimm_bus->dev);
369 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
375 put_device(&nvdimm_bus->dev);
378 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
380 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
384 device_unregister(&nvdimm_bus->dev);
386 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
388 static int child_unregister(struct device *dev, void *data)
391 * the singular ndctl class device per bus needs to be
392 * "device_destroy"ed, so skip it here
394 * i.e. remove classless children
399 if (is_nvdimm(dev)) {
400 struct nvdimm *nvdimm = to_nvdimm(dev);
401 bool dev_put = false;
403 /* We are shutting down. Make state frozen artificially. */
404 nvdimm_bus_lock(dev);
405 set_bit(NVDIMM_SECURITY_FROZEN, &nvdimm->sec.flags);
406 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
408 nvdimm_bus_unlock(dev);
409 cancel_delayed_work_sync(&nvdimm->dwork);
413 nd_device_unregister(dev, ND_SYNC);
418 static void free_badrange_list(struct list_head *badrange_list)
420 struct badrange_entry *bre, *next;
422 list_for_each_entry_safe(bre, next, badrange_list, list) {
423 list_del(&bre->list);
426 list_del_init(badrange_list);
429 static void nd_bus_remove(struct device *dev)
431 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
433 mutex_lock(&nvdimm_bus_list_mutex);
434 list_del_init(&nvdimm_bus->list);
435 mutex_unlock(&nvdimm_bus_list_mutex);
437 wait_event(nvdimm_bus->wait,
438 atomic_read(&nvdimm_bus->ioctl_active) == 0);
441 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
443 spin_lock(&nvdimm_bus->badrange.lock);
444 free_badrange_list(&nvdimm_bus->badrange.list);
445 spin_unlock(&nvdimm_bus->badrange.lock);
447 nvdimm_bus_destroy_ndctl(nvdimm_bus);
450 static int nd_bus_probe(struct device *dev)
452 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
455 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
459 mutex_lock(&nvdimm_bus_list_mutex);
460 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
461 mutex_unlock(&nvdimm_bus_list_mutex);
463 /* enable bus provider attributes to look up their local context */
464 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
469 static struct nd_device_driver nd_bus_driver = {
470 .probe = nd_bus_probe,
471 .remove = nd_bus_remove,
474 .suppress_bind_attrs = true,
475 .bus = &nvdimm_bus_type,
476 .owner = THIS_MODULE,
477 .mod_name = KBUILD_MODNAME,
481 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
483 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
485 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
488 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
491 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
493 void nd_synchronize(void)
495 async_synchronize_full_domain(&nd_async_domain);
497 EXPORT_SYMBOL_GPL(nd_synchronize);
499 static void nd_async_device_register(void *d, async_cookie_t cookie)
501 struct device *dev = d;
503 if (device_add(dev) != 0) {
504 dev_err(dev, "%s: failed\n", __func__);
509 put_device(dev->parent);
512 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
514 struct device *dev = d;
516 /* flush bus operations before delete */
517 nvdimm_bus_lock(dev);
518 nvdimm_bus_unlock(dev);
520 device_unregister(dev);
524 void __nd_device_register(struct device *dev)
530 * Ensure that region devices always have their NUMA node set as
531 * early as possible. This way we are able to make certain that
532 * any memory associated with the creation and the creation
533 * itself of the region is associated with the correct node.
535 if (is_nd_region(dev))
536 set_dev_node(dev, to_nd_region(dev)->numa_node);
538 dev->bus = &nvdimm_bus_type;
540 get_device(dev->parent);
541 if (dev_to_node(dev) == NUMA_NO_NODE)
542 set_dev_node(dev, dev_to_node(dev->parent));
546 async_schedule_dev_domain(nd_async_device_register, dev,
550 void nd_device_register(struct device *dev)
552 device_initialize(dev);
553 __nd_device_register(dev);
555 EXPORT_SYMBOL(nd_device_register);
557 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
564 * In the async case this is being triggered with the
565 * device lock held and the unregistration work needs to
566 * be moved out of line iff this is thread has won the
567 * race to schedule the deletion.
569 if (!kill_device(dev))
573 async_schedule_domain(nd_async_device_unregister, dev,
578 * In the sync case the device is being unregistered due
579 * to a state change of the parent. Claim the kill state
580 * to synchronize against other unregistration requests,
581 * or otherwise let the async path handle it if the
582 * unregistration was already queued.
585 killed = kill_device(dev);
586 nd_device_unlock(dev);
592 device_unregister(dev);
596 EXPORT_SYMBOL(nd_device_unregister);
599 * __nd_driver_register() - register a region or a namespace driver
600 * @nd_drv: driver to register
601 * @owner: automatically set by nd_driver_register() macro
602 * @mod_name: automatically set by nd_driver_register() macro
604 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
605 const char *mod_name)
607 struct device_driver *drv = &nd_drv->drv;
610 pr_debug("driver type bitmask not set (%ps)\n",
611 __builtin_return_address(0));
615 if (!nd_drv->probe) {
616 pr_debug("%s ->probe() must be specified\n", mod_name);
620 drv->bus = &nvdimm_bus_type;
622 drv->mod_name = mod_name;
624 return driver_register(drv);
626 EXPORT_SYMBOL(__nd_driver_register);
628 void nvdimm_check_and_set_ro(struct gendisk *disk)
630 struct device *dev = disk_to_dev(disk)->parent;
631 struct nd_region *nd_region = to_nd_region(dev->parent);
632 int disk_ro = get_disk_ro(disk);
634 /* catch the disk up with the region ro state */
635 if (disk_ro == nd_region->ro)
638 dev_info(dev, "%s read-%s, marking %s read-%s\n",
639 dev_name(&nd_region->dev), nd_region->ro ? "only" : "write",
640 disk->disk_name, nd_region->ro ? "only" : "write");
641 set_disk_ro(disk, nd_region->ro);
643 EXPORT_SYMBOL(nvdimm_check_and_set_ro);
645 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
648 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
649 to_nd_device_type(dev));
651 static DEVICE_ATTR_RO(modalias);
653 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
656 return sprintf(buf, "%s\n", dev->type->name);
658 static DEVICE_ATTR_RO(devtype);
660 static struct attribute *nd_device_attributes[] = {
661 &dev_attr_modalias.attr,
662 &dev_attr_devtype.attr,
667 * nd_device_attribute_group - generic attributes for all devices on an nd bus
669 const struct attribute_group nd_device_attribute_group = {
670 .attrs = nd_device_attributes,
673 static ssize_t numa_node_show(struct device *dev,
674 struct device_attribute *attr, char *buf)
676 return sprintf(buf, "%d\n", dev_to_node(dev));
678 static DEVICE_ATTR_RO(numa_node);
680 static int nvdimm_dev_to_target_node(struct device *dev)
682 struct device *parent = dev->parent;
683 struct nd_region *nd_region = NULL;
685 if (is_nd_region(dev))
686 nd_region = to_nd_region(dev);
687 else if (parent && is_nd_region(parent))
688 nd_region = to_nd_region(parent);
692 return nd_region->target_node;
695 static ssize_t target_node_show(struct device *dev,
696 struct device_attribute *attr, char *buf)
698 return sprintf(buf, "%d\n", nvdimm_dev_to_target_node(dev));
700 static DEVICE_ATTR_RO(target_node);
702 static struct attribute *nd_numa_attributes[] = {
703 &dev_attr_numa_node.attr,
704 &dev_attr_target_node.attr,
708 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
711 struct device *dev = container_of(kobj, typeof(*dev), kobj);
713 if (!IS_ENABLED(CONFIG_NUMA))
716 if (a == &dev_attr_target_node.attr &&
717 nvdimm_dev_to_target_node(dev) == NUMA_NO_NODE)
724 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
726 const struct attribute_group nd_numa_attribute_group = {
727 .attrs = nd_numa_attributes,
728 .is_visible = nd_numa_attr_visible,
731 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
733 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
736 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
737 "ndctl%d", nvdimm_bus->id);
740 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
741 nvdimm_bus->id, PTR_ERR(dev));
742 return PTR_ERR_OR_ZERO(dev);
745 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
747 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
750 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
751 [ND_CMD_IMPLEMENTED] = { },
754 .out_sizes = { 4, 128, },
756 [ND_CMD_SMART_THRESHOLD] = {
758 .out_sizes = { 4, 8, },
760 [ND_CMD_DIMM_FLAGS] = {
762 .out_sizes = { 4, 4 },
764 [ND_CMD_GET_CONFIG_SIZE] = {
766 .out_sizes = { 4, 4, 4, },
768 [ND_CMD_GET_CONFIG_DATA] = {
770 .in_sizes = { 4, 4, },
772 .out_sizes = { 4, UINT_MAX, },
774 [ND_CMD_SET_CONFIG_DATA] = {
776 .in_sizes = { 4, 4, UINT_MAX, },
782 .in_sizes = { 4, 4, UINT_MAX, },
784 .out_sizes = { 4, 4, UINT_MAX, },
788 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
790 .out_sizes = { UINT_MAX, },
794 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
796 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
797 return &__nd_cmd_dimm_descs[cmd];
800 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
802 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
803 [ND_CMD_IMPLEMENTED] = { },
806 .in_sizes = { 8, 8, },
808 .out_sizes = { 4, 4, 4, 4, },
810 [ND_CMD_ARS_START] = {
812 .in_sizes = { 8, 8, 2, 1, 5, },
814 .out_sizes = { 4, 4, },
816 [ND_CMD_ARS_STATUS] = {
818 .out_sizes = { 4, 4, UINT_MAX, },
820 [ND_CMD_CLEAR_ERROR] = {
822 .in_sizes = { 8, 8, },
824 .out_sizes = { 4, 4, 8, },
828 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
830 .out_sizes = { UINT_MAX, },
834 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
836 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
837 return &__nd_cmd_bus_descs[cmd];
840 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
842 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
843 const struct nd_cmd_desc *desc, int idx, void *buf)
845 if (idx >= desc->in_num)
848 if (desc->in_sizes[idx] < UINT_MAX)
849 return desc->in_sizes[idx];
851 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
852 struct nd_cmd_set_config_hdr *hdr = buf;
854 return hdr->in_length;
855 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
856 struct nd_cmd_vendor_hdr *hdr = buf;
858 return hdr->in_length;
859 } else if (cmd == ND_CMD_CALL) {
860 struct nd_cmd_pkg *pkg = buf;
862 return pkg->nd_size_in;
867 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
869 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
870 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
871 const u32 *out_field, unsigned long remainder)
873 if (idx >= desc->out_num)
876 if (desc->out_sizes[idx] < UINT_MAX)
877 return desc->out_sizes[idx];
879 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
881 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
883 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
885 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
886 * "Size of Output Buffer in bytes, including this
889 if (out_field[1] < 4)
892 * ACPI 6.1 is ambiguous if 'status' is included in the
893 * output size. If we encounter an output size that
894 * overshoots the remainder by 4 bytes, assume it was
895 * including 'status'.
897 if (out_field[1] - 4 == remainder)
899 return out_field[1] - 8;
900 } else if (cmd == ND_CMD_CALL) {
901 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
903 return pkg->nd_size_out;
909 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
911 void wait_nvdimm_bus_probe_idle(struct device *dev)
913 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
916 if (nvdimm_bus->probe_active == 0)
918 nvdimm_bus_unlock(dev);
919 nd_device_unlock(dev);
920 wait_event(nvdimm_bus->wait,
921 nvdimm_bus->probe_active == 0);
923 nvdimm_bus_lock(dev);
927 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
929 struct nd_cmd_clear_error *clear_err =
930 (struct nd_cmd_clear_error *)data;
931 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
932 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
933 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
934 struct nd_namespace_common *ndns = NULL;
935 struct nd_namespace_io *nsio;
936 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
938 if (nd_dax || !dev->driver)
941 start = clear_err->address;
942 end = clear_err->address + clear_err->cleared - 1;
944 if (nd_btt || nd_pfn || nd_dax) {
950 ndns = nd_dax->nd_pfn.ndns;
957 nsio = to_nd_namespace_io(&ndns->dev);
958 pstart = nsio->res.start + offset;
959 pend = nsio->res.end - end_trunc;
961 if ((pstart >= start) && (pend <= end))
968 static int nd_ns_forget_poison_check(struct device *dev, void *data)
970 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
973 /* set_config requires an idle interleave set */
974 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
975 struct nvdimm *nvdimm, unsigned int cmd, void *data)
977 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
979 /* ask the bus provider if it would like to block this request */
980 if (nd_desc->clear_to_send) {
981 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
987 /* require clear error to go through the pmem driver */
988 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
989 return device_for_each_child(&nvdimm_bus->dev, data,
990 nd_ns_forget_poison_check);
992 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
995 /* prevent label manipulation while the kernel owns label updates */
996 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
997 if (atomic_read(&nvdimm->busy))
1002 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
1003 int read_only, unsigned int ioctl_cmd, unsigned long arg)
1005 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
1006 const struct nd_cmd_desc *desc = NULL;
1007 unsigned int cmd = _IOC_NR(ioctl_cmd);
1008 struct device *dev = &nvdimm_bus->dev;
1009 void __user *p = (void __user *) arg;
1010 char *out_env = NULL, *in_env = NULL;
1011 const char *cmd_name, *dimm_name;
1012 u32 in_len = 0, out_len = 0;
1013 unsigned int func = cmd;
1014 unsigned long cmd_mask;
1015 struct nd_cmd_pkg pkg;
1021 desc = nd_cmd_dimm_desc(cmd);
1022 cmd_name = nvdimm_cmd_name(cmd);
1023 cmd_mask = nvdimm->cmd_mask;
1024 dimm_name = dev_name(&nvdimm->dev);
1026 desc = nd_cmd_bus_desc(cmd);
1027 cmd_name = nvdimm_bus_cmd_name(cmd);
1028 cmd_mask = nd_desc->cmd_mask;
1032 /* Validate command family support against bus declared support */
1033 if (cmd == ND_CMD_CALL) {
1034 unsigned long *mask;
1036 if (copy_from_user(&pkg, p, sizeof(pkg)))
1040 if (pkg.nd_family > NVDIMM_FAMILY_MAX)
1042 mask = &nd_desc->dimm_family_mask;
1044 if (pkg.nd_family > NVDIMM_BUS_FAMILY_MAX)
1046 mask = &nd_desc->bus_family_mask;
1049 if (!test_bit(pkg.nd_family, mask))
1054 (desc->out_num + desc->in_num == 0) ||
1055 cmd > ND_CMD_CALL ||
1056 !test_bit(cmd, &cmd_mask))
1059 /* fail write commands (when read-only) */
1063 case ND_CMD_SET_CONFIG_DATA:
1064 case ND_CMD_ARS_START:
1065 case ND_CMD_CLEAR_ERROR:
1067 dev_dbg(dev, "'%s' command while read-only.\n",
1068 nvdimm ? nvdimm_cmd_name(cmd)
1069 : nvdimm_bus_cmd_name(cmd));
1075 /* process an input envelope */
1076 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1079 for (i = 0; i < desc->in_num; i++) {
1082 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1083 if (in_size == UINT_MAX) {
1084 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1085 __func__, dimm_name, cmd_name, i);
1089 if (in_len < ND_CMD_MAX_ENVELOPE)
1090 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1093 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1100 if (cmd == ND_CMD_CALL) {
1101 func = pkg.nd_command;
1102 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1103 dimm_name, pkg.nd_command,
1104 in_len, out_len, buf_len);
1107 /* process an output envelope */
1108 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1114 for (i = 0; i < desc->out_num; i++) {
1115 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1116 (u32 *) in_env, (u32 *) out_env, 0);
1119 if (out_size == UINT_MAX) {
1120 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1121 dimm_name, cmd_name, i);
1125 if (out_len < ND_CMD_MAX_ENVELOPE)
1126 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1129 if (copy && copy_from_user(&out_env[out_len],
1130 p + in_len + out_len, copy)) {
1134 out_len += out_size;
1137 buf_len = (u64) out_len + (u64) in_len;
1138 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1139 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1140 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1145 buf = vmalloc(buf_len);
1151 if (copy_from_user(buf, p, buf_len)) {
1156 nd_device_lock(dev);
1157 nvdimm_bus_lock(dev);
1158 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1162 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1166 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1167 struct nd_cmd_clear_error *clear_err = buf;
1169 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1170 clear_err->cleared);
1173 if (copy_to_user(p, buf, buf_len))
1177 nvdimm_bus_unlock(dev);
1178 nd_device_unlock(dev);
1186 enum nd_ioctl_mode {
1191 static int match_dimm(struct device *dev, void *data)
1193 long id = (long) data;
1195 if (is_nvdimm(dev)) {
1196 struct nvdimm *nvdimm = to_nvdimm(dev);
1198 return nvdimm->id == id;
1204 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1205 enum nd_ioctl_mode mode)
1208 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1209 long id = (long) file->private_data;
1210 struct nvdimm *nvdimm = NULL;
1213 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1214 mutex_lock(&nvdimm_bus_list_mutex);
1215 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1216 if (mode == DIMM_IOCTL) {
1219 dev = device_find_child(&nvdimm_bus->dev,
1220 file->private_data, match_dimm);
1223 nvdimm = to_nvdimm(dev);
1225 } else if (nvdimm_bus->id == id) {
1230 atomic_inc(&nvdimm_bus->ioctl_active);
1234 mutex_unlock(&nvdimm_bus_list_mutex);
1240 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1243 put_device(&nvdimm->dev);
1244 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1245 wake_up(&nvdimm_bus->wait);
1250 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1252 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1255 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1257 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1260 static int nd_open(struct inode *inode, struct file *file)
1262 long minor = iminor(inode);
1264 file->private_data = (void *) minor;
1268 static const struct file_operations nvdimm_bus_fops = {
1269 .owner = THIS_MODULE,
1271 .unlocked_ioctl = bus_ioctl,
1272 .compat_ioctl = compat_ptr_ioctl,
1273 .llseek = noop_llseek,
1276 static const struct file_operations nvdimm_fops = {
1277 .owner = THIS_MODULE,
1279 .unlocked_ioctl = dimm_ioctl,
1280 .compat_ioctl = compat_ptr_ioctl,
1281 .llseek = noop_llseek,
1284 int __init nvdimm_bus_init(void)
1288 rc = bus_register(&nvdimm_bus_type);
1292 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1294 goto err_bus_chrdev;
1295 nvdimm_bus_major = rc;
1297 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1299 goto err_dimm_chrdev;
1302 nd_class = class_create(THIS_MODULE, "nd");
1303 if (IS_ERR(nd_class)) {
1304 rc = PTR_ERR(nd_class);
1308 rc = driver_register(&nd_bus_driver.drv);
1315 class_destroy(nd_class);
1317 unregister_chrdev(nvdimm_major, "dimmctl");
1319 unregister_chrdev(nvdimm_bus_major, "ndctl");
1321 bus_unregister(&nvdimm_bus_type);
1326 void nvdimm_bus_exit(void)
1328 driver_unregister(&nd_bus_driver.drv);
1329 class_destroy(nd_class);
1330 unregister_chrdev(nvdimm_bus_major, "ndctl");
1331 unregister_chrdev(nvdimm_major, "dimmctl");
1332 bus_unregister(&nvdimm_bus_type);
1333 ida_destroy(&nd_ida);