2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/vmalloc.h>
15 #include <linux/uaccess.h>
16 #include <linux/module.h>
17 #include <linux/blkdev.h>
18 #include <linux/fcntl.h>
19 #include <linux/async.h>
20 #include <linux/genhd.h>
21 #include <linux/ndctl.h>
22 #include <linux/sched.h>
23 #include <linux/slab.h>
32 static int nvdimm_bus_major;
33 static struct class *nd_class;
35 static int to_nd_device_type(struct device *dev)
38 return ND_DEVICE_DIMM;
39 else if (is_nd_pmem(dev))
40 return ND_DEVICE_REGION_PMEM;
41 else if (is_nd_blk(dev))
42 return ND_DEVICE_REGION_BLK;
43 else if (is_nd_pmem(dev->parent) || is_nd_blk(dev->parent))
44 return nd_region_to_nstype(to_nd_region(dev->parent));
49 static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
52 * Ensure that region devices always have their numa node set as
55 if (is_nd_pmem(dev) || is_nd_blk(dev))
56 set_dev_node(dev, to_nd_region(dev)->numa_node);
57 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
58 to_nd_device_type(dev));
61 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
63 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
65 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
68 static struct module *to_bus_provider(struct device *dev)
70 /* pin bus providers while regions are enabled */
71 if (is_nd_pmem(dev) || is_nd_blk(dev)) {
72 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
74 return nvdimm_bus->module;
79 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
81 nvdimm_bus_lock(&nvdimm_bus->dev);
82 nvdimm_bus->probe_active++;
83 nvdimm_bus_unlock(&nvdimm_bus->dev);
86 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
88 nvdimm_bus_lock(&nvdimm_bus->dev);
89 if (--nvdimm_bus->probe_active == 0)
90 wake_up(&nvdimm_bus->probe_wait);
91 nvdimm_bus_unlock(&nvdimm_bus->dev);
94 static int nvdimm_bus_probe(struct device *dev)
96 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
97 struct module *provider = to_bus_provider(dev);
98 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
101 if (!try_module_get(provider))
104 nvdimm_bus_probe_start(nvdimm_bus);
105 rc = nd_drv->probe(dev);
107 nd_region_probe_success(nvdimm_bus, dev);
109 nd_region_disable(nvdimm_bus, dev);
110 nvdimm_bus_probe_end(nvdimm_bus);
112 dev_dbg(&nvdimm_bus->dev, "%s.probe(%s) = %d\n", dev->driver->name,
116 module_put(provider);
120 static int nvdimm_bus_remove(struct device *dev)
122 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
123 struct module *provider = to_bus_provider(dev);
124 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
127 rc = nd_drv->remove(dev);
128 nd_region_disable(nvdimm_bus, dev);
130 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
132 module_put(provider);
136 void nd_device_notify(struct device *dev, enum nvdimm_event event)
140 struct nd_device_driver *nd_drv;
142 nd_drv = to_nd_device_driver(dev->driver);
144 nd_drv->notify(dev, event);
148 EXPORT_SYMBOL(nd_device_notify);
150 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
152 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
157 /* caller is responsible for holding a reference on the device */
158 nd_device_notify(&nd_region->dev, event);
160 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
162 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
165 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
166 struct nvdimm_bus_descriptor *nd_desc;
167 struct nd_cmd_clear_error clear_err;
168 struct nd_cmd_ars_cap ars_cap;
169 u32 clear_err_unit, mask;
175 nd_desc = nvdimm_bus->nd_desc;
179 memset(&ars_cap, 0, sizeof(ars_cap));
180 ars_cap.address = phys;
181 ars_cap.length = len;
182 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
183 sizeof(ars_cap), &cmd_rc);
188 clear_err_unit = ars_cap.clear_err_unit;
189 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
192 mask = clear_err_unit - 1;
193 if ((phys | len) & mask)
195 memset(&clear_err, 0, sizeof(clear_err));
196 clear_err.address = phys;
197 clear_err.length = len;
198 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
199 sizeof(clear_err), &cmd_rc);
204 return clear_err.cleared;
206 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
208 static struct bus_type nvdimm_bus_type = {
210 .uevent = nvdimm_bus_uevent,
211 .match = nvdimm_bus_match,
212 .probe = nvdimm_bus_probe,
213 .remove = nvdimm_bus_remove,
216 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
218 void nd_synchronize(void)
220 async_synchronize_full_domain(&nd_async_domain);
222 EXPORT_SYMBOL_GPL(nd_synchronize);
224 static void nd_async_device_register(void *d, async_cookie_t cookie)
226 struct device *dev = d;
228 if (device_add(dev) != 0) {
229 dev_err(dev, "%s: failed\n", __func__);
235 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
237 struct device *dev = d;
239 /* flush bus operations before delete */
240 nvdimm_bus_lock(dev);
241 nvdimm_bus_unlock(dev);
243 device_unregister(dev);
247 void __nd_device_register(struct device *dev)
249 dev->bus = &nvdimm_bus_type;
251 async_schedule_domain(nd_async_device_register, dev,
255 void nd_device_register(struct device *dev)
257 device_initialize(dev);
258 __nd_device_register(dev);
260 EXPORT_SYMBOL(nd_device_register);
262 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
267 async_schedule_domain(nd_async_device_unregister, dev,
272 device_unregister(dev);
276 EXPORT_SYMBOL(nd_device_unregister);
279 * __nd_driver_register() - register a region or a namespace driver
280 * @nd_drv: driver to register
281 * @owner: automatically set by nd_driver_register() macro
282 * @mod_name: automatically set by nd_driver_register() macro
284 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
285 const char *mod_name)
287 struct device_driver *drv = &nd_drv->drv;
290 pr_debug("driver type bitmask not set (%pf)\n",
291 __builtin_return_address(0));
295 if (!nd_drv->probe || !nd_drv->remove) {
296 pr_debug("->probe() and ->remove() must be specified\n");
300 drv->bus = &nvdimm_bus_type;
302 drv->mod_name = mod_name;
304 return driver_register(drv);
306 EXPORT_SYMBOL(__nd_driver_register);
308 int nvdimm_revalidate_disk(struct gendisk *disk)
310 struct device *dev = disk->driverfs_dev;
311 struct nd_region *nd_region = to_nd_region(dev->parent);
312 const char *pol = nd_region->ro ? "only" : "write";
314 if (nd_region->ro == get_disk_ro(disk))
317 dev_info(dev, "%s read-%s, marking %s read-%s\n",
318 dev_name(&nd_region->dev), pol, disk->disk_name, pol);
319 set_disk_ro(disk, nd_region->ro);
324 EXPORT_SYMBOL(nvdimm_revalidate_disk);
326 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
329 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
330 to_nd_device_type(dev));
332 static DEVICE_ATTR_RO(modalias);
334 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
337 return sprintf(buf, "%s\n", dev->type->name);
339 static DEVICE_ATTR_RO(devtype);
341 static struct attribute *nd_device_attributes[] = {
342 &dev_attr_modalias.attr,
343 &dev_attr_devtype.attr,
348 * nd_device_attribute_group - generic attributes for all devices on an nd bus
350 struct attribute_group nd_device_attribute_group = {
351 .attrs = nd_device_attributes,
353 EXPORT_SYMBOL_GPL(nd_device_attribute_group);
355 static ssize_t numa_node_show(struct device *dev,
356 struct device_attribute *attr, char *buf)
358 return sprintf(buf, "%d\n", dev_to_node(dev));
360 static DEVICE_ATTR_RO(numa_node);
362 static struct attribute *nd_numa_attributes[] = {
363 &dev_attr_numa_node.attr,
367 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
370 if (!IS_ENABLED(CONFIG_NUMA))
377 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
379 struct attribute_group nd_numa_attribute_group = {
380 .attrs = nd_numa_attributes,
381 .is_visible = nd_numa_attr_visible,
383 EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
385 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
387 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
390 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
391 "ndctl%d", nvdimm_bus->id);
394 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
395 nvdimm_bus->id, PTR_ERR(dev));
401 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
403 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
406 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
407 [ND_CMD_IMPLEMENTED] = { },
410 .out_sizes = { 4, 128, },
412 [ND_CMD_SMART_THRESHOLD] = {
414 .out_sizes = { 4, 8, },
416 [ND_CMD_DIMM_FLAGS] = {
418 .out_sizes = { 4, 4 },
420 [ND_CMD_GET_CONFIG_SIZE] = {
422 .out_sizes = { 4, 4, 4, },
424 [ND_CMD_GET_CONFIG_DATA] = {
426 .in_sizes = { 4, 4, },
428 .out_sizes = { 4, UINT_MAX, },
430 [ND_CMD_SET_CONFIG_DATA] = {
432 .in_sizes = { 4, 4, UINT_MAX, },
438 .in_sizes = { 4, 4, UINT_MAX, },
440 .out_sizes = { 4, 4, UINT_MAX, },
444 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
446 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
447 return &__nd_cmd_dimm_descs[cmd];
450 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
452 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
453 [ND_CMD_IMPLEMENTED] = { },
456 .in_sizes = { 8, 8, },
458 .out_sizes = { 4, 4, 4, 4, },
460 [ND_CMD_ARS_START] = {
462 .in_sizes = { 8, 8, 2, 1, 5, },
464 .out_sizes = { 4, 4, },
466 [ND_CMD_ARS_STATUS] = {
468 .out_sizes = { 4, 4, UINT_MAX, },
470 [ND_CMD_CLEAR_ERROR] = {
472 .in_sizes = { 8, 8, },
474 .out_sizes = { 4, 4, 8, },
478 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
480 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
481 return &__nd_cmd_bus_descs[cmd];
484 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
486 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
487 const struct nd_cmd_desc *desc, int idx, void *buf)
489 if (idx >= desc->in_num)
492 if (desc->in_sizes[idx] < UINT_MAX)
493 return desc->in_sizes[idx];
495 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
496 struct nd_cmd_set_config_hdr *hdr = buf;
498 return hdr->in_length;
499 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
500 struct nd_cmd_vendor_hdr *hdr = buf;
502 return hdr->in_length;
507 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
509 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
510 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
511 const u32 *out_field)
513 if (idx >= desc->out_num)
516 if (desc->out_sizes[idx] < UINT_MAX)
517 return desc->out_sizes[idx];
519 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
521 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
523 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2)
524 return out_field[1] - 8;
528 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
530 void wait_nvdimm_bus_probe_idle(struct device *dev)
532 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
535 if (nvdimm_bus->probe_active == 0)
537 nvdimm_bus_unlock(&nvdimm_bus->dev);
538 wait_event(nvdimm_bus->probe_wait,
539 nvdimm_bus->probe_active == 0);
540 nvdimm_bus_lock(&nvdimm_bus->dev);
544 static int pmem_active(struct device *dev, void *data)
546 if (is_nd_pmem(dev) && dev->driver)
551 /* set_config requires an idle interleave set */
552 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
553 struct nvdimm *nvdimm, unsigned int cmd)
555 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
557 /* ask the bus provider if it would like to block this request */
558 if (nd_desc->clear_to_send) {
559 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd);
565 /* require clear error to go through the pmem driver */
566 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
567 return device_for_each_child(&nvdimm_bus->dev, NULL,
570 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
573 /* prevent label manipulation while the kernel owns label updates */
574 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
575 if (atomic_read(&nvdimm->busy))
580 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
581 int read_only, unsigned int ioctl_cmd, unsigned long arg)
583 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
584 size_t buf_len = 0, in_len = 0, out_len = 0;
585 static char out_env[ND_CMD_MAX_ENVELOPE];
586 static char in_env[ND_CMD_MAX_ENVELOPE];
587 const struct nd_cmd_desc *desc = NULL;
588 unsigned int cmd = _IOC_NR(ioctl_cmd);
589 void __user *p = (void __user *) arg;
590 struct device *dev = &nvdimm_bus->dev;
591 const char *cmd_name, *dimm_name;
592 unsigned long dsm_mask;
597 desc = nd_cmd_dimm_desc(cmd);
598 cmd_name = nvdimm_cmd_name(cmd);
599 dsm_mask = nvdimm->dsm_mask ? *(nvdimm->dsm_mask) : 0;
600 dimm_name = dev_name(&nvdimm->dev);
602 desc = nd_cmd_bus_desc(cmd);
603 cmd_name = nvdimm_bus_cmd_name(cmd);
604 dsm_mask = nd_desc->dsm_mask;
608 if (!desc || (desc->out_num + desc->in_num == 0) ||
609 !test_bit(cmd, &dsm_mask))
612 /* fail write commands (when read-only) */
616 case ND_CMD_SET_CONFIG_DATA:
617 case ND_CMD_ARS_START:
618 case ND_CMD_CLEAR_ERROR:
619 dev_dbg(&nvdimm_bus->dev, "'%s' command while read-only.\n",
620 nvdimm ? nvdimm_cmd_name(cmd)
621 : nvdimm_bus_cmd_name(cmd));
627 /* process an input envelope */
628 for (i = 0; i < desc->in_num; i++) {
631 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
632 if (in_size == UINT_MAX) {
633 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
634 __func__, dimm_name, cmd_name, i);
637 if (in_len < sizeof(in_env))
638 copy = min_t(u32, sizeof(in_env) - in_len, in_size);
641 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy))
646 /* process an output envelope */
647 for (i = 0; i < desc->out_num; i++) {
648 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
649 (u32 *) in_env, (u32 *) out_env);
652 if (out_size == UINT_MAX) {
653 dev_dbg(dev, "%s:%s unknown output size cmd: %s field: %d\n",
654 __func__, dimm_name, cmd_name, i);
657 if (out_len < sizeof(out_env))
658 copy = min_t(u32, sizeof(out_env) - out_len, out_size);
661 if (copy && copy_from_user(&out_env[out_len],
662 p + in_len + out_len, copy))
667 buf_len = out_len + in_len;
668 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
669 dev_dbg(dev, "%s:%s cmd: %s buf_len: %zu > %d\n", __func__,
670 dimm_name, cmd_name, buf_len,
671 ND_IOCTL_MAX_BUFLEN);
675 buf = vmalloc(buf_len);
679 if (copy_from_user(buf, p, buf_len)) {
684 nvdimm_bus_lock(&nvdimm_bus->dev);
685 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, cmd);
689 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, NULL);
692 if (copy_to_user(p, buf, buf_len))
695 nvdimm_bus_unlock(&nvdimm_bus->dev);
701 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
703 long id = (long) file->private_data;
705 struct nvdimm_bus *nvdimm_bus;
707 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
708 mutex_lock(&nvdimm_bus_list_mutex);
709 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
710 if (nvdimm_bus->id == id) {
711 rc = __nd_ioctl(nvdimm_bus, NULL, ro, cmd, arg);
715 mutex_unlock(&nvdimm_bus_list_mutex);
720 static int match_dimm(struct device *dev, void *data)
722 long id = (long) data;
724 if (is_nvdimm(dev)) {
725 struct nvdimm *nvdimm = to_nvdimm(dev);
727 return nvdimm->id == id;
733 static long nvdimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
736 struct nvdimm_bus *nvdimm_bus;
738 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
739 mutex_lock(&nvdimm_bus_list_mutex);
740 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
741 struct device *dev = device_find_child(&nvdimm_bus->dev,
742 file->private_data, match_dimm);
743 struct nvdimm *nvdimm;
748 nvdimm = to_nvdimm(dev);
749 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
753 mutex_unlock(&nvdimm_bus_list_mutex);
758 static int nd_open(struct inode *inode, struct file *file)
760 long minor = iminor(inode);
762 file->private_data = (void *) minor;
766 static const struct file_operations nvdimm_bus_fops = {
767 .owner = THIS_MODULE,
769 .unlocked_ioctl = nd_ioctl,
770 .compat_ioctl = nd_ioctl,
771 .llseek = noop_llseek,
774 static const struct file_operations nvdimm_fops = {
775 .owner = THIS_MODULE,
777 .unlocked_ioctl = nvdimm_ioctl,
778 .compat_ioctl = nvdimm_ioctl,
779 .llseek = noop_llseek,
782 int __init nvdimm_bus_init(void)
786 rc = bus_register(&nvdimm_bus_type);
790 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
793 nvdimm_bus_major = rc;
795 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
797 goto err_dimm_chrdev;
800 nd_class = class_create(THIS_MODULE, "nd");
801 if (IS_ERR(nd_class)) {
802 rc = PTR_ERR(nd_class);
809 unregister_chrdev(nvdimm_major, "dimmctl");
811 unregister_chrdev(nvdimm_bus_major, "ndctl");
813 bus_unregister(&nvdimm_bus_type);
818 void nvdimm_bus_exit(void)
820 class_destroy(nd_class);
821 unregister_chrdev(nvdimm_bus_major, "ndctl");
822 unregister_chrdev(nvdimm_major, "dimmctl");
823 bus_unregister(&nvdimm_bus_type);