drivers/nvdimm/bus.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   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.
   7  *
   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.
  12  */
  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>
  24 #include <linux/fs.h>
  25 #include <linux/io.h>
  26 #include <linux/mm.h>
  27 #include <linux/nd.h>
  28 #include "nd-core.h"
  29 #include "nd.h"
  30
  31 int nvdimm_major;
  32 static int nvdimm_bus_major;
  33 static struct class *nd_class;
  34 static DEFINE_IDA(nd_ida);
  35
  36 static int to_nd_device_type(struct device *dev)
  37 {
  38         if (is_nvdimm(dev))
  39                 return ND_DEVICE_DIMM;
  40         else if (is_nd_pmem(dev))
  41                 return ND_DEVICE_REGION_PMEM;
  42         else if (is_nd_blk(dev))
  43                 return ND_DEVICE_REGION_BLK;
  44         else if (is_nd_dax(dev))
  45                 return ND_DEVICE_DAX_PMEM;
  46         else if (is_nd_pmem(dev->parent) || is_nd_blk(dev->parent))
  47                 return nd_region_to_nstype(to_nd_region(dev->parent));
  48
  49         return 0;
  50 }
  51
  52 static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
  53 {
  54         /*
  55          * Ensure that region devices always have their numa node set as
  56          * early as possible.
  57          */
  58         if (is_nd_pmem(dev) || is_nd_blk(dev))
  59                 set_dev_node(dev, to_nd_region(dev)->numa_node);
  60         return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
  61                         to_nd_device_type(dev));
  62 }
  63
  64 static struct module *to_bus_provider(struct device *dev)
  65 {
  66         /* pin bus providers while regions are enabled */
  67         if (is_nd_pmem(dev) || is_nd_blk(dev)) {
  68                 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
  69
  70                 return nvdimm_bus->nd_desc->module;
  71         }
  72         return NULL;
  73 }
  74
  75 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
  76 {
  77         nvdimm_bus_lock(&nvdimm_bus->dev);
  78         nvdimm_bus->probe_active++;
  79         nvdimm_bus_unlock(&nvdimm_bus->dev);
  80 }
  81
  82 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
  83 {
  84         nvdimm_bus_lock(&nvdimm_bus->dev);
  85         if (--nvdimm_bus->probe_active == 0)
  86                 wake_up(&nvdimm_bus->probe_wait);
  87         nvdimm_bus_unlock(&nvdimm_bus->dev);
  88 }
  89
  90 static int nvdimm_bus_probe(struct device *dev)
  91 {
  92         struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
  93         struct module *provider = to_bus_provider(dev);
  94         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
  95         int rc;
  96
  97         if (!try_module_get(provider))
  98                 return -ENXIO;
  99
 100         nvdimm_bus_probe_start(nvdimm_bus);
 101         rc = nd_drv->probe(dev);
 102         if (rc == 0)
 103                 nd_region_probe_success(nvdimm_bus, dev);
 104         else
 105                 nd_region_disable(nvdimm_bus, dev);
 106         nvdimm_bus_probe_end(nvdimm_bus);
 107
 108         dev_dbg(&nvdimm_bus->dev, "%s.probe(%s) = %d\n", dev->driver->name,
 109                         dev_name(dev), rc);
 110
 111         if (rc != 0)
 112                 module_put(provider);
 113         return rc;
 114 }
 115
 116 static int nvdimm_bus_remove(struct device *dev)
 117 {
 118         struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
 119         struct module *provider = to_bus_provider(dev);
 120         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 121         int rc = 0;
 122
 123         if (nd_drv->remove)
 124                 rc = nd_drv->remove(dev);
 125         nd_region_disable(nvdimm_bus, dev);
 126
 127         dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
 128                         dev_name(dev), rc);
 129         module_put(provider);
 130         return rc;
 131 }
 132
 133 static void nvdimm_bus_shutdown(struct device *dev)
 134 {
 135         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 136         struct nd_device_driver *nd_drv = NULL;
 137
 138         if (dev->driver)
 139                 nd_drv = to_nd_device_driver(dev->driver);
 140
 141         if (nd_drv && nd_drv->shutdown) {
 142                 nd_drv->shutdown(dev);
 143                 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
 144                                 dev->driver->name, dev_name(dev));
 145         }
 146 }
 147
 148 void nd_device_notify(struct device *dev, enum nvdimm_event event)
 149 {
 150         device_lock(dev);
 151         if (dev->driver) {
 152                 struct nd_device_driver *nd_drv;
 153
 154                 nd_drv = to_nd_device_driver(dev->driver);
 155                 if (nd_drv->notify)
 156                         nd_drv->notify(dev, event);
 157         }
 158         device_unlock(dev);
 159 }
 160 EXPORT_SYMBOL(nd_device_notify);
 161
 162 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
 163 {
 164         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
 165
 166         if (!nvdimm_bus)
 167                 return;
 168
 169         /* caller is responsible for holding a reference on the device */
 170         nd_device_notify(&nd_region->dev, event);
 171 }
 172 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
 173
 174 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
 175                 unsigned int len)
 176 {
 177         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 178         struct nvdimm_bus_descriptor *nd_desc;
 179         struct nd_cmd_clear_error clear_err;
 180         struct nd_cmd_ars_cap ars_cap;
 181         u32 clear_err_unit, mask;
 182         int cmd_rc, rc;
 183
 184         if (!nvdimm_bus)
 185                 return -ENXIO;
 186
 187         nd_desc = nvdimm_bus->nd_desc;
 188         /*
 189          * if ndctl does not exist, it's PMEM_LEGACY and
 190          * we want to just pretend everything is handled.
 191          */
 192         if (!nd_desc->ndctl)
 193                 return len;
 194
 195         memset(&ars_cap, 0, sizeof(ars_cap));
 196         ars_cap.address = phys;
 197         ars_cap.length = len;
 198         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
 199                         sizeof(ars_cap), &cmd_rc);
 200         if (rc < 0)
 201                 return rc;
 202         if (cmd_rc < 0)
 203                 return cmd_rc;
 204         clear_err_unit = ars_cap.clear_err_unit;
 205         if (!clear_err_unit || !is_power_of_2(clear_err_unit))
 206                 return -ENXIO;
 207
 208         mask = clear_err_unit - 1;
 209         if ((phys | len) & mask)
 210                 return -ENXIO;
 211         memset(&clear_err, 0, sizeof(clear_err));
 212         clear_err.address = phys;
 213         clear_err.length = len;
 214         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
 215                         sizeof(clear_err), &cmd_rc);
 216         if (rc < 0)
 217                 return rc;
 218         if (cmd_rc < 0)
 219                 return cmd_rc;
 220
 221         nvdimm_clear_from_poison_list(nvdimm_bus, phys, len);
 222         return clear_err.cleared;
 223 }
 224 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
 225
 226 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
 227
 228 static struct bus_type nvdimm_bus_type = {
 229         .name = "nd",
 230         .uevent = nvdimm_bus_uevent,
 231         .match = nvdimm_bus_match,
 232         .probe = nvdimm_bus_probe,
 233         .remove = nvdimm_bus_remove,
 234         .shutdown = nvdimm_bus_shutdown,
 235 };
 236
 237 static void nvdimm_bus_release(struct device *dev)
 238 {
 239         struct nvdimm_bus *nvdimm_bus;
 240
 241         nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
 242         ida_simple_remove(&nd_ida, nvdimm_bus->id);
 243         kfree(nvdimm_bus);
 244 }
 245
 246 static bool is_nvdimm_bus(struct device *dev)
 247 {
 248         return dev->release == nvdimm_bus_release;
 249 }
 250
 251 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
 252 {
 253         struct device *dev;
 254
 255         for (dev = nd_dev; dev; dev = dev->parent)
 256                 if (is_nvdimm_bus(dev))
 257                         break;
 258         dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
 259         if (dev)
 260                 return to_nvdimm_bus(dev);
 261         return NULL;
 262 }
 263
 264 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
 265 {
 266         struct nvdimm_bus *nvdimm_bus;
 267
 268         nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
 269         WARN_ON(!is_nvdimm_bus(dev));
 270         return nvdimm_bus;
 271 }
 272 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
 273
 274 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
 275                 struct nvdimm_bus_descriptor *nd_desc)
 276 {
 277         struct nvdimm_bus *nvdimm_bus;
 278         int rc;
 279
 280         nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
 281         if (!nvdimm_bus)
 282                 return NULL;
 283         INIT_LIST_HEAD(&nvdimm_bus->list);
 284         INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
 285         INIT_LIST_HEAD(&nvdimm_bus->poison_list);
 286         init_waitqueue_head(&nvdimm_bus->probe_wait);
 287         nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
 288         mutex_init(&nvdimm_bus->reconfig_mutex);
 289         if (nvdimm_bus->id < 0) {
 290                 kfree(nvdimm_bus);
 291                 return NULL;
 292         }
 293         nvdimm_bus->nd_desc = nd_desc;
 294         nvdimm_bus->dev.parent = parent;
 295         nvdimm_bus->dev.release = nvdimm_bus_release;
 296         nvdimm_bus->dev.groups = nd_desc->attr_groups;
 297         nvdimm_bus->dev.bus = &nvdimm_bus_type;
 298         dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
 299         rc = device_register(&nvdimm_bus->dev);
 300         if (rc) {
 301                 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
 302                 goto err;
 303         }
 304
 305         return nvdimm_bus;
 306  err:
 307         put_device(&nvdimm_bus->dev);
 308         return NULL;
 309 }
 310 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
 311
 312 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
 313 {
 314         if (!nvdimm_bus)
 315                 return;
 316         device_unregister(&nvdimm_bus->dev);
 317 }
 318 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
 319
 320 static int child_unregister(struct device *dev, void *data)
 321 {
 322         /*
 323          * the singular ndctl class device per bus needs to be
 324          * "device_destroy"ed, so skip it here
 325          *
 326          * i.e. remove classless children
 327          */
 328         if (dev->class)
 329                 /* pass */;
 330         else
 331                 nd_device_unregister(dev, ND_SYNC);
 332         return 0;
 333 }
 334
 335 static void free_poison_list(struct list_head *poison_list)
 336 {
 337         struct nd_poison *pl, *next;
 338
 339         list_for_each_entry_safe(pl, next, poison_list, list) {
 340                 list_del(&pl->list);
 341                 kfree(pl);
 342         }
 343         list_del_init(poison_list);
 344 }
 345
 346 static int nd_bus_remove(struct device *dev)
 347 {
 348         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 349
 350         mutex_lock(&nvdimm_bus_list_mutex);
 351         list_del_init(&nvdimm_bus->list);
 352         mutex_unlock(&nvdimm_bus_list_mutex);
 353
 354         nd_synchronize();
 355         device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
 356
 357         nvdimm_bus_lock(&nvdimm_bus->dev);
 358         free_poison_list(&nvdimm_bus->poison_list);
 359         nvdimm_bus_unlock(&nvdimm_bus->dev);
 360
 361         nvdimm_bus_destroy_ndctl(nvdimm_bus);
 362
 363         return 0;
 364 }
 365
 366 static int nd_bus_probe(struct device *dev)
 367 {
 368         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 369         int rc;
 370
 371         rc = nvdimm_bus_create_ndctl(nvdimm_bus);
 372         if (rc)
 373                 return rc;
 374
 375         mutex_lock(&nvdimm_bus_list_mutex);
 376         list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
 377         mutex_unlock(&nvdimm_bus_list_mutex);
 378
 379         /* enable bus provider attributes to look up their local context */
 380         dev_set_drvdata(dev, nvdimm_bus->nd_desc);
 381
 382         return 0;
 383 }
 384
 385 static struct nd_device_driver nd_bus_driver = {
 386         .probe = nd_bus_probe,
 387         .remove = nd_bus_remove,
 388         .drv = {
 389                 .name = "nd_bus",
 390                 .suppress_bind_attrs = true,
 391                 .bus = &nvdimm_bus_type,
 392                 .owner = THIS_MODULE,
 393                 .mod_name = KBUILD_MODNAME,
 394         },
 395 };
 396
 397 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
 398 {
 399         struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
 400
 401         if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
 402                 return true;
 403
 404         return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
 405 }
 406
 407 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
 408
 409 void nd_synchronize(void)
 410 {
 411         async_synchronize_full_domain(&nd_async_domain);
 412 }
 413 EXPORT_SYMBOL_GPL(nd_synchronize);
 414
 415 static void nd_async_device_register(void *d, async_cookie_t cookie)
 416 {
 417         struct device *dev = d;
 418
 419         if (device_add(dev) != 0) {
 420                 dev_err(dev, "%s: failed\n", __func__);
 421                 put_device(dev);
 422         }
 423         put_device(dev);
 424 }
 425
 426 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
 427 {
 428         struct device *dev = d;
 429
 430         /* flush bus operations before delete */
 431         nvdimm_bus_lock(dev);
 432         nvdimm_bus_unlock(dev);
 433
 434         device_unregister(dev);
 435         put_device(dev);
 436 }
 437
 438 void __nd_device_register(struct device *dev)
 439 {
 440         if (!dev)
 441                 return;
 442         dev->bus = &nvdimm_bus_type;
 443         get_device(dev);
 444         async_schedule_domain(nd_async_device_register, dev,
 445                         &nd_async_domain);
 446 }
 447
 448 void nd_device_register(struct device *dev)
 449 {
 450         device_initialize(dev);
 451         __nd_device_register(dev);
 452 }
 453 EXPORT_SYMBOL(nd_device_register);
 454
 455 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
 456 {
 457         switch (mode) {
 458         case ND_ASYNC:
 459                 get_device(dev);
 460                 async_schedule_domain(nd_async_device_unregister, dev,
 461                                 &nd_async_domain);
 462                 break;
 463         case ND_SYNC:
 464                 nd_synchronize();
 465                 device_unregister(dev);
 466                 break;
 467         }
 468 }
 469 EXPORT_SYMBOL(nd_device_unregister);
 470
 471 /**
 472  * __nd_driver_register() - register a region or a namespace driver
 473  * @nd_drv: driver to register
 474  * @owner: automatically set by nd_driver_register() macro
 475  * @mod_name: automatically set by nd_driver_register() macro
 476  */
 477 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
 478                 const char *mod_name)
 479 {
 480         struct device_driver *drv = &nd_drv->drv;
 481
 482         if (!nd_drv->type) {
 483                 pr_debug("driver type bitmask not set (%pf)\n",
 484                                 __builtin_return_address(0));
 485                 return -EINVAL;
 486         }
 487
 488         if (!nd_drv->probe) {
 489                 pr_debug("%s ->probe() must be specified\n", mod_name);
 490                 return -EINVAL;
 491         }
 492
 493         drv->bus = &nvdimm_bus_type;
 494         drv->owner = owner;
 495         drv->mod_name = mod_name;
 496
 497         return driver_register(drv);
 498 }
 499 EXPORT_SYMBOL(__nd_driver_register);
 500
 501 int nvdimm_revalidate_disk(struct gendisk *disk)
 502 {
 503         struct device *dev = disk_to_dev(disk)->parent;
 504         struct nd_region *nd_region = to_nd_region(dev->parent);
 505         const char *pol = nd_region->ro ? "only" : "write";
 506
 507         if (nd_region->ro == get_disk_ro(disk))
 508                 return 0;
 509
 510         dev_info(dev, "%s read-%s, marking %s read-%s\n",
 511                         dev_name(&nd_region->dev), pol, disk->disk_name, pol);
 512         set_disk_ro(disk, nd_region->ro);
 513
 514         return 0;
 515
 516 }
 517 EXPORT_SYMBOL(nvdimm_revalidate_disk);
 518
 519 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
 520                 char *buf)
 521 {
 522         return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
 523                         to_nd_device_type(dev));
 524 }
 525 static DEVICE_ATTR_RO(modalias);
 526
 527 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
 528                 char *buf)
 529 {
 530         return sprintf(buf, "%s\n", dev->type->name);
 531 }
 532 static DEVICE_ATTR_RO(devtype);
 533
 534 static struct attribute *nd_device_attributes[] = {
 535         &dev_attr_modalias.attr,
 536         &dev_attr_devtype.attr,
 537         NULL,
 538 };
 539
 540 /**
 541  * nd_device_attribute_group - generic attributes for all devices on an nd bus
 542  */
 543 struct attribute_group nd_device_attribute_group = {
 544         .attrs = nd_device_attributes,
 545 };
 546 EXPORT_SYMBOL_GPL(nd_device_attribute_group);
 547
 548 static ssize_t numa_node_show(struct device *dev,
 549                 struct device_attribute *attr, char *buf)
 550 {
 551         return sprintf(buf, "%d\n", dev_to_node(dev));
 552 }
 553 static DEVICE_ATTR_RO(numa_node);
 554
 555 static struct attribute *nd_numa_attributes[] = {
 556         &dev_attr_numa_node.attr,
 557         NULL,
 558 };
 559
 560 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
 561                 int n)
 562 {
 563         if (!IS_ENABLED(CONFIG_NUMA))
 564                 return 0;
 565
 566         return a->mode;
 567 }
 568
 569 /**
 570  * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
 571  */
 572 struct attribute_group nd_numa_attribute_group = {
 573         .attrs = nd_numa_attributes,
 574         .is_visible = nd_numa_attr_visible,
 575 };
 576 EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
 577
 578 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
 579 {
 580         dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
 581         struct device *dev;
 582
 583         dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
 584                         "ndctl%d", nvdimm_bus->id);
 585
 586         if (IS_ERR(dev))
 587                 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
 588                                 nvdimm_bus->id, PTR_ERR(dev));
 589         return PTR_ERR_OR_ZERO(dev);
 590 }
 591
 592 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
 593 {
 594         device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
 595 }
 596
 597 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
 598         [ND_CMD_IMPLEMENTED] = { },
 599         [ND_CMD_SMART] = {
 600                 .out_num = 2,
 601                 .out_sizes = { 4, 128, },
 602         },
 603         [ND_CMD_SMART_THRESHOLD] = {
 604                 .out_num = 2,
 605                 .out_sizes = { 4, 8, },
 606         },
 607         [ND_CMD_DIMM_FLAGS] = {
 608                 .out_num = 2,
 609                 .out_sizes = { 4, 4 },
 610         },
 611         [ND_CMD_GET_CONFIG_SIZE] = {
 612                 .out_num = 3,
 613                 .out_sizes = { 4, 4, 4, },
 614         },
 615         [ND_CMD_GET_CONFIG_DATA] = {
 616                 .in_num = 2,
 617                 .in_sizes = { 4, 4, },
 618                 .out_num = 2,
 619                 .out_sizes = { 4, UINT_MAX, },
 620         },
 621         [ND_CMD_SET_CONFIG_DATA] = {
 622                 .in_num = 3,
 623                 .in_sizes = { 4, 4, UINT_MAX, },
 624                 .out_num = 1,
 625                 .out_sizes = { 4, },
 626         },
 627         [ND_CMD_VENDOR] = {
 628                 .in_num = 3,
 629                 .in_sizes = { 4, 4, UINT_MAX, },
 630                 .out_num = 3,
 631                 .out_sizes = { 4, 4, UINT_MAX, },
 632         },
 633         [ND_CMD_CALL] = {
 634                 .in_num = 2,
 635                 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
 636                 .out_num = 1,
 637                 .out_sizes = { UINT_MAX, },
 638         },
 639 };
 640
 641 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
 642 {
 643         if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
 644                 return &__nd_cmd_dimm_descs[cmd];
 645         return NULL;
 646 }
 647 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
 648
 649 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
 650         [ND_CMD_IMPLEMENTED] = { },
 651         [ND_CMD_ARS_CAP] = {
 652                 .in_num = 2,
 653                 .in_sizes = { 8, 8, },
 654                 .out_num = 4,
 655                 .out_sizes = { 4, 4, 4, 4, },
 656         },
 657         [ND_CMD_ARS_START] = {
 658                 .in_num = 5,
 659                 .in_sizes = { 8, 8, 2, 1, 5, },
 660                 .out_num = 2,
 661                 .out_sizes = { 4, 4, },
 662         },
 663         [ND_CMD_ARS_STATUS] = {
 664                 .out_num = 3,
 665                 .out_sizes = { 4, 4, UINT_MAX, },
 666         },
 667         [ND_CMD_CLEAR_ERROR] = {
 668                 .in_num = 2,
 669                 .in_sizes = { 8, 8, },
 670                 .out_num = 3,
 671                 .out_sizes = { 4, 4, 8, },
 672         },
 673         [ND_CMD_CALL] = {
 674                 .in_num = 2,
 675                 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
 676                 .out_num = 1,
 677                 .out_sizes = { UINT_MAX, },
 678         },
 679 };
 680
 681 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
 682 {
 683         if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
 684                 return &__nd_cmd_bus_descs[cmd];
 685         return NULL;
 686 }
 687 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
 688
 689 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
 690                 const struct nd_cmd_desc *desc, int idx, void *buf)
 691 {
 692         if (idx >= desc->in_num)
 693                 return UINT_MAX;
 694
 695         if (desc->in_sizes[idx] < UINT_MAX)
 696                 return desc->in_sizes[idx];
 697
 698         if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
 699                 struct nd_cmd_set_config_hdr *hdr = buf;
 700
 701                 return hdr->in_length;
 702         } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
 703                 struct nd_cmd_vendor_hdr *hdr = buf;
 704
 705                 return hdr->in_length;
 706         } else if (cmd == ND_CMD_CALL) {
 707                 struct nd_cmd_pkg *pkg = buf;
 708
 709                 return pkg->nd_size_in;
 710         }
 711
 712         return UINT_MAX;
 713 }
 714 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
 715
 716 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
 717                 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
 718                 const u32 *out_field)
 719 {
 720         if (idx >= desc->out_num)
 721                 return UINT_MAX;
 722
 723         if (desc->out_sizes[idx] < UINT_MAX)
 724                 return desc->out_sizes[idx];
 725
 726         if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
 727                 return in_field[1];
 728         else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
 729                 return out_field[1];
 730         else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2)
 731                 return out_field[1] - 8;
 732         else if (cmd == ND_CMD_CALL) {
 733                 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
 734
 735                 return pkg->nd_size_out;
 736         }
 737
 738
 739         return UINT_MAX;
 740 }
 741 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
 742
 743 void wait_nvdimm_bus_probe_idle(struct device *dev)
 744 {
 745         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 746
 747         do {
 748                 if (nvdimm_bus->probe_active == 0)
 749                         break;
 750                 nvdimm_bus_unlock(&nvdimm_bus->dev);
 751                 wait_event(nvdimm_bus->probe_wait,
 752                                 nvdimm_bus->probe_active == 0);
 753                 nvdimm_bus_lock(&nvdimm_bus->dev);
 754         } while (true);
 755 }
 756
 757 static int pmem_active(struct device *dev, void *data)
 758 {
 759         if (is_nd_pmem(dev) && dev->driver)
 760                 return -EBUSY;
 761         return 0;
 762 }
 763
 764 /* set_config requires an idle interleave set */
 765 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
 766                 struct nvdimm *nvdimm, unsigned int cmd)
 767 {
 768         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 769
 770         /* ask the bus provider if it would like to block this request */
 771         if (nd_desc->clear_to_send) {
 772                 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd);
 773
 774                 if (rc)
 775                         return rc;
 776         }
 777
 778         /* require clear error to go through the pmem driver */
 779         if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
 780                 return device_for_each_child(&nvdimm_bus->dev, NULL,
 781                                 pmem_active);
 782
 783         if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
 784                 return 0;
 785
 786         /* prevent label manipulation while the kernel owns label updates */
 787         wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
 788         if (atomic_read(&nvdimm->busy))
 789                 return -EBUSY;
 790         return 0;
 791 }
 792
 793 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
 794                 int read_only, unsigned int ioctl_cmd, unsigned long arg)
 795 {
 796         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 797         size_t buf_len = 0, in_len = 0, out_len = 0;
 798         static char out_env[ND_CMD_MAX_ENVELOPE];
 799         static char in_env[ND_CMD_MAX_ENVELOPE];
 800         const struct nd_cmd_desc *desc = NULL;
 801         unsigned int cmd = _IOC_NR(ioctl_cmd);
 802         void __user *p = (void __user *) arg;
 803         struct device *dev = &nvdimm_bus->dev;
 804         struct nd_cmd_pkg pkg;
 805         const char *cmd_name, *dimm_name;
 806         unsigned long cmd_mask;
 807         void *buf;
 808         int rc, i;
 809
 810         if (nvdimm) {
 811                 desc = nd_cmd_dimm_desc(cmd);
 812                 cmd_name = nvdimm_cmd_name(cmd);
 813                 cmd_mask = nvdimm->cmd_mask;
 814                 dimm_name = dev_name(&nvdimm->dev);
 815         } else {
 816                 desc = nd_cmd_bus_desc(cmd);
 817                 cmd_name = nvdimm_bus_cmd_name(cmd);
 818                 cmd_mask = nd_desc->cmd_mask;
 819                 dimm_name = "bus";
 820         }
 821
 822         if (cmd == ND_CMD_CALL) {
 823                 if (copy_from_user(&pkg, p, sizeof(pkg)))
 824                         return -EFAULT;
 825         }
 826
 827         if (!desc || (desc->out_num + desc->in_num == 0) ||
 828                         !test_bit(cmd, &cmd_mask))
 829                 return -ENOTTY;
 830
 831         /* fail write commands (when read-only) */
 832         if (read_only)
 833                 switch (cmd) {
 834                 case ND_CMD_VENDOR:
 835                 case ND_CMD_SET_CONFIG_DATA:
 836                 case ND_CMD_ARS_START:
 837                 case ND_CMD_CLEAR_ERROR:
 838                 case ND_CMD_CALL:
 839                         dev_dbg(&nvdimm_bus->dev, "'%s' command while read-only.\n",
 840                                         nvdimm ? nvdimm_cmd_name(cmd)
 841                                         : nvdimm_bus_cmd_name(cmd));
 842                         return -EPERM;
 843                 default:
 844                         break;
 845                 }
 846
 847         /* process an input envelope */
 848         for (i = 0; i < desc->in_num; i++) {
 849                 u32 in_size, copy;
 850
 851                 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
 852                 if (in_size == UINT_MAX) {
 853                         dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
 854                                         __func__, dimm_name, cmd_name, i);
 855                         return -ENXIO;
 856                 }
 857                 if (in_len < sizeof(in_env))
 858                         copy = min_t(u32, sizeof(in_env) - in_len, in_size);
 859                 else
 860                         copy = 0;
 861                 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy))
 862                         return -EFAULT;
 863                 in_len += in_size;
 864         }
 865
 866         if (cmd == ND_CMD_CALL) {
 867                 dev_dbg(dev, "%s:%s, idx: %llu, in: %zu, out: %zu, len %zu\n",
 868                                 __func__, dimm_name, pkg.nd_command,
 869                                 in_len, out_len, buf_len);
 870
 871                 for (i = 0; i < ARRAY_SIZE(pkg.nd_reserved2); i++)
 872                         if (pkg.nd_reserved2[i])
 873                                 return -EINVAL;
 874         }
 875
 876         /* process an output envelope */
 877         for (i = 0; i < desc->out_num; i++) {
 878                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
 879                                 (u32 *) in_env, (u32 *) out_env);
 880                 u32 copy;
 881
 882                 if (out_size == UINT_MAX) {
 883                         dev_dbg(dev, "%s:%s unknown output size cmd: %s field: %d\n",
 884                                         __func__, dimm_name, cmd_name, i);
 885                         return -EFAULT;
 886                 }
 887                 if (out_len < sizeof(out_env))
 888                         copy = min_t(u32, sizeof(out_env) - out_len, out_size);
 889                 else
 890                         copy = 0;
 891                 if (copy && copy_from_user(&out_env[out_len],
 892                                         p + in_len + out_len, copy))
 893                         return -EFAULT;
 894                 out_len += out_size;
 895         }
 896
 897         buf_len = out_len + in_len;
 898         if (buf_len > ND_IOCTL_MAX_BUFLEN) {
 899                 dev_dbg(dev, "%s:%s cmd: %s buf_len: %zu > %d\n", __func__,
 900                                 dimm_name, cmd_name, buf_len,
 901                                 ND_IOCTL_MAX_BUFLEN);
 902                 return -EINVAL;
 903         }
 904
 905         buf = vmalloc(buf_len);
 906         if (!buf)
 907                 return -ENOMEM;
 908
 909         if (copy_from_user(buf, p, buf_len)) {
 910                 rc = -EFAULT;
 911                 goto out;
 912         }
 913
 914         nvdimm_bus_lock(&nvdimm_bus->dev);
 915         rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, cmd);
 916         if (rc)
 917                 goto out_unlock;
 918
 919         rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, NULL);
 920         if (rc < 0)
 921                 goto out_unlock;
 922         if (copy_to_user(p, buf, buf_len))
 923                 rc = -EFAULT;
 924  out_unlock:
 925         nvdimm_bus_unlock(&nvdimm_bus->dev);
 926  out:
 927         vfree(buf);
 928         return rc;
 929 }
 930
 931 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 932 {
 933         long id = (long) file->private_data;
 934         int rc = -ENXIO, ro;
 935         struct nvdimm_bus *nvdimm_bus;
 936
 937         ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
 938         mutex_lock(&nvdimm_bus_list_mutex);
 939         list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
 940                 if (nvdimm_bus->id == id) {
 941                         rc = __nd_ioctl(nvdimm_bus, NULL, ro, cmd, arg);
 942                         break;
 943                 }
 944         }
 945         mutex_unlock(&nvdimm_bus_list_mutex);
 946
 947         return rc;
 948 }
 949
 950 static int match_dimm(struct device *dev, void *data)
 951 {
 952         long id = (long) data;
 953
 954         if (is_nvdimm(dev)) {
 955                 struct nvdimm *nvdimm = to_nvdimm(dev);
 956
 957                 return nvdimm->id == id;
 958         }
 959
 960         return 0;
 961 }
 962
 963 static long nvdimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 964 {
 965         int rc = -ENXIO, ro;
 966         struct nvdimm_bus *nvdimm_bus;
 967
 968         ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
 969         mutex_lock(&nvdimm_bus_list_mutex);
 970         list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
 971                 struct device *dev = device_find_child(&nvdimm_bus->dev,
 972                                 file->private_data, match_dimm);
 973                 struct nvdimm *nvdimm;
 974
 975                 if (!dev)
 976                         continue;
 977
 978                 nvdimm = to_nvdimm(dev);
 979                 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
 980                 put_device(dev);
 981                 break;
 982         }
 983         mutex_unlock(&nvdimm_bus_list_mutex);
 984
 985         return rc;
 986 }
 987
 988 static int nd_open(struct inode *inode, struct file *file)
 989 {
 990         long minor = iminor(inode);
 991
 992         file->private_data = (void *) minor;
 993         return 0;
 994 }
 995
 996 static const struct file_operations nvdimm_bus_fops = {
 997         .owner = THIS_MODULE,
 998         .open = nd_open,
 999         .unlocked_ioctl = nd_ioctl,
1000         .compat_ioctl = nd_ioctl,
1001         .llseek = noop_llseek,
1002 };
1003
1004 static const struct file_operations nvdimm_fops = {
1005         .owner = THIS_MODULE,
1006         .open = nd_open,
1007         .unlocked_ioctl = nvdimm_ioctl,
1008         .compat_ioctl = nvdimm_ioctl,
1009         .llseek = noop_llseek,
1010 };
1011
1012 int __init nvdimm_bus_init(void)
1013 {
1014         int rc;
1015
1016         BUILD_BUG_ON(sizeof(struct nd_smart_payload) != 128);
1017         BUILD_BUG_ON(sizeof(struct nd_smart_threshold_payload) != 8);
1018
1019         rc = bus_register(&nvdimm_bus_type);
1020         if (rc)
1021                 return rc;
1022
1023         rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1024         if (rc < 0)
1025                 goto err_bus_chrdev;
1026         nvdimm_bus_major = rc;
1027
1028         rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1029         if (rc < 0)
1030                 goto err_dimm_chrdev;
1031         nvdimm_major = rc;
1032
1033         nd_class = class_create(THIS_MODULE, "nd");
1034         if (IS_ERR(nd_class)) {
1035                 rc = PTR_ERR(nd_class);
1036                 goto err_class;
1037         }
1038
1039         rc = driver_register(&nd_bus_driver.drv);
1040         if (rc)
1041                 goto err_nd_bus;
1042
1043         return 0;
1044
1045  err_nd_bus:
1046         class_destroy(nd_class);
1047  err_class:
1048         unregister_chrdev(nvdimm_major, "dimmctl");
1049  err_dimm_chrdev:
1050         unregister_chrdev(nvdimm_bus_major, "ndctl");
1051  err_bus_chrdev:
1052         bus_unregister(&nvdimm_bus_type);
1053
1054         return rc;
1055 }
1056
1057 void nvdimm_bus_exit(void)
1058 {
1059         driver_unregister(&nd_bus_driver.drv);
1060         class_destroy(nd_class);
1061         unregister_chrdev(nvdimm_bus_major, "ndctl");
1062         unregister_chrdev(nvdimm_major, "dimmctl");
1063         bus_unregister(&nvdimm_bus_type);
1064         ida_destroy(&nd_ida);
1065 }