drivers/nvdimm/dimm_devs.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/device.h>
  16 #include <linux/ndctl.h>
  17 #include <linux/slab.h>
  18 #include <linux/io.h>
  19 #include <linux/fs.h>
  20 #include <linux/mm.h>
  21 #include "nd-core.h"
  22 #include "label.h"
  23 #include "nd.h"
  24
  25 static DEFINE_IDA(dimm_ida);
  26
  27 /*
  28  * Retrieve bus and dimm handle and return if this bus supports
  29  * get_config_data commands
  30  */
  31 static int __validate_dimm(struct nvdimm_drvdata *ndd)
  32 {
  33         struct nvdimm *nvdimm;
  34
  35         if (!ndd)
  36                 return -EINVAL;
  37
  38         nvdimm = to_nvdimm(ndd->dev);
  39
  40         if (!nvdimm->dsm_mask)
  41                 return -ENXIO;
  42         if (!test_bit(ND_CMD_GET_CONFIG_DATA, nvdimm->dsm_mask))
  43                 return -ENXIO;
  44
  45         return 0;
  46 }
  47
  48 static int validate_dimm(struct nvdimm_drvdata *ndd)
  49 {
  50         int rc = __validate_dimm(ndd);
  51
  52         if (rc && ndd)
  53                 dev_dbg(ndd->dev, "%pf: %s error: %d\n",
  54                                 __builtin_return_address(0), __func__, rc);
  55         return rc;
  56 }
  57
  58 /**
  59  * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
  60  * @nvdimm: dimm to initialize
  61  */
  62 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
  63 {
  64         struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
  65         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
  66         struct nvdimm_bus_descriptor *nd_desc;
  67         int rc = validate_dimm(ndd);
  68
  69         if (rc)
  70                 return rc;
  71
  72         if (cmd->config_size)
  73                 return 0; /* already valid */
  74
  75         memset(cmd, 0, sizeof(*cmd));
  76         nd_desc = nvdimm_bus->nd_desc;
  77         return nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
  78                         ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd));
  79 }
  80
  81 int nvdimm_init_config_data(struct nvdimm_drvdata *ndd)
  82 {
  83         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
  84         struct nd_cmd_get_config_data_hdr *cmd;
  85         struct nvdimm_bus_descriptor *nd_desc;
  86         int rc = validate_dimm(ndd);
  87         u32 max_cmd_size, config_size;
  88         size_t offset;
  89
  90         if (rc)
  91                 return rc;
  92
  93         if (ndd->data)
  94                 return 0;
  95
  96         if (ndd->nsarea.status || ndd->nsarea.max_xfer == 0
  97                         || ndd->nsarea.config_size < ND_LABEL_MIN_SIZE) {
  98                 dev_dbg(ndd->dev, "failed to init config data area: (%d:%d)\n",
  99                                 ndd->nsarea.max_xfer, ndd->nsarea.config_size);
 100                 return -ENXIO;
 101         }
 102
 103         ndd->data = kmalloc(ndd->nsarea.config_size, GFP_KERNEL);
 104         if (!ndd->data)
 105                 ndd->data = vmalloc(ndd->nsarea.config_size);
 106
 107         if (!ndd->data)
 108                 return -ENOMEM;
 109
 110         max_cmd_size = min_t(u32, PAGE_SIZE, ndd->nsarea.max_xfer);
 111         cmd = kzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
 112         if (!cmd)
 113                 return -ENOMEM;
 114
 115         nd_desc = nvdimm_bus->nd_desc;
 116         for (config_size = ndd->nsarea.config_size, offset = 0;
 117                         config_size; config_size -= cmd->in_length,
 118                         offset += cmd->in_length) {
 119                 cmd->in_length = min(config_size, max_cmd_size);
 120                 cmd->in_offset = offset;
 121                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
 122                                 ND_CMD_GET_CONFIG_DATA, cmd,
 123                                 cmd->in_length + sizeof(*cmd));
 124                 if (rc || cmd->status) {
 125                         rc = -ENXIO;
 126                         break;
 127                 }
 128                 memcpy(ndd->data + offset, cmd->out_buf, cmd->in_length);
 129         }
 130         dev_dbg(ndd->dev, "%s: len: %zu rc: %d\n", __func__, offset, rc);
 131         kfree(cmd);
 132
 133         return rc;
 134 }
 135
 136 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
 137                 void *buf, size_t len)
 138 {
 139         int rc = validate_dimm(ndd);
 140         size_t max_cmd_size, buf_offset;
 141         struct nd_cmd_set_config_hdr *cmd;
 142         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
 143         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 144
 145         if (rc)
 146                 return rc;
 147
 148         if (!ndd->data)
 149                 return -ENXIO;
 150
 151         if (offset + len > ndd->nsarea.config_size)
 152                 return -ENXIO;
 153
 154         max_cmd_size = min_t(u32, PAGE_SIZE, len);
 155         max_cmd_size = min_t(u32, max_cmd_size, ndd->nsarea.max_xfer);
 156         cmd = kzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
 157         if (!cmd)
 158                 return -ENOMEM;
 159
 160         for (buf_offset = 0; len; len -= cmd->in_length,
 161                         buf_offset += cmd->in_length) {
 162                 size_t cmd_size;
 163                 u32 *status;
 164
 165                 cmd->in_offset = offset + buf_offset;
 166                 cmd->in_length = min(max_cmd_size, len);
 167                 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
 168
 169                 /* status is output in the last 4-bytes of the command buffer */
 170                 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
 171                 status = ((void *) cmd) + cmd_size - sizeof(u32);
 172
 173                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
 174                                 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size);
 175                 if (rc || *status) {
 176                         rc = rc ? rc : -ENXIO;
 177                         break;
 178                 }
 179         }
 180         kfree(cmd);
 181
 182         return rc;
 183 }
 184
 185 static void nvdimm_release(struct device *dev)
 186 {
 187         struct nvdimm *nvdimm = to_nvdimm(dev);
 188
 189         ida_simple_remove(&dimm_ida, nvdimm->id);
 190         kfree(nvdimm);
 191 }
 192
 193 static struct device_type nvdimm_device_type = {
 194         .name = "nvdimm",
 195         .release = nvdimm_release,
 196 };
 197
 198 bool is_nvdimm(struct device *dev)
 199 {
 200         return dev->type == &nvdimm_device_type;
 201 }
 202
 203 struct nvdimm *to_nvdimm(struct device *dev)
 204 {
 205         struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
 206
 207         WARN_ON(!is_nvdimm(dev));
 208         return nvdimm;
 209 }
 210 EXPORT_SYMBOL_GPL(to_nvdimm);
 211
 212 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
 213 {
 214         struct nd_region *nd_region = &ndbr->nd_region;
 215         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 216
 217         return nd_mapping->nvdimm;
 218 }
 219 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
 220
 221 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
 222 {
 223         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 224
 225         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
 226
 227         return dev_get_drvdata(&nvdimm->dev);
 228 }
 229 EXPORT_SYMBOL(to_ndd);
 230
 231 void nvdimm_drvdata_release(struct kref *kref)
 232 {
 233         struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
 234         struct device *dev = ndd->dev;
 235         struct resource *res, *_r;
 236
 237         dev_dbg(dev, "%s\n", __func__);
 238
 239         nvdimm_bus_lock(dev);
 240         for_each_dpa_resource_safe(ndd, res, _r)
 241                 nvdimm_free_dpa(ndd, res);
 242         nvdimm_bus_unlock(dev);
 243
 244         if (ndd->data && is_vmalloc_addr(ndd->data))
 245                 vfree(ndd->data);
 246         else
 247                 kfree(ndd->data);
 248         kfree(ndd);
 249         put_device(dev);
 250 }
 251
 252 void get_ndd(struct nvdimm_drvdata *ndd)
 253 {
 254         kref_get(&ndd->kref);
 255 }
 256
 257 void put_ndd(struct nvdimm_drvdata *ndd)
 258 {
 259         if (ndd)
 260                 kref_put(&ndd->kref, nvdimm_drvdata_release);
 261 }
 262
 263 const char *nvdimm_name(struct nvdimm *nvdimm)
 264 {
 265         return dev_name(&nvdimm->dev);
 266 }
 267 EXPORT_SYMBOL_GPL(nvdimm_name);
 268
 269 void *nvdimm_provider_data(struct nvdimm *nvdimm)
 270 {
 271         if (nvdimm)
 272                 return nvdimm->provider_data;
 273         return NULL;
 274 }
 275 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
 276
 277 static ssize_t commands_show(struct device *dev,
 278                 struct device_attribute *attr, char *buf)
 279 {
 280         struct nvdimm *nvdimm = to_nvdimm(dev);
 281         int cmd, len = 0;
 282
 283         if (!nvdimm->dsm_mask)
 284                 return sprintf(buf, "\n");
 285
 286         for_each_set_bit(cmd, nvdimm->dsm_mask, BITS_PER_LONG)
 287                 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
 288         len += sprintf(buf + len, "\n");
 289         return len;
 290 }
 291 static DEVICE_ATTR_RO(commands);
 292
 293 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 294                 char *buf)
 295 {
 296         struct nvdimm *nvdimm = to_nvdimm(dev);
 297
 298         /*
 299          * The state may be in the process of changing, userspace should
 300          * quiesce probing if it wants a static answer
 301          */
 302         nvdimm_bus_lock(dev);
 303         nvdimm_bus_unlock(dev);
 304         return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
 305                         ? "active" : "idle");
 306 }
 307 static DEVICE_ATTR_RO(state);
 308
 309 static ssize_t available_slots_show(struct device *dev,
 310                 struct device_attribute *attr, char *buf)
 311 {
 312         struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
 313         ssize_t rc;
 314         u32 nfree;
 315
 316         if (!ndd)
 317                 return -ENXIO;
 318
 319         nvdimm_bus_lock(dev);
 320         nfree = nd_label_nfree(ndd);
 321         if (nfree - 1 > nfree) {
 322                 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
 323                 nfree = 0;
 324         } else
 325                 nfree--;
 326         rc = sprintf(buf, "%d\n", nfree);
 327         nvdimm_bus_unlock(dev);
 328         return rc;
 329 }
 330 static DEVICE_ATTR_RO(available_slots);
 331
 332 static struct attribute *nvdimm_attributes[] = {
 333         &dev_attr_state.attr,
 334         &dev_attr_commands.attr,
 335         &dev_attr_available_slots.attr,
 336         NULL,
 337 };
 338
 339 struct attribute_group nvdimm_attribute_group = {
 340         .attrs = nvdimm_attributes,
 341 };
 342 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
 343
 344 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
 345                 const struct attribute_group **groups, unsigned long flags,
 346                 unsigned long *dsm_mask)
 347 {
 348         struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
 349         struct device *dev;
 350
 351         if (!nvdimm)
 352                 return NULL;
 353
 354         nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
 355         if (nvdimm->id < 0) {
 356                 kfree(nvdimm);
 357                 return NULL;
 358         }
 359         nvdimm->provider_data = provider_data;
 360         nvdimm->flags = flags;
 361         nvdimm->dsm_mask = dsm_mask;
 362         atomic_set(&nvdimm->busy, 0);
 363         dev = &nvdimm->dev;
 364         dev_set_name(dev, "nmem%d", nvdimm->id);
 365         dev->parent = &nvdimm_bus->dev;
 366         dev->type = &nvdimm_device_type;
 367         dev->devt = MKDEV(nvdimm_major, nvdimm->id);
 368         dev->groups = groups;
 369         nd_device_register(dev);
 370
 371         return nvdimm;
 372 }
 373 EXPORT_SYMBOL_GPL(nvdimm_create);
 374
 375 /**
 376  * nd_blk_available_dpa - account the unused dpa of BLK region
 377  * @nd_mapping: container of dpa-resource-root + labels
 378  *
 379  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges.
 380  */
 381 resource_size_t nd_blk_available_dpa(struct nd_mapping *nd_mapping)
 382 {
 383         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 384         resource_size_t map_end, busy = 0, available;
 385         struct resource *res;
 386
 387         if (!ndd)
 388                 return 0;
 389
 390         map_end = nd_mapping->start + nd_mapping->size - 1;
 391         for_each_dpa_resource(ndd, res)
 392                 if (res->start >= nd_mapping->start && res->start < map_end) {
 393                         resource_size_t end = min(map_end, res->end);
 394
 395                         busy += end - res->start + 1;
 396                 } else if (res->end >= nd_mapping->start
 397                                 && res->end <= map_end) {
 398                         busy += res->end - nd_mapping->start;
 399                 } else if (nd_mapping->start > res->start
 400                                 && nd_mapping->start < res->end) {
 401                         /* total eclipse of the BLK region mapping */
 402                         busy += nd_mapping->size;
 403                 }
 404
 405         available = map_end - nd_mapping->start + 1;
 406         if (busy < available)
 407                 return available - busy;
 408         return 0;
 409 }
 410
 411 /**
 412  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
 413  * @nd_mapping: container of dpa-resource-root + labels
 414  * @nd_region: constrain available space check to this reference region
 415  * @overlap: calculate available space assuming this level of overlap
 416  *
 417  * Validate that a PMEM label, if present, aligns with the start of an
 418  * interleave set and truncate the available size at the lowest BLK
 419  * overlap point.
 420  *
 421  * The expectation is that this routine is called multiple times as it
 422  * probes for the largest BLK encroachment for any single member DIMM of
 423  * the interleave set.  Once that value is determined the PMEM-limit for
 424  * the set can be established.
 425  */
 426 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
 427                 struct nd_mapping *nd_mapping, resource_size_t *overlap)
 428 {
 429         resource_size_t map_start, map_end, busy = 0, available, blk_start;
 430         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 431         struct resource *res;
 432         const char *reason;
 433
 434         if (!ndd)
 435                 return 0;
 436
 437         map_start = nd_mapping->start;
 438         map_end = map_start + nd_mapping->size - 1;
 439         blk_start = max(map_start, map_end + 1 - *overlap);
 440         for_each_dpa_resource(ndd, res)
 441                 if (res->start >= map_start && res->start < map_end) {
 442                         if (strncmp(res->name, "blk", 3) == 0)
 443                                 blk_start = min(blk_start, res->start);
 444                         else if (res->start != map_start) {
 445                                 reason = "misaligned to iset";
 446                                 goto err;
 447                         } else {
 448                                 if (busy) {
 449                                         reason = "duplicate overlapping PMEM reservations?";
 450                                         goto err;
 451                                 }
 452                                 busy += resource_size(res);
 453                                 continue;
 454                         }
 455                 } else if (res->end >= map_start && res->end <= map_end) {
 456                         if (strncmp(res->name, "blk", 3) == 0) {
 457                                 /*
 458                                  * If a BLK allocation overlaps the start of
 459                                  * PMEM the entire interleave set may now only
 460                                  * be used for BLK.
 461                                  */
 462                                 blk_start = map_start;
 463                         } else {
 464                                 reason = "misaligned to iset";
 465                                 goto err;
 466                         }
 467                 } else if (map_start > res->start && map_start < res->end) {
 468                         /* total eclipse of the mapping */
 469                         busy += nd_mapping->size;
 470                         blk_start = map_start;
 471                 }
 472
 473         *overlap = map_end + 1 - blk_start;
 474         available = blk_start - map_start;
 475         if (busy < available)
 476                 return available - busy;
 477         return 0;
 478
 479  err:
 480         /*
 481          * Something is wrong, PMEM must align with the start of the
 482          * interleave set, and there can only be one allocation per set.
 483          */
 484         nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
 485         return 0;
 486 }
 487
 488 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
 489 {
 490         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
 491         kfree(res->name);
 492         __release_region(&ndd->dpa, res->start, resource_size(res));
 493 }
 494
 495 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
 496                 struct nd_label_id *label_id, resource_size_t start,
 497                 resource_size_t n)
 498 {
 499         char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
 500         struct resource *res;
 501
 502         if (!name)
 503                 return NULL;
 504
 505         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
 506         res = __request_region(&ndd->dpa, start, n, name, 0);
 507         if (!res)
 508                 kfree(name);
 509         return res;
 510 }
 511
 512 /**
 513  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
 514  * @nvdimm: container of dpa-resource-root + labels
 515  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
 516  */
 517 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
 518                 struct nd_label_id *label_id)
 519 {
 520         resource_size_t allocated = 0;
 521         struct resource *res;
 522
 523         for_each_dpa_resource(ndd, res)
 524                 if (strcmp(res->name, label_id->id) == 0)
 525                         allocated += resource_size(res);
 526
 527         return allocated;
 528 }
 529
 530 static int count_dimms(struct device *dev, void *c)
 531 {
 532         int *count = c;
 533
 534         if (is_nvdimm(dev))
 535                 (*count)++;
 536         return 0;
 537 }
 538
 539 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
 540 {
 541         int count = 0;
 542         /* Flush any possible dimm registration failures */
 543         nd_synchronize();
 544
 545         device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
 546         dev_dbg(&nvdimm_bus->dev, "%s: count: %d\n", __func__, count);
 547         if (count != dimm_count)
 548                 return -ENXIO;
 549         return 0;
 550 }
 551 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);