| 1 | |
| 2 | |
| 3 | * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
|
| 4 |
|
| 5 | #include <linux/scatterlist.h>
|
| 6 | #include <linux/memregion.h>
|
| 7 | #include <linux/highmem.h>
|
| 8 | #include <linux/kstrtox.h>
|
| 9 | #include <linux/sched.h>
|
| 10 | #include <linux/slab.h>
|
| 11 | #include <linux/hash.h>
|
| 12 | #include <linux/sort.h>
|
| 13 | #include <linux/io.h>
|
| 14 | #include <linux/nd.h>
|
| 15 | #include "nd-core.h"
|
| 16 | #include "nd.h"
|
| 17 |
|
| 18 | |
| 19 | * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
|
| 20 | * irrelevant.
|
| 21 |
|
| 22 | #include <linux/io-64-nonatomic-hi-lo.h>
|
| 23 |
|
| 24 | static DEFINE_PER_CPU(int, flush_idx);
|
| 25 |
|
| 26 | static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
|
| 27 | struct nd_region_data *ndrd)
|
| 28 | {
|
| 29 | int i, j;
|
| 30 |
|
| 31 | dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
|
| 32 | nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
|
| 33 | for (i = 0; i < (1 << ndrd->hints_shift); i++) {
|
| 34 | struct resource *res = &nvdimm->flush_wpq[i];
|
| 35 | unsigned long pfn = PHYS_PFN(res->start);
|
| 36 | void __iomem *flush_page;
|
| 37 |
|
| 38 |
|
| 39 | for (j = 0; j < i; j++) {
|
| 40 | struct resource *res_j = &nvdimm->flush_wpq[j];
|
| 41 | unsigned long pfn_j = PHYS_PFN(res_j->start);
|
| 42 |
|
| 43 | if (pfn == pfn_j)
|
| 44 | break;
|
| 45 | }
|
| 46 |
|
| 47 | if (j < i)
|
| 48 | flush_page = (void __iomem *) ((unsigned long)
|
| 49 | ndrd_get_flush_wpq(ndrd, dimm, j)
|
| 50 | & PAGE_MASK);
|
| 51 | else
|
| 52 | flush_page = devm_nvdimm_ioremap(dev,
|
| 53 | PFN_PHYS(pfn), PAGE_SIZE);
|
| 54 | if (!flush_page)
|
| 55 | return -ENXIO;
|
| 56 | ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
|
| 57 | + (res->start & ~PAGE_MASK));
|
| 58 | }
|
| 59 |
|
| 60 | return 0;
|
| 61 | }
|
| 62 |
|
| 63 | static int nd_region_invalidate_memregion(struct nd_region *nd_region)
|
| 64 | {
|
| 65 | int i, incoherent = 0;
|
| 66 |
|
| 67 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 68 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 69 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 70 |
|
| 71 | if (test_bit(NDD_INCOHERENT, &nvdimm->flags)) {
|
| 72 | incoherent++;
|
| 73 | break;
|
| 74 | }
|
| 75 | }
|
| 76 |
|
| 77 | if (!incoherent)
|
| 78 | return 0;
|
| 79 |
|
| 80 | if (!cpu_cache_has_invalidate_memregion()) {
|
| 81 | if (IS_ENABLED(CONFIG_NVDIMM_SECURITY_TEST)) {
|
| 82 | dev_warn(
|
| 83 | &nd_region->dev,
|
| 84 | "Bypassing cpu_cache_invalidate_memergion() for testing!\n");
|
| 85 | goto out;
|
| 86 | } else {
|
| 87 | dev_err(&nd_region->dev,
|
| 88 | "Failed to synchronize CPU cache state\n");
|
| 89 | return -ENXIO;
|
| 90 | }
|
| 91 | }
|
| 92 |
|
| 93 | cpu_cache_invalidate_all();
|
| 94 | out:
|
| 95 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 96 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 97 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 98 |
|
| 99 | clear_bit(NDD_INCOHERENT, &nvdimm->flags);
|
| 100 | }
|
| 101 |
|
| 102 | return 0;
|
| 103 | }
|
| 104 |
|
| 105 | static int get_flush_data(struct nd_region *nd_region, size_t *size, int *num_flush)
|
| 106 | {
|
| 107 | size_t flush_data_size = sizeof(void *);
|
| 108 | int _num_flush = 0;
|
| 109 | int i;
|
| 110 |
|
| 111 | guard(nvdimm_bus)(&nd_region->dev);
|
| 112 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 113 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 114 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 115 |
|
| 116 | if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags))
|
| 117 | return -EBUSY;
|
| 118 |
|
| 119 |
|
| 120 | flush_data_size += sizeof(void *);
|
| 121 | _num_flush = min_not_zero(_num_flush, nvdimm->num_flush);
|
| 122 | if (!nvdimm->num_flush)
|
| 123 | continue;
|
| 124 | flush_data_size += nvdimm->num_flush * sizeof(void *);
|
| 125 | }
|
| 126 |
|
| 127 | *size = flush_data_size;
|
| 128 | *num_flush = _num_flush;
|
| 129 |
|
| 130 | return 0;
|
| 131 | }
|
| 132 |
|
| 133 | int nd_region_activate(struct nd_region *nd_region)
|
| 134 | {
|
| 135 | int i, j, rc, num_flush;
|
| 136 | struct nd_region_data *ndrd;
|
| 137 | struct device *dev = &nd_region->dev;
|
| 138 | size_t flush_data_size;
|
| 139 |
|
| 140 | rc = get_flush_data(nd_region, &flush_data_size, &num_flush);
|
| 141 | if (rc)
|
| 142 | return rc;
|
| 143 |
|
| 144 | rc = nd_region_invalidate_memregion(nd_region);
|
| 145 | if (rc)
|
| 146 | return rc;
|
| 147 |
|
| 148 | ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
|
| 149 | if (!ndrd)
|
| 150 | return -ENOMEM;
|
| 151 | dev_set_drvdata(dev, ndrd);
|
| 152 |
|
| 153 | if (!num_flush)
|
| 154 | return 0;
|
| 155 |
|
| 156 | ndrd->hints_shift = ilog2(num_flush);
|
| 157 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 158 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 159 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 160 | int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
|
| 161 |
|
| 162 | if (rc)
|
| 163 | return rc;
|
| 164 | }
|
| 165 |
|
| 166 | |
| 167 | * Clear out entries that are duplicates. This should prevent the
|
| 168 | * extra flushings.
|
| 169 |
|
| 170 | for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
|
| 171 |
|
| 172 | if (!ndrd_get_flush_wpq(ndrd, i, 0))
|
| 173 | continue;
|
| 174 |
|
| 175 | for (j = i + 1; j < nd_region->ndr_mappings; j++)
|
| 176 | if (ndrd_get_flush_wpq(ndrd, i, 0) ==
|
| 177 | ndrd_get_flush_wpq(ndrd, j, 0))
|
| 178 | ndrd_set_flush_wpq(ndrd, j, 0, NULL);
|
| 179 | }
|
| 180 |
|
| 181 | return 0;
|
| 182 | }
|
| 183 |
|
| 184 | static void nd_region_release(struct device *dev)
|
| 185 | {
|
| 186 | struct nd_region *nd_region = to_nd_region(dev);
|
| 187 | u16 i;
|
| 188 |
|
| 189 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 190 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 191 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 192 |
|
| 193 | put_device(&nvdimm->dev);
|
| 194 | }
|
| 195 | for (i = 0; i < nd_region->num_lanes; i++)
|
| 196 | mutex_destroy(&nd_region->lane[i].lock);
|
| 197 | kfree(nd_region->lane);
|
| 198 | if (!test_bit(ND_REGION_CXL, &nd_region->flags))
|
| 199 | memregion_free(nd_region->id);
|
| 200 | kfree(nd_region);
|
| 201 | }
|
| 202 |
|
| 203 | struct nd_region *to_nd_region(struct device *dev)
|
| 204 | {
|
| 205 | struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
|
| 206 |
|
| 207 | WARN_ON(dev->type->release != nd_region_release);
|
| 208 | return nd_region;
|
| 209 | }
|
| 210 | EXPORT_SYMBOL_GPL(to_nd_region);
|
| 211 |
|
| 212 | struct device *nd_region_dev(struct nd_region *nd_region)
|
| 213 | {
|
| 214 | if (!nd_region)
|
| 215 | return NULL;
|
| 216 | return &nd_region->dev;
|
| 217 | }
|
| 218 | EXPORT_SYMBOL_GPL(nd_region_dev);
|
| 219 |
|
| 220 | void *nd_region_provider_data(struct nd_region *nd_region)
|
| 221 | {
|
| 222 | return nd_region->provider_data;
|
| 223 | }
|
| 224 | EXPORT_SYMBOL_GPL(nd_region_provider_data);
|
| 225 |
|
| 226 | |
| 227 | * nd_region_to_nstype() - region to an integer namespace type
|
| 228 | * @nd_region: region-device to interrogate
|
| 229 | *
|
| 230 | * This is the 'nstype' attribute of a region as well, an input to the
|
| 231 | * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
|
| 232 | * namespace devices with namespace drivers.
|
| 233 |
|
| 234 | int nd_region_to_nstype(struct nd_region *nd_region)
|
| 235 | {
|
| 236 | if (is_memory(&nd_region->dev)) {
|
| 237 | u16 i, label;
|
| 238 |
|
| 239 | for (i = 0, label = 0; i < nd_region->ndr_mappings; i++) {
|
| 240 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 241 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 242 |
|
| 243 | if (test_bit(NDD_LABELING, &nvdimm->flags))
|
| 244 | label++;
|
| 245 | }
|
| 246 | if (label)
|
| 247 | return ND_DEVICE_NAMESPACE_PMEM;
|
| 248 | else
|
| 249 | return ND_DEVICE_NAMESPACE_IO;
|
| 250 | }
|
| 251 |
|
| 252 | return 0;
|
| 253 | }
|
| 254 | EXPORT_SYMBOL(nd_region_to_nstype);
|
| 255 |
|
| 256 | static unsigned long long region_size(struct nd_region *nd_region)
|
| 257 | {
|
| 258 | if (is_memory(&nd_region->dev)) {
|
| 259 | return nd_region->ndr_size;
|
| 260 | } else if (nd_region->ndr_mappings == 1) {
|
| 261 | struct nd_mapping *nd_mapping = &nd_region->mapping[0];
|
| 262 |
|
| 263 | return nd_mapping->size;
|
| 264 | }
|
| 265 |
|
| 266 | return 0;
|
| 267 | }
|
| 268 |
|
| 269 | static ssize_t size_show(struct device *dev,
|
| 270 | struct device_attribute *attr, char *buf)
|
| 271 | {
|
| 272 | struct nd_region *nd_region = to_nd_region(dev);
|
| 273 |
|
| 274 | return sprintf(buf, "%llu\n", region_size(nd_region));
|
| 275 | }
|
| 276 | static DEVICE_ATTR_RO(size);
|
| 277 |
|
| 278 | static ssize_t deep_flush_show(struct device *dev,
|
| 279 | struct device_attribute *attr, char *buf)
|
| 280 | {
|
| 281 | struct nd_region *nd_region = to_nd_region(dev);
|
| 282 |
|
| 283 | |
| 284 | * NOTE: in the nvdimm_has_flush() error case this attribute is
|
| 285 | * not visible.
|
| 286 |
|
| 287 | return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
|
| 288 | }
|
| 289 |
|
| 290 | static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
|
| 291 | const char *buf, size_t len)
|
| 292 | {
|
| 293 | bool flush;
|
| 294 | int rc = kstrtobool(buf, &flush);
|
| 295 | struct nd_region *nd_region = to_nd_region(dev);
|
| 296 |
|
| 297 | if (rc)
|
| 298 | return rc;
|
| 299 | if (!flush)
|
| 300 | return -EINVAL;
|
| 301 | rc = nvdimm_flush(nd_region, NULL);
|
| 302 | if (rc)
|
| 303 | return rc;
|
| 304 |
|
| 305 | return len;
|
| 306 | }
|
| 307 | static DEVICE_ATTR_RW(deep_flush);
|
| 308 |
|
| 309 | static ssize_t mappings_show(struct device *dev,
|
| 310 | struct device_attribute *attr, char *buf)
|
| 311 | {
|
| 312 | struct nd_region *nd_region = to_nd_region(dev);
|
| 313 |
|
| 314 | return sprintf(buf, "%d\n", nd_region->ndr_mappings);
|
| 315 | }
|
| 316 | static DEVICE_ATTR_RO(mappings);
|
| 317 |
|
| 318 | static ssize_t nstype_show(struct device *dev,
|
| 319 | struct device_attribute *attr, char *buf)
|
| 320 | {
|
| 321 | struct nd_region *nd_region = to_nd_region(dev);
|
| 322 |
|
| 323 | return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
|
| 324 | }
|
| 325 | static DEVICE_ATTR_RO(nstype);
|
| 326 |
|
| 327 | static ssize_t set_cookie_show(struct device *dev,
|
| 328 | struct device_attribute *attr, char *buf)
|
| 329 | {
|
| 330 | struct nd_region *nd_region = to_nd_region(dev);
|
| 331 | struct nd_interleave_set *nd_set = nd_region->nd_set;
|
| 332 | ssize_t rc = 0;
|
| 333 |
|
| 334 | if (is_memory(dev) && nd_set)
|
| 335 | ;
|
| 336 | else
|
| 337 | return -ENXIO;
|
| 338 |
|
| 339 | |
| 340 | * The cookie to show depends on which specification of the
|
| 341 | * labels we are using. If there are not labels then default to
|
| 342 | * the v1.1 namespace label cookie definition. To read all this
|
| 343 | * data we need to wait for probing to settle.
|
| 344 |
|
| 345 | guard(device)(dev);
|
| 346 | guard(nvdimm_bus)(dev);
|
| 347 | wait_nvdimm_bus_probe_idle(dev);
|
| 348 | if (nd_region->ndr_mappings) {
|
| 349 | struct nd_mapping *nd_mapping = &nd_region->mapping[0];
|
| 350 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
|
| 351 |
|
| 352 | if (ndd) {
|
| 353 | struct nd_namespace_index *nsindex;
|
| 354 |
|
| 355 | nsindex = to_namespace_index(ndd, ndd->ns_current);
|
| 356 | rc = sprintf(buf, "%#llx\n",
|
| 357 | nd_region_interleave_set_cookie(nd_region,
|
| 358 | nsindex));
|
| 359 | }
|
| 360 | }
|
| 361 |
|
| 362 | if (rc)
|
| 363 | return rc;
|
| 364 | return sprintf(buf, "%#llx\n", nd_set->cookie1);
|
| 365 | }
|
| 366 | static DEVICE_ATTR_RO(set_cookie);
|
| 367 |
|
| 368 | resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
|
| 369 | {
|
| 370 | resource_size_t available;
|
| 371 | int i;
|
| 372 |
|
| 373 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
|
| 374 |
|
| 375 | available = 0;
|
| 376 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 377 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 378 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
|
| 379 |
|
| 380 |
|
| 381 | if (!ndd)
|
| 382 | return 0;
|
| 383 |
|
| 384 | available += nd_pmem_available_dpa(nd_region, nd_mapping);
|
| 385 | }
|
| 386 |
|
| 387 | return available;
|
| 388 | }
|
| 389 |
|
| 390 | resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
|
| 391 | {
|
| 392 | resource_size_t avail = 0;
|
| 393 | int i;
|
| 394 |
|
| 395 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
|
| 396 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 397 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 398 |
|
| 399 | avail = min_not_zero(avail, nd_pmem_max_contiguous_dpa(
|
| 400 | nd_region, nd_mapping));
|
| 401 | }
|
| 402 | return avail * nd_region->ndr_mappings;
|
| 403 | }
|
| 404 |
|
| 405 | static ssize_t available_size_show(struct device *dev,
|
| 406 | struct device_attribute *attr, char *buf)
|
| 407 | {
|
| 408 | struct nd_region *nd_region = to_nd_region(dev);
|
| 409 |
|
| 410 | |
| 411 | * Flush in-flight updates and grab a snapshot of the available
|
| 412 | * size. Of course, this value is potentially invalidated the
|
| 413 | * memory nvdimm_bus_lock() is dropped, but that's userspace's
|
| 414 | * problem to not race itself.
|
| 415 |
|
| 416 | guard(device)(dev);
|
| 417 | guard(nvdimm_bus)(dev);
|
| 418 | wait_nvdimm_bus_probe_idle(dev);
|
| 419 |
|
| 420 | return sprintf(buf, "%llu\n", nd_region_available_dpa(nd_region));
|
| 421 | }
|
| 422 | static DEVICE_ATTR_RO(available_size);
|
| 423 |
|
| 424 | static ssize_t max_available_extent_show(struct device *dev,
|
| 425 | struct device_attribute *attr, char *buf)
|
| 426 | {
|
| 427 | struct nd_region *nd_region = to_nd_region(dev);
|
| 428 |
|
| 429 | guard(device)(dev);
|
| 430 | guard(nvdimm_bus)(dev);
|
| 431 | wait_nvdimm_bus_probe_idle(dev);
|
| 432 |
|
| 433 | return sprintf(buf, "%llu\n", nd_region_allocatable_dpa(nd_region));
|
| 434 | }
|
| 435 | static DEVICE_ATTR_RO(max_available_extent);
|
| 436 |
|
| 437 | static ssize_t init_namespaces_show(struct device *dev,
|
| 438 | struct device_attribute *attr, char *buf)
|
| 439 | {
|
| 440 | struct nd_region_data *ndrd = dev_get_drvdata(dev);
|
| 441 |
|
| 442 | guard(nvdimm_bus)(dev);
|
| 443 | if (!ndrd)
|
| 444 | return -ENXIO;
|
| 445 |
|
| 446 | return sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
|
| 447 | }
|
| 448 | static DEVICE_ATTR_RO(init_namespaces);
|
| 449 |
|
| 450 | static ssize_t namespace_seed_show(struct device *dev,
|
| 451 | struct device_attribute *attr, char *buf)
|
| 452 | {
|
| 453 | struct nd_region *nd_region = to_nd_region(dev);
|
| 454 |
|
| 455 | guard(nvdimm_bus)(dev);
|
| 456 | if (nd_region->ns_seed)
|
| 457 | return sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
|
| 458 |
|
| 459 | return sprintf(buf, "\n");
|
| 460 | }
|
| 461 | static DEVICE_ATTR_RO(namespace_seed);
|
| 462 |
|
| 463 | static ssize_t btt_seed_show(struct device *dev,
|
| 464 | struct device_attribute *attr, char *buf)
|
| 465 | {
|
| 466 | struct nd_region *nd_region = to_nd_region(dev);
|
| 467 |
|
| 468 | guard(nvdimm_bus)(dev);
|
| 469 | if (nd_region->btt_seed)
|
| 470 | return sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
|
| 471 |
|
| 472 | return sprintf(buf, "\n");
|
| 473 | }
|
| 474 | static DEVICE_ATTR_RO(btt_seed);
|
| 475 |
|
| 476 | static ssize_t pfn_seed_show(struct device *dev,
|
| 477 | struct device_attribute *attr, char *buf)
|
| 478 | {
|
| 479 | struct nd_region *nd_region = to_nd_region(dev);
|
| 480 |
|
| 481 | guard(nvdimm_bus)(dev);
|
| 482 | if (nd_region->pfn_seed)
|
| 483 | return sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
|
| 484 |
|
| 485 | return sprintf(buf, "\n");
|
| 486 | }
|
| 487 | static DEVICE_ATTR_RO(pfn_seed);
|
| 488 |
|
| 489 | static ssize_t dax_seed_show(struct device *dev,
|
| 490 | struct device_attribute *attr, char *buf)
|
| 491 | {
|
| 492 | struct nd_region *nd_region = to_nd_region(dev);
|
| 493 |
|
| 494 | guard(nvdimm_bus)(dev);
|
| 495 | if (nd_region->dax_seed)
|
| 496 | return sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
|
| 497 |
|
| 498 | return sprintf(buf, "\n");
|
| 499 | }
|
| 500 | static DEVICE_ATTR_RO(dax_seed);
|
| 501 |
|
| 502 | static ssize_t read_only_show(struct device *dev,
|
| 503 | struct device_attribute *attr, char *buf)
|
| 504 | {
|
| 505 | struct nd_region *nd_region = to_nd_region(dev);
|
| 506 |
|
| 507 | return sprintf(buf, "%d\n", nd_region->ro);
|
| 508 | }
|
| 509 |
|
| 510 | static int revalidate_read_only(struct device *dev, void *data)
|
| 511 | {
|
| 512 | nd_device_notify(dev, NVDIMM_REVALIDATE_REGION);
|
| 513 | return 0;
|
| 514 | }
|
| 515 |
|
| 516 | static ssize_t read_only_store(struct device *dev,
|
| 517 | struct device_attribute *attr, const char *buf, size_t len)
|
| 518 | {
|
| 519 | bool ro;
|
| 520 | int rc = kstrtobool(buf, &ro);
|
| 521 | struct nd_region *nd_region = to_nd_region(dev);
|
| 522 |
|
| 523 | if (rc)
|
| 524 | return rc;
|
| 525 |
|
| 526 | nd_region->ro = ro;
|
| 527 | device_for_each_child(dev, NULL, revalidate_read_only);
|
| 528 | return len;
|
| 529 | }
|
| 530 | static DEVICE_ATTR_RW(read_only);
|
| 531 |
|
| 532 | static ssize_t align_show(struct device *dev,
|
| 533 | struct device_attribute *attr, char *buf)
|
| 534 | {
|
| 535 | struct nd_region *nd_region = to_nd_region(dev);
|
| 536 |
|
| 537 | return sprintf(buf, "%#lx\n", nd_region->align);
|
| 538 | }
|
| 539 |
|
| 540 | static ssize_t align_store(struct device *dev,
|
| 541 | struct device_attribute *attr, const char *buf, size_t len)
|
| 542 | {
|
| 543 | struct nd_region *nd_region = to_nd_region(dev);
|
| 544 | unsigned long val, dpa;
|
| 545 | u32 mappings, remainder;
|
| 546 | int rc;
|
| 547 |
|
| 548 | rc = kstrtoul(buf, 0, &val);
|
| 549 | if (rc)
|
| 550 | return rc;
|
| 551 |
|
| 552 | |
| 553 | * Ensure space-align is evenly divisible by the region
|
| 554 | * interleave-width because the kernel typically has no facility
|
| 555 | * to determine which DIMM(s), dimm-physical-addresses, would
|
| 556 | * contribute to the tail capacity in system-physical-address
|
| 557 | * space for the namespace.
|
| 558 |
|
| 559 | mappings = max_t(u32, 1, nd_region->ndr_mappings);
|
| 560 | dpa = div_u64_rem(val, mappings, &remainder);
|
| 561 | if (!is_power_of_2(dpa) || dpa < PAGE_SIZE
|
| 562 | || val > region_size(nd_region) || remainder)
|
| 563 | return -EINVAL;
|
| 564 |
|
| 565 | |
| 566 | * Given that space allocation consults this value multiple
|
| 567 | * times ensure it does not change for the duration of the
|
| 568 | * allocation.
|
| 569 |
|
| 570 | guard(nvdimm_bus)(dev);
|
| 571 | nd_region->align = val;
|
| 572 |
|
| 573 | return len;
|
| 574 | }
|
| 575 | static DEVICE_ATTR_RW(align);
|
| 576 |
|
| 577 | static ssize_t region_badblocks_show(struct device *dev,
|
| 578 | struct device_attribute *attr, char *buf)
|
| 579 | {
|
| 580 | struct nd_region *nd_region = to_nd_region(dev);
|
| 581 | ssize_t rc;
|
| 582 |
|
| 583 | device_lock(dev);
|
| 584 | if (dev->driver)
|
| 585 | rc = badblocks_show(&nd_region->bb, buf, 0);
|
| 586 | else
|
| 587 | rc = -ENXIO;
|
| 588 | device_unlock(dev);
|
| 589 |
|
| 590 | return rc;
|
| 591 | }
|
| 592 | static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
|
| 593 |
|
| 594 | static ssize_t resource_show(struct device *dev,
|
| 595 | struct device_attribute *attr, char *buf)
|
| 596 | {
|
| 597 | struct nd_region *nd_region = to_nd_region(dev);
|
| 598 |
|
| 599 | return sprintf(buf, "%#llx\n", nd_region->ndr_start);
|
| 600 | }
|
| 601 | static DEVICE_ATTR_ADMIN_RO(resource);
|
| 602 |
|
| 603 | static ssize_t persistence_domain_show(struct device *dev,
|
| 604 | struct device_attribute *attr, char *buf)
|
| 605 | {
|
| 606 | struct nd_region *nd_region = to_nd_region(dev);
|
| 607 |
|
| 608 | if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
|
| 609 | return sprintf(buf, "cpu_cache\n");
|
| 610 | else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
|
| 611 | return sprintf(buf, "memory_controller\n");
|
| 612 | else
|
| 613 | return sprintf(buf, "\n");
|
| 614 | }
|
| 615 | static DEVICE_ATTR_RO(persistence_domain);
|
| 616 |
|
| 617 | static struct attribute *nd_region_attributes[] = {
|
| 618 | &dev_attr_size.attr,
|
| 619 | &dev_attr_align.attr,
|
| 620 | &dev_attr_nstype.attr,
|
| 621 | &dev_attr_mappings.attr,
|
| 622 | &dev_attr_btt_seed.attr,
|
| 623 | &dev_attr_pfn_seed.attr,
|
| 624 | &dev_attr_dax_seed.attr,
|
| 625 | &dev_attr_deep_flush.attr,
|
| 626 | &dev_attr_read_only.attr,
|
| 627 | &dev_attr_set_cookie.attr,
|
| 628 | &dev_attr_available_size.attr,
|
| 629 | &dev_attr_max_available_extent.attr,
|
| 630 | &dev_attr_namespace_seed.attr,
|
| 631 | &dev_attr_init_namespaces.attr,
|
| 632 | &dev_attr_badblocks.attr,
|
| 633 | &dev_attr_resource.attr,
|
| 634 | &dev_attr_persistence_domain.attr,
|
| 635 | NULL,
|
| 636 | };
|
| 637 |
|
| 638 | static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
|
| 639 | {
|
| 640 | struct device *dev = container_of(kobj, typeof(*dev), kobj);
|
| 641 | struct nd_region *nd_region = to_nd_region(dev);
|
| 642 | struct nd_interleave_set *nd_set = nd_region->nd_set;
|
| 643 | int type = nd_region_to_nstype(nd_region);
|
| 644 |
|
| 645 | if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
|
| 646 | return 0;
|
| 647 |
|
| 648 | if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
|
| 649 | return 0;
|
| 650 |
|
| 651 | if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
|
| 652 | return 0;
|
| 653 |
|
| 654 | if (a == &dev_attr_resource.attr && !is_memory(dev))
|
| 655 | return 0;
|
| 656 |
|
| 657 | if (a == &dev_attr_deep_flush.attr) {
|
| 658 | int has_flush = nvdimm_has_flush(nd_region);
|
| 659 |
|
| 660 | if (has_flush == 1)
|
| 661 | return a->mode;
|
| 662 | else if (has_flush == 0)
|
| 663 | return 0444;
|
| 664 | else
|
| 665 | return 0;
|
| 666 | }
|
| 667 |
|
| 668 | if (a == &dev_attr_persistence_domain.attr) {
|
| 669 | if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
|
| 670 | | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
|
| 671 | return 0;
|
| 672 | return a->mode;
|
| 673 | }
|
| 674 |
|
| 675 | if (a == &dev_attr_align.attr)
|
| 676 | return a->mode;
|
| 677 |
|
| 678 | if (a != &dev_attr_set_cookie.attr
|
| 679 | && a != &dev_attr_available_size.attr)
|
| 680 | return a->mode;
|
| 681 |
|
| 682 | if (type == ND_DEVICE_NAMESPACE_PMEM &&
|
| 683 | a == &dev_attr_available_size.attr)
|
| 684 | return a->mode;
|
| 685 | else if (is_memory(dev) && nd_set)
|
| 686 | return a->mode;
|
| 687 |
|
| 688 | return 0;
|
| 689 | }
|
| 690 |
|
| 691 | static ssize_t mappingN(struct device *dev, char *buf, int n)
|
| 692 | {
|
| 693 | struct nd_region *nd_region = to_nd_region(dev);
|
| 694 | struct nd_mapping *nd_mapping;
|
| 695 | struct nvdimm *nvdimm;
|
| 696 |
|
| 697 | if (n >= nd_region->ndr_mappings)
|
| 698 | return -ENXIO;
|
| 699 | nd_mapping = &nd_region->mapping[n];
|
| 700 | nvdimm = nd_mapping->nvdimm;
|
| 701 |
|
| 702 | return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
|
| 703 | nd_mapping->start, nd_mapping->size,
|
| 704 | nd_mapping->position);
|
| 705 | }
|
| 706 |
|
| 707 | #define REGION_MAPPING(idx) \
|
| 708 | static ssize_t mapping##idx##_show(struct device *dev, \
|
| 709 | struct device_attribute *attr, char *buf) \
|
| 710 | { \
|
| 711 | return mappingN(dev, buf, idx); \
|
| 712 | } \
|
| 713 | static DEVICE_ATTR_RO(mapping##idx)
|
| 714 |
|
| 715 | |
| 716 | * 32 should be enough for a while, even in the presence of socket
|
| 717 | * interleave a 32-way interleave set is a degenerate case.
|
| 718 |
|
| 719 | REGION_MAPPING(0);
|
| 720 | REGION_MAPPING(1);
|
| 721 | REGION_MAPPING(2);
|
| 722 | REGION_MAPPING(3);
|
| 723 | REGION_MAPPING(4);
|
| 724 | REGION_MAPPING(5);
|
| 725 | REGION_MAPPING(6);
|
| 726 | REGION_MAPPING(7);
|
| 727 | REGION_MAPPING(8);
|
| 728 | REGION_MAPPING(9);
|
| 729 | REGION_MAPPING(10);
|
| 730 | REGION_MAPPING(11);
|
| 731 | REGION_MAPPING(12);
|
| 732 | REGION_MAPPING(13);
|
| 733 | REGION_MAPPING(14);
|
| 734 | REGION_MAPPING(15);
|
| 735 | REGION_MAPPING(16);
|
| 736 | REGION_MAPPING(17);
|
| 737 | REGION_MAPPING(18);
|
| 738 | REGION_MAPPING(19);
|
| 739 | REGION_MAPPING(20);
|
| 740 | REGION_MAPPING(21);
|
| 741 | REGION_MAPPING(22);
|
| 742 | REGION_MAPPING(23);
|
| 743 | REGION_MAPPING(24);
|
| 744 | REGION_MAPPING(25);
|
| 745 | REGION_MAPPING(26);
|
| 746 | REGION_MAPPING(27);
|
| 747 | REGION_MAPPING(28);
|
| 748 | REGION_MAPPING(29);
|
| 749 | REGION_MAPPING(30);
|
| 750 | REGION_MAPPING(31);
|
| 751 |
|
| 752 | static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
|
| 753 | {
|
| 754 | struct device *dev = container_of(kobj, struct device, kobj);
|
| 755 | struct nd_region *nd_region = to_nd_region(dev);
|
| 756 |
|
| 757 | if (n < nd_region->ndr_mappings)
|
| 758 | return a->mode;
|
| 759 | return 0;
|
| 760 | }
|
| 761 |
|
| 762 | static struct attribute *mapping_attributes[] = {
|
| 763 | &dev_attr_mapping0.attr,
|
| 764 | &dev_attr_mapping1.attr,
|
| 765 | &dev_attr_mapping2.attr,
|
| 766 | &dev_attr_mapping3.attr,
|
| 767 | &dev_attr_mapping4.attr,
|
| 768 | &dev_attr_mapping5.attr,
|
| 769 | &dev_attr_mapping6.attr,
|
| 770 | &dev_attr_mapping7.attr,
|
| 771 | &dev_attr_mapping8.attr,
|
| 772 | &dev_attr_mapping9.attr,
|
| 773 | &dev_attr_mapping10.attr,
|
| 774 | &dev_attr_mapping11.attr,
|
| 775 | &dev_attr_mapping12.attr,
|
| 776 | &dev_attr_mapping13.attr,
|
| 777 | &dev_attr_mapping14.attr,
|
| 778 | &dev_attr_mapping15.attr,
|
| 779 | &dev_attr_mapping16.attr,
|
| 780 | &dev_attr_mapping17.attr,
|
| 781 | &dev_attr_mapping18.attr,
|
| 782 | &dev_attr_mapping19.attr,
|
| 783 | &dev_attr_mapping20.attr,
|
| 784 | &dev_attr_mapping21.attr,
|
| 785 | &dev_attr_mapping22.attr,
|
| 786 | &dev_attr_mapping23.attr,
|
| 787 | &dev_attr_mapping24.attr,
|
| 788 | &dev_attr_mapping25.attr,
|
| 789 | &dev_attr_mapping26.attr,
|
| 790 | &dev_attr_mapping27.attr,
|
| 791 | &dev_attr_mapping28.attr,
|
| 792 | &dev_attr_mapping29.attr,
|
| 793 | &dev_attr_mapping30.attr,
|
| 794 | &dev_attr_mapping31.attr,
|
| 795 | NULL,
|
| 796 | };
|
| 797 |
|
| 798 | static const struct attribute_group nd_mapping_attribute_group = {
|
| 799 | .is_visible = mapping_visible,
|
| 800 | .attrs = mapping_attributes,
|
| 801 | };
|
| 802 |
|
| 803 | static const struct attribute_group nd_region_attribute_group = {
|
| 804 | .attrs = nd_region_attributes,
|
| 805 | .is_visible = region_visible,
|
| 806 | };
|
| 807 |
|
| 808 | static const struct attribute_group *nd_region_attribute_groups[] = {
|
| 809 | &nd_device_attribute_group,
|
| 810 | &nd_region_attribute_group,
|
| 811 | &nd_numa_attribute_group,
|
| 812 | &nd_mapping_attribute_group,
|
| 813 | NULL,
|
| 814 | };
|
| 815 |
|
| 816 | static const struct device_type nd_pmem_device_type = {
|
| 817 | .name = "nd_pmem",
|
| 818 | .release = nd_region_release,
|
| 819 | .groups = nd_region_attribute_groups,
|
| 820 | };
|
| 821 |
|
| 822 | static const struct device_type nd_volatile_device_type = {
|
| 823 | .name = "nd_volatile",
|
| 824 | .release = nd_region_release,
|
| 825 | .groups = nd_region_attribute_groups,
|
| 826 | };
|
| 827 |
|
| 828 | bool is_nd_pmem(const struct device *dev)
|
| 829 | {
|
| 830 | return dev ? dev->type == &nd_pmem_device_type : false;
|
| 831 | }
|
| 832 |
|
| 833 | bool is_nd_volatile(const struct device *dev)
|
| 834 | {
|
| 835 | return dev ? dev->type == &nd_volatile_device_type : false;
|
| 836 | }
|
| 837 |
|
| 838 | u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
|
| 839 | struct nd_namespace_index *nsindex)
|
| 840 | {
|
| 841 | struct nd_interleave_set *nd_set = nd_region->nd_set;
|
| 842 |
|
| 843 | if (!nd_set)
|
| 844 | return 0;
|
| 845 |
|
| 846 | if (nsindex && __le16_to_cpu(nsindex->major) == 1
|
| 847 | && __le16_to_cpu(nsindex->minor) == 1)
|
| 848 | return nd_set->cookie1;
|
| 849 | return nd_set->cookie2;
|
| 850 | }
|
| 851 |
|
| 852 | u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
|
| 853 | {
|
| 854 | struct nd_interleave_set *nd_set = nd_region->nd_set;
|
| 855 |
|
| 856 | if (nd_set)
|
| 857 | return nd_set->altcookie;
|
| 858 | return 0;
|
| 859 | }
|
| 860 |
|
| 861 | void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
|
| 862 | {
|
| 863 | struct nd_label_ent *label_ent, *e;
|
| 864 |
|
| 865 | lockdep_assert_held(&nd_mapping->lock);
|
| 866 | list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
|
| 867 | list_del(&label_ent->list);
|
| 868 | kfree(label_ent);
|
| 869 | }
|
| 870 | }
|
| 871 |
|
| 872 | |
| 873 | * When a namespace is activated create new seeds for the next
|
| 874 | * namespace, or namespace-personality to be configured.
|
| 875 |
|
| 876 | void nd_region_advance_seeds(struct nd_region *nd_region, struct device *dev)
|
| 877 | {
|
| 878 | guard(nvdimm_bus)(dev);
|
| 879 | if (nd_region->ns_seed == dev) {
|
| 880 | nd_region_create_ns_seed(nd_region);
|
| 881 | } else if (is_nd_btt(dev)) {
|
| 882 | struct nd_btt *nd_btt = to_nd_btt(dev);
|
| 883 |
|
| 884 | if (nd_region->btt_seed == dev)
|
| 885 | nd_region_create_btt_seed(nd_region);
|
| 886 | if (nd_region->ns_seed == &nd_btt->ndns->dev)
|
| 887 | nd_region_create_ns_seed(nd_region);
|
| 888 | } else if (is_nd_pfn(dev)) {
|
| 889 | struct nd_pfn *nd_pfn = to_nd_pfn(dev);
|
| 890 |
|
| 891 | if (nd_region->pfn_seed == dev)
|
| 892 | nd_region_create_pfn_seed(nd_region);
|
| 893 | if (nd_region->ns_seed == &nd_pfn->ndns->dev)
|
| 894 | nd_region_create_ns_seed(nd_region);
|
| 895 | } else if (is_nd_dax(dev)) {
|
| 896 | struct nd_dax *nd_dax = to_nd_dax(dev);
|
| 897 |
|
| 898 | if (nd_region->dax_seed == dev)
|
| 899 | nd_region_create_dax_seed(nd_region);
|
| 900 | if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
|
| 901 | nd_region_create_ns_seed(nd_region);
|
| 902 | }
|
| 903 | }
|
| 904 |
|
| 905 | |
| 906 | * nd_region_acquire_lane - allocate and lock a lane
|
| 907 | * @nd_region: region id and number of lanes possible
|
| 908 | *
|
| 909 | * A lane correlates to a log slot in the BTT. Lanes are shared across
|
| 910 | * CPUs using a static lane = cpu % num_lanes mapping, with a per-lane
|
| 911 | * mutex to serialize access.
|
| 912 | *
|
| 913 | * Callers must be in sleepable context. The only in-tree caller is
|
| 914 | * BTT's ->submit_bio handler (btt_read_pg / btt_write_pg).
|
| 915 |
|
| 916 | unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
|
| 917 | __acquires(&nd_region->lane[lane].lock)
|
| 918 | {
|
| 919 | unsigned int lane;
|
| 920 |
|
| 921 | might_sleep();
|
| 922 |
|
| 923 | lane = raw_smp_processor_id() % nd_region->num_lanes;
|
| 924 | mutex_lock(&nd_region->lane[lane].lock);
|
| 925 | return lane;
|
| 926 | }
|
| 927 | EXPORT_SYMBOL(nd_region_acquire_lane);
|
| 928 |
|
| 929 | void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
|
| 930 | __releases(&nd_region->lane[lane].lock)
|
| 931 | {
|
| 932 | mutex_unlock(&nd_region->lane[lane].lock);
|
| 933 | }
|
| 934 | EXPORT_SYMBOL(nd_region_release_lane);
|
| 935 |
|
| 936 | |
| 937 | * PowerPC requires this alignment for memremap_pages(). All other archs
|
| 938 | * should be ok with SUBSECTION_SIZE (see memremap_compat_align()).
|
| 939 |
|
| 940 | #define MEMREMAP_COMPAT_ALIGN_MAX SZ_16M
|
| 941 |
|
| 942 | static unsigned long default_align(struct nd_region *nd_region)
|
| 943 | {
|
| 944 | unsigned long align;
|
| 945 | u32 remainder;
|
| 946 | int mappings;
|
| 947 |
|
| 948 | align = MEMREMAP_COMPAT_ALIGN_MAX;
|
| 949 | if (nd_region->ndr_size < MEMREMAP_COMPAT_ALIGN_MAX)
|
| 950 | align = PAGE_SIZE;
|
| 951 |
|
| 952 | mappings = max_t(u16, 1, nd_region->ndr_mappings);
|
| 953 | div_u64_rem(align, mappings, &remainder);
|
| 954 | if (remainder)
|
| 955 | align *= mappings;
|
| 956 |
|
| 957 | return align;
|
| 958 | }
|
| 959 |
|
| 960 | static struct lock_class_key nvdimm_region_key;
|
| 961 |
|
| 962 | static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
|
| 963 | struct nd_region_desc *ndr_desc,
|
| 964 | const struct device_type *dev_type, const char *caller)
|
| 965 | {
|
| 966 | struct nd_region *nd_region;
|
| 967 | struct device *dev;
|
| 968 | unsigned int i;
|
| 969 | int ro = 0;
|
| 970 |
|
| 971 | for (i = 0; i < ndr_desc->num_mappings; i++) {
|
| 972 | struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
|
| 973 | struct nvdimm *nvdimm = mapping->nvdimm;
|
| 974 |
|
| 975 | if ((mapping->start | mapping->size) % PAGE_SIZE) {
|
| 976 | dev_err(&nvdimm_bus->dev,
|
| 977 | "%s: %s mapping%d is not %ld aligned\n",
|
| 978 | caller, dev_name(&nvdimm->dev), i, PAGE_SIZE);
|
| 979 | return NULL;
|
| 980 | }
|
| 981 |
|
| 982 | if (test_bit(NDD_UNARMED, &nvdimm->flags))
|
| 983 | ro = 1;
|
| 984 |
|
| 985 | }
|
| 986 |
|
| 987 | nd_region =
|
| 988 | kzalloc_flex(*nd_region, mapping, ndr_desc->num_mappings);
|
| 989 |
|
| 990 | if (!nd_region)
|
| 991 | return NULL;
|
| 992 | nd_region->ndr_mappings = ndr_desc->num_mappings;
|
| 993 |
|
| 994 | if (test_bit(ND_REGION_CXL, &ndr_desc->flags)) {
|
| 995 | nd_region->id = ndr_desc->memregion;
|
| 996 | } else {
|
| 997 | nd_region->id = memregion_alloc(GFP_KERNEL);
|
| 998 | if (nd_region->id < 0)
|
| 999 | goto err_id;
|
| 1000 | }
|
| 1001 |
|
| 1002 | nd_region->num_lanes = ndr_desc->num_lanes;
|
| 1003 | if (!nd_region->num_lanes)
|
| 1004 | goto err_percpu;
|
| 1005 | nd_region->lane = kzalloc_objs(*nd_region->lane, nd_region->num_lanes);
|
| 1006 | if (!nd_region->lane)
|
| 1007 | goto err_percpu;
|
| 1008 |
|
| 1009 | for (i = 0; i < nd_region->num_lanes; i++)
|
| 1010 | mutex_init(&nd_region->lane[i].lock);
|
| 1011 |
|
| 1012 | for (i = 0; i < ndr_desc->num_mappings; i++) {
|
| 1013 | struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
|
| 1014 | struct nvdimm *nvdimm = mapping->nvdimm;
|
| 1015 |
|
| 1016 | nd_region->mapping[i].nvdimm = nvdimm;
|
| 1017 | nd_region->mapping[i].start = mapping->start;
|
| 1018 | nd_region->mapping[i].size = mapping->size;
|
| 1019 | nd_region->mapping[i].position = mapping->position;
|
| 1020 | INIT_LIST_HEAD(&nd_region->mapping[i].labels);
|
| 1021 | mutex_init(&nd_region->mapping[i].lock);
|
| 1022 |
|
| 1023 | get_device(&nvdimm->dev);
|
| 1024 | }
|
| 1025 | nd_region->provider_data = ndr_desc->provider_data;
|
| 1026 | nd_region->nd_set = ndr_desc->nd_set;
|
| 1027 | nd_region->flags = ndr_desc->flags;
|
| 1028 | nd_region->ro = ro;
|
| 1029 | nd_region->numa_node = ndr_desc->numa_node;
|
| 1030 | nd_region->target_node = ndr_desc->target_node;
|
| 1031 | ida_init(&nd_region->ns_ida);
|
| 1032 | ida_init(&nd_region->btt_ida);
|
| 1033 | ida_init(&nd_region->pfn_ida);
|
| 1034 | ida_init(&nd_region->dax_ida);
|
| 1035 | dev = &nd_region->dev;
|
| 1036 | dev_set_name(dev, "region%d", nd_region->id);
|
| 1037 | dev->parent = &nvdimm_bus->dev;
|
| 1038 | dev->type = dev_type;
|
| 1039 | dev->groups = ndr_desc->attr_groups;
|
| 1040 | dev->of_node = ndr_desc->of_node;
|
| 1041 | nd_region->ndr_size = resource_size(ndr_desc->res);
|
| 1042 | nd_region->ndr_start = ndr_desc->res->start;
|
| 1043 | nd_region->align = default_align(nd_region);
|
| 1044 | if (ndr_desc->flush)
|
| 1045 | nd_region->flush = ndr_desc->flush;
|
| 1046 | else
|
| 1047 | nd_region->flush = NULL;
|
| 1048 |
|
| 1049 | device_initialize(dev);
|
| 1050 | lockdep_set_class(&dev->mutex, &nvdimm_region_key);
|
| 1051 | nd_device_register(dev);
|
| 1052 |
|
| 1053 | return nd_region;
|
| 1054 |
|
| 1055 | err_percpu:
|
| 1056 | if (!test_bit(ND_REGION_CXL, &ndr_desc->flags))
|
| 1057 | memregion_free(nd_region->id);
|
| 1058 | err_id:
|
| 1059 | kfree(nd_region);
|
| 1060 | return NULL;
|
| 1061 | }
|
| 1062 |
|
| 1063 | struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
|
| 1064 | struct nd_region_desc *ndr_desc)
|
| 1065 | {
|
| 1066 | ndr_desc->num_lanes = ND_MAX_LANES;
|
| 1067 | return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
|
| 1068 | __func__);
|
| 1069 | }
|
| 1070 | EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
|
| 1071 |
|
| 1072 | struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
|
| 1073 | struct nd_region_desc *ndr_desc)
|
| 1074 | {
|
| 1075 | ndr_desc->num_lanes = ND_MAX_LANES;
|
| 1076 | return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
|
| 1077 | __func__);
|
| 1078 | }
|
| 1079 | EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
|
| 1080 |
|
| 1081 | void nvdimm_region_delete(struct nd_region *nd_region)
|
| 1082 | {
|
| 1083 | if (nd_region)
|
| 1084 | nd_device_unregister(&nd_region->dev, ND_SYNC);
|
| 1085 | }
|
| 1086 | EXPORT_SYMBOL_GPL(nvdimm_region_delete);
|
| 1087 |
|
| 1088 | int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
|
| 1089 | {
|
| 1090 | int rc = 0;
|
| 1091 |
|
| 1092 | if (!nd_region->flush)
|
| 1093 | rc = generic_nvdimm_flush(nd_region);
|
| 1094 | else {
|
| 1095 | rc = nd_region->flush(nd_region, bio);
|
| 1096 | if (rc > 0)
|
| 1097 | return rc;
|
| 1098 | if (rc && rc != -ENOMEM)
|
| 1099 | rc = -EIO;
|
| 1100 | }
|
| 1101 |
|
| 1102 | return rc;
|
| 1103 | }
|
| 1104 | |
| 1105 | * generic_nvdimm_flush() - flush any posted write queues between the cpu and pmem media
|
| 1106 | * @nd_region: interleaved pmem region
|
| 1107 |
|
| 1108 | int generic_nvdimm_flush(struct nd_region *nd_region)
|
| 1109 | {
|
| 1110 | struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
|
| 1111 | int i, idx;
|
| 1112 |
|
| 1113 | |
| 1114 | * Try to encourage some diversity in flush hint addresses
|
| 1115 | * across cpus assuming a limited number of flush hints.
|
| 1116 |
|
| 1117 | idx = this_cpu_read(flush_idx);
|
| 1118 | idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
|
| 1119 |
|
| 1120 | |
| 1121 | * The pmem_wmb() is needed to 'sfence' all
|
| 1122 | * previous writes such that they are architecturally visible for
|
| 1123 | * the platform buffer flush. Note that we've already arranged for pmem
|
| 1124 | * writes to avoid the cache via memcpy_flushcache(). The final
|
| 1125 | * wmb() ensures ordering for the NVDIMM flush write.
|
| 1126 |
|
| 1127 | pmem_wmb();
|
| 1128 | for (i = 0; i < nd_region->ndr_mappings; i++)
|
| 1129 | if (ndrd_get_flush_wpq(ndrd, i, 0))
|
| 1130 | writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
|
| 1131 | wmb();
|
| 1132 |
|
| 1133 | return 0;
|
| 1134 | }
|
| 1135 | EXPORT_SYMBOL_GPL(nvdimm_flush);
|
| 1136 |
|
| 1137 | |
| 1138 | * nvdimm_has_flush - determine write flushing requirements
|
| 1139 | * @nd_region: interleaved pmem region
|
| 1140 | *
|
| 1141 | * Returns 1 if writes require flushing
|
| 1142 | * Returns 0 if writes do not require flushing
|
| 1143 | * Returns -ENXIO if flushing capability can not be determined
|
| 1144 |
|
| 1145 | int nvdimm_has_flush(struct nd_region *nd_region)
|
| 1146 | {
|
| 1147 | int i;
|
| 1148 |
|
| 1149 |
|
| 1150 | if (nd_region->ndr_mappings == 0
|
| 1151 | || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
|
| 1152 | return -ENXIO;
|
| 1153 |
|
| 1154 |
|
| 1155 | if (test_bit(ND_REGION_ASYNC, &nd_region->flags) && nd_region->flush)
|
| 1156 | return 1;
|
| 1157 |
|
| 1158 |
|
| 1159 | for (i = 0; i < nd_region->ndr_mappings; i++) {
|
| 1160 | struct nd_mapping *nd_mapping = &nd_region->mapping[i];
|
| 1161 | struct nvdimm *nvdimm = nd_mapping->nvdimm;
|
| 1162 |
|
| 1163 |
|
| 1164 | if (nvdimm->num_flush)
|
| 1165 | return 1;
|
| 1166 | }
|
| 1167 |
|
| 1168 | |
| 1169 | * The platform defines dimm devices without hints nor explicit flush,
|
| 1170 | * assume platform persistence mechanism like ADR
|
| 1171 |
|
| 1172 | return 0;
|
| 1173 | }
|
| 1174 | EXPORT_SYMBOL_GPL(nvdimm_has_flush);
|
| 1175 |
|
| 1176 | int nvdimm_has_cache(struct nd_region *nd_region)
|
| 1177 | {
|
| 1178 | return is_nd_pmem(&nd_region->dev) &&
|
| 1179 | !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
|
| 1180 | }
|
| 1181 | EXPORT_SYMBOL_GPL(nvdimm_has_cache);
|
| 1182 |
|
| 1183 | bool is_nvdimm_sync(struct nd_region *nd_region)
|
| 1184 | {
|
| 1185 | if (is_nd_volatile(&nd_region->dev))
|
| 1186 | return true;
|
| 1187 |
|
| 1188 | return is_nd_pmem(&nd_region->dev) &&
|
| 1189 | !test_bit(ND_REGION_ASYNC, &nd_region->flags);
|
| 1190 | }
|
| 1191 | EXPORT_SYMBOL_GPL(is_nvdimm_sync);
|
| 1192 |
|
| 1193 | MODULE_IMPORT_NS("DEVMEM");
|