1 /* 2 * Persistent Memory Driver 3 * 4 * Copyright (c) 2014-2015, Intel Corporation. 5 * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>. 6 * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>. 7 * 8 * This program is free software; you can redistribute it and/or modify it 9 * under the terms and conditions of the GNU General Public License, 10 * version 2, as published by the Free Software Foundation. 11 * 12 * This program is distributed in the hope it will be useful, but WITHOUT 13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 15 * more details. 16 */ 17 18 #include <asm/cacheflush.h> 19 #include <linux/blkdev.h> 20 #include <linux/hdreg.h> 21 #include <linux/init.h> 22 #include <linux/platform_device.h> 23 #include <linux/module.h> 24 #include <linux/moduleparam.h> 25 #include <linux/badblocks.h> 26 #include <linux/memremap.h> 27 #include <linux/vmalloc.h> 28 #include <linux/blk-mq.h> 29 #include <linux/pfn_t.h> 30 #include <linux/slab.h> 31 #include <linux/pmem.h> 32 #include <linux/nd.h> 33 #include "pmem.h" 34 #include "pfn.h" 35 #include "nd.h" 36 37 static struct device *to_dev(struct pmem_device *pmem) 38 { 39 /* 40 * nvdimm bus services need a 'dev' parameter, and we record the device 41 * at init in bb.dev. 42 */ 43 return pmem->bb.dev; 44 } 45 46 static struct nd_region *to_region(struct pmem_device *pmem) 47 { 48 return to_nd_region(to_dev(pmem)->parent); 49 } 50 51 static int pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset, 52 unsigned int len) 53 { 54 struct device *dev = to_dev(pmem); 55 sector_t sector; 56 long cleared; 57 int rc = 0; 58 59 sector = (offset - pmem->data_offset) / 512; 60 61 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len); 62 if (cleared < len) 63 rc = -EIO; 64 if (cleared > 0 && cleared / 512) { 65 cleared /= 512; 66 dev_dbg(dev, "%s: %#llx clear %ld sector%s\n", __func__, 67 (unsigned long long) sector, cleared, 68 cleared > 1 ? "s" : ""); 69 badblocks_clear(&pmem->bb, sector, cleared); 70 } 71 72 invalidate_pmem(pmem->virt_addr + offset, len); 73 74 return rc; 75 } 76 77 static void write_pmem(void *pmem_addr, struct page *page, 78 unsigned int off, unsigned int len) 79 { 80 void *mem = kmap_atomic(page); 81 82 memcpy_to_pmem(pmem_addr, mem + off, len); 83 kunmap_atomic(mem); 84 } 85 86 static int read_pmem(struct page *page, unsigned int off, 87 void *pmem_addr, unsigned int len) 88 { 89 int rc; 90 void *mem = kmap_atomic(page); 91 92 rc = memcpy_from_pmem(mem + off, pmem_addr, len); 93 kunmap_atomic(mem); 94 if (rc) 95 return -EIO; 96 return 0; 97 } 98 99 static int pmem_do_bvec(struct pmem_device *pmem, struct page *page, 100 unsigned int len, unsigned int off, bool is_write, 101 sector_t sector) 102 { 103 int rc = 0; 104 bool bad_pmem = false; 105 phys_addr_t pmem_off = sector * 512 + pmem->data_offset; 106 void *pmem_addr = pmem->virt_addr + pmem_off; 107 108 if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) 109 bad_pmem = true; 110 111 if (!is_write) { 112 if (unlikely(bad_pmem)) 113 rc = -EIO; 114 else { 115 rc = read_pmem(page, off, pmem_addr, len); 116 flush_dcache_page(page); 117 } 118 } else { 119 /* 120 * Note that we write the data both before and after 121 * clearing poison. The write before clear poison 122 * handles situations where the latest written data is 123 * preserved and the clear poison operation simply marks 124 * the address range as valid without changing the data. 125 * In this case application software can assume that an 126 * interrupted write will either return the new good 127 * data or an error. 128 * 129 * However, if pmem_clear_poison() leaves the data in an 130 * indeterminate state we need to perform the write 131 * after clear poison. 132 */ 133 flush_dcache_page(page); 134 write_pmem(pmem_addr, page, off, len); 135 if (unlikely(bad_pmem)) { 136 rc = pmem_clear_poison(pmem, pmem_off, len); 137 write_pmem(pmem_addr, page, off, len); 138 } 139 } 140 141 return rc; 142 } 143 144 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */ 145 #ifndef REQ_FLUSH 146 #define REQ_FLUSH REQ_PREFLUSH 147 #endif 148 149 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio) 150 { 151 int rc = 0; 152 bool do_acct; 153 unsigned long start; 154 struct bio_vec bvec; 155 struct bvec_iter iter; 156 struct pmem_device *pmem = q->queuedata; 157 struct nd_region *nd_region = to_region(pmem); 158 159 if (bio->bi_opf & REQ_FLUSH) 160 nvdimm_flush(nd_region); 161 162 do_acct = nd_iostat_start(bio, &start); 163 bio_for_each_segment(bvec, bio, iter) { 164 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, 165 bvec.bv_offset, op_is_write(bio_op(bio)), 166 iter.bi_sector); 167 if (rc) { 168 bio->bi_error = rc; 169 break; 170 } 171 } 172 if (do_acct) 173 nd_iostat_end(bio, start); 174 175 if (bio->bi_opf & REQ_FUA) 176 nvdimm_flush(nd_region); 177 178 bio_endio(bio); 179 return BLK_QC_T_NONE; 180 } 181 182 static int pmem_rw_page(struct block_device *bdev, sector_t sector, 183 struct page *page, bool is_write) 184 { 185 struct pmem_device *pmem = bdev->bd_queue->queuedata; 186 int rc; 187 188 rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, is_write, sector); 189 190 /* 191 * The ->rw_page interface is subtle and tricky. The core 192 * retries on any error, so we can only invoke page_endio() in 193 * the successful completion case. Otherwise, we'll see crashes 194 * caused by double completion. 195 */ 196 if (rc == 0) 197 page_endio(page, is_write, 0); 198 199 return rc; 200 } 201 202 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */ 203 __weak long pmem_direct_access(struct block_device *bdev, sector_t sector, 204 void **kaddr, pfn_t *pfn, long size) 205 { 206 struct pmem_device *pmem = bdev->bd_queue->queuedata; 207 resource_size_t offset = sector * 512 + pmem->data_offset; 208 209 if (unlikely(is_bad_pmem(&pmem->bb, sector, size))) 210 return -EIO; 211 *kaddr = pmem->virt_addr + offset; 212 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags); 213 214 /* 215 * If badblocks are present, limit known good range to the 216 * requested range. 217 */ 218 if (unlikely(pmem->bb.count)) 219 return size; 220 return pmem->size - pmem->pfn_pad - offset; 221 } 222 223 static const struct block_device_operations pmem_fops = { 224 .owner = THIS_MODULE, 225 .rw_page = pmem_rw_page, 226 .direct_access = pmem_direct_access, 227 .revalidate_disk = nvdimm_revalidate_disk, 228 }; 229 230 static void pmem_release_queue(void *q) 231 { 232 blk_cleanup_queue(q); 233 } 234 235 static void pmem_freeze_queue(void *q) 236 { 237 blk_freeze_queue_start(q); 238 } 239 240 static void pmem_release_disk(void *disk) 241 { 242 del_gendisk(disk); 243 put_disk(disk); 244 } 245 246 static int pmem_attach_disk(struct device *dev, 247 struct nd_namespace_common *ndns) 248 { 249 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); 250 struct nd_region *nd_region = to_nd_region(dev->parent); 251 struct vmem_altmap __altmap, *altmap = NULL; 252 struct resource *res = &nsio->res; 253 struct nd_pfn *nd_pfn = NULL; 254 int nid = dev_to_node(dev); 255 struct nd_pfn_sb *pfn_sb; 256 struct pmem_device *pmem; 257 struct resource pfn_res; 258 struct request_queue *q; 259 struct gendisk *disk; 260 void *addr; 261 262 /* while nsio_rw_bytes is active, parse a pfn info block if present */ 263 if (is_nd_pfn(dev)) { 264 nd_pfn = to_nd_pfn(dev); 265 altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap); 266 if (IS_ERR(altmap)) 267 return PTR_ERR(altmap); 268 } 269 270 /* we're attaching a block device, disable raw namespace access */ 271 devm_nsio_disable(dev, nsio); 272 273 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); 274 if (!pmem) 275 return -ENOMEM; 276 277 dev_set_drvdata(dev, pmem); 278 pmem->phys_addr = res->start; 279 pmem->size = resource_size(res); 280 if (nvdimm_has_flush(nd_region) < 0) 281 dev_warn(dev, "unable to guarantee persistence of writes\n"); 282 283 if (!devm_request_mem_region(dev, res->start, resource_size(res), 284 dev_name(&ndns->dev))) { 285 dev_warn(dev, "could not reserve region %pR\n", res); 286 return -EBUSY; 287 } 288 289 q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev)); 290 if (!q) 291 return -ENOMEM; 292 293 if (devm_add_action_or_reset(dev, pmem_release_queue, q)) 294 return -ENOMEM; 295 296 pmem->pfn_flags = PFN_DEV; 297 if (is_nd_pfn(dev)) { 298 addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter, 299 altmap); 300 pfn_sb = nd_pfn->pfn_sb; 301 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff); 302 pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res); 303 pmem->pfn_flags |= PFN_MAP; 304 res = &pfn_res; /* for badblocks populate */ 305 res->start += pmem->data_offset; 306 } else if (pmem_should_map_pages(dev)) { 307 addr = devm_memremap_pages(dev, &nsio->res, 308 &q->q_usage_counter, NULL); 309 pmem->pfn_flags |= PFN_MAP; 310 } else 311 addr = devm_memremap(dev, pmem->phys_addr, 312 pmem->size, ARCH_MEMREMAP_PMEM); 313 314 /* 315 * At release time the queue must be frozen before 316 * devm_memremap_pages is unwound 317 */ 318 if (devm_add_action_or_reset(dev, pmem_freeze_queue, q)) 319 return -ENOMEM; 320 321 if (IS_ERR(addr)) 322 return PTR_ERR(addr); 323 pmem->virt_addr = addr; 324 325 blk_queue_write_cache(q, true, true); 326 blk_queue_make_request(q, pmem_make_request); 327 blk_queue_physical_block_size(q, PAGE_SIZE); 328 blk_queue_max_hw_sectors(q, UINT_MAX); 329 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 330 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); 331 queue_flag_set_unlocked(QUEUE_FLAG_DAX, q); 332 q->queuedata = pmem; 333 334 disk = alloc_disk_node(0, nid); 335 if (!disk) 336 return -ENOMEM; 337 338 disk->fops = &pmem_fops; 339 disk->queue = q; 340 disk->flags = GENHD_FL_EXT_DEVT; 341 nvdimm_namespace_disk_name(ndns, disk->disk_name); 342 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset) 343 / 512); 344 if (devm_init_badblocks(dev, &pmem->bb)) 345 return -ENOMEM; 346 nvdimm_badblocks_populate(nd_region, &pmem->bb, res); 347 disk->bb = &pmem->bb; 348 device_add_disk(dev, disk); 349 350 if (devm_add_action_or_reset(dev, pmem_release_disk, disk)) 351 return -ENOMEM; 352 353 revalidate_disk(disk); 354 355 return 0; 356 } 357 358 static int nd_pmem_probe(struct device *dev) 359 { 360 struct nd_namespace_common *ndns; 361 362 ndns = nvdimm_namespace_common_probe(dev); 363 if (IS_ERR(ndns)) 364 return PTR_ERR(ndns); 365 366 if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev))) 367 return -ENXIO; 368 369 if (is_nd_btt(dev)) 370 return nvdimm_namespace_attach_btt(ndns); 371 372 if (is_nd_pfn(dev)) 373 return pmem_attach_disk(dev, ndns); 374 375 /* if we find a valid info-block we'll come back as that personality */ 376 if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0 377 || nd_dax_probe(dev, ndns) == 0) 378 return -ENXIO; 379 380 /* ...otherwise we're just a raw pmem device */ 381 return pmem_attach_disk(dev, ndns); 382 } 383 384 static int nd_pmem_remove(struct device *dev) 385 { 386 if (is_nd_btt(dev)) 387 nvdimm_namespace_detach_btt(to_nd_btt(dev)); 388 nvdimm_flush(to_nd_region(dev->parent)); 389 390 return 0; 391 } 392 393 static void nd_pmem_shutdown(struct device *dev) 394 { 395 nvdimm_flush(to_nd_region(dev->parent)); 396 } 397 398 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event) 399 { 400 struct pmem_device *pmem = dev_get_drvdata(dev); 401 struct nd_region *nd_region = to_region(pmem); 402 resource_size_t offset = 0, end_trunc = 0; 403 struct nd_namespace_common *ndns; 404 struct nd_namespace_io *nsio; 405 struct resource res; 406 407 if (event != NVDIMM_REVALIDATE_POISON) 408 return; 409 410 if (is_nd_btt(dev)) { 411 struct nd_btt *nd_btt = to_nd_btt(dev); 412 413 ndns = nd_btt->ndns; 414 } else if (is_nd_pfn(dev)) { 415 struct nd_pfn *nd_pfn = to_nd_pfn(dev); 416 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; 417 418 ndns = nd_pfn->ndns; 419 offset = pmem->data_offset + __le32_to_cpu(pfn_sb->start_pad); 420 end_trunc = __le32_to_cpu(pfn_sb->end_trunc); 421 } else 422 ndns = to_ndns(dev); 423 424 nsio = to_nd_namespace_io(&ndns->dev); 425 res.start = nsio->res.start + offset; 426 res.end = nsio->res.end - end_trunc; 427 nvdimm_badblocks_populate(nd_region, &pmem->bb, &res); 428 } 429 430 MODULE_ALIAS("pmem"); 431 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); 432 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); 433 static struct nd_device_driver nd_pmem_driver = { 434 .probe = nd_pmem_probe, 435 .remove = nd_pmem_remove, 436 .notify = nd_pmem_notify, 437 .shutdown = nd_pmem_shutdown, 438 .drv = { 439 .name = "nd_pmem", 440 }, 441 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, 442 }; 443 444 static int __init pmem_init(void) 445 { 446 return nd_driver_register(&nd_pmem_driver); 447 } 448 module_init(pmem_init); 449 450 static void pmem_exit(void) 451 { 452 driver_unregister(&nd_pmem_driver.drv); 453 } 454 module_exit(pmem_exit); 455 456 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); 457 MODULE_LICENSE("GPL v2"); 458