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/memory_hotplug.h> 25 #include <linux/moduleparam.h> 26 #include <linux/vmalloc.h> 27 #include <linux/slab.h> 28 #include <linux/pmem.h> 29 #include <linux/nd.h> 30 #include "pfn.h" 31 #include "nd.h" 32 33 struct pmem_device { 34 struct request_queue *pmem_queue; 35 struct gendisk *pmem_disk; 36 struct nd_namespace_common *ndns; 37 38 /* One contiguous memory region per device */ 39 phys_addr_t phys_addr; 40 /* when non-zero this device is hosting a 'pfn' instance */ 41 phys_addr_t data_offset; 42 void __pmem *virt_addr; 43 size_t size; 44 }; 45 46 static int pmem_major; 47 48 static void pmem_do_bvec(struct pmem_device *pmem, struct page *page, 49 unsigned int len, unsigned int off, int rw, 50 sector_t sector) 51 { 52 void *mem = kmap_atomic(page); 53 phys_addr_t pmem_off = sector * 512 + pmem->data_offset; 54 void __pmem *pmem_addr = pmem->virt_addr + pmem_off; 55 56 if (rw == READ) { 57 memcpy_from_pmem(mem + off, pmem_addr, len); 58 flush_dcache_page(page); 59 } else { 60 flush_dcache_page(page); 61 memcpy_to_pmem(pmem_addr, mem + off, len); 62 } 63 64 kunmap_atomic(mem); 65 } 66 67 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio) 68 { 69 bool do_acct; 70 unsigned long start; 71 struct bio_vec bvec; 72 struct bvec_iter iter; 73 struct block_device *bdev = bio->bi_bdev; 74 struct pmem_device *pmem = bdev->bd_disk->private_data; 75 76 do_acct = nd_iostat_start(bio, &start); 77 bio_for_each_segment(bvec, bio, iter) 78 pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, bvec.bv_offset, 79 bio_data_dir(bio), iter.bi_sector); 80 if (do_acct) 81 nd_iostat_end(bio, start); 82 83 if (bio_data_dir(bio)) 84 wmb_pmem(); 85 86 bio_endio(bio); 87 return BLK_QC_T_NONE; 88 } 89 90 static int pmem_rw_page(struct block_device *bdev, sector_t sector, 91 struct page *page, int rw) 92 { 93 struct pmem_device *pmem = bdev->bd_disk->private_data; 94 95 pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector); 96 if (rw & WRITE) 97 wmb_pmem(); 98 page_endio(page, rw & WRITE, 0); 99 100 return 0; 101 } 102 103 static long pmem_direct_access(struct block_device *bdev, sector_t sector, 104 void __pmem **kaddr, unsigned long *pfn) 105 { 106 struct pmem_device *pmem = bdev->bd_disk->private_data; 107 resource_size_t offset = sector * 512 + pmem->data_offset; 108 109 *kaddr = pmem->virt_addr + offset; 110 *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT; 111 112 return pmem->size - offset; 113 } 114 115 static const struct block_device_operations pmem_fops = { 116 .owner = THIS_MODULE, 117 .rw_page = pmem_rw_page, 118 .direct_access = pmem_direct_access, 119 .revalidate_disk = nvdimm_revalidate_disk, 120 }; 121 122 static struct pmem_device *pmem_alloc(struct device *dev, 123 struct resource *res, int id) 124 { 125 struct pmem_device *pmem; 126 127 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); 128 if (!pmem) 129 return ERR_PTR(-ENOMEM); 130 131 pmem->phys_addr = res->start; 132 pmem->size = resource_size(res); 133 if (!arch_has_wmb_pmem()) 134 dev_warn(dev, "unable to guarantee persistence of writes\n"); 135 136 if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size, 137 dev_name(dev))) { 138 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n", 139 &pmem->phys_addr, pmem->size); 140 return ERR_PTR(-EBUSY); 141 } 142 143 if (pmem_should_map_pages(dev)) 144 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res); 145 else 146 pmem->virt_addr = (void __pmem *) devm_memremap(dev, 147 pmem->phys_addr, pmem->size, 148 ARCH_MEMREMAP_PMEM); 149 150 if (IS_ERR(pmem->virt_addr)) 151 return (void __force *) pmem->virt_addr; 152 153 return pmem; 154 } 155 156 static void pmem_detach_disk(struct pmem_device *pmem) 157 { 158 if (!pmem->pmem_disk) 159 return; 160 161 del_gendisk(pmem->pmem_disk); 162 put_disk(pmem->pmem_disk); 163 blk_cleanup_queue(pmem->pmem_queue); 164 } 165 166 static int pmem_attach_disk(struct device *dev, 167 struct nd_namespace_common *ndns, struct pmem_device *pmem) 168 { 169 int nid = dev_to_node(dev); 170 struct gendisk *disk; 171 172 pmem->pmem_queue = blk_alloc_queue_node(GFP_KERNEL, nid); 173 if (!pmem->pmem_queue) 174 return -ENOMEM; 175 176 blk_queue_make_request(pmem->pmem_queue, pmem_make_request); 177 blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE); 178 blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX); 179 blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY); 180 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue); 181 182 disk = alloc_disk_node(0, nid); 183 if (!disk) { 184 blk_cleanup_queue(pmem->pmem_queue); 185 return -ENOMEM; 186 } 187 188 disk->major = pmem_major; 189 disk->first_minor = 0; 190 disk->fops = &pmem_fops; 191 disk->private_data = pmem; 192 disk->queue = pmem->pmem_queue; 193 disk->flags = GENHD_FL_EXT_DEVT; 194 nvdimm_namespace_disk_name(ndns, disk->disk_name); 195 disk->driverfs_dev = dev; 196 set_capacity(disk, (pmem->size - pmem->data_offset) / 512); 197 pmem->pmem_disk = disk; 198 199 add_disk(disk); 200 revalidate_disk(disk); 201 202 return 0; 203 } 204 205 static int pmem_rw_bytes(struct nd_namespace_common *ndns, 206 resource_size_t offset, void *buf, size_t size, int rw) 207 { 208 struct pmem_device *pmem = dev_get_drvdata(ndns->claim); 209 210 if (unlikely(offset + size > pmem->size)) { 211 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n"); 212 return -EFAULT; 213 } 214 215 if (rw == READ) 216 memcpy_from_pmem(buf, pmem->virt_addr + offset, size); 217 else { 218 memcpy_to_pmem(pmem->virt_addr + offset, buf, size); 219 wmb_pmem(); 220 } 221 222 return 0; 223 } 224 225 static int nd_pfn_init(struct nd_pfn *nd_pfn) 226 { 227 struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL); 228 struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev); 229 struct nd_namespace_common *ndns = nd_pfn->ndns; 230 struct nd_region *nd_region; 231 unsigned long npfns; 232 phys_addr_t offset; 233 u64 checksum; 234 int rc; 235 236 if (!pfn_sb) 237 return -ENOMEM; 238 239 nd_pfn->pfn_sb = pfn_sb; 240 rc = nd_pfn_validate(nd_pfn); 241 if (rc == 0 || rc == -EBUSY) 242 return rc; 243 244 /* section alignment for simple hotplug */ 245 if (nvdimm_namespace_capacity(ndns) < ND_PFN_ALIGN 246 || pmem->phys_addr & ND_PFN_MASK) 247 return -ENODEV; 248 249 nd_region = to_nd_region(nd_pfn->dev.parent); 250 if (nd_region->ro) { 251 dev_info(&nd_pfn->dev, 252 "%s is read-only, unable to init metadata\n", 253 dev_name(&nd_region->dev)); 254 goto err; 255 } 256 257 memset(pfn_sb, 0, sizeof(*pfn_sb)); 258 npfns = (pmem->size - SZ_8K) / SZ_4K; 259 /* 260 * Note, we use 64 here for the standard size of struct page, 261 * debugging options may cause it to be larger in which case the 262 * implementation will limit the pfns advertised through 263 * ->direct_access() to those that are included in the memmap. 264 */ 265 if (nd_pfn->mode == PFN_MODE_PMEM) 266 offset = ALIGN(SZ_8K + 64 * npfns, PMD_SIZE); 267 else if (nd_pfn->mode == PFN_MODE_RAM) 268 offset = SZ_8K; 269 else 270 goto err; 271 272 npfns = (pmem->size - offset) / SZ_4K; 273 pfn_sb->mode = cpu_to_le32(nd_pfn->mode); 274 pfn_sb->dataoff = cpu_to_le64(offset); 275 pfn_sb->npfns = cpu_to_le64(npfns); 276 memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN); 277 memcpy(pfn_sb->uuid, nd_pfn->uuid, 16); 278 pfn_sb->version_major = cpu_to_le16(1); 279 checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb); 280 pfn_sb->checksum = cpu_to_le64(checksum); 281 282 rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb)); 283 if (rc) 284 goto err; 285 286 return 0; 287 err: 288 nd_pfn->pfn_sb = NULL; 289 kfree(pfn_sb); 290 return -ENXIO; 291 } 292 293 static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns) 294 { 295 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); 296 struct pmem_device *pmem; 297 298 /* free pmem disk */ 299 pmem = dev_get_drvdata(&nd_pfn->dev); 300 pmem_detach_disk(pmem); 301 302 /* release nd_pfn resources */ 303 kfree(nd_pfn->pfn_sb); 304 nd_pfn->pfn_sb = NULL; 305 306 return 0; 307 } 308 309 static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns) 310 { 311 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); 312 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); 313 struct device *dev = &nd_pfn->dev; 314 struct vmem_altmap *altmap; 315 struct nd_region *nd_region; 316 struct nd_pfn_sb *pfn_sb; 317 struct pmem_device *pmem; 318 phys_addr_t offset; 319 int rc; 320 321 if (!nd_pfn->uuid || !nd_pfn->ndns) 322 return -ENODEV; 323 324 nd_region = to_nd_region(dev->parent); 325 rc = nd_pfn_init(nd_pfn); 326 if (rc) 327 return rc; 328 329 if (PAGE_SIZE != SZ_4K) { 330 dev_err(dev, "only supported on systems with 4K PAGE_SIZE\n"); 331 return -ENXIO; 332 } 333 if (nsio->res.start & ND_PFN_MASK) { 334 dev_err(dev, "%s not memory hotplug section aligned\n", 335 dev_name(&ndns->dev)); 336 return -ENXIO; 337 } 338 339 pfn_sb = nd_pfn->pfn_sb; 340 offset = le64_to_cpu(pfn_sb->dataoff); 341 nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode); 342 if (nd_pfn->mode == PFN_MODE_RAM) { 343 if (offset != SZ_8K) 344 return -EINVAL; 345 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns); 346 altmap = NULL; 347 } else { 348 rc = -ENXIO; 349 goto err; 350 } 351 352 /* establish pfn range for lookup, and switch to direct map */ 353 pmem = dev_get_drvdata(dev); 354 devm_memunmap(dev, (void __force *) pmem->virt_addr); 355 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &nsio->res); 356 if (IS_ERR(pmem->virt_addr)) { 357 rc = PTR_ERR(pmem->virt_addr); 358 goto err; 359 } 360 361 /* attach pmem disk in "pfn-mode" */ 362 pmem->data_offset = offset; 363 rc = pmem_attach_disk(dev, ndns, pmem); 364 if (rc) 365 goto err; 366 367 return rc; 368 err: 369 nvdimm_namespace_detach_pfn(ndns); 370 return rc; 371 } 372 373 static int nd_pmem_probe(struct device *dev) 374 { 375 struct nd_region *nd_region = to_nd_region(dev->parent); 376 struct nd_namespace_common *ndns; 377 struct nd_namespace_io *nsio; 378 struct pmem_device *pmem; 379 380 ndns = nvdimm_namespace_common_probe(dev); 381 if (IS_ERR(ndns)) 382 return PTR_ERR(ndns); 383 384 nsio = to_nd_namespace_io(&ndns->dev); 385 pmem = pmem_alloc(dev, &nsio->res, nd_region->id); 386 if (IS_ERR(pmem)) 387 return PTR_ERR(pmem); 388 389 pmem->ndns = ndns; 390 dev_set_drvdata(dev, pmem); 391 ndns->rw_bytes = pmem_rw_bytes; 392 393 if (is_nd_btt(dev)) 394 return nvdimm_namespace_attach_btt(ndns); 395 396 if (is_nd_pfn(dev)) 397 return nvdimm_namespace_attach_pfn(ndns); 398 399 if (nd_btt_probe(ndns, pmem) == 0) { 400 /* we'll come back as btt-pmem */ 401 return -ENXIO; 402 } 403 404 if (nd_pfn_probe(ndns, pmem) == 0) { 405 /* we'll come back as pfn-pmem */ 406 return -ENXIO; 407 } 408 409 return pmem_attach_disk(dev, ndns, pmem); 410 } 411 412 static int nd_pmem_remove(struct device *dev) 413 { 414 struct pmem_device *pmem = dev_get_drvdata(dev); 415 416 if (is_nd_btt(dev)) 417 nvdimm_namespace_detach_btt(pmem->ndns); 418 else if (is_nd_pfn(dev)) 419 nvdimm_namespace_detach_pfn(pmem->ndns); 420 else 421 pmem_detach_disk(pmem); 422 423 return 0; 424 } 425 426 MODULE_ALIAS("pmem"); 427 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); 428 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); 429 static struct nd_device_driver nd_pmem_driver = { 430 .probe = nd_pmem_probe, 431 .remove = nd_pmem_remove, 432 .drv = { 433 .name = "nd_pmem", 434 }, 435 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, 436 }; 437 438 static int __init pmem_init(void) 439 { 440 int error; 441 442 pmem_major = register_blkdev(0, "pmem"); 443 if (pmem_major < 0) 444 return pmem_major; 445 446 error = nd_driver_register(&nd_pmem_driver); 447 if (error) { 448 unregister_blkdev(pmem_major, "pmem"); 449 return error; 450 } 451 452 return 0; 453 } 454 module_init(pmem_init); 455 456 static void pmem_exit(void) 457 { 458 driver_unregister(&nd_pmem_driver.drv); 459 unregister_blkdev(pmem_major, "pmem"); 460 } 461 module_exit(pmem_exit); 462 463 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); 464 MODULE_LICENSE("GPL v2"); 465