drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/slab.h>
  22 #include <linux/types.h>
  23 #include <linux/pr.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/nvme_ioctl.h>
  26 #include <linux/t10-pi.h>
  27 #include <scsi/sg.h>
  28 #include <asm/unaligned.h>
  29
  30 #include "nvme.h"
  31
  32 DEFINE_SPINLOCK(dev_list_lock);
  33
  34 static void nvme_free_ns(struct kref *kref)
  35 {
  36         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
  37
  38         if (ns->type == NVME_NS_LIGHTNVM)
  39                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
  40
  41         spin_lock(&dev_list_lock);
  42         ns->disk->private_data = NULL;
  43         spin_unlock(&dev_list_lock);
  44
  45         nvme_put_ctrl(ns->ctrl);
  46         put_disk(ns->disk);
  47         kfree(ns);
  48 }
  49
  50 void nvme_put_ns(struct nvme_ns *ns)
  51 {
  52         kref_put(&ns->kref, nvme_free_ns);
  53 }
  54
  55 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  56 {
  57         struct nvme_ns *ns;
  58
  59         spin_lock(&dev_list_lock);
  60         ns = disk->private_data;
  61         if (ns && !kref_get_unless_zero(&ns->kref))
  62                 ns = NULL;
  63         spin_unlock(&dev_list_lock);
  64
  65         return ns;
  66 }
  67
  68 struct request *nvme_alloc_request(struct request_queue *q,
  69                 struct nvme_command *cmd, unsigned int flags)
  70 {
  71         bool write = cmd->common.opcode & 1;
  72         struct request *req;
  73
  74         req = blk_mq_alloc_request(q, write, flags);
  75         if (IS_ERR(req))
  76                 return req;
  77
  78         req->cmd_type = REQ_TYPE_DRV_PRIV;
  79         req->cmd_flags |= REQ_FAILFAST_DRIVER;
  80         req->__data_len = 0;
  81         req->__sector = (sector_t) -1;
  82         req->bio = req->biotail = NULL;
  83
  84         req->cmd = (unsigned char *)cmd;
  85         req->cmd_len = sizeof(struct nvme_command);
  86         req->special = (void *)0;
  87
  88         return req;
  89 }
  90
  91 /*
  92  * Returns 0 on success.  If the result is negative, it's a Linux error code;
  93  * if the result is positive, it's an NVM Express status code
  94  */
  95 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
  96                 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
  97 {
  98         struct request *req;
  99         int ret;
 100
 101         req = nvme_alloc_request(q, cmd, 0);
 102         if (IS_ERR(req))
 103                 return PTR_ERR(req);
 104
 105         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 106
 107         if (buffer && bufflen) {
 108                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 109                 if (ret)
 110                         goto out;
 111         }
 112
 113         blk_execute_rq(req->q, NULL, req, 0);
 114         if (result)
 115                 *result = (u32)(uintptr_t)req->special;
 116         ret = req->errors;
 117  out:
 118         blk_mq_free_request(req);
 119         return ret;
 120 }
 121
 122 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 123                 void *buffer, unsigned bufflen)
 124 {
 125         return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
 126 }
 127
 128 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 129                 void __user *ubuffer, unsigned bufflen,
 130                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 131                 u32 *result, unsigned timeout)
 132 {
 133         bool write = cmd->common.opcode & 1;
 134         struct nvme_ns *ns = q->queuedata;
 135         struct gendisk *disk = ns ? ns->disk : NULL;
 136         struct request *req;
 137         struct bio *bio = NULL;
 138         void *meta = NULL;
 139         int ret;
 140
 141         req = nvme_alloc_request(q, cmd, 0);
 142         if (IS_ERR(req))
 143                 return PTR_ERR(req);
 144
 145         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 146
 147         if (ubuffer && bufflen) {
 148                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 149                                 GFP_KERNEL);
 150                 if (ret)
 151                         goto out;
 152                 bio = req->bio;
 153
 154                 if (!disk)
 155                         goto submit;
 156                 bio->bi_bdev = bdget_disk(disk, 0);
 157                 if (!bio->bi_bdev) {
 158                         ret = -ENODEV;
 159                         goto out_unmap;
 160                 }
 161
 162                 if (meta_buffer) {
 163                         struct bio_integrity_payload *bip;
 164
 165                         meta = kmalloc(meta_len, GFP_KERNEL);
 166                         if (!meta) {
 167                                 ret = -ENOMEM;
 168                                 goto out_unmap;
 169                         }
 170
 171                         if (write) {
 172                                 if (copy_from_user(meta, meta_buffer,
 173                                                 meta_len)) {
 174                                         ret = -EFAULT;
 175                                         goto out_free_meta;
 176                                 }
 177                         }
 178
 179                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 180                         if (!bip) {
 181                                 ret = -ENOMEM;
 182                                 goto out_free_meta;
 183                         }
 184
 185                         bip->bip_iter.bi_size = meta_len;
 186                         bip->bip_iter.bi_sector = meta_seed;
 187
 188                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 189                                         meta_len, offset_in_page(meta));
 190                         if (ret != meta_len) {
 191                                 ret = -ENOMEM;
 192                                 goto out_free_meta;
 193                         }
 194                 }
 195         }
 196  submit:
 197         blk_execute_rq(req->q, disk, req, 0);
 198         ret = req->errors;
 199         if (result)
 200                 *result = (u32)(uintptr_t)req->special;
 201         if (meta && !ret && !write) {
 202                 if (copy_to_user(meta_buffer, meta, meta_len))
 203                         ret = -EFAULT;
 204         }
 205  out_free_meta:
 206         kfree(meta);
 207  out_unmap:
 208         if (bio) {
 209                 if (disk && bio->bi_bdev)
 210                         bdput(bio->bi_bdev);
 211                 blk_rq_unmap_user(bio);
 212         }
 213  out:
 214         blk_mq_free_request(req);
 215         return ret;
 216 }
 217
 218 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 219                 void __user *ubuffer, unsigned bufflen, u32 *result,
 220                 unsigned timeout)
 221 {
 222         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 223                         result, timeout);
 224 }
 225
 226 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 227 {
 228         struct nvme_command c = { };
 229         int error;
 230
 231         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 232         c.identify.opcode = nvme_admin_identify;
 233         c.identify.cns = cpu_to_le32(1);
 234
 235         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 236         if (!*id)
 237                 return -ENOMEM;
 238
 239         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 240                         sizeof(struct nvme_id_ctrl));
 241         if (error)
 242                 kfree(*id);
 243         return error;
 244 }
 245
 246 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 247                 struct nvme_id_ns **id)
 248 {
 249         struct nvme_command c = { };
 250         int error;
 251
 252         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 253         c.identify.opcode = nvme_admin_identify,
 254         c.identify.nsid = cpu_to_le32(nsid),
 255
 256         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 257         if (!*id)
 258                 return -ENOMEM;
 259
 260         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 261                         sizeof(struct nvme_id_ns));
 262         if (error)
 263                 kfree(*id);
 264         return error;
 265 }
 266
 267 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 268                                         dma_addr_t dma_addr, u32 *result)
 269 {
 270         struct nvme_command c;
 271
 272         memset(&c, 0, sizeof(c));
 273         c.features.opcode = nvme_admin_get_features;
 274         c.features.nsid = cpu_to_le32(nsid);
 275         c.features.prp1 = cpu_to_le64(dma_addr);
 276         c.features.fid = cpu_to_le32(fid);
 277
 278         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 279 }
 280
 281 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 282                                         dma_addr_t dma_addr, u32 *result)
 283 {
 284         struct nvme_command c;
 285
 286         memset(&c, 0, sizeof(c));
 287         c.features.opcode = nvme_admin_set_features;
 288         c.features.prp1 = cpu_to_le64(dma_addr);
 289         c.features.fid = cpu_to_le32(fid);
 290         c.features.dword11 = cpu_to_le32(dword11);
 291
 292         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 293 }
 294
 295 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 296 {
 297         struct nvme_command c = { };
 298         int error;
 299
 300         c.common.opcode = nvme_admin_get_log_page,
 301         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 302         c.common.cdw10[0] = cpu_to_le32(
 303                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 304                          NVME_LOG_SMART),
 305
 306         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 307         if (!*log)
 308                 return -ENOMEM;
 309
 310         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 311                         sizeof(struct nvme_smart_log));
 312         if (error)
 313                 kfree(*log);
 314         return error;
 315 }
 316
 317 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 318 {
 319         struct nvme_user_io io;
 320         struct nvme_command c;
 321         unsigned length, meta_len;
 322         void __user *metadata;
 323
 324         if (copy_from_user(&io, uio, sizeof(io)))
 325                 return -EFAULT;
 326
 327         switch (io.opcode) {
 328         case nvme_cmd_write:
 329         case nvme_cmd_read:
 330         case nvme_cmd_compare:
 331                 break;
 332         default:
 333                 return -EINVAL;
 334         }
 335
 336         length = (io.nblocks + 1) << ns->lba_shift;
 337         meta_len = (io.nblocks + 1) * ns->ms;
 338         metadata = (void __user *)(uintptr_t)io.metadata;
 339
 340         if (ns->ext) {
 341                 length += meta_len;
 342                 meta_len = 0;
 343         } else if (meta_len) {
 344                 if ((io.metadata & 3) || !io.metadata)
 345                         return -EINVAL;
 346         }
 347
 348         memset(&c, 0, sizeof(c));
 349         c.rw.opcode = io.opcode;
 350         c.rw.flags = io.flags;
 351         c.rw.nsid = cpu_to_le32(ns->ns_id);
 352         c.rw.slba = cpu_to_le64(io.slba);
 353         c.rw.length = cpu_to_le16(io.nblocks);
 354         c.rw.control = cpu_to_le16(io.control);
 355         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 356         c.rw.reftag = cpu_to_le32(io.reftag);
 357         c.rw.apptag = cpu_to_le16(io.apptag);
 358         c.rw.appmask = cpu_to_le16(io.appmask);
 359
 360         return __nvme_submit_user_cmd(ns->queue, &c,
 361                         (void __user *)(uintptr_t)io.addr, length,
 362                         metadata, meta_len, io.slba, NULL, 0);
 363 }
 364
 365 int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 366                         struct nvme_passthru_cmd __user *ucmd)
 367 {
 368         struct nvme_passthru_cmd cmd;
 369         struct nvme_command c;
 370         unsigned timeout = 0;
 371         int status;
 372
 373         if (!capable(CAP_SYS_ADMIN))
 374                 return -EACCES;
 375         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 376                 return -EFAULT;
 377
 378         memset(&c, 0, sizeof(c));
 379         c.common.opcode = cmd.opcode;
 380         c.common.flags = cmd.flags;
 381         c.common.nsid = cpu_to_le32(cmd.nsid);
 382         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 383         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 384         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 385         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 386         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 387         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 388         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 389         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 390
 391         if (cmd.timeout_ms)
 392                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 393
 394         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 395                         (void __user *)cmd.addr, cmd.data_len,
 396                         &cmd.result, timeout);
 397         if (status >= 0) {
 398                 if (put_user(cmd.result, &ucmd->result))
 399                         return -EFAULT;
 400         }
 401
 402         return status;
 403 }
 404
 405 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 406                 unsigned int cmd, unsigned long arg)
 407 {
 408         struct nvme_ns *ns = bdev->bd_disk->private_data;
 409
 410         switch (cmd) {
 411         case NVME_IOCTL_ID:
 412                 force_successful_syscall_return();
 413                 return ns->ns_id;
 414         case NVME_IOCTL_ADMIN_CMD:
 415                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 416         case NVME_IOCTL_IO_CMD:
 417                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 418         case NVME_IOCTL_SUBMIT_IO:
 419                 return nvme_submit_io(ns, (void __user *)arg);
 420         case SG_GET_VERSION_NUM:
 421                 return nvme_sg_get_version_num((void __user *)arg);
 422         case SG_IO:
 423                 return nvme_sg_io(ns, (void __user *)arg);
 424         default:
 425                 return -ENOTTY;
 426         }
 427 }
 428
 429 #ifdef CONFIG_COMPAT
 430 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 431                         unsigned int cmd, unsigned long arg)
 432 {
 433         switch (cmd) {
 434         case SG_IO:
 435                 return -ENOIOCTLCMD;
 436         }
 437         return nvme_ioctl(bdev, mode, cmd, arg);
 438 }
 439 #else
 440 #define nvme_compat_ioctl       NULL
 441 #endif
 442
 443 static int nvme_open(struct block_device *bdev, fmode_t mode)
 444 {
 445         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 446 }
 447
 448 static void nvme_release(struct gendisk *disk, fmode_t mode)
 449 {
 450         nvme_put_ns(disk->private_data);
 451 }
 452
 453 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 454 {
 455         /* some standard values */
 456         geo->heads = 1 << 6;
 457         geo->sectors = 1 << 5;
 458         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 459         return 0;
 460 }
 461
 462 #ifdef CONFIG_BLK_DEV_INTEGRITY
 463 static void nvme_init_integrity(struct nvme_ns *ns)
 464 {
 465         struct blk_integrity integrity;
 466
 467         switch (ns->pi_type) {
 468         case NVME_NS_DPS_PI_TYPE3:
 469                 integrity.profile = &t10_pi_type3_crc;
 470                 break;
 471         case NVME_NS_DPS_PI_TYPE1:
 472         case NVME_NS_DPS_PI_TYPE2:
 473                 integrity.profile = &t10_pi_type1_crc;
 474                 break;
 475         default:
 476                 integrity.profile = NULL;
 477                 break;
 478         }
 479         integrity.tuple_size = ns->ms;
 480         blk_integrity_register(ns->disk, &integrity);
 481         blk_queue_max_integrity_segments(ns->queue, 1);
 482 }
 483 #else
 484 static void nvme_init_integrity(struct nvme_ns *ns)
 485 {
 486 }
 487 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 488
 489 static void nvme_config_discard(struct nvme_ns *ns)
 490 {
 491         u32 logical_block_size = queue_logical_block_size(ns->queue);
 492         ns->queue->limits.discard_zeroes_data = 0;
 493         ns->queue->limits.discard_alignment = logical_block_size;
 494         ns->queue->limits.discard_granularity = logical_block_size;
 495         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 496         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 497 }
 498
 499 int nvme_revalidate_disk(struct gendisk *disk)
 500 {
 501         struct nvme_ns *ns = disk->private_data;
 502         struct nvme_id_ns *id;
 503         u8 lbaf, pi_type;
 504         u16 old_ms;
 505         unsigned short bs;
 506
 507         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 508                 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
 509                                 __func__, ns->ctrl->instance, ns->ns_id);
 510                 return -ENODEV;
 511         }
 512         if (id->ncap == 0) {
 513                 kfree(id);
 514                 return -ENODEV;
 515         }
 516
 517         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 518                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 519                         dev_warn(ns->ctrl->dev,
 520                                 "%s: LightNVM init failure\n", __func__);
 521                         kfree(id);
 522                         return -ENODEV;
 523                 }
 524                 ns->type = NVME_NS_LIGHTNVM;
 525         }
 526
 527         old_ms = ns->ms;
 528         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 529         ns->lba_shift = id->lbaf[lbaf].ds;
 530         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 531         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 532
 533         /*
 534          * If identify namespace failed, use default 512 byte block size so
 535          * block layer can use before failing read/write for 0 capacity.
 536          */
 537         if (ns->lba_shift == 0)
 538                 ns->lba_shift = 9;
 539         bs = 1 << ns->lba_shift;
 540
 541         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 542         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 543                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 544
 545         blk_mq_freeze_queue(disk->queue);
 546         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 547                                 ns->ms != old_ms ||
 548                                 bs != queue_logical_block_size(disk->queue) ||
 549                                 (ns->ms && ns->ext)))
 550                 blk_integrity_unregister(disk);
 551
 552         ns->pi_type = pi_type;
 553         blk_queue_logical_block_size(ns->queue, bs);
 554
 555         if (ns->ms && !ns->ext)
 556                 nvme_init_integrity(ns);
 557
 558         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 559                 set_capacity(disk, 0);
 560         else
 561                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 562
 563         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 564                 nvme_config_discard(ns);
 565         blk_mq_unfreeze_queue(disk->queue);
 566
 567         kfree(id);
 568         return 0;
 569 }
 570
 571 static char nvme_pr_type(enum pr_type type)
 572 {
 573         switch (type) {
 574         case PR_WRITE_EXCLUSIVE:
 575                 return 1;
 576         case PR_EXCLUSIVE_ACCESS:
 577                 return 2;
 578         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 579                 return 3;
 580         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 581                 return 4;
 582         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 583                 return 5;
 584         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 585                 return 6;
 586         default:
 587                 return 0;
 588         }
 589 };
 590
 591 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 592                                 u64 key, u64 sa_key, u8 op)
 593 {
 594         struct nvme_ns *ns = bdev->bd_disk->private_data;
 595         struct nvme_command c;
 596         u8 data[16] = { 0, };
 597
 598         put_unaligned_le64(key, &data[0]);
 599         put_unaligned_le64(sa_key, &data[8]);
 600
 601         memset(&c, 0, sizeof(c));
 602         c.common.opcode = op;
 603         c.common.nsid = cpu_to_le32(ns->ns_id);
 604         c.common.cdw10[0] = cpu_to_le32(cdw10);
 605
 606         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 607 }
 608
 609 static int nvme_pr_register(struct block_device *bdev, u64 old,
 610                 u64 new, unsigned flags)
 611 {
 612         u32 cdw10;
 613
 614         if (flags & ~PR_FL_IGNORE_KEY)
 615                 return -EOPNOTSUPP;
 616
 617         cdw10 = old ? 2 : 0;
 618         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 619         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 620         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 621 }
 622
 623 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 624                 enum pr_type type, unsigned flags)
 625 {
 626         u32 cdw10;
 627
 628         if (flags & ~PR_FL_IGNORE_KEY)
 629                 return -EOPNOTSUPP;
 630
 631         cdw10 = nvme_pr_type(type) << 8;
 632         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 633         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 634 }
 635
 636 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 637                 enum pr_type type, bool abort)
 638 {
 639         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 640         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 641 }
 642
 643 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 644 {
 645         u32 cdw10 = 1 | key ? 1 << 3 : 0;
 646         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 647 }
 648
 649 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 650 {
 651         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 652         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 653 }
 654
 655 static const struct pr_ops nvme_pr_ops = {
 656         .pr_register    = nvme_pr_register,
 657         .pr_reserve     = nvme_pr_reserve,
 658         .pr_release     = nvme_pr_release,
 659         .pr_preempt     = nvme_pr_preempt,
 660         .pr_clear       = nvme_pr_clear,
 661 };
 662
 663 const struct block_device_operations nvme_fops = {
 664         .owner          = THIS_MODULE,
 665         .ioctl          = nvme_ioctl,
 666         .compat_ioctl   = nvme_compat_ioctl,
 667         .open           = nvme_open,
 668         .release        = nvme_release,
 669         .getgeo         = nvme_getgeo,
 670         .revalidate_disk= nvme_revalidate_disk,
 671         .pr_ops         = &nvme_pr_ops,
 672 };
 673
 674 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 675 {
 676         unsigned long timeout =
 677                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 678         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 679         int ret;
 680
 681         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 682                 if ((csts & NVME_CSTS_RDY) == bit)
 683                         break;
 684
 685                 msleep(100);
 686                 if (fatal_signal_pending(current))
 687                         return -EINTR;
 688                 if (time_after(jiffies, timeout)) {
 689                         dev_err(ctrl->dev,
 690                                 "Device not ready; aborting %s\n", enabled ?
 691                                                 "initialisation" : "reset");
 692                         return -ENODEV;
 693                 }
 694         }
 695
 696         return ret;
 697 }
 698
 699 /*
 700  * If the device has been passed off to us in an enabled state, just clear
 701  * the enabled bit.  The spec says we should set the 'shutdown notification
 702  * bits', but doing so may cause the device to complete commands to the
 703  * admin queue ... and we don't know what memory that might be pointing at!
 704  */
 705 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 706 {
 707         int ret;
 708
 709         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 710         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 711
 712         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 713         if (ret)
 714                 return ret;
 715         return nvme_wait_ready(ctrl, cap, false);
 716 }
 717
 718 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 719 {
 720         /*
 721          * Default to a 4K page size, with the intention to update this
 722          * path in the future to accomodate architectures with differing
 723          * kernel and IO page sizes.
 724          */
 725         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 726         int ret;
 727
 728         if (page_shift < dev_page_min) {
 729                 dev_err(ctrl->dev,
 730                         "Minimum device page size %u too large for host (%u)\n",
 731                         1 << dev_page_min, 1 << page_shift);
 732                 return -ENODEV;
 733         }
 734
 735         ctrl->page_size = 1 << page_shift;
 736
 737         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 738         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 739         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 740         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 741         ctrl->ctrl_config |= NVME_CC_ENABLE;
 742
 743         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 744         if (ret)
 745                 return ret;
 746         return nvme_wait_ready(ctrl, cap, true);
 747 }
 748
 749 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 750 {
 751         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 752         u32 csts;
 753         int ret;
 754
 755         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 756         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 757
 758         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 759         if (ret)
 760                 return ret;
 761
 762         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 763                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 764                         break;
 765
 766                 msleep(100);
 767                 if (fatal_signal_pending(current))
 768                         return -EINTR;
 769                 if (time_after(jiffies, timeout)) {
 770                         dev_err(ctrl->dev,
 771                                 "Device shutdown incomplete; abort shutdown\n");
 772                         return -ENODEV;
 773                 }
 774         }
 775
 776         return ret;
 777 }
 778
 779 /*
 780  * Initialize the cached copies of the Identify data and various controller
 781  * register in our nvme_ctrl structure.  This should be called as soon as
 782  * the admin queue is fully up and running.
 783  */
 784 int nvme_init_identify(struct nvme_ctrl *ctrl)
 785 {
 786         struct nvme_id_ctrl *id;
 787         u64 cap;
 788         int ret, page_shift;
 789
 790         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 791         if (ret) {
 792                 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
 793                 return ret;
 794         }
 795         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 796
 797         ret = nvme_identify_ctrl(ctrl, &id);
 798         if (ret) {
 799                 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
 800                 return -EIO;
 801         }
 802
 803         ctrl->oncs = le16_to_cpup(&id->oncs);
 804         ctrl->abort_limit = id->acl + 1;
 805         ctrl->vwc = id->vwc;
 806         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 807         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 808         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 809         if (id->mdts)
 810                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 811         else
 812                 ctrl->max_hw_sectors = UINT_MAX;
 813
 814         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 815                 unsigned int max_hw_sectors;
 816
 817                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 818                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 819                 if (ctrl->max_hw_sectors) {
 820                         ctrl->max_hw_sectors = min(max_hw_sectors,
 821                                                         ctrl->max_hw_sectors);
 822                 } else {
 823                         ctrl->max_hw_sectors = max_hw_sectors;
 824                 }
 825         }
 826
 827         kfree(id);
 828         return 0;
 829 }
 830
 831 static void nvme_free_ctrl(struct kref *kref)
 832 {
 833         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
 834
 835         ctrl->ops->free_ctrl(ctrl);
 836 }
 837
 838 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
 839 {
 840         kref_put(&ctrl->kref, nvme_free_ctrl);
 841 }
 842