drivers/s390/block/scm_blk.c

   1 /*
   2  * Block driver for s390 storage class memory.
   3  *
   4  * Copyright IBM Corp. 2012
   5  * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
   6  */
   7
   8 #define KMSG_COMPONENT "scm_block"
   9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  10
  11 #include <linux/interrupt.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/mempool.h>
  14 #include <linux/module.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/genhd.h>
  17 #include <linux/slab.h>
  18 #include <linux/list.h>
  19 #include <asm/eadm.h>
  20 #include "scm_blk.h"
  21
  22 debug_info_t *scm_debug;
  23 static int scm_major;
  24 static mempool_t *aidaw_pool;
  25 static DEFINE_SPINLOCK(list_lock);
  26 static LIST_HEAD(inactive_requests);
  27 static unsigned int nr_requests = 64;
  28 static atomic_t nr_devices = ATOMIC_INIT(0);
  29 module_param(nr_requests, uint, S_IRUGO);
  30 MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
  31
  32 MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
  33 MODULE_LICENSE("GPL");
  34 MODULE_ALIAS("scm:scmdev*");
  35
  36 static void __scm_free_rq(struct scm_request *scmrq)
  37 {
  38         struct aob_rq_header *aobrq = to_aobrq(scmrq);
  39
  40         free_page((unsigned long) scmrq->aob);
  41         __scm_free_rq_cluster(scmrq);
  42         kfree(aobrq);
  43 }
  44
  45 static void scm_free_rqs(void)
  46 {
  47         struct list_head *iter, *safe;
  48         struct scm_request *scmrq;
  49
  50         spin_lock_irq(&list_lock);
  51         list_for_each_safe(iter, safe, &inactive_requests) {
  52                 scmrq = list_entry(iter, struct scm_request, list);
  53                 list_del(&scmrq->list);
  54                 __scm_free_rq(scmrq);
  55         }
  56         spin_unlock_irq(&list_lock);
  57
  58         mempool_destroy(aidaw_pool);
  59 }
  60
  61 static int __scm_alloc_rq(void)
  62 {
  63         struct aob_rq_header *aobrq;
  64         struct scm_request *scmrq;
  65
  66         aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
  67         if (!aobrq)
  68                 return -ENOMEM;
  69
  70         scmrq = (void *) aobrq->data;
  71         scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
  72         if (!scmrq->aob) {
  73                 __scm_free_rq(scmrq);
  74                 return -ENOMEM;
  75         }
  76
  77         if (__scm_alloc_rq_cluster(scmrq)) {
  78                 __scm_free_rq(scmrq);
  79                 return -ENOMEM;
  80         }
  81
  82         INIT_LIST_HEAD(&scmrq->list);
  83         spin_lock_irq(&list_lock);
  84         list_add(&scmrq->list, &inactive_requests);
  85         spin_unlock_irq(&list_lock);
  86
  87         return 0;
  88 }
  89
  90 static int scm_alloc_rqs(unsigned int nrqs)
  91 {
  92         int ret = 0;
  93
  94         aidaw_pool = mempool_create_page_pool(max(nrqs/8, 1U), 0);
  95         if (!aidaw_pool)
  96                 return -ENOMEM;
  97
  98         while (nrqs-- && !ret)
  99                 ret = __scm_alloc_rq();
 100
 101         return ret;
 102 }
 103
 104 static struct scm_request *scm_request_fetch(void)
 105 {
 106         struct scm_request *scmrq = NULL;
 107
 108         spin_lock(&list_lock);
 109         if (list_empty(&inactive_requests))
 110                 goto out;
 111         scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
 112         list_del(&scmrq->list);
 113 out:
 114         spin_unlock(&list_lock);
 115         return scmrq;
 116 }
 117
 118 static void scm_request_done(struct scm_request *scmrq)
 119 {
 120         struct msb *msb = &scmrq->aob->msb[0];
 121         u64 aidaw = msb->data_addr;
 122         unsigned long flags;
 123
 124         if ((msb->flags & MSB_FLAG_IDA) && aidaw &&
 125             IS_ALIGNED(aidaw, PAGE_SIZE))
 126                 mempool_free(virt_to_page(aidaw), aidaw_pool);
 127
 128         spin_lock_irqsave(&list_lock, flags);
 129         list_add(&scmrq->list, &inactive_requests);
 130         spin_unlock_irqrestore(&list_lock, flags);
 131 }
 132
 133 static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
 134 {
 135         return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
 136 }
 137
 138 static inline struct aidaw *scm_aidaw_alloc(void)
 139 {
 140         struct page *page = mempool_alloc(aidaw_pool, GFP_ATOMIC);
 141
 142         return page ? page_address(page) : NULL;
 143 }
 144
 145 static inline unsigned long scm_aidaw_bytes(struct aidaw *aidaw)
 146 {
 147         unsigned long _aidaw = (unsigned long) aidaw;
 148         unsigned long bytes = ALIGN(_aidaw, PAGE_SIZE) - _aidaw;
 149
 150         return (bytes / sizeof(*aidaw)) * PAGE_SIZE;
 151 }
 152
 153 struct aidaw *scm_aidaw_fetch(struct scm_request *scmrq, unsigned int bytes)
 154 {
 155         struct aidaw *aidaw;
 156
 157         if (scm_aidaw_bytes(scmrq->next_aidaw) >= bytes)
 158                 return scmrq->next_aidaw;
 159
 160         aidaw = scm_aidaw_alloc();
 161         if (aidaw)
 162                 memset(aidaw, 0, PAGE_SIZE);
 163         return aidaw;
 164 }
 165
 166 static int scm_request_prepare(struct scm_request *scmrq)
 167 {
 168         struct scm_blk_dev *bdev = scmrq->bdev;
 169         struct scm_device *scmdev = bdev->gendisk->private_data;
 170         struct msb *msb = &scmrq->aob->msb[0];
 171         struct req_iterator iter;
 172         struct aidaw *aidaw;
 173         struct bio_vec bv;
 174
 175         aidaw = scm_aidaw_fetch(scmrq, blk_rq_bytes(scmrq->request));
 176         if (!aidaw)
 177                 return -ENOMEM;
 178
 179         msb->bs = MSB_BS_4K;
 180         scmrq->aob->request.msb_count = 1;
 181         msb->scm_addr = scmdev->address +
 182                 ((u64) blk_rq_pos(scmrq->request) << 9);
 183         msb->oc = (rq_data_dir(scmrq->request) == READ) ?
 184                 MSB_OC_READ : MSB_OC_WRITE;
 185         msb->flags |= MSB_FLAG_IDA;
 186         msb->data_addr = (u64) aidaw;
 187
 188         rq_for_each_segment(bv, scmrq->request, iter) {
 189                 WARN_ON(bv.bv_offset);
 190                 msb->blk_count += bv.bv_len >> 12;
 191                 aidaw->data_addr = (u64) page_address(bv.bv_page);
 192                 aidaw++;
 193         }
 194
 195         return 0;
 196 }
 197
 198 static inline void scm_request_init(struct scm_blk_dev *bdev,
 199                                     struct scm_request *scmrq,
 200                                     struct request *req)
 201 {
 202         struct aob_rq_header *aobrq = to_aobrq(scmrq);
 203         struct aob *aob = scmrq->aob;
 204
 205         memset(aob, 0, sizeof(*aob));
 206         aobrq->scmdev = bdev->scmdev;
 207         aob->request.cmd_code = ARQB_CMD_MOVE;
 208         aob->request.data = (u64) aobrq;
 209         scmrq->request = req;
 210         scmrq->bdev = bdev;
 211         scmrq->retries = 4;
 212         scmrq->error = 0;
 213         /* We don't use all msbs - place aidaws at the end of the aob page. */
 214         scmrq->next_aidaw = (void *) &aob->msb[1];
 215         scm_request_cluster_init(scmrq);
 216 }
 217
 218 static void scm_ensure_queue_restart(struct scm_blk_dev *bdev)
 219 {
 220         if (atomic_read(&bdev->queued_reqs)) {
 221                 /* Queue restart is triggered by the next interrupt. */
 222                 return;
 223         }
 224         blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY);
 225 }
 226
 227 void scm_request_requeue(struct scm_request *scmrq)
 228 {
 229         struct scm_blk_dev *bdev = scmrq->bdev;
 230
 231         scm_release_cluster(scmrq);
 232         blk_requeue_request(bdev->rq, scmrq->request);
 233         atomic_dec(&bdev->queued_reqs);
 234         scm_request_done(scmrq);
 235         scm_ensure_queue_restart(bdev);
 236 }
 237
 238 void scm_request_finish(struct scm_request *scmrq)
 239 {
 240         struct scm_blk_dev *bdev = scmrq->bdev;
 241
 242         scm_release_cluster(scmrq);
 243         blk_end_request_all(scmrq->request, scmrq->error);
 244         atomic_dec(&bdev->queued_reqs);
 245         scm_request_done(scmrq);
 246 }
 247
 248 static void scm_blk_request(struct request_queue *rq)
 249 {
 250         struct scm_device *scmdev = rq->queuedata;
 251         struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
 252         struct scm_request *scmrq;
 253         struct request *req;
 254         int ret;
 255
 256         while ((req = blk_peek_request(rq))) {
 257                 if (req->cmd_type != REQ_TYPE_FS) {
 258                         blk_start_request(req);
 259                         blk_dump_rq_flags(req, KMSG_COMPONENT " bad request");
 260                         blk_end_request_all(req, -EIO);
 261                         continue;
 262                 }
 263
 264                 if (!scm_permit_request(bdev, req)) {
 265                         scm_ensure_queue_restart(bdev);
 266                         return;
 267                 }
 268                 scmrq = scm_request_fetch();
 269                 if (!scmrq) {
 270                         SCM_LOG(5, "no request");
 271                         scm_ensure_queue_restart(bdev);
 272                         return;
 273                 }
 274                 scm_request_init(bdev, scmrq, req);
 275                 if (!scm_reserve_cluster(scmrq)) {
 276                         SCM_LOG(5, "cluster busy");
 277                         scm_request_done(scmrq);
 278                         return;
 279                 }
 280                 if (scm_need_cluster_request(scmrq)) {
 281                         atomic_inc(&bdev->queued_reqs);
 282                         blk_start_request(req);
 283                         scm_initiate_cluster_request(scmrq);
 284                         return;
 285                 }
 286
 287                 if (scm_request_prepare(scmrq)) {
 288                         SCM_LOG(5, "no aidaw");
 289                         scm_release_cluster(scmrq);
 290                         scm_request_done(scmrq);
 291                         scm_ensure_queue_restart(bdev);
 292                         return;
 293                 }
 294
 295                 atomic_inc(&bdev->queued_reqs);
 296                 blk_start_request(req);
 297
 298                 ret = eadm_start_aob(scmrq->aob);
 299                 if (ret) {
 300                         SCM_LOG(5, "no subchannel");
 301                         scm_request_requeue(scmrq);
 302                         return;
 303                 }
 304         }
 305 }
 306
 307 static void __scmrq_log_error(struct scm_request *scmrq)
 308 {
 309         struct aob *aob = scmrq->aob;
 310
 311         if (scmrq->error == -ETIMEDOUT)
 312                 SCM_LOG(1, "Request timeout");
 313         else {
 314                 SCM_LOG(1, "Request error");
 315                 SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
 316         }
 317         if (scmrq->retries)
 318                 SCM_LOG(1, "Retry request");
 319         else
 320                 pr_err("An I/O operation to SCM failed with rc=%d\n",
 321                        scmrq->error);
 322 }
 323
 324 void scm_blk_irq(struct scm_device *scmdev, void *data, int error)
 325 {
 326         struct scm_request *scmrq = data;
 327         struct scm_blk_dev *bdev = scmrq->bdev;
 328
 329         scmrq->error = error;
 330         if (error)
 331                 __scmrq_log_error(scmrq);
 332
 333         spin_lock(&bdev->lock);
 334         list_add_tail(&scmrq->list, &bdev->finished_requests);
 335         spin_unlock(&bdev->lock);
 336         tasklet_hi_schedule(&bdev->tasklet);
 337 }
 338
 339 static void scm_blk_handle_error(struct scm_request *scmrq)
 340 {
 341         struct scm_blk_dev *bdev = scmrq->bdev;
 342         unsigned long flags;
 343
 344         if (scmrq->error != -EIO)
 345                 goto restart;
 346
 347         /* For -EIO the response block is valid. */
 348         switch (scmrq->aob->response.eqc) {
 349         case EQC_WR_PROHIBIT:
 350                 spin_lock_irqsave(&bdev->lock, flags);
 351                 if (bdev->state != SCM_WR_PROHIBIT)
 352                         pr_info("%lx: Write access to the SCM increment is suspended\n",
 353                                 (unsigned long) bdev->scmdev->address);
 354                 bdev->state = SCM_WR_PROHIBIT;
 355                 spin_unlock_irqrestore(&bdev->lock, flags);
 356                 goto requeue;
 357         default:
 358                 break;
 359         }
 360
 361 restart:
 362         if (!eadm_start_aob(scmrq->aob))
 363                 return;
 364
 365 requeue:
 366         spin_lock_irqsave(&bdev->rq_lock, flags);
 367         scm_request_requeue(scmrq);
 368         spin_unlock_irqrestore(&bdev->rq_lock, flags);
 369 }
 370
 371 static void scm_blk_tasklet(struct scm_blk_dev *bdev)
 372 {
 373         struct scm_request *scmrq;
 374         unsigned long flags;
 375
 376         spin_lock_irqsave(&bdev->lock, flags);
 377         while (!list_empty(&bdev->finished_requests)) {
 378                 scmrq = list_first_entry(&bdev->finished_requests,
 379                                          struct scm_request, list);
 380                 list_del(&scmrq->list);
 381                 spin_unlock_irqrestore(&bdev->lock, flags);
 382
 383                 if (scmrq->error && scmrq->retries-- > 0) {
 384                         scm_blk_handle_error(scmrq);
 385
 386                         /* Request restarted or requeued, handle next. */
 387                         spin_lock_irqsave(&bdev->lock, flags);
 388                         continue;
 389                 }
 390
 391                 if (scm_test_cluster_request(scmrq)) {
 392                         scm_cluster_request_irq(scmrq);
 393                         spin_lock_irqsave(&bdev->lock, flags);
 394                         continue;
 395                 }
 396
 397                 scm_request_finish(scmrq);
 398                 spin_lock_irqsave(&bdev->lock, flags);
 399         }
 400         spin_unlock_irqrestore(&bdev->lock, flags);
 401         /* Look out for more requests. */
 402         blk_run_queue(bdev->rq);
 403 }
 404
 405 static const struct block_device_operations scm_blk_devops = {
 406         .owner = THIS_MODULE,
 407 };
 408
 409 int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
 410 {
 411         struct request_queue *rq;
 412         int len, ret = -ENOMEM;
 413         unsigned int devindex, nr_max_blk;
 414
 415         devindex = atomic_inc_return(&nr_devices) - 1;
 416         /* scma..scmz + scmaa..scmzz */
 417         if (devindex > 701) {
 418                 ret = -ENODEV;
 419                 goto out;
 420         }
 421
 422         bdev->scmdev = scmdev;
 423         bdev->state = SCM_OPER;
 424         spin_lock_init(&bdev->rq_lock);
 425         spin_lock_init(&bdev->lock);
 426         INIT_LIST_HEAD(&bdev->finished_requests);
 427         atomic_set(&bdev->queued_reqs, 0);
 428         tasklet_init(&bdev->tasklet,
 429                      (void (*)(unsigned long)) scm_blk_tasklet,
 430                      (unsigned long) bdev);
 431
 432         rq = blk_init_queue(scm_blk_request, &bdev->rq_lock);
 433         if (!rq)
 434                 goto out;
 435
 436         bdev->rq = rq;
 437         nr_max_blk = min(scmdev->nr_max_block,
 438                          (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
 439
 440         blk_queue_logical_block_size(rq, 1 << 12);
 441         blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
 442         blk_queue_max_segments(rq, nr_max_blk);
 443         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
 444         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, rq);
 445         scm_blk_dev_cluster_setup(bdev);
 446
 447         bdev->gendisk = alloc_disk(SCM_NR_PARTS);
 448         if (!bdev->gendisk)
 449                 goto out_queue;
 450
 451         rq->queuedata = scmdev;
 452         bdev->gendisk->driverfs_dev = &scmdev->dev;
 453         bdev->gendisk->private_data = scmdev;
 454         bdev->gendisk->fops = &scm_blk_devops;
 455         bdev->gendisk->queue = rq;
 456         bdev->gendisk->major = scm_major;
 457         bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
 458
 459         len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
 460         if (devindex > 25) {
 461                 len += snprintf(bdev->gendisk->disk_name + len,
 462                                 DISK_NAME_LEN - len, "%c",
 463                                 'a' + (devindex / 26) - 1);
 464                 devindex = devindex % 26;
 465         }
 466         snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
 467                  'a' + devindex);
 468
 469         /* 512 byte sectors */
 470         set_capacity(bdev->gendisk, scmdev->size >> 9);
 471         add_disk(bdev->gendisk);
 472         return 0;
 473
 474 out_queue:
 475         blk_cleanup_queue(rq);
 476 out:
 477         atomic_dec(&nr_devices);
 478         return ret;
 479 }
 480
 481 void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
 482 {
 483         tasklet_kill(&bdev->tasklet);
 484         del_gendisk(bdev->gendisk);
 485         blk_cleanup_queue(bdev->gendisk->queue);
 486         put_disk(bdev->gendisk);
 487 }
 488
 489 void scm_blk_set_available(struct scm_blk_dev *bdev)
 490 {
 491         unsigned long flags;
 492
 493         spin_lock_irqsave(&bdev->lock, flags);
 494         if (bdev->state == SCM_WR_PROHIBIT)
 495                 pr_info("%lx: Write access to the SCM increment is restored\n",
 496                         (unsigned long) bdev->scmdev->address);
 497         bdev->state = SCM_OPER;
 498         spin_unlock_irqrestore(&bdev->lock, flags);
 499 }
 500
 501 static int __init scm_blk_init(void)
 502 {
 503         int ret = -EINVAL;
 504
 505         if (!scm_cluster_size_valid())
 506                 goto out;
 507
 508         ret = register_blkdev(0, "scm");
 509         if (ret < 0)
 510                 goto out;
 511
 512         scm_major = ret;
 513         ret = scm_alloc_rqs(nr_requests);
 514         if (ret)
 515                 goto out_free;
 516
 517         scm_debug = debug_register("scm_log", 16, 1, 16);
 518         if (!scm_debug) {
 519                 ret = -ENOMEM;
 520                 goto out_free;
 521         }
 522
 523         debug_register_view(scm_debug, &debug_hex_ascii_view);
 524         debug_set_level(scm_debug, 2);
 525
 526         ret = scm_drv_init();
 527         if (ret)
 528                 goto out_dbf;
 529
 530         return ret;
 531
 532 out_dbf:
 533         debug_unregister(scm_debug);
 534 out_free:
 535         scm_free_rqs();
 536         unregister_blkdev(scm_major, "scm");
 537 out:
 538         return ret;
 539 }
 540 module_init(scm_blk_init);
 541
 542 static void __exit scm_blk_cleanup(void)
 543 {
 544         scm_drv_cleanup();
 545         debug_unregister(scm_debug);
 546         scm_free_rqs();
 547         unregister_blkdev(scm_major, "scm");
 548 }
 549 module_exit(scm_blk_cleanup);