arch/powerpc/platforms/pseries/papr_scm.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #define pr_fmt(fmt)     "papr-scm: " fmt
   4
   5 #include <linux/of.h>
   6 #include <linux/kernel.h>
   7 #include <linux/module.h>
   8 #include <linux/ioport.h>
   9 #include <linux/slab.h>
  10 #include <linux/ndctl.h>
  11 #include <linux/sched.h>
  12 #include <linux/libnvdimm.h>
  13 #include <linux/platform_device.h>
  14 #include <linux/delay.h>
  15 #include <linux/seq_buf.h>
  16 #include <linux/nd.h>
  17
  18 #include <asm/plpar_wrappers.h>
  19 #include <asm/papr_pdsm.h>
  20 #include <asm/mce.h>
  21 #include <asm/unaligned.h>
  22 #include <linux/perf_event.h>
  23
  24 #define BIND_ANY_ADDR (~0ul)
  25
  26 #define PAPR_SCM_DIMM_CMD_MASK \
  27         ((1ul << ND_CMD_GET_CONFIG_SIZE) | \
  28          (1ul << ND_CMD_GET_CONFIG_DATA) | \
  29          (1ul << ND_CMD_SET_CONFIG_DATA) | \
  30          (1ul << ND_CMD_CALL))
  31
  32 /* DIMM health bitmap bitmap indicators */
  33 /* SCM device is unable to persist memory contents */
  34 #define PAPR_PMEM_UNARMED                   (1ULL << (63 - 0))
  35 /* SCM device failed to persist memory contents */
  36 #define PAPR_PMEM_SHUTDOWN_DIRTY            (1ULL << (63 - 1))
  37 /* SCM device contents are persisted from previous IPL */
  38 #define PAPR_PMEM_SHUTDOWN_CLEAN            (1ULL << (63 - 2))
  39 /* SCM device contents are not persisted from previous IPL */
  40 #define PAPR_PMEM_EMPTY                     (1ULL << (63 - 3))
  41 /* SCM device memory life remaining is critically low */
  42 #define PAPR_PMEM_HEALTH_CRITICAL           (1ULL << (63 - 4))
  43 /* SCM device will be garded off next IPL due to failure */
  44 #define PAPR_PMEM_HEALTH_FATAL              (1ULL << (63 - 5))
  45 /* SCM contents cannot persist due to current platform health status */
  46 #define PAPR_PMEM_HEALTH_UNHEALTHY          (1ULL << (63 - 6))
  47 /* SCM device is unable to persist memory contents in certain conditions */
  48 #define PAPR_PMEM_HEALTH_NON_CRITICAL       (1ULL << (63 - 7))
  49 /* SCM device is encrypted */
  50 #define PAPR_PMEM_ENCRYPTED                 (1ULL << (63 - 8))
  51 /* SCM device has been scrubbed and locked */
  52 #define PAPR_PMEM_SCRUBBED_AND_LOCKED       (1ULL << (63 - 9))
  53
  54 /* Bits status indicators for health bitmap indicating unarmed dimm */
  55 #define PAPR_PMEM_UNARMED_MASK (PAPR_PMEM_UNARMED |             \
  56                                 PAPR_PMEM_HEALTH_UNHEALTHY)
  57
  58 /* Bits status indicators for health bitmap indicating unflushed dimm */
  59 #define PAPR_PMEM_BAD_SHUTDOWN_MASK (PAPR_PMEM_SHUTDOWN_DIRTY)
  60
  61 /* Bits status indicators for health bitmap indicating unrestored dimm */
  62 #define PAPR_PMEM_BAD_RESTORE_MASK  (PAPR_PMEM_EMPTY)
  63
  64 /* Bit status indicators for smart event notification */
  65 #define PAPR_PMEM_SMART_EVENT_MASK (PAPR_PMEM_HEALTH_CRITICAL | \
  66                                     PAPR_PMEM_HEALTH_FATAL |    \
  67                                     PAPR_PMEM_HEALTH_UNHEALTHY)
  68
  69 #define PAPR_SCM_PERF_STATS_EYECATCHER __stringify(SCMSTATS)
  70 #define PAPR_SCM_PERF_STATS_VERSION 0x1
  71
  72 /* Struct holding a single performance metric */
  73 struct papr_scm_perf_stat {
  74         u8 stat_id[8];
  75         __be64 stat_val;
  76 } __packed;
  77
  78 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */
  79 struct papr_scm_perf_stats {
  80         u8 eye_catcher[8];
  81         /* Should be PAPR_SCM_PERF_STATS_VERSION */
  82         __be32 stats_version;
  83         /* Number of stats following */
  84         __be32 num_statistics;
  85         /* zero or more performance matrics */
  86         struct papr_scm_perf_stat scm_statistic[];
  87 } __packed;
  88
  89 /* private struct associated with each region */
  90 struct papr_scm_priv {
  91         struct platform_device *pdev;
  92         struct device_node *dn;
  93         uint32_t drc_index;
  94         uint64_t blocks;
  95         uint64_t block_size;
  96         int metadata_size;
  97         bool is_volatile;
  98         bool hcall_flush_required;
  99
 100         uint64_t bound_addr;
 101
 102         struct nvdimm_bus_descriptor bus_desc;
 103         struct nvdimm_bus *bus;
 104         struct nvdimm *nvdimm;
 105         struct resource res;
 106         struct nd_region *region;
 107         struct nd_interleave_set nd_set;
 108         struct list_head region_list;
 109
 110         /* Protect dimm health data from concurrent read/writes */
 111         struct mutex health_mutex;
 112
 113         /* Last time the health information of the dimm was updated */
 114         unsigned long lasthealth_jiffies;
 115
 116         /* Health information for the dimm */
 117         u64 health_bitmap;
 118
 119         /* Holds the last known dirty shutdown counter value */
 120         u64 dirty_shutdown_counter;
 121
 122         /* length of the stat buffer as expected by phyp */
 123         size_t stat_buffer_len;
 124
 125         /* The bits which needs to be overridden */
 126         u64 health_bitmap_inject_mask;
 127
 128          /* array to have event_code and stat_id mappings */
 129         char **nvdimm_events_map;
 130 };
 131
 132 static int papr_scm_pmem_flush(struct nd_region *nd_region,
 133                                struct bio *bio __maybe_unused)
 134 {
 135         struct papr_scm_priv *p = nd_region_provider_data(nd_region);
 136         unsigned long ret_buf[PLPAR_HCALL_BUFSIZE], token = 0;
 137         long rc;
 138
 139         dev_dbg(&p->pdev->dev, "flush drc 0x%x", p->drc_index);
 140
 141         do {
 142                 rc = plpar_hcall(H_SCM_FLUSH, ret_buf, p->drc_index, token);
 143                 token = ret_buf[0];
 144
 145                 /* Check if we are stalled for some time */
 146                 if (H_IS_LONG_BUSY(rc)) {
 147                         msleep(get_longbusy_msecs(rc));
 148                         rc = H_BUSY;
 149                 } else if (rc == H_BUSY) {
 150                         cond_resched();
 151                 }
 152         } while (rc == H_BUSY);
 153
 154         if (rc) {
 155                 dev_err(&p->pdev->dev, "flush error: %ld", rc);
 156                 rc = -EIO;
 157         } else {
 158                 dev_dbg(&p->pdev->dev, "flush drc 0x%x complete", p->drc_index);
 159         }
 160
 161         return rc;
 162 }
 163
 164 static LIST_HEAD(papr_nd_regions);
 165 static DEFINE_MUTEX(papr_ndr_lock);
 166
 167 static int drc_pmem_bind(struct papr_scm_priv *p)
 168 {
 169         unsigned long ret[PLPAR_HCALL_BUFSIZE];
 170         uint64_t saved = 0;
 171         uint64_t token;
 172         int64_t rc;
 173
 174         /*
 175          * When the hypervisor cannot map all the requested memory in a single
 176          * hcall it returns H_BUSY and we call again with the token until
 177          * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS
 178          * leave the system in an undefined state, so we wait.
 179          */
 180         token = 0;
 181
 182         do {
 183                 rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0,
 184                                 p->blocks, BIND_ANY_ADDR, token);
 185                 token = ret[0];
 186                 if (!saved)
 187                         saved = ret[1];
 188                 cond_resched();
 189         } while (rc == H_BUSY);
 190
 191         if (rc)
 192                 return rc;
 193
 194         p->bound_addr = saved;
 195         dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n",
 196                 p->drc_index, (unsigned long)saved);
 197         return rc;
 198 }
 199
 200 static void drc_pmem_unbind(struct papr_scm_priv *p)
 201 {
 202         unsigned long ret[PLPAR_HCALL_BUFSIZE];
 203         uint64_t token = 0;
 204         int64_t rc;
 205
 206         dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index);
 207
 208         /* NB: unbind has the same retry requirements as drc_pmem_bind() */
 209         do {
 210
 211                 /* Unbind of all SCM resources associated with drcIndex */
 212                 rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC,
 213                                  p->drc_index, token);
 214                 token = ret[0];
 215
 216                 /* Check if we are stalled for some time */
 217                 if (H_IS_LONG_BUSY(rc)) {
 218                         msleep(get_longbusy_msecs(rc));
 219                         rc = H_BUSY;
 220                 } else if (rc == H_BUSY) {
 221                         cond_resched();
 222                 }
 223
 224         } while (rc == H_BUSY);
 225
 226         if (rc)
 227                 dev_err(&p->pdev->dev, "unbind error: %lld\n", rc);
 228         else
 229                 dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n",
 230                         p->drc_index);
 231
 232         return;
 233 }
 234
 235 static int drc_pmem_query_n_bind(struct papr_scm_priv *p)
 236 {
 237         unsigned long start_addr;
 238         unsigned long end_addr;
 239         unsigned long ret[PLPAR_HCALL_BUFSIZE];
 240         int64_t rc;
 241
 242
 243         rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
 244                          p->drc_index, 0);
 245         if (rc)
 246                 goto err_out;
 247         start_addr = ret[0];
 248
 249         /* Make sure the full region is bound. */
 250         rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret,
 251                          p->drc_index, p->blocks - 1);
 252         if (rc)
 253                 goto err_out;
 254         end_addr = ret[0];
 255
 256         if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size))
 257                 goto err_out;
 258
 259         p->bound_addr = start_addr;
 260         dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr);
 261         return rc;
 262
 263 err_out:
 264         dev_info(&p->pdev->dev,
 265                  "Failed to query, trying an unbind followed by bind");
 266         drc_pmem_unbind(p);
 267         return drc_pmem_bind(p);
 268 }
 269
 270 /*
 271  * Query the Dimm performance stats from PHYP and copy them (if returned) to
 272  * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast
 273  * (num_stats + header) bytes.
 274  * - If buff_stats == NULL the return value is the size in bytes of the buffer
 275  * needed to hold all supported performance-statistics.
 276  * - If buff_stats != NULL and num_stats == 0 then we copy all known
 277  * performance-statistics to 'buff_stat' and expect to be large enough to
 278  * hold them.
 279  * - if buff_stats != NULL and num_stats > 0 then copy the requested
 280  * performance-statistics to buff_stats.
 281  */
 282 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
 283                                     struct papr_scm_perf_stats *buff_stats,
 284                                     unsigned int num_stats)
 285 {
 286         unsigned long ret[PLPAR_HCALL_BUFSIZE];
 287         size_t size;
 288         s64 rc;
 289
 290         /* Setup the out buffer */
 291         if (buff_stats) {
 292                 memcpy(buff_stats->eye_catcher,
 293                        PAPR_SCM_PERF_STATS_EYECATCHER, 8);
 294                 buff_stats->stats_version =
 295                         cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION);
 296                 buff_stats->num_statistics =
 297                         cpu_to_be32(num_stats);
 298
 299                 /*
 300                  * Calculate the buffer size based on num-stats provided
 301                  * or use the prefetched max buffer length
 302                  */
 303                 if (num_stats)
 304                         /* Calculate size from the num_stats */
 305                         size = sizeof(struct papr_scm_perf_stats) +
 306                                 num_stats * sizeof(struct papr_scm_perf_stat);
 307                 else
 308                         size = p->stat_buffer_len;
 309         } else {
 310                 /* In case of no out buffer ignore the size */
 311                 size = 0;
 312         }
 313
 314         /* Do the HCALL asking PHYP for info */
 315         rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index,
 316                          buff_stats ? virt_to_phys(buff_stats) : 0,
 317                          size);
 318
 319         /* Check if the error was due to an unknown stat-id */
 320         if (rc == H_PARTIAL) {
 321                 dev_err(&p->pdev->dev,
 322                         "Unknown performance stats, Err:0x%016lX\n", ret[0]);
 323                 return -ENOENT;
 324         } else if (rc == H_AUTHORITY) {
 325                 dev_info(&p->pdev->dev,
 326                          "Permission denied while accessing performance stats");
 327                 return -EPERM;
 328         } else if (rc == H_UNSUPPORTED) {
 329                 dev_dbg(&p->pdev->dev, "Performance stats unsupported\n");
 330                 return -EOPNOTSUPP;
 331         } else if (rc != H_SUCCESS) {
 332                 dev_err(&p->pdev->dev,
 333                         "Failed to query performance stats, Err:%lld\n", rc);
 334                 return -EIO;
 335
 336         } else if (!size) {
 337                 /* Handle case where stat buffer size was requested */
 338                 dev_dbg(&p->pdev->dev,
 339                         "Performance stats size %ld\n", ret[0]);
 340                 return ret[0];
 341         }
 342
 343         /* Successfully fetched the requested stats from phyp */
 344         dev_dbg(&p->pdev->dev,
 345                 "Performance stats returned %d stats\n",
 346                 be32_to_cpu(buff_stats->num_statistics));
 347         return 0;
 348 }
 349
 350 #ifdef CONFIG_PERF_EVENTS
 351 #define to_nvdimm_pmu(_pmu)     container_of(_pmu, struct nvdimm_pmu, pmu)
 352
 353 static int papr_scm_pmu_get_value(struct perf_event *event, struct device *dev, u64 *count)
 354 {
 355         struct papr_scm_perf_stat *stat;
 356         struct papr_scm_perf_stats *stats;
 357         struct papr_scm_priv *p = (struct papr_scm_priv *)dev->driver_data;
 358         int rc, size;
 359
 360         /* Allocate request buffer enough to hold single performance stat */
 361         size = sizeof(struct papr_scm_perf_stats) +
 362                 sizeof(struct papr_scm_perf_stat);
 363
 364         if (!p || !p->nvdimm_events_map)
 365                 return -EINVAL;
 366
 367         stats = kzalloc(size, GFP_KERNEL);
 368         if (!stats)
 369                 return -ENOMEM;
 370
 371         stat = &stats->scm_statistic[0];
 372         memcpy(&stat->stat_id,
 373                p->nvdimm_events_map[event->attr.config],
 374                 sizeof(stat->stat_id));
 375         stat->stat_val = 0;
 376
 377         rc = drc_pmem_query_stats(p, stats, 1);
 378         if (rc < 0) {
 379                 kfree(stats);
 380                 return rc;
 381         }
 382
 383         *count = be64_to_cpu(stat->stat_val);
 384         kfree(stats);
 385         return 0;
 386 }
 387
 388 static int papr_scm_pmu_event_init(struct perf_event *event)
 389 {
 390         struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
 391         struct papr_scm_priv *p;
 392
 393         if (!nd_pmu)
 394                 return -EINVAL;
 395
 396         /* test the event attr type for PMU enumeration */
 397         if (event->attr.type != event->pmu->type)
 398                 return -ENOENT;
 399
 400         /* it does not support event sampling mode */
 401         if (is_sampling_event(event))
 402                 return -EOPNOTSUPP;
 403
 404         /* no branch sampling */
 405         if (has_branch_stack(event))
 406                 return -EOPNOTSUPP;
 407
 408         p = (struct papr_scm_priv *)nd_pmu->dev->driver_data;
 409         if (!p)
 410                 return -EINVAL;
 411
 412         /* Invalid eventcode */
 413         if (event->attr.config == 0 || event->attr.config > 16)
 414                 return -EINVAL;
 415
 416         return 0;
 417 }
 418
 419 static int papr_scm_pmu_add(struct perf_event *event, int flags)
 420 {
 421         u64 count;
 422         int rc;
 423         struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
 424
 425         if (!nd_pmu)
 426                 return -EINVAL;
 427
 428         if (flags & PERF_EF_START) {
 429                 rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &count);
 430                 if (rc)
 431                         return rc;
 432
 433                 local64_set(&event->hw.prev_count, count);
 434         }
 435
 436         return 0;
 437 }
 438
 439 static void papr_scm_pmu_read(struct perf_event *event)
 440 {
 441         u64 prev, now;
 442         int rc;
 443         struct nvdimm_pmu *nd_pmu = to_nvdimm_pmu(event->pmu);
 444
 445         if (!nd_pmu)
 446                 return;
 447
 448         rc = papr_scm_pmu_get_value(event, nd_pmu->dev, &now);
 449         if (rc)
 450                 return;
 451
 452         prev = local64_xchg(&event->hw.prev_count, now);
 453         local64_add(now - prev, &event->count);
 454 }
 455
 456 static void papr_scm_pmu_del(struct perf_event *event, int flags)
 457 {
 458         papr_scm_pmu_read(event);
 459 }
 460
 461 static int papr_scm_pmu_check_events(struct papr_scm_priv *p, struct nvdimm_pmu *nd_pmu)
 462 {
 463         struct papr_scm_perf_stat *stat;
 464         struct papr_scm_perf_stats *stats;
 465         int index, rc, count;
 466         u32 available_events;
 467
 468         if (!p->stat_buffer_len)
 469                 return -ENOENT;
 470
 471         available_events = (p->stat_buffer_len  - sizeof(struct papr_scm_perf_stats))
 472                         / sizeof(struct papr_scm_perf_stat);
 473
 474         /* Allocate the buffer for phyp where stats are written */
 475         stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
 476         if (!stats) {
 477                 rc = -ENOMEM;
 478                 return rc;
 479         }
 480
 481         /* Allocate memory to nvdimm_event_map */
 482         p->nvdimm_events_map = kcalloc(available_events, sizeof(char *), GFP_KERNEL);
 483         if (!p->nvdimm_events_map) {
 484                 rc = -ENOMEM;
 485                 goto out_stats;
 486         }
 487
 488         /* Called to get list of events supported */
 489         rc = drc_pmem_query_stats(p, stats, 0);
 490         if (rc)
 491                 goto out_nvdimm_events_map;
 492
 493         for (index = 0, stat = stats->scm_statistic, count = 0;
 494                      index < available_events; index++, ++stat) {
 495                 p->nvdimm_events_map[count] = kmemdup_nul(stat->stat_id, 8, GFP_KERNEL);
 496                 if (!p->nvdimm_events_map[count]) {
 497                         rc = -ENOMEM;
 498                         goto out_nvdimm_events_map;
 499                 }
 500
 501                 count++;
 502         }
 503         p->nvdimm_events_map[count] = NULL;
 504         kfree(stats);
 505         return 0;
 506
 507 out_nvdimm_events_map:
 508         kfree(p->nvdimm_events_map);
 509 out_stats:
 510         kfree(stats);
 511         return rc;
 512 }
 513
 514 static void papr_scm_pmu_register(struct papr_scm_priv *p)
 515 {
 516         struct nvdimm_pmu *nd_pmu;
 517         int rc, nodeid;
 518
 519         nd_pmu = kzalloc(sizeof(*nd_pmu), GFP_KERNEL);
 520         if (!nd_pmu) {
 521                 rc = -ENOMEM;
 522                 goto pmu_err_print;
 523         }
 524
 525         rc = papr_scm_pmu_check_events(p, nd_pmu);
 526         if (rc)
 527                 goto pmu_check_events_err;
 528
 529         nd_pmu->pmu.task_ctx_nr = perf_invalid_context;
 530         nd_pmu->pmu.name = nvdimm_name(p->nvdimm);
 531         nd_pmu->pmu.event_init = papr_scm_pmu_event_init;
 532         nd_pmu->pmu.read = papr_scm_pmu_read;
 533         nd_pmu->pmu.add = papr_scm_pmu_add;
 534         nd_pmu->pmu.del = papr_scm_pmu_del;
 535
 536         nd_pmu->pmu.capabilities = PERF_PMU_CAP_NO_INTERRUPT |
 537                                 PERF_PMU_CAP_NO_EXCLUDE;
 538
 539         /*updating the cpumask variable */
 540         nodeid = numa_map_to_online_node(dev_to_node(&p->pdev->dev));
 541         nd_pmu->arch_cpumask = *cpumask_of_node(nodeid);
 542
 543         rc = register_nvdimm_pmu(nd_pmu, p->pdev);
 544         if (rc)
 545                 goto pmu_register_err;
 546
 547         /*
 548          * Set archdata.priv value to nvdimm_pmu structure, to handle the
 549          * unregistering of pmu device.
 550          */
 551         p->pdev->archdata.priv = nd_pmu;
 552         return;
 553
 554 pmu_register_err:
 555         kfree(p->nvdimm_events_map);
 556 pmu_check_events_err:
 557         kfree(nd_pmu);
 558 pmu_err_print:
 559         dev_info(&p->pdev->dev, "nvdimm pmu didn't register rc=%d\n", rc);
 560 }
 561
 562 #else
 563 static void papr_scm_pmu_register(struct papr_scm_priv *p) { }
 564 #endif
 565
 566 /*
 567  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
 568  * health information.
 569  */
 570 static int __drc_pmem_query_health(struct papr_scm_priv *p)
 571 {
 572         unsigned long ret[PLPAR_HCALL_BUFSIZE];
 573         u64 bitmap = 0;
 574         long rc;
 575
 576         /* issue the hcall */
 577         rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index);
 578         if (rc == H_SUCCESS)
 579                 bitmap = ret[0] & ret[1];
 580         else if (rc == H_FUNCTION)
 581                 dev_info_once(&p->pdev->dev,
 582                               "Hcall H_SCM_HEALTH not implemented, assuming empty health bitmap");
 583         else {
 584
 585                 dev_err(&p->pdev->dev,
 586                         "Failed to query health information, Err:%ld\n", rc);
 587                 return -ENXIO;
 588         }
 589
 590         p->lasthealth_jiffies = jiffies;
 591         /* Allow injecting specific health bits via inject mask. */
 592         if (p->health_bitmap_inject_mask)
 593                 bitmap = (bitmap & ~p->health_bitmap_inject_mask) |
 594                         p->health_bitmap_inject_mask;
 595         WRITE_ONCE(p->health_bitmap, bitmap);
 596         dev_dbg(&p->pdev->dev,
 597                 "Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n",
 598                 ret[0], ret[1]);
 599
 600         return 0;
 601 }
 602
 603 /* Min interval in seconds for assuming stable dimm health */
 604 #define MIN_HEALTH_QUERY_INTERVAL 60
 605
 606 /* Query cached health info and if needed call drc_pmem_query_health */
 607 static int drc_pmem_query_health(struct papr_scm_priv *p)
 608 {
 609         unsigned long cache_timeout;
 610         int rc;
 611
 612         /* Protect concurrent modifications to papr_scm_priv */
 613         rc = mutex_lock_interruptible(&p->health_mutex);
 614         if (rc)
 615                 return rc;
 616
 617         /* Jiffies offset for which the health data is assumed to be same */
 618         cache_timeout = p->lasthealth_jiffies +
 619                 msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000);
 620
 621         /* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */
 622         if (time_after(jiffies, cache_timeout))
 623                 rc = __drc_pmem_query_health(p);
 624         else
 625                 /* Assume cached health data is valid */
 626                 rc = 0;
 627
 628         mutex_unlock(&p->health_mutex);
 629         return rc;
 630 }
 631
 632 static int papr_scm_meta_get(struct papr_scm_priv *p,
 633                              struct nd_cmd_get_config_data_hdr *hdr)
 634 {
 635         unsigned long data[PLPAR_HCALL_BUFSIZE];
 636         unsigned long offset, data_offset;
 637         int len, read;
 638         int64_t ret;
 639
 640         if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
 641                 return -EINVAL;
 642
 643         for (len = hdr->in_length; len; len -= read) {
 644
 645                 data_offset = hdr->in_length - len;
 646                 offset = hdr->in_offset + data_offset;
 647
 648                 if (len >= 8)
 649                         read = 8;
 650                 else if (len >= 4)
 651                         read = 4;
 652                 else if (len >= 2)
 653                         read = 2;
 654                 else
 655                         read = 1;
 656
 657                 ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index,
 658                                   offset, read);
 659
 660                 if (ret == H_PARAMETER) /* bad DRC index */
 661                         return -ENODEV;
 662                 if (ret)
 663                         return -EINVAL; /* other invalid parameter */
 664
 665                 switch (read) {
 666                 case 8:
 667                         *(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]);
 668                         break;
 669                 case 4:
 670                         *(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff);
 671                         break;
 672
 673                 case 2:
 674                         *(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff);
 675                         break;
 676
 677                 case 1:
 678                         *(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff);
 679                         break;
 680                 }
 681         }
 682         return 0;
 683 }
 684
 685 static int papr_scm_meta_set(struct papr_scm_priv *p,
 686                              struct nd_cmd_set_config_hdr *hdr)
 687 {
 688         unsigned long offset, data_offset;
 689         int len, wrote;
 690         unsigned long data;
 691         __be64 data_be;
 692         int64_t ret;
 693
 694         if ((hdr->in_offset + hdr->in_length) > p->metadata_size)
 695                 return -EINVAL;
 696
 697         for (len = hdr->in_length; len; len -= wrote) {
 698
 699                 data_offset = hdr->in_length - len;
 700                 offset = hdr->in_offset + data_offset;
 701
 702                 if (len >= 8) {
 703                         data = *(uint64_t *)(hdr->in_buf + data_offset);
 704                         data_be = cpu_to_be64(data);
 705                         wrote = 8;
 706                 } else if (len >= 4) {
 707                         data = *(uint32_t *)(hdr->in_buf + data_offset);
 708                         data &= 0xffffffff;
 709                         data_be = cpu_to_be32(data);
 710                         wrote = 4;
 711                 } else if (len >= 2) {
 712                         data = *(uint16_t *)(hdr->in_buf + data_offset);
 713                         data &= 0xffff;
 714                         data_be = cpu_to_be16(data);
 715                         wrote = 2;
 716                 } else {
 717                         data_be = *(uint8_t *)(hdr->in_buf + data_offset);
 718                         data_be &= 0xff;
 719                         wrote = 1;
 720                 }
 721
 722                 ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index,
 723                                          offset, data_be, wrote);
 724                 if (ret == H_PARAMETER) /* bad DRC index */
 725                         return -ENODEV;
 726                 if (ret)
 727                         return -EINVAL; /* other invalid parameter */
 728         }
 729
 730         return 0;
 731 }
 732
 733 /*
 734  * Do a sanity checks on the inputs args to dimm-control function and return
 735  * '0' if valid. Validation of PDSM payloads happens later in
 736  * papr_scm_service_pdsm.
 737  */
 738 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
 739                         unsigned int buf_len)
 740 {
 741         unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK;
 742         struct nd_cmd_pkg *nd_cmd;
 743         struct papr_scm_priv *p;
 744         enum papr_pdsm pdsm;
 745
 746         /* Only dimm-specific calls are supported atm */
 747         if (!nvdimm)
 748                 return -EINVAL;
 749
 750         /* get the provider data from struct nvdimm */
 751         p = nvdimm_provider_data(nvdimm);
 752
 753         if (!test_bit(cmd, &cmd_mask)) {
 754                 dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd);
 755                 return -EINVAL;
 756         }
 757
 758         /* For CMD_CALL verify pdsm request */
 759         if (cmd == ND_CMD_CALL) {
 760                 /* Verify the envelope and envelop size */
 761                 if (!buf ||
 762                     buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) {
 763                         dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n",
 764                                 buf_len);
 765                         return -EINVAL;
 766                 }
 767
 768                 /* Verify that the nd_cmd_pkg.nd_family is correct */
 769                 nd_cmd = (struct nd_cmd_pkg *)buf;
 770
 771                 if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) {
 772                         dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n",
 773                                 nd_cmd->nd_family);
 774                         return -EINVAL;
 775                 }
 776
 777                 pdsm = (enum papr_pdsm)nd_cmd->nd_command;
 778
 779                 /* Verify if the pdsm command is valid */
 780                 if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) {
 781                         dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n",
 782                                 pdsm);
 783                         return -EINVAL;
 784                 }
 785
 786                 /* Have enough space to hold returned 'nd_pkg_pdsm' header */
 787                 if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) {
 788                         dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n",
 789                                 pdsm);
 790                         return -EINVAL;
 791                 }
 792         }
 793
 794         /* Let the command be further processed */
 795         return 0;
 796 }
 797
 798 static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p,
 799                                 union nd_pdsm_payload *payload)
 800 {
 801         int rc, size;
 802         u64 statval;
 803         struct papr_scm_perf_stat *stat;
 804         struct papr_scm_perf_stats *stats;
 805
 806         /* Silently fail if fetching performance metrics isn't  supported */
 807         if (!p->stat_buffer_len)
 808                 return 0;
 809
 810         /* Allocate request buffer enough to hold single performance stat */
 811         size = sizeof(struct papr_scm_perf_stats) +
 812                 sizeof(struct papr_scm_perf_stat);
 813
 814         stats = kzalloc(size, GFP_KERNEL);
 815         if (!stats)
 816                 return -ENOMEM;
 817
 818         stat = &stats->scm_statistic[0];
 819         memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id));
 820         stat->stat_val = 0;
 821
 822         /* Fetch the fuel gauge and populate it in payload */
 823         rc = drc_pmem_query_stats(p, stats, 1);
 824         if (rc < 0) {
 825                 dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc);
 826                 goto free_stats;
 827         }
 828
 829         statval = be64_to_cpu(stat->stat_val);
 830         dev_dbg(&p->pdev->dev,
 831                 "Fetched fuel-gauge %llu", statval);
 832         payload->health.extension_flags |=
 833                 PDSM_DIMM_HEALTH_RUN_GAUGE_VALID;
 834         payload->health.dimm_fuel_gauge = statval;
 835
 836         rc = sizeof(struct nd_papr_pdsm_health);
 837
 838 free_stats:
 839         kfree(stats);
 840         return rc;
 841 }
 842
 843 /* Add the dirty-shutdown-counter value to the pdsm */
 844 static int papr_pdsm_dsc(struct papr_scm_priv *p,
 845                          union nd_pdsm_payload *payload)
 846 {
 847         payload->health.extension_flags |= PDSM_DIMM_DSC_VALID;
 848         payload->health.dimm_dsc = p->dirty_shutdown_counter;
 849
 850         return sizeof(struct nd_papr_pdsm_health);
 851 }
 852
 853 /* Fetch the DIMM health info and populate it in provided package. */
 854 static int papr_pdsm_health(struct papr_scm_priv *p,
 855                             union nd_pdsm_payload *payload)
 856 {
 857         int rc;
 858
 859         /* Ensure dimm health mutex is taken preventing concurrent access */
 860         rc = mutex_lock_interruptible(&p->health_mutex);
 861         if (rc)
 862                 goto out;
 863
 864         /* Always fetch upto date dimm health data ignoring cached values */
 865         rc = __drc_pmem_query_health(p);
 866         if (rc) {
 867                 mutex_unlock(&p->health_mutex);
 868                 goto out;
 869         }
 870
 871         /* update health struct with various flags derived from health bitmap */
 872         payload->health = (struct nd_papr_pdsm_health) {
 873                 .extension_flags = 0,
 874                 .dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK),
 875                 .dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK),
 876                 .dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK),
 877                 .dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
 878                 .dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED),
 879                 .dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED),
 880                 .dimm_health = PAPR_PDSM_DIMM_HEALTHY,
 881         };
 882
 883         /* Update field dimm_health based on health_bitmap flags */
 884         if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL)
 885                 payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL;
 886         else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL)
 887                 payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL;
 888         else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY)
 889                 payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY;
 890
 891         /* struct populated hence can release the mutex now */
 892         mutex_unlock(&p->health_mutex);
 893
 894         /* Populate the fuel gauge meter in the payload */
 895         papr_pdsm_fuel_gauge(p, payload);
 896         /* Populate the dirty-shutdown-counter field */
 897         papr_pdsm_dsc(p, payload);
 898
 899         rc = sizeof(struct nd_papr_pdsm_health);
 900
 901 out:
 902         return rc;
 903 }
 904
 905 /* Inject a smart error Add the dirty-shutdown-counter value to the pdsm */
 906 static int papr_pdsm_smart_inject(struct papr_scm_priv *p,
 907                                   union nd_pdsm_payload *payload)
 908 {
 909         int rc;
 910         u32 supported_flags = 0;
 911         u64 inject_mask = 0, clear_mask = 0;
 912         u64 mask;
 913
 914         /* Check for individual smart error flags and update inject/clear masks */
 915         if (payload->smart_inject.flags & PDSM_SMART_INJECT_HEALTH_FATAL) {
 916                 supported_flags |= PDSM_SMART_INJECT_HEALTH_FATAL;
 917                 if (payload->smart_inject.fatal_enable)
 918                         inject_mask |= PAPR_PMEM_HEALTH_FATAL;
 919                 else
 920                         clear_mask |= PAPR_PMEM_HEALTH_FATAL;
 921         }
 922
 923         if (payload->smart_inject.flags & PDSM_SMART_INJECT_BAD_SHUTDOWN) {
 924                 supported_flags |= PDSM_SMART_INJECT_BAD_SHUTDOWN;
 925                 if (payload->smart_inject.unsafe_shutdown_enable)
 926                         inject_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
 927                 else
 928                         clear_mask |= PAPR_PMEM_SHUTDOWN_DIRTY;
 929         }
 930
 931         dev_dbg(&p->pdev->dev, "[Smart-inject] inject_mask=%#llx clear_mask=%#llx\n",
 932                 inject_mask, clear_mask);
 933
 934         /* Prevent concurrent access to dimm health bitmap related members */
 935         rc = mutex_lock_interruptible(&p->health_mutex);
 936         if (rc)
 937                 return rc;
 938
 939         /* Use inject/clear masks to set health_bitmap_inject_mask */
 940         mask = READ_ONCE(p->health_bitmap_inject_mask);
 941         mask = (mask & ~clear_mask) | inject_mask;
 942         WRITE_ONCE(p->health_bitmap_inject_mask, mask);
 943
 944         /* Invalidate cached health bitmap */
 945         p->lasthealth_jiffies = 0;
 946
 947         mutex_unlock(&p->health_mutex);
 948
 949         /* Return the supported flags back to userspace */
 950         payload->smart_inject.flags = supported_flags;
 951
 952         return sizeof(struct nd_papr_pdsm_health);
 953 }
 954
 955 /*
 956  * 'struct pdsm_cmd_desc'
 957  * Identifies supported PDSMs' expected length of in/out payloads
 958  * and pdsm service function.
 959  *
 960  * size_in      : Size of input payload if any in the PDSM request.
 961  * size_out     : Size of output payload if any in the PDSM request.
 962  * service      : Service function for the PDSM request. Return semantics:
 963  *                rc < 0 : Error servicing PDSM and rc indicates the error.
 964  *                rc >=0 : Serviced successfully and 'rc' indicate number of
 965  *                      bytes written to payload.
 966  */
 967 struct pdsm_cmd_desc {
 968         u32 size_in;
 969         u32 size_out;
 970         int (*service)(struct papr_scm_priv *dimm,
 971                        union nd_pdsm_payload *payload);
 972 };
 973
 974 /* Holds all supported PDSMs' command descriptors */
 975 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = {
 976         [PAPR_PDSM_MIN] = {
 977                 .size_in = 0,
 978                 .size_out = 0,
 979                 .service = NULL,
 980         },
 981         /* New PDSM command descriptors to be added below */
 982
 983         [PAPR_PDSM_HEALTH] = {
 984                 .size_in = 0,
 985                 .size_out = sizeof(struct nd_papr_pdsm_health),
 986                 .service = papr_pdsm_health,
 987         },
 988
 989         [PAPR_PDSM_SMART_INJECT] = {
 990                 .size_in = sizeof(struct nd_papr_pdsm_smart_inject),
 991                 .size_out = sizeof(struct nd_papr_pdsm_smart_inject),
 992                 .service = papr_pdsm_smart_inject,
 993         },
 994         /* Empty */
 995         [PAPR_PDSM_MAX] = {
 996                 .size_in = 0,
 997                 .size_out = 0,
 998                 .service = NULL,
 999         },
1000 };
1001
1002 /* Given a valid pdsm cmd return its command descriptor else return NULL */
1003 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd)
1004 {
1005         if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors))
1006                 return &__pdsm_cmd_descriptors[cmd];
1007
1008         return NULL;
1009 }
1010
1011 /*
1012  * For a given pdsm request call an appropriate service function.
1013  * Returns errors if any while handling the pdsm command package.
1014  */
1015 static int papr_scm_service_pdsm(struct papr_scm_priv *p,
1016                                  struct nd_cmd_pkg *pkg)
1017 {
1018         /* Get the PDSM header and PDSM command */
1019         struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload;
1020         enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command;
1021         const struct pdsm_cmd_desc *pdsc;
1022         int rc;
1023
1024         /* Fetch corresponding pdsm descriptor for validation and servicing */
1025         pdsc = pdsm_cmd_desc(pdsm);
1026
1027         /* Validate pdsm descriptor */
1028         /* Ensure that reserved fields are 0 */
1029         if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) {
1030                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n",
1031                         pdsm);
1032                 return -EINVAL;
1033         }
1034
1035         /* If pdsm expects some input, then ensure that the size_in matches */
1036         if (pdsc->size_in &&
1037             pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) {
1038                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n",
1039                         pdsm, pkg->nd_size_in);
1040                 return -EINVAL;
1041         }
1042
1043         /* If pdsm wants to return data, then ensure that  size_out matches */
1044         if (pdsc->size_out &&
1045             pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) {
1046                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n",
1047                         pdsm, pkg->nd_size_out);
1048                 return -EINVAL;
1049         }
1050
1051         /* Service the pdsm */
1052         if (pdsc->service) {
1053                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm);
1054
1055                 rc = pdsc->service(p, &pdsm_pkg->payload);
1056
1057                 if (rc < 0) {
1058                         /* error encountered while servicing pdsm */
1059                         pdsm_pkg->cmd_status = rc;
1060                         pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
1061                 } else {
1062                         /* pdsm serviced and 'rc' bytes written to payload */
1063                         pdsm_pkg->cmd_status = 0;
1064                         pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc;
1065                 }
1066         } else {
1067                 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n",
1068                         pdsm);
1069                 pdsm_pkg->cmd_status = -ENOENT;
1070                 pkg->nd_fw_size = ND_PDSM_HDR_SIZE;
1071         }
1072
1073         return pdsm_pkg->cmd_status;
1074 }
1075
1076 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
1077                           struct nvdimm *nvdimm, unsigned int cmd, void *buf,
1078                           unsigned int buf_len, int *cmd_rc)
1079 {
1080         struct nd_cmd_get_config_size *get_size_hdr;
1081         struct nd_cmd_pkg *call_pkg = NULL;
1082         struct papr_scm_priv *p;
1083         int rc;
1084
1085         rc = is_cmd_valid(nvdimm, cmd, buf, buf_len);
1086         if (rc) {
1087                 pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc);
1088                 return rc;
1089         }
1090
1091         /* Use a local variable in case cmd_rc pointer is NULL */
1092         if (!cmd_rc)
1093                 cmd_rc = &rc;
1094
1095         p = nvdimm_provider_data(nvdimm);
1096
1097         switch (cmd) {
1098         case ND_CMD_GET_CONFIG_SIZE:
1099                 get_size_hdr = buf;
1100
1101                 get_size_hdr->status = 0;
1102                 get_size_hdr->max_xfer = 8;
1103                 get_size_hdr->config_size = p->metadata_size;
1104                 *cmd_rc = 0;
1105                 break;
1106
1107         case ND_CMD_GET_CONFIG_DATA:
1108                 *cmd_rc = papr_scm_meta_get(p, buf);
1109                 break;
1110
1111         case ND_CMD_SET_CONFIG_DATA:
1112                 *cmd_rc = papr_scm_meta_set(p, buf);
1113                 break;
1114
1115         case ND_CMD_CALL:
1116                 call_pkg = (struct nd_cmd_pkg *)buf;
1117                 *cmd_rc = papr_scm_service_pdsm(p, call_pkg);
1118                 break;
1119
1120         default:
1121                 dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd);
1122                 return -EINVAL;
1123         }
1124
1125         dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc);
1126
1127         return 0;
1128 }
1129
1130 static ssize_t health_bitmap_inject_show(struct device *dev,
1131                                          struct device_attribute *attr,
1132                                          char *buf)
1133 {
1134         struct nvdimm *dimm = to_nvdimm(dev);
1135         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1136
1137         return sprintf(buf, "%#llx\n",
1138                        READ_ONCE(p->health_bitmap_inject_mask));
1139 }
1140
1141 static DEVICE_ATTR_ADMIN_RO(health_bitmap_inject);
1142
1143 static ssize_t perf_stats_show(struct device *dev,
1144                                struct device_attribute *attr, char *buf)
1145 {
1146         int index;
1147         ssize_t rc;
1148         struct seq_buf s;
1149         struct papr_scm_perf_stat *stat;
1150         struct papr_scm_perf_stats *stats;
1151         struct nvdimm *dimm = to_nvdimm(dev);
1152         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1153
1154         if (!p->stat_buffer_len)
1155                 return -ENOENT;
1156
1157         /* Allocate the buffer for phyp where stats are written */
1158         stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
1159         if (!stats)
1160                 return -ENOMEM;
1161
1162         /* Ask phyp to return all dimm perf stats */
1163         rc = drc_pmem_query_stats(p, stats, 0);
1164         if (rc)
1165                 goto free_stats;
1166         /*
1167          * Go through the returned output buffer and print stats and
1168          * values. Since stat_id is essentially a char string of
1169          * 8 bytes, simply use the string format specifier to print it.
1170          */
1171         seq_buf_init(&s, buf, PAGE_SIZE);
1172         for (index = 0, stat = stats->scm_statistic;
1173              index < be32_to_cpu(stats->num_statistics);
1174              ++index, ++stat) {
1175                 seq_buf_printf(&s, "%.8s = 0x%016llX\n",
1176                                stat->stat_id,
1177                                be64_to_cpu(stat->stat_val));
1178         }
1179
1180 free_stats:
1181         kfree(stats);
1182         return rc ? rc : (ssize_t)seq_buf_used(&s);
1183 }
1184 static DEVICE_ATTR_ADMIN_RO(perf_stats);
1185
1186 static ssize_t flags_show(struct device *dev,
1187                           struct device_attribute *attr, char *buf)
1188 {
1189         struct nvdimm *dimm = to_nvdimm(dev);
1190         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1191         struct seq_buf s;
1192         u64 health;
1193         int rc;
1194
1195         rc = drc_pmem_query_health(p);
1196         if (rc)
1197                 return rc;
1198
1199         /* Copy health_bitmap locally, check masks & update out buffer */
1200         health = READ_ONCE(p->health_bitmap);
1201
1202         seq_buf_init(&s, buf, PAGE_SIZE);
1203         if (health & PAPR_PMEM_UNARMED_MASK)
1204                 seq_buf_printf(&s, "not_armed ");
1205
1206         if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK)
1207                 seq_buf_printf(&s, "flush_fail ");
1208
1209         if (health & PAPR_PMEM_BAD_RESTORE_MASK)
1210                 seq_buf_printf(&s, "restore_fail ");
1211
1212         if (health & PAPR_PMEM_ENCRYPTED)
1213                 seq_buf_printf(&s, "encrypted ");
1214
1215         if (health & PAPR_PMEM_SMART_EVENT_MASK)
1216                 seq_buf_printf(&s, "smart_notify ");
1217
1218         if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED)
1219                 seq_buf_printf(&s, "scrubbed locked ");
1220
1221         if (seq_buf_used(&s))
1222                 seq_buf_printf(&s, "\n");
1223
1224         return seq_buf_used(&s);
1225 }
1226 DEVICE_ATTR_RO(flags);
1227
1228 static ssize_t dirty_shutdown_show(struct device *dev,
1229                           struct device_attribute *attr, char *buf)
1230 {
1231         struct nvdimm *dimm = to_nvdimm(dev);
1232         struct papr_scm_priv *p = nvdimm_provider_data(dimm);
1233
1234         return sysfs_emit(buf, "%llu\n", p->dirty_shutdown_counter);
1235 }
1236 DEVICE_ATTR_RO(dirty_shutdown);
1237
1238 static umode_t papr_nd_attribute_visible(struct kobject *kobj,
1239                                          struct attribute *attr, int n)
1240 {
1241         struct device *dev = kobj_to_dev(kobj);
1242         struct nvdimm *nvdimm = to_nvdimm(dev);
1243         struct papr_scm_priv *p = nvdimm_provider_data(nvdimm);
1244
1245         /* For if perf-stats not available remove perf_stats sysfs */
1246         if (attr == &dev_attr_perf_stats.attr && p->stat_buffer_len == 0)
1247                 return 0;
1248
1249         return attr->mode;
1250 }
1251
1252 /* papr_scm specific dimm attributes */
1253 static struct attribute *papr_nd_attributes[] = {
1254         &dev_attr_flags.attr,
1255         &dev_attr_perf_stats.attr,
1256         &dev_attr_dirty_shutdown.attr,
1257         &dev_attr_health_bitmap_inject.attr,
1258         NULL,
1259 };
1260
1261 static const struct attribute_group papr_nd_attribute_group = {
1262         .name = "papr",
1263         .is_visible = papr_nd_attribute_visible,
1264         .attrs = papr_nd_attributes,
1265 };
1266
1267 static const struct attribute_group *papr_nd_attr_groups[] = {
1268         &papr_nd_attribute_group,
1269         NULL,
1270 };
1271
1272 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
1273 {
1274         struct device *dev = &p->pdev->dev;
1275         struct nd_mapping_desc mapping;
1276         struct nd_region_desc ndr_desc;
1277         unsigned long dimm_flags;
1278         int target_nid, online_nid;
1279
1280         p->bus_desc.ndctl = papr_scm_ndctl;
1281         p->bus_desc.module = THIS_MODULE;
1282         p->bus_desc.of_node = p->pdev->dev.of_node;
1283         p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL);
1284
1285         /* Set the dimm command family mask to accept PDSMs */
1286         set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask);
1287
1288         if (!p->bus_desc.provider_name)
1289                 return -ENOMEM;
1290
1291         p->bus = nvdimm_bus_register(NULL, &p->bus_desc);
1292         if (!p->bus) {
1293                 dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn);
1294                 kfree(p->bus_desc.provider_name);
1295                 return -ENXIO;
1296         }
1297
1298         dimm_flags = 0;
1299         set_bit(NDD_LABELING, &dimm_flags);
1300
1301         /*
1302          * Check if the nvdimm is unarmed. No locking needed as we are still
1303          * initializing. Ignore error encountered if any.
1304          */
1305         __drc_pmem_query_health(p);
1306
1307         if (p->health_bitmap & PAPR_PMEM_UNARMED_MASK)
1308                 set_bit(NDD_UNARMED, &dimm_flags);
1309
1310         p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups,
1311                                   dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL);
1312         if (!p->nvdimm) {
1313                 dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn);
1314                 goto err;
1315         }
1316
1317         if (nvdimm_bus_check_dimm_count(p->bus, 1))
1318                 goto err;
1319
1320         /* now add the region */
1321
1322         memset(&mapping, 0, sizeof(mapping));
1323         mapping.nvdimm = p->nvdimm;
1324         mapping.start = 0;
1325         mapping.size = p->blocks * p->block_size; // XXX: potential overflow?
1326
1327         memset(&ndr_desc, 0, sizeof(ndr_desc));
1328         target_nid = dev_to_node(&p->pdev->dev);
1329         online_nid = numa_map_to_online_node(target_nid);
1330         ndr_desc.numa_node = online_nid;
1331         ndr_desc.target_node = target_nid;
1332         ndr_desc.res = &p->res;
1333         ndr_desc.of_node = p->dn;
1334         ndr_desc.provider_data = p;
1335         ndr_desc.mapping = &mapping;
1336         ndr_desc.num_mappings = 1;
1337         ndr_desc.nd_set = &p->nd_set;
1338
1339         if (p->hcall_flush_required) {
1340                 set_bit(ND_REGION_ASYNC, &ndr_desc.flags);
1341                 ndr_desc.flush = papr_scm_pmem_flush;
1342         }
1343
1344         if (p->is_volatile)
1345                 p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc);
1346         else {
1347                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
1348                 p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc);
1349         }
1350         if (!p->region) {
1351                 dev_err(dev, "Error registering region %pR from %pOF\n",
1352                                 ndr_desc.res, p->dn);
1353                 goto err;
1354         }
1355         if (target_nid != online_nid)
1356                 dev_info(dev, "Region registered with target node %d and online node %d",
1357                          target_nid, online_nid);
1358
1359         mutex_lock(&papr_ndr_lock);
1360         list_add_tail(&p->region_list, &papr_nd_regions);
1361         mutex_unlock(&papr_ndr_lock);
1362
1363         return 0;
1364
1365 err:    nvdimm_bus_unregister(p->bus);
1366         kfree(p->bus_desc.provider_name);
1367         return -ENXIO;
1368 }
1369
1370 static void papr_scm_add_badblock(struct nd_region *region,
1371                                   struct nvdimm_bus *bus, u64 phys_addr)
1372 {
1373         u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES);
1374
1375         if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) {
1376                 pr_err("Bad block registration for 0x%llx failed\n", phys_addr);
1377                 return;
1378         }
1379
1380         pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n",
1381                  aligned_addr, aligned_addr + L1_CACHE_BYTES);
1382
1383         nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON);
1384 }
1385
1386 static int handle_mce_ue(struct notifier_block *nb, unsigned long val,
1387                          void *data)
1388 {
1389         struct machine_check_event *evt = data;
1390         struct papr_scm_priv *p;
1391         u64 phys_addr;
1392         bool found = false;
1393
1394         if (evt->error_type != MCE_ERROR_TYPE_UE)
1395                 return NOTIFY_DONE;
1396
1397         if (list_empty(&papr_nd_regions))
1398                 return NOTIFY_DONE;
1399
1400         /*
1401          * The physical address obtained here is PAGE_SIZE aligned, so get the
1402          * exact address from the effective address
1403          */
1404         phys_addr = evt->u.ue_error.physical_address +
1405                         (evt->u.ue_error.effective_address & ~PAGE_MASK);
1406
1407         if (!evt->u.ue_error.physical_address_provided ||
1408             !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT)))
1409                 return NOTIFY_DONE;
1410
1411         /* mce notifier is called from a process context, so mutex is safe */
1412         mutex_lock(&papr_ndr_lock);
1413         list_for_each_entry(p, &papr_nd_regions, region_list) {
1414                 if (phys_addr >= p->res.start && phys_addr <= p->res.end) {
1415                         found = true;
1416                         break;
1417                 }
1418         }
1419
1420         if (found)
1421                 papr_scm_add_badblock(p->region, p->bus, phys_addr);
1422
1423         mutex_unlock(&papr_ndr_lock);
1424
1425         return found ? NOTIFY_OK : NOTIFY_DONE;
1426 }
1427
1428 static struct notifier_block mce_ue_nb = {
1429         .notifier_call = handle_mce_ue
1430 };
1431
1432 static int papr_scm_probe(struct platform_device *pdev)
1433 {
1434         struct device_node *dn = pdev->dev.of_node;
1435         u32 drc_index, metadata_size;
1436         u64 blocks, block_size;
1437         struct papr_scm_priv *p;
1438         u8 uuid_raw[UUID_SIZE];
1439         const char *uuid_str;
1440         ssize_t stat_size;
1441         uuid_t uuid;
1442         int rc;
1443
1444         /* check we have all the required DT properties */
1445         if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) {
1446                 dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn);
1447                 return -ENODEV;
1448         }
1449
1450         if (of_property_read_u64(dn, "ibm,block-size", &block_size)) {
1451                 dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn);
1452                 return -ENODEV;
1453         }
1454
1455         if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) {
1456                 dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn);
1457                 return -ENODEV;
1458         }
1459
1460         if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) {
1461                 dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn);
1462                 return -ENODEV;
1463         }
1464
1465
1466         p = kzalloc(sizeof(*p), GFP_KERNEL);
1467         if (!p)
1468                 return -ENOMEM;
1469
1470         /* Initialize the dimm mutex */
1471         mutex_init(&p->health_mutex);
1472
1473         /* optional DT properties */
1474         of_property_read_u32(dn, "ibm,metadata-size", &metadata_size);
1475
1476         p->dn = dn;
1477         p->drc_index = drc_index;
1478         p->block_size = block_size;
1479         p->blocks = blocks;
1480         p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required");
1481         p->hcall_flush_required = of_property_read_bool(dn, "ibm,hcall-flush-required");
1482
1483         if (of_property_read_u64(dn, "ibm,persistence-failed-count",
1484                                  &p->dirty_shutdown_counter))
1485                 p->dirty_shutdown_counter = 0;
1486
1487         /* We just need to ensure that set cookies are unique across */
1488         uuid_parse(uuid_str, &uuid);
1489
1490         /*
1491          * The cookie1 and cookie2 are not really little endian.
1492          * We store a raw buffer representation of the
1493          * uuid string so that we can compare this with the label
1494          * area cookie irrespective of the endian configuration
1495          * with which the kernel is built.
1496          *
1497          * Historically we stored the cookie in the below format.
1498          * for a uuid string 72511b67-0b3b-42fd-8d1d-5be3cae8bcaa
1499          *      cookie1 was 0xfd423b0b671b5172
1500          *      cookie2 was 0xaabce8cae35b1d8d
1501          */
1502         export_uuid(uuid_raw, &uuid);
1503         p->nd_set.cookie1 = get_unaligned_le64(&uuid_raw[0]);
1504         p->nd_set.cookie2 = get_unaligned_le64(&uuid_raw[8]);
1505
1506         /* might be zero */
1507         p->metadata_size = metadata_size;
1508         p->pdev = pdev;
1509
1510         /* request the hypervisor to bind this region to somewhere in memory */
1511         rc = drc_pmem_bind(p);
1512
1513         /* If phyp says drc memory still bound then force unbound and retry */
1514         if (rc == H_OVERLAP)
1515                 rc = drc_pmem_query_n_bind(p);
1516
1517         if (rc != H_SUCCESS) {
1518                 dev_err(&p->pdev->dev, "bind err: %d\n", rc);
1519                 rc = -ENXIO;
1520                 goto err;
1521         }
1522
1523         /* setup the resource for the newly bound range */
1524         p->res.start = p->bound_addr;
1525         p->res.end   = p->bound_addr + p->blocks * p->block_size - 1;
1526         p->res.name  = pdev->name;
1527         p->res.flags = IORESOURCE_MEM;
1528
1529         /* Try retrieving the stat buffer and see if its supported */
1530         stat_size = drc_pmem_query_stats(p, NULL, 0);
1531         if (stat_size > 0) {
1532                 p->stat_buffer_len = stat_size;
1533                 dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
1534                         p->stat_buffer_len);
1535         }
1536
1537         rc = papr_scm_nvdimm_init(p);
1538         if (rc)
1539                 goto err2;
1540
1541         platform_set_drvdata(pdev, p);
1542         papr_scm_pmu_register(p);
1543
1544         return 0;
1545
1546 err2:   drc_pmem_unbind(p);
1547 err:    kfree(p);
1548         return rc;
1549 }
1550
1551 static int papr_scm_remove(struct platform_device *pdev)
1552 {
1553         struct papr_scm_priv *p = platform_get_drvdata(pdev);
1554
1555         mutex_lock(&papr_ndr_lock);
1556         list_del(&p->region_list);
1557         mutex_unlock(&papr_ndr_lock);
1558
1559         nvdimm_bus_unregister(p->bus);
1560         drc_pmem_unbind(p);
1561
1562         if (pdev->archdata.priv)
1563                 unregister_nvdimm_pmu(pdev->archdata.priv);
1564
1565         pdev->archdata.priv = NULL;
1566         kfree(p->nvdimm_events_map);
1567         kfree(p->bus_desc.provider_name);
1568         kfree(p);
1569
1570         return 0;
1571 }
1572
1573 static const struct of_device_id papr_scm_match[] = {
1574         { .compatible = "ibm,pmemory" },
1575         { .compatible = "ibm,pmemory-v2" },
1576         { },
1577 };
1578
1579 static struct platform_driver papr_scm_driver = {
1580         .probe = papr_scm_probe,
1581         .remove = papr_scm_remove,
1582         .driver = {
1583                 .name = "papr_scm",
1584                 .of_match_table = papr_scm_match,
1585         },
1586 };
1587
1588 static int __init papr_scm_init(void)
1589 {
1590         int ret;
1591
1592         ret = platform_driver_register(&papr_scm_driver);
1593         if (!ret)
1594                 mce_register_notifier(&mce_ue_nb);
1595
1596         return ret;
1597 }
1598 module_init(papr_scm_init);
1599
1600 static void __exit papr_scm_exit(void)
1601 {
1602         mce_unregister_notifier(&mce_ue_nb);
1603         platform_driver_unregister(&papr_scm_driver);
1604 }
1605 module_exit(papr_scm_exit);
1606
1607 MODULE_DEVICE_TABLE(of, papr_scm_match);
1608 MODULE_LICENSE("GPL");
1609 MODULE_AUTHOR("IBM Corporation");