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 22 #define BIND_ANY_ADDR (~0ul) 23 24 #define PAPR_SCM_DIMM_CMD_MASK \ 25 ((1ul << ND_CMD_GET_CONFIG_SIZE) | \ 26 (1ul << ND_CMD_GET_CONFIG_DATA) | \ 27 (1ul << ND_CMD_SET_CONFIG_DATA) | \ 28 (1ul << ND_CMD_CALL)) 29 30 /* DIMM health bitmap bitmap indicators */ 31 /* SCM device is unable to persist memory contents */ 32 #define PAPR_PMEM_UNARMED (1ULL << (63 - 0)) 33 /* SCM device failed to persist memory contents */ 34 #define PAPR_PMEM_SHUTDOWN_DIRTY (1ULL << (63 - 1)) 35 /* SCM device contents are persisted from previous IPL */ 36 #define PAPR_PMEM_SHUTDOWN_CLEAN (1ULL << (63 - 2)) 37 /* SCM device contents are not persisted from previous IPL */ 38 #define PAPR_PMEM_EMPTY (1ULL << (63 - 3)) 39 /* SCM device memory life remaining is critically low */ 40 #define PAPR_PMEM_HEALTH_CRITICAL (1ULL << (63 - 4)) 41 /* SCM device will be garded off next IPL due to failure */ 42 #define PAPR_PMEM_HEALTH_FATAL (1ULL << (63 - 5)) 43 /* SCM contents cannot persist due to current platform health status */ 44 #define PAPR_PMEM_HEALTH_UNHEALTHY (1ULL << (63 - 6)) 45 /* SCM device is unable to persist memory contents in certain conditions */ 46 #define PAPR_PMEM_HEALTH_NON_CRITICAL (1ULL << (63 - 7)) 47 /* SCM device is encrypted */ 48 #define PAPR_PMEM_ENCRYPTED (1ULL << (63 - 8)) 49 /* SCM device has been scrubbed and locked */ 50 #define PAPR_PMEM_SCRUBBED_AND_LOCKED (1ULL << (63 - 9)) 51 52 /* Bits status indicators for health bitmap indicating unarmed dimm */ 53 #define PAPR_PMEM_UNARMED_MASK (PAPR_PMEM_UNARMED | \ 54 PAPR_PMEM_HEALTH_UNHEALTHY) 55 56 /* Bits status indicators for health bitmap indicating unflushed dimm */ 57 #define PAPR_PMEM_BAD_SHUTDOWN_MASK (PAPR_PMEM_SHUTDOWN_DIRTY) 58 59 /* Bits status indicators for health bitmap indicating unrestored dimm */ 60 #define PAPR_PMEM_BAD_RESTORE_MASK (PAPR_PMEM_EMPTY) 61 62 /* Bit status indicators for smart event notification */ 63 #define PAPR_PMEM_SMART_EVENT_MASK (PAPR_PMEM_HEALTH_CRITICAL | \ 64 PAPR_PMEM_HEALTH_FATAL | \ 65 PAPR_PMEM_HEALTH_UNHEALTHY) 66 67 #define PAPR_SCM_PERF_STATS_EYECATCHER __stringify(SCMSTATS) 68 #define PAPR_SCM_PERF_STATS_VERSION 0x1 69 70 /* Struct holding a single performance metric */ 71 struct papr_scm_perf_stat { 72 u8 stat_id[8]; 73 __be64 stat_val; 74 } __packed; 75 76 /* Struct exchanged between kernel and PHYP for fetching drc perf stats */ 77 struct papr_scm_perf_stats { 78 u8 eye_catcher[8]; 79 /* Should be PAPR_SCM_PERF_STATS_VERSION */ 80 __be32 stats_version; 81 /* Number of stats following */ 82 __be32 num_statistics; 83 /* zero or more performance matrics */ 84 struct papr_scm_perf_stat scm_statistic[]; 85 } __packed; 86 87 /* private struct associated with each region */ 88 struct papr_scm_priv { 89 struct platform_device *pdev; 90 struct device_node *dn; 91 uint32_t drc_index; 92 uint64_t blocks; 93 uint64_t block_size; 94 int metadata_size; 95 bool is_volatile; 96 bool hcall_flush_required; 97 98 uint64_t bound_addr; 99 100 struct nvdimm_bus_descriptor bus_desc; 101 struct nvdimm_bus *bus; 102 struct nvdimm *nvdimm; 103 struct resource res; 104 struct nd_region *region; 105 struct nd_interleave_set nd_set; 106 struct list_head region_list; 107 108 /* Protect dimm health data from concurrent read/writes */ 109 struct mutex health_mutex; 110 111 /* Last time the health information of the dimm was updated */ 112 unsigned long lasthealth_jiffies; 113 114 /* Health information for the dimm */ 115 u64 health_bitmap; 116 117 /* length of the stat buffer as expected by phyp */ 118 size_t stat_buffer_len; 119 }; 120 121 static int papr_scm_pmem_flush(struct nd_region *nd_region, 122 struct bio *bio __maybe_unused) 123 { 124 struct papr_scm_priv *p = nd_region_provider_data(nd_region); 125 unsigned long ret_buf[PLPAR_HCALL_BUFSIZE], token = 0; 126 long rc; 127 128 dev_dbg(&p->pdev->dev, "flush drc 0x%x", p->drc_index); 129 130 do { 131 rc = plpar_hcall(H_SCM_FLUSH, ret_buf, p->drc_index, token); 132 token = ret_buf[0]; 133 134 /* Check if we are stalled for some time */ 135 if (H_IS_LONG_BUSY(rc)) { 136 msleep(get_longbusy_msecs(rc)); 137 rc = H_BUSY; 138 } else if (rc == H_BUSY) { 139 cond_resched(); 140 } 141 } while (rc == H_BUSY); 142 143 if (rc) { 144 dev_err(&p->pdev->dev, "flush error: %ld", rc); 145 rc = -EIO; 146 } else { 147 dev_dbg(&p->pdev->dev, "flush drc 0x%x complete", p->drc_index); 148 } 149 150 return rc; 151 } 152 153 static LIST_HEAD(papr_nd_regions); 154 static DEFINE_MUTEX(papr_ndr_lock); 155 156 static int drc_pmem_bind(struct papr_scm_priv *p) 157 { 158 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 159 uint64_t saved = 0; 160 uint64_t token; 161 int64_t rc; 162 163 /* 164 * When the hypervisor cannot map all the requested memory in a single 165 * hcall it returns H_BUSY and we call again with the token until 166 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS 167 * leave the system in an undefined state, so we wait. 168 */ 169 token = 0; 170 171 do { 172 rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0, 173 p->blocks, BIND_ANY_ADDR, token); 174 token = ret[0]; 175 if (!saved) 176 saved = ret[1]; 177 cond_resched(); 178 } while (rc == H_BUSY); 179 180 if (rc) 181 return rc; 182 183 p->bound_addr = saved; 184 dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", 185 p->drc_index, (unsigned long)saved); 186 return rc; 187 } 188 189 static void drc_pmem_unbind(struct papr_scm_priv *p) 190 { 191 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 192 uint64_t token = 0; 193 int64_t rc; 194 195 dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index); 196 197 /* NB: unbind has the same retry requirements as drc_pmem_bind() */ 198 do { 199 200 /* Unbind of all SCM resources associated with drcIndex */ 201 rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC, 202 p->drc_index, token); 203 token = ret[0]; 204 205 /* Check if we are stalled for some time */ 206 if (H_IS_LONG_BUSY(rc)) { 207 msleep(get_longbusy_msecs(rc)); 208 rc = H_BUSY; 209 } else if (rc == H_BUSY) { 210 cond_resched(); 211 } 212 213 } while (rc == H_BUSY); 214 215 if (rc) 216 dev_err(&p->pdev->dev, "unbind error: %lld\n", rc); 217 else 218 dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n", 219 p->drc_index); 220 221 return; 222 } 223 224 static int drc_pmem_query_n_bind(struct papr_scm_priv *p) 225 { 226 unsigned long start_addr; 227 unsigned long end_addr; 228 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 229 int64_t rc; 230 231 232 rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret, 233 p->drc_index, 0); 234 if (rc) 235 goto err_out; 236 start_addr = ret[0]; 237 238 /* Make sure the full region is bound. */ 239 rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret, 240 p->drc_index, p->blocks - 1); 241 if (rc) 242 goto err_out; 243 end_addr = ret[0]; 244 245 if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size)) 246 goto err_out; 247 248 p->bound_addr = start_addr; 249 dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr); 250 return rc; 251 252 err_out: 253 dev_info(&p->pdev->dev, 254 "Failed to query, trying an unbind followed by bind"); 255 drc_pmem_unbind(p); 256 return drc_pmem_bind(p); 257 } 258 259 /* 260 * Query the Dimm performance stats from PHYP and copy them (if returned) to 261 * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast 262 * (num_stats + header) bytes. 263 * - If buff_stats == NULL the return value is the size in byes of the buffer 264 * needed to hold all supported performance-statistics. 265 * - If buff_stats != NULL and num_stats == 0 then we copy all known 266 * performance-statistics to 'buff_stat' and expect to be large enough to 267 * hold them. 268 * - if buff_stats != NULL and num_stats > 0 then copy the requested 269 * performance-statistics to buff_stats. 270 */ 271 static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p, 272 struct papr_scm_perf_stats *buff_stats, 273 unsigned int num_stats) 274 { 275 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 276 size_t size; 277 s64 rc; 278 279 /* Setup the out buffer */ 280 if (buff_stats) { 281 memcpy(buff_stats->eye_catcher, 282 PAPR_SCM_PERF_STATS_EYECATCHER, 8); 283 buff_stats->stats_version = 284 cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION); 285 buff_stats->num_statistics = 286 cpu_to_be32(num_stats); 287 288 /* 289 * Calculate the buffer size based on num-stats provided 290 * or use the prefetched max buffer length 291 */ 292 if (num_stats) 293 /* Calculate size from the num_stats */ 294 size = sizeof(struct papr_scm_perf_stats) + 295 num_stats * sizeof(struct papr_scm_perf_stat); 296 else 297 size = p->stat_buffer_len; 298 } else { 299 /* In case of no out buffer ignore the size */ 300 size = 0; 301 } 302 303 /* Do the HCALL asking PHYP for info */ 304 rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index, 305 buff_stats ? virt_to_phys(buff_stats) : 0, 306 size); 307 308 /* Check if the error was due to an unknown stat-id */ 309 if (rc == H_PARTIAL) { 310 dev_err(&p->pdev->dev, 311 "Unknown performance stats, Err:0x%016lX\n", ret[0]); 312 return -ENOENT; 313 } else if (rc == H_AUTHORITY) { 314 dev_info(&p->pdev->dev, 315 "Permission denied while accessing performance stats"); 316 return -EPERM; 317 } else if (rc == H_UNSUPPORTED) { 318 dev_dbg(&p->pdev->dev, "Performance stats unsupported\n"); 319 return -EOPNOTSUPP; 320 } else if (rc != H_SUCCESS) { 321 dev_err(&p->pdev->dev, 322 "Failed to query performance stats, Err:%lld\n", rc); 323 return -EIO; 324 325 } else if (!size) { 326 /* Handle case where stat buffer size was requested */ 327 dev_dbg(&p->pdev->dev, 328 "Performance stats size %ld\n", ret[0]); 329 return ret[0]; 330 } 331 332 /* Successfully fetched the requested stats from phyp */ 333 dev_dbg(&p->pdev->dev, 334 "Performance stats returned %d stats\n", 335 be32_to_cpu(buff_stats->num_statistics)); 336 return 0; 337 } 338 339 /* 340 * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the 341 * health information. 342 */ 343 static int __drc_pmem_query_health(struct papr_scm_priv *p) 344 { 345 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 346 long rc; 347 348 /* issue the hcall */ 349 rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index); 350 if (rc != H_SUCCESS) { 351 dev_err(&p->pdev->dev, 352 "Failed to query health information, Err:%ld\n", rc); 353 return -ENXIO; 354 } 355 356 p->lasthealth_jiffies = jiffies; 357 p->health_bitmap = ret[0] & ret[1]; 358 359 dev_dbg(&p->pdev->dev, 360 "Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n", 361 ret[0], ret[1]); 362 363 return 0; 364 } 365 366 /* Min interval in seconds for assuming stable dimm health */ 367 #define MIN_HEALTH_QUERY_INTERVAL 60 368 369 /* Query cached health info and if needed call drc_pmem_query_health */ 370 static int drc_pmem_query_health(struct papr_scm_priv *p) 371 { 372 unsigned long cache_timeout; 373 int rc; 374 375 /* Protect concurrent modifications to papr_scm_priv */ 376 rc = mutex_lock_interruptible(&p->health_mutex); 377 if (rc) 378 return rc; 379 380 /* Jiffies offset for which the health data is assumed to be same */ 381 cache_timeout = p->lasthealth_jiffies + 382 msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000); 383 384 /* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */ 385 if (time_after(jiffies, cache_timeout)) 386 rc = __drc_pmem_query_health(p); 387 else 388 /* Assume cached health data is valid */ 389 rc = 0; 390 391 mutex_unlock(&p->health_mutex); 392 return rc; 393 } 394 395 static int papr_scm_meta_get(struct papr_scm_priv *p, 396 struct nd_cmd_get_config_data_hdr *hdr) 397 { 398 unsigned long data[PLPAR_HCALL_BUFSIZE]; 399 unsigned long offset, data_offset; 400 int len, read; 401 int64_t ret; 402 403 if ((hdr->in_offset + hdr->in_length) > p->metadata_size) 404 return -EINVAL; 405 406 for (len = hdr->in_length; len; len -= read) { 407 408 data_offset = hdr->in_length - len; 409 offset = hdr->in_offset + data_offset; 410 411 if (len >= 8) 412 read = 8; 413 else if (len >= 4) 414 read = 4; 415 else if (len >= 2) 416 read = 2; 417 else 418 read = 1; 419 420 ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index, 421 offset, read); 422 423 if (ret == H_PARAMETER) /* bad DRC index */ 424 return -ENODEV; 425 if (ret) 426 return -EINVAL; /* other invalid parameter */ 427 428 switch (read) { 429 case 8: 430 *(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]); 431 break; 432 case 4: 433 *(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff); 434 break; 435 436 case 2: 437 *(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff); 438 break; 439 440 case 1: 441 *(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff); 442 break; 443 } 444 } 445 return 0; 446 } 447 448 static int papr_scm_meta_set(struct papr_scm_priv *p, 449 struct nd_cmd_set_config_hdr *hdr) 450 { 451 unsigned long offset, data_offset; 452 int len, wrote; 453 unsigned long data; 454 __be64 data_be; 455 int64_t ret; 456 457 if ((hdr->in_offset + hdr->in_length) > p->metadata_size) 458 return -EINVAL; 459 460 for (len = hdr->in_length; len; len -= wrote) { 461 462 data_offset = hdr->in_length - len; 463 offset = hdr->in_offset + data_offset; 464 465 if (len >= 8) { 466 data = *(uint64_t *)(hdr->in_buf + data_offset); 467 data_be = cpu_to_be64(data); 468 wrote = 8; 469 } else if (len >= 4) { 470 data = *(uint32_t *)(hdr->in_buf + data_offset); 471 data &= 0xffffffff; 472 data_be = cpu_to_be32(data); 473 wrote = 4; 474 } else if (len >= 2) { 475 data = *(uint16_t *)(hdr->in_buf + data_offset); 476 data &= 0xffff; 477 data_be = cpu_to_be16(data); 478 wrote = 2; 479 } else { 480 data_be = *(uint8_t *)(hdr->in_buf + data_offset); 481 data_be &= 0xff; 482 wrote = 1; 483 } 484 485 ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index, 486 offset, data_be, wrote); 487 if (ret == H_PARAMETER) /* bad DRC index */ 488 return -ENODEV; 489 if (ret) 490 return -EINVAL; /* other invalid parameter */ 491 } 492 493 return 0; 494 } 495 496 /* 497 * Do a sanity checks on the inputs args to dimm-control function and return 498 * '0' if valid. Validation of PDSM payloads happens later in 499 * papr_scm_service_pdsm. 500 */ 501 static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf, 502 unsigned int buf_len) 503 { 504 unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK; 505 struct nd_cmd_pkg *nd_cmd; 506 struct papr_scm_priv *p; 507 enum papr_pdsm pdsm; 508 509 /* Only dimm-specific calls are supported atm */ 510 if (!nvdimm) 511 return -EINVAL; 512 513 /* get the provider data from struct nvdimm */ 514 p = nvdimm_provider_data(nvdimm); 515 516 if (!test_bit(cmd, &cmd_mask)) { 517 dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd); 518 return -EINVAL; 519 } 520 521 /* For CMD_CALL verify pdsm request */ 522 if (cmd == ND_CMD_CALL) { 523 /* Verify the envelope and envelop size */ 524 if (!buf || 525 buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) { 526 dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n", 527 buf_len); 528 return -EINVAL; 529 } 530 531 /* Verify that the nd_cmd_pkg.nd_family is correct */ 532 nd_cmd = (struct nd_cmd_pkg *)buf; 533 534 if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) { 535 dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n", 536 nd_cmd->nd_family); 537 return -EINVAL; 538 } 539 540 pdsm = (enum papr_pdsm)nd_cmd->nd_command; 541 542 /* Verify if the pdsm command is valid */ 543 if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) { 544 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n", 545 pdsm); 546 return -EINVAL; 547 } 548 549 /* Have enough space to hold returned 'nd_pkg_pdsm' header */ 550 if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) { 551 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n", 552 pdsm); 553 return -EINVAL; 554 } 555 } 556 557 /* Let the command be further processed */ 558 return 0; 559 } 560 561 static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p, 562 union nd_pdsm_payload *payload) 563 { 564 int rc, size; 565 u64 statval; 566 struct papr_scm_perf_stat *stat; 567 struct papr_scm_perf_stats *stats; 568 569 /* Silently fail if fetching performance metrics isn't supported */ 570 if (!p->stat_buffer_len) 571 return 0; 572 573 /* Allocate request buffer enough to hold single performance stat */ 574 size = sizeof(struct papr_scm_perf_stats) + 575 sizeof(struct papr_scm_perf_stat); 576 577 stats = kzalloc(size, GFP_KERNEL); 578 if (!stats) 579 return -ENOMEM; 580 581 stat = &stats->scm_statistic[0]; 582 memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id)); 583 stat->stat_val = 0; 584 585 /* Fetch the fuel gauge and populate it in payload */ 586 rc = drc_pmem_query_stats(p, stats, 1); 587 if (rc < 0) { 588 dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc); 589 goto free_stats; 590 } 591 592 statval = be64_to_cpu(stat->stat_val); 593 dev_dbg(&p->pdev->dev, 594 "Fetched fuel-gauge %llu", statval); 595 payload->health.extension_flags |= 596 PDSM_DIMM_HEALTH_RUN_GAUGE_VALID; 597 payload->health.dimm_fuel_gauge = statval; 598 599 rc = sizeof(struct nd_papr_pdsm_health); 600 601 free_stats: 602 kfree(stats); 603 return rc; 604 } 605 606 /* Fetch the DIMM health info and populate it in provided package. */ 607 static int papr_pdsm_health(struct papr_scm_priv *p, 608 union nd_pdsm_payload *payload) 609 { 610 int rc; 611 612 /* Ensure dimm health mutex is taken preventing concurrent access */ 613 rc = mutex_lock_interruptible(&p->health_mutex); 614 if (rc) 615 goto out; 616 617 /* Always fetch upto date dimm health data ignoring cached values */ 618 rc = __drc_pmem_query_health(p); 619 if (rc) { 620 mutex_unlock(&p->health_mutex); 621 goto out; 622 } 623 624 /* update health struct with various flags derived from health bitmap */ 625 payload->health = (struct nd_papr_pdsm_health) { 626 .extension_flags = 0, 627 .dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK), 628 .dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK), 629 .dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK), 630 .dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED), 631 .dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED), 632 .dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED), 633 .dimm_health = PAPR_PDSM_DIMM_HEALTHY, 634 }; 635 636 /* Update field dimm_health based on health_bitmap flags */ 637 if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL) 638 payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL; 639 else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL) 640 payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL; 641 else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY) 642 payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY; 643 644 /* struct populated hence can release the mutex now */ 645 mutex_unlock(&p->health_mutex); 646 647 /* Populate the fuel gauge meter in the payload */ 648 papr_pdsm_fuel_gauge(p, payload); 649 650 rc = sizeof(struct nd_papr_pdsm_health); 651 652 out: 653 return rc; 654 } 655 656 /* 657 * 'struct pdsm_cmd_desc' 658 * Identifies supported PDSMs' expected length of in/out payloads 659 * and pdsm service function. 660 * 661 * size_in : Size of input payload if any in the PDSM request. 662 * size_out : Size of output payload if any in the PDSM request. 663 * service : Service function for the PDSM request. Return semantics: 664 * rc < 0 : Error servicing PDSM and rc indicates the error. 665 * rc >=0 : Serviced successfully and 'rc' indicate number of 666 * bytes written to payload. 667 */ 668 struct pdsm_cmd_desc { 669 u32 size_in; 670 u32 size_out; 671 int (*service)(struct papr_scm_priv *dimm, 672 union nd_pdsm_payload *payload); 673 }; 674 675 /* Holds all supported PDSMs' command descriptors */ 676 static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = { 677 [PAPR_PDSM_MIN] = { 678 .size_in = 0, 679 .size_out = 0, 680 .service = NULL, 681 }, 682 /* New PDSM command descriptors to be added below */ 683 684 [PAPR_PDSM_HEALTH] = { 685 .size_in = 0, 686 .size_out = sizeof(struct nd_papr_pdsm_health), 687 .service = papr_pdsm_health, 688 }, 689 /* Empty */ 690 [PAPR_PDSM_MAX] = { 691 .size_in = 0, 692 .size_out = 0, 693 .service = NULL, 694 }, 695 }; 696 697 /* Given a valid pdsm cmd return its command descriptor else return NULL */ 698 static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd) 699 { 700 if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors)) 701 return &__pdsm_cmd_descriptors[cmd]; 702 703 return NULL; 704 } 705 706 /* 707 * For a given pdsm request call an appropriate service function. 708 * Returns errors if any while handling the pdsm command package. 709 */ 710 static int papr_scm_service_pdsm(struct papr_scm_priv *p, 711 struct nd_cmd_pkg *pkg) 712 { 713 /* Get the PDSM header and PDSM command */ 714 struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload; 715 enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command; 716 const struct pdsm_cmd_desc *pdsc; 717 int rc; 718 719 /* Fetch corresponding pdsm descriptor for validation and servicing */ 720 pdsc = pdsm_cmd_desc(pdsm); 721 722 /* Validate pdsm descriptor */ 723 /* Ensure that reserved fields are 0 */ 724 if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) { 725 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n", 726 pdsm); 727 return -EINVAL; 728 } 729 730 /* If pdsm expects some input, then ensure that the size_in matches */ 731 if (pdsc->size_in && 732 pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) { 733 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n", 734 pdsm, pkg->nd_size_in); 735 return -EINVAL; 736 } 737 738 /* If pdsm wants to return data, then ensure that size_out matches */ 739 if (pdsc->size_out && 740 pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) { 741 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n", 742 pdsm, pkg->nd_size_out); 743 return -EINVAL; 744 } 745 746 /* Service the pdsm */ 747 if (pdsc->service) { 748 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm); 749 750 rc = pdsc->service(p, &pdsm_pkg->payload); 751 752 if (rc < 0) { 753 /* error encountered while servicing pdsm */ 754 pdsm_pkg->cmd_status = rc; 755 pkg->nd_fw_size = ND_PDSM_HDR_SIZE; 756 } else { 757 /* pdsm serviced and 'rc' bytes written to payload */ 758 pdsm_pkg->cmd_status = 0; 759 pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc; 760 } 761 } else { 762 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n", 763 pdsm); 764 pdsm_pkg->cmd_status = -ENOENT; 765 pkg->nd_fw_size = ND_PDSM_HDR_SIZE; 766 } 767 768 return pdsm_pkg->cmd_status; 769 } 770 771 static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc, 772 struct nvdimm *nvdimm, unsigned int cmd, void *buf, 773 unsigned int buf_len, int *cmd_rc) 774 { 775 struct nd_cmd_get_config_size *get_size_hdr; 776 struct nd_cmd_pkg *call_pkg = NULL; 777 struct papr_scm_priv *p; 778 int rc; 779 780 rc = is_cmd_valid(nvdimm, cmd, buf, buf_len); 781 if (rc) { 782 pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc); 783 return rc; 784 } 785 786 /* Use a local variable in case cmd_rc pointer is NULL */ 787 if (!cmd_rc) 788 cmd_rc = &rc; 789 790 p = nvdimm_provider_data(nvdimm); 791 792 switch (cmd) { 793 case ND_CMD_GET_CONFIG_SIZE: 794 get_size_hdr = buf; 795 796 get_size_hdr->status = 0; 797 get_size_hdr->max_xfer = 8; 798 get_size_hdr->config_size = p->metadata_size; 799 *cmd_rc = 0; 800 break; 801 802 case ND_CMD_GET_CONFIG_DATA: 803 *cmd_rc = papr_scm_meta_get(p, buf); 804 break; 805 806 case ND_CMD_SET_CONFIG_DATA: 807 *cmd_rc = papr_scm_meta_set(p, buf); 808 break; 809 810 case ND_CMD_CALL: 811 call_pkg = (struct nd_cmd_pkg *)buf; 812 *cmd_rc = papr_scm_service_pdsm(p, call_pkg); 813 break; 814 815 default: 816 dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd); 817 return -EINVAL; 818 } 819 820 dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc); 821 822 return 0; 823 } 824 825 static ssize_t perf_stats_show(struct device *dev, 826 struct device_attribute *attr, char *buf) 827 { 828 int index; 829 ssize_t rc; 830 struct seq_buf s; 831 struct papr_scm_perf_stat *stat; 832 struct papr_scm_perf_stats *stats; 833 struct nvdimm *dimm = to_nvdimm(dev); 834 struct papr_scm_priv *p = nvdimm_provider_data(dimm); 835 836 if (!p->stat_buffer_len) 837 return -ENOENT; 838 839 /* Allocate the buffer for phyp where stats are written */ 840 stats = kzalloc(p->stat_buffer_len, GFP_KERNEL); 841 if (!stats) 842 return -ENOMEM; 843 844 /* Ask phyp to return all dimm perf stats */ 845 rc = drc_pmem_query_stats(p, stats, 0); 846 if (rc) 847 goto free_stats; 848 /* 849 * Go through the returned output buffer and print stats and 850 * values. Since stat_id is essentially a char string of 851 * 8 bytes, simply use the string format specifier to print it. 852 */ 853 seq_buf_init(&s, buf, PAGE_SIZE); 854 for (index = 0, stat = stats->scm_statistic; 855 index < be32_to_cpu(stats->num_statistics); 856 ++index, ++stat) { 857 seq_buf_printf(&s, "%.8s = 0x%016llX\n", 858 stat->stat_id, 859 be64_to_cpu(stat->stat_val)); 860 } 861 862 free_stats: 863 kfree(stats); 864 return rc ? rc : (ssize_t)seq_buf_used(&s); 865 } 866 static DEVICE_ATTR_ADMIN_RO(perf_stats); 867 868 static ssize_t flags_show(struct device *dev, 869 struct device_attribute *attr, char *buf) 870 { 871 struct nvdimm *dimm = to_nvdimm(dev); 872 struct papr_scm_priv *p = nvdimm_provider_data(dimm); 873 struct seq_buf s; 874 u64 health; 875 int rc; 876 877 rc = drc_pmem_query_health(p); 878 if (rc) 879 return rc; 880 881 /* Copy health_bitmap locally, check masks & update out buffer */ 882 health = READ_ONCE(p->health_bitmap); 883 884 seq_buf_init(&s, buf, PAGE_SIZE); 885 if (health & PAPR_PMEM_UNARMED_MASK) 886 seq_buf_printf(&s, "not_armed "); 887 888 if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK) 889 seq_buf_printf(&s, "flush_fail "); 890 891 if (health & PAPR_PMEM_BAD_RESTORE_MASK) 892 seq_buf_printf(&s, "restore_fail "); 893 894 if (health & PAPR_PMEM_ENCRYPTED) 895 seq_buf_printf(&s, "encrypted "); 896 897 if (health & PAPR_PMEM_SMART_EVENT_MASK) 898 seq_buf_printf(&s, "smart_notify "); 899 900 if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED) 901 seq_buf_printf(&s, "scrubbed locked "); 902 903 if (seq_buf_used(&s)) 904 seq_buf_printf(&s, "\n"); 905 906 return seq_buf_used(&s); 907 } 908 DEVICE_ATTR_RO(flags); 909 910 /* papr_scm specific dimm attributes */ 911 static struct attribute *papr_nd_attributes[] = { 912 &dev_attr_flags.attr, 913 &dev_attr_perf_stats.attr, 914 NULL, 915 }; 916 917 static struct attribute_group papr_nd_attribute_group = { 918 .name = "papr", 919 .attrs = papr_nd_attributes, 920 }; 921 922 static const struct attribute_group *papr_nd_attr_groups[] = { 923 &papr_nd_attribute_group, 924 NULL, 925 }; 926 927 static int papr_scm_nvdimm_init(struct papr_scm_priv *p) 928 { 929 struct device *dev = &p->pdev->dev; 930 struct nd_mapping_desc mapping; 931 struct nd_region_desc ndr_desc; 932 unsigned long dimm_flags; 933 int target_nid, online_nid; 934 ssize_t stat_size; 935 936 p->bus_desc.ndctl = papr_scm_ndctl; 937 p->bus_desc.module = THIS_MODULE; 938 p->bus_desc.of_node = p->pdev->dev.of_node; 939 p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL); 940 941 /* Set the dimm command family mask to accept PDSMs */ 942 set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask); 943 944 if (!p->bus_desc.provider_name) 945 return -ENOMEM; 946 947 p->bus = nvdimm_bus_register(NULL, &p->bus_desc); 948 if (!p->bus) { 949 dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn); 950 kfree(p->bus_desc.provider_name); 951 return -ENXIO; 952 } 953 954 dimm_flags = 0; 955 set_bit(NDD_LABELING, &dimm_flags); 956 957 /* 958 * Check if the nvdimm is unarmed. No locking needed as we are still 959 * initializing. Ignore error encountered if any. 960 */ 961 __drc_pmem_query_health(p); 962 963 if (p->health_bitmap & PAPR_PMEM_UNARMED_MASK) 964 set_bit(NDD_UNARMED, &dimm_flags); 965 966 p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups, 967 dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL); 968 if (!p->nvdimm) { 969 dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn); 970 goto err; 971 } 972 973 if (nvdimm_bus_check_dimm_count(p->bus, 1)) 974 goto err; 975 976 /* now add the region */ 977 978 memset(&mapping, 0, sizeof(mapping)); 979 mapping.nvdimm = p->nvdimm; 980 mapping.start = 0; 981 mapping.size = p->blocks * p->block_size; // XXX: potential overflow? 982 983 memset(&ndr_desc, 0, sizeof(ndr_desc)); 984 target_nid = dev_to_node(&p->pdev->dev); 985 online_nid = numa_map_to_online_node(target_nid); 986 ndr_desc.numa_node = online_nid; 987 ndr_desc.target_node = target_nid; 988 ndr_desc.res = &p->res; 989 ndr_desc.of_node = p->dn; 990 ndr_desc.provider_data = p; 991 ndr_desc.mapping = &mapping; 992 ndr_desc.num_mappings = 1; 993 ndr_desc.nd_set = &p->nd_set; 994 995 if (p->hcall_flush_required) { 996 set_bit(ND_REGION_ASYNC, &ndr_desc.flags); 997 ndr_desc.flush = papr_scm_pmem_flush; 998 } 999 1000 if (p->is_volatile) 1001 p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc); 1002 else { 1003 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags); 1004 p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc); 1005 } 1006 if (!p->region) { 1007 dev_err(dev, "Error registering region %pR from %pOF\n", 1008 ndr_desc.res, p->dn); 1009 goto err; 1010 } 1011 if (target_nid != online_nid) 1012 dev_info(dev, "Region registered with target node %d and online node %d", 1013 target_nid, online_nid); 1014 1015 mutex_lock(&papr_ndr_lock); 1016 list_add_tail(&p->region_list, &papr_nd_regions); 1017 mutex_unlock(&papr_ndr_lock); 1018 1019 /* Try retriving the stat buffer and see if its supported */ 1020 stat_size = drc_pmem_query_stats(p, NULL, 0); 1021 if (stat_size > 0) { 1022 p->stat_buffer_len = stat_size; 1023 dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n", 1024 p->stat_buffer_len); 1025 } else { 1026 dev_info(&p->pdev->dev, "Dimm performance stats unavailable\n"); 1027 } 1028 1029 return 0; 1030 1031 err: nvdimm_bus_unregister(p->bus); 1032 kfree(p->bus_desc.provider_name); 1033 return -ENXIO; 1034 } 1035 1036 static void papr_scm_add_badblock(struct nd_region *region, 1037 struct nvdimm_bus *bus, u64 phys_addr) 1038 { 1039 u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES); 1040 1041 if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) { 1042 pr_err("Bad block registration for 0x%llx failed\n", phys_addr); 1043 return; 1044 } 1045 1046 pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n", 1047 aligned_addr, aligned_addr + L1_CACHE_BYTES); 1048 1049 nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON); 1050 } 1051 1052 static int handle_mce_ue(struct notifier_block *nb, unsigned long val, 1053 void *data) 1054 { 1055 struct machine_check_event *evt = data; 1056 struct papr_scm_priv *p; 1057 u64 phys_addr; 1058 bool found = false; 1059 1060 if (evt->error_type != MCE_ERROR_TYPE_UE) 1061 return NOTIFY_DONE; 1062 1063 if (list_empty(&papr_nd_regions)) 1064 return NOTIFY_DONE; 1065 1066 /* 1067 * The physical address obtained here is PAGE_SIZE aligned, so get the 1068 * exact address from the effective address 1069 */ 1070 phys_addr = evt->u.ue_error.physical_address + 1071 (evt->u.ue_error.effective_address & ~PAGE_MASK); 1072 1073 if (!evt->u.ue_error.physical_address_provided || 1074 !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT))) 1075 return NOTIFY_DONE; 1076 1077 /* mce notifier is called from a process context, so mutex is safe */ 1078 mutex_lock(&papr_ndr_lock); 1079 list_for_each_entry(p, &papr_nd_regions, region_list) { 1080 if (phys_addr >= p->res.start && phys_addr <= p->res.end) { 1081 found = true; 1082 break; 1083 } 1084 } 1085 1086 if (found) 1087 papr_scm_add_badblock(p->region, p->bus, phys_addr); 1088 1089 mutex_unlock(&papr_ndr_lock); 1090 1091 return found ? NOTIFY_OK : NOTIFY_DONE; 1092 } 1093 1094 static struct notifier_block mce_ue_nb = { 1095 .notifier_call = handle_mce_ue 1096 }; 1097 1098 static int papr_scm_probe(struct platform_device *pdev) 1099 { 1100 struct device_node *dn = pdev->dev.of_node; 1101 u32 drc_index, metadata_size; 1102 u64 blocks, block_size; 1103 struct papr_scm_priv *p; 1104 const char *uuid_str; 1105 u64 uuid[2]; 1106 int rc; 1107 1108 /* check we have all the required DT properties */ 1109 if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) { 1110 dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn); 1111 return -ENODEV; 1112 } 1113 1114 if (of_property_read_u64(dn, "ibm,block-size", &block_size)) { 1115 dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn); 1116 return -ENODEV; 1117 } 1118 1119 if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) { 1120 dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn); 1121 return -ENODEV; 1122 } 1123 1124 if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) { 1125 dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn); 1126 return -ENODEV; 1127 } 1128 1129 1130 p = kzalloc(sizeof(*p), GFP_KERNEL); 1131 if (!p) 1132 return -ENOMEM; 1133 1134 /* Initialize the dimm mutex */ 1135 mutex_init(&p->health_mutex); 1136 1137 /* optional DT properties */ 1138 of_property_read_u32(dn, "ibm,metadata-size", &metadata_size); 1139 1140 p->dn = dn; 1141 p->drc_index = drc_index; 1142 p->block_size = block_size; 1143 p->blocks = blocks; 1144 p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required"); 1145 p->hcall_flush_required = of_property_read_bool(dn, "ibm,hcall-flush-required"); 1146 1147 /* We just need to ensure that set cookies are unique across */ 1148 uuid_parse(uuid_str, (uuid_t *) uuid); 1149 /* 1150 * cookie1 and cookie2 are not really little endian 1151 * we store a little endian representation of the 1152 * uuid str so that we can compare this with the label 1153 * area cookie irrespective of the endian config with which 1154 * the kernel is built. 1155 */ 1156 p->nd_set.cookie1 = cpu_to_le64(uuid[0]); 1157 p->nd_set.cookie2 = cpu_to_le64(uuid[1]); 1158 1159 /* might be zero */ 1160 p->metadata_size = metadata_size; 1161 p->pdev = pdev; 1162 1163 /* request the hypervisor to bind this region to somewhere in memory */ 1164 rc = drc_pmem_bind(p); 1165 1166 /* If phyp says drc memory still bound then force unbound and retry */ 1167 if (rc == H_OVERLAP) 1168 rc = drc_pmem_query_n_bind(p); 1169 1170 if (rc != H_SUCCESS) { 1171 dev_err(&p->pdev->dev, "bind err: %d\n", rc); 1172 rc = -ENXIO; 1173 goto err; 1174 } 1175 1176 /* setup the resource for the newly bound range */ 1177 p->res.start = p->bound_addr; 1178 p->res.end = p->bound_addr + p->blocks * p->block_size - 1; 1179 p->res.name = pdev->name; 1180 p->res.flags = IORESOURCE_MEM; 1181 1182 rc = papr_scm_nvdimm_init(p); 1183 if (rc) 1184 goto err2; 1185 1186 platform_set_drvdata(pdev, p); 1187 1188 return 0; 1189 1190 err2: drc_pmem_unbind(p); 1191 err: kfree(p); 1192 return rc; 1193 } 1194 1195 static int papr_scm_remove(struct platform_device *pdev) 1196 { 1197 struct papr_scm_priv *p = platform_get_drvdata(pdev); 1198 1199 mutex_lock(&papr_ndr_lock); 1200 list_del(&p->region_list); 1201 mutex_unlock(&papr_ndr_lock); 1202 1203 nvdimm_bus_unregister(p->bus); 1204 drc_pmem_unbind(p); 1205 kfree(p->bus_desc.provider_name); 1206 kfree(p); 1207 1208 return 0; 1209 } 1210 1211 static const struct of_device_id papr_scm_match[] = { 1212 { .compatible = "ibm,pmemory" }, 1213 { .compatible = "ibm,pmemory-v2" }, 1214 { }, 1215 }; 1216 1217 static struct platform_driver papr_scm_driver = { 1218 .probe = papr_scm_probe, 1219 .remove = papr_scm_remove, 1220 .driver = { 1221 .name = "papr_scm", 1222 .of_match_table = papr_scm_match, 1223 }, 1224 }; 1225 1226 static int __init papr_scm_init(void) 1227 { 1228 int ret; 1229 1230 ret = platform_driver_register(&papr_scm_driver); 1231 if (!ret) 1232 mce_register_notifier(&mce_ue_nb); 1233 1234 return ret; 1235 } 1236 module_init(papr_scm_init); 1237 1238 static void __exit papr_scm_exit(void) 1239 { 1240 mce_unregister_notifier(&mce_ue_nb); 1241 platform_driver_unregister(&papr_scm_driver); 1242 } 1243 module_exit(papr_scm_exit); 1244 1245 MODULE_DEVICE_TABLE(of, papr_scm_match); 1246 MODULE_LICENSE("GPL"); 1247 MODULE_AUTHOR("IBM Corporation"); 1248