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 23#define BIND_ANY_ADDR (~0ul) 24 25#define PAPR_SCM_DIMM_CMD_MASK \ 26 ((1ul << ND_CMD_GET_CONFIG_SIZE) | \ 27 (1ul << ND_CMD_GET_CONFIG_DATA) | \ 28 (1ul << ND_CMD_SET_CONFIG_DATA) | \ 29 (1ul << ND_CMD_CALL)) 30 31/* DIMM health bitmap bitmap indicators */ 32/* SCM device is unable to persist memory contents */ 33#define PAPR_PMEM_UNARMED (1ULL << (63 - 0)) 34/* SCM device failed to persist memory contents */ 35#define PAPR_PMEM_SHUTDOWN_DIRTY (1ULL << (63 - 1)) 36/* SCM device contents are persisted from previous IPL */ 37#define PAPR_PMEM_SHUTDOWN_CLEAN (1ULL << (63 - 2)) 38/* SCM device contents are not persisted from previous IPL */ 39#define PAPR_PMEM_EMPTY (1ULL << (63 - 3)) 40/* SCM device memory life remaining is critically low */ 41#define PAPR_PMEM_HEALTH_CRITICAL (1ULL << (63 - 4)) 42/* SCM device will be garded off next IPL due to failure */ 43#define PAPR_PMEM_HEALTH_FATAL (1ULL << (63 - 5)) 44/* SCM contents cannot persist due to current platform health status */ 45#define PAPR_PMEM_HEALTH_UNHEALTHY (1ULL << (63 - 6)) 46/* SCM device is unable to persist memory contents in certain conditions */ 47#define PAPR_PMEM_HEALTH_NON_CRITICAL (1ULL << (63 - 7)) 48/* SCM device is encrypted */ 49#define PAPR_PMEM_ENCRYPTED (1ULL << (63 - 8)) 50/* SCM device has been scrubbed and locked */ 51#define PAPR_PMEM_SCRUBBED_AND_LOCKED (1ULL << (63 - 9)) 52 53/* Bits status indicators for health bitmap indicating unarmed dimm */ 54#define PAPR_PMEM_UNARMED_MASK (PAPR_PMEM_UNARMED | \ 55 PAPR_PMEM_HEALTH_UNHEALTHY) 56 57/* Bits status indicators for health bitmap indicating unflushed dimm */ 58#define PAPR_PMEM_BAD_SHUTDOWN_MASK (PAPR_PMEM_SHUTDOWN_DIRTY) 59 60/* Bits status indicators for health bitmap indicating unrestored dimm */ 61#define PAPR_PMEM_BAD_RESTORE_MASK (PAPR_PMEM_EMPTY) 62 63/* Bit status indicators for smart event notification */ 64#define PAPR_PMEM_SMART_EVENT_MASK (PAPR_PMEM_HEALTH_CRITICAL | \ 65 PAPR_PMEM_HEALTH_FATAL | \ 66 PAPR_PMEM_HEALTH_UNHEALTHY) 67 68#define PAPR_SCM_PERF_STATS_EYECATCHER __stringify(SCMSTATS) 69#define PAPR_SCM_PERF_STATS_VERSION 0x1 70 71/* Struct holding a single performance metric */ 72struct papr_scm_perf_stat { 73 u8 stat_id[8]; 74 __be64 stat_val; 75} __packed; 76 77/* Struct exchanged between kernel and PHYP for fetching drc perf stats */ 78struct papr_scm_perf_stats { 79 u8 eye_catcher[8]; 80 /* Should be PAPR_SCM_PERF_STATS_VERSION */ 81 __be32 stats_version; 82 /* Number of stats following */ 83 __be32 num_statistics; 84 /* zero or more performance matrics */ 85 struct papr_scm_perf_stat scm_statistic[]; 86} __packed; 87 88/* private struct associated with each region */ 89struct papr_scm_priv { 90 struct platform_device *pdev; 91 struct device_node *dn; 92 uint32_t drc_index; 93 uint64_t blocks; 94 uint64_t block_size; 95 int metadata_size; 96 bool is_volatile; 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 121static LIST_HEAD(papr_nd_regions); 122static DEFINE_MUTEX(papr_ndr_lock); 123 124static int drc_pmem_bind(struct papr_scm_priv *p) 125{ 126 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 127 uint64_t saved = 0; 128 uint64_t token; 129 int64_t rc; 130 131 /* 132 * When the hypervisor cannot map all the requested memory in a single 133 * hcall it returns H_BUSY and we call again with the token until 134 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS 135 * leave the system in an undefined state, so we wait. 136 */ 137 token = 0; 138 139 do { 140 rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0, 141 p->blocks, BIND_ANY_ADDR, token); 142 token = ret[0]; 143 if (!saved) 144 saved = ret[1]; 145 cond_resched(); 146 } while (rc == H_BUSY); 147 148 if (rc) 149 return rc; 150 151 p->bound_addr = saved; 152 dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", 153 p->drc_index, (unsigned long)saved); 154 return rc; 155} 156 157static void drc_pmem_unbind(struct papr_scm_priv *p) 158{ 159 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 160 uint64_t token = 0; 161 int64_t rc; 162 163 dev_dbg(&p->pdev->dev, "unbind drc 0x%x\n", p->drc_index); 164 165 /* NB: unbind has the same retry requirements as drc_pmem_bind() */ 166 do { 167 168 /* Unbind of all SCM resources associated with drcIndex */ 169 rc = plpar_hcall(H_SCM_UNBIND_ALL, ret, H_UNBIND_SCOPE_DRC, 170 p->drc_index, token); 171 token = ret[0]; 172 173 /* Check if we are stalled for some time */ 174 if (H_IS_LONG_BUSY(rc)) { 175 msleep(get_longbusy_msecs(rc)); 176 rc = H_BUSY; 177 } else if (rc == H_BUSY) { 178 cond_resched(); 179 } 180 181 } while (rc == H_BUSY); 182 183 if (rc) 184 dev_err(&p->pdev->dev, "unbind error: %lld\n", rc); 185 else 186 dev_dbg(&p->pdev->dev, "unbind drc 0x%x complete\n", 187 p->drc_index); 188 189 return; 190} 191 192static int drc_pmem_query_n_bind(struct papr_scm_priv *p) 193{ 194 unsigned long start_addr; 195 unsigned long end_addr; 196 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 197 int64_t rc; 198 199 200 rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret, 201 p->drc_index, 0); 202 if (rc) 203 goto err_out; 204 start_addr = ret[0]; 205 206 /* Make sure the full region is bound. */ 207 rc = plpar_hcall(H_SCM_QUERY_BLOCK_MEM_BINDING, ret, 208 p->drc_index, p->blocks - 1); 209 if (rc) 210 goto err_out; 211 end_addr = ret[0]; 212 213 if ((end_addr - start_addr) != ((p->blocks - 1) * p->block_size)) 214 goto err_out; 215 216 p->bound_addr = start_addr; 217 dev_dbg(&p->pdev->dev, "bound drc 0x%x to 0x%lx\n", p->drc_index, start_addr); 218 return rc; 219 220err_out: 221 dev_info(&p->pdev->dev, 222 "Failed to query, trying an unbind followed by bind"); 223 drc_pmem_unbind(p); 224 return drc_pmem_bind(p); 225} 226 227/* 228 * Query the Dimm performance stats from PHYP and copy them (if returned) to 229 * provided struct papr_scm_perf_stats instance 'stats' that can hold atleast 230 * (num_stats + header) bytes. 231 * - If buff_stats == NULL the return value is the size in byes of the buffer 232 * needed to hold all supported performance-statistics. 233 * - If buff_stats != NULL and num_stats == 0 then we copy all known 234 * performance-statistics to 'buff_stat' and expect to be large enough to 235 * hold them. 236 * - if buff_stats != NULL and num_stats > 0 then copy the requested 237 * performance-statistics to buff_stats. 238 */ 239static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p, 240 struct papr_scm_perf_stats *buff_stats, 241 unsigned int num_stats) 242{ 243 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 244 size_t size; 245 s64 rc; 246 247 /* Setup the out buffer */ 248 if (buff_stats) { 249 memcpy(buff_stats->eye_catcher, 250 PAPR_SCM_PERF_STATS_EYECATCHER, 8); 251 buff_stats->stats_version = 252 cpu_to_be32(PAPR_SCM_PERF_STATS_VERSION); 253 buff_stats->num_statistics = 254 cpu_to_be32(num_stats); 255 256 /* 257 * Calculate the buffer size based on num-stats provided 258 * or use the prefetched max buffer length 259 */ 260 if (num_stats) 261 /* Calculate size from the num_stats */ 262 size = sizeof(struct papr_scm_perf_stats) + 263 num_stats * sizeof(struct papr_scm_perf_stat); 264 else 265 size = p->stat_buffer_len; 266 } else { 267 /* In case of no out buffer ignore the size */ 268 size = 0; 269 } 270 271 /* Do the HCALL asking PHYP for info */ 272 rc = plpar_hcall(H_SCM_PERFORMANCE_STATS, ret, p->drc_index, 273 buff_stats ? virt_to_phys(buff_stats) : 0, 274 size); 275 276 /* Check if the error was due to an unknown stat-id */ 277 if (rc == H_PARTIAL) { 278 dev_err(&p->pdev->dev, 279 "Unknown performance stats, Err:0x%016lX\n", ret[0]); 280 return -ENOENT; 281 } else if (rc != H_SUCCESS) { 282 dev_err(&p->pdev->dev, 283 "Failed to query performance stats, Err:%lld\n", rc); 284 return -EIO; 285 286 } else if (!size) { 287 /* Handle case where stat buffer size was requested */ 288 dev_dbg(&p->pdev->dev, 289 "Performance stats size %ld\n", ret[0]); 290 return ret[0]; 291 } 292 293 /* Successfully fetched the requested stats from phyp */ 294 dev_dbg(&p->pdev->dev, 295 "Performance stats returned %d stats\n", 296 be32_to_cpu(buff_stats->num_statistics)); 297 return 0; 298} 299 300/* 301 * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the 302 * health information. 303 */ 304static int __drc_pmem_query_health(struct papr_scm_priv *p) 305{ 306 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 307 long rc; 308 309 /* issue the hcall */ 310 rc = plpar_hcall(H_SCM_HEALTH, ret, p->drc_index); 311 if (rc != H_SUCCESS) { 312 dev_err(&p->pdev->dev, 313 "Failed to query health information, Err:%ld\n", rc); 314 return -ENXIO; 315 } 316 317 p->lasthealth_jiffies = jiffies; 318 p->health_bitmap = ret[0] & ret[1]; 319 320 dev_dbg(&p->pdev->dev, 321 "Queried dimm health info. Bitmap:0x%016lx Mask:0x%016lx\n", 322 ret[0], ret[1]); 323 324 return 0; 325} 326 327/* Min interval in seconds for assuming stable dimm health */ 328#define MIN_HEALTH_QUERY_INTERVAL 60 329 330/* Query cached health info and if needed call drc_pmem_query_health */ 331static int drc_pmem_query_health(struct papr_scm_priv *p) 332{ 333 unsigned long cache_timeout; 334 int rc; 335 336 /* Protect concurrent modifications to papr_scm_priv */ 337 rc = mutex_lock_interruptible(&p->health_mutex); 338 if (rc) 339 return rc; 340 341 /* Jiffies offset for which the health data is assumed to be same */ 342 cache_timeout = p->lasthealth_jiffies + 343 msecs_to_jiffies(MIN_HEALTH_QUERY_INTERVAL * 1000); 344 345 /* Fetch new health info is its older than MIN_HEALTH_QUERY_INTERVAL */ 346 if (time_after(jiffies, cache_timeout)) 347 rc = __drc_pmem_query_health(p); 348 else 349 /* Assume cached health data is valid */ 350 rc = 0; 351 352 mutex_unlock(&p->health_mutex); 353 return rc; 354} 355 356static int papr_scm_meta_get(struct papr_scm_priv *p, 357 struct nd_cmd_get_config_data_hdr *hdr) 358{ 359 unsigned long data[PLPAR_HCALL_BUFSIZE]; 360 unsigned long offset, data_offset; 361 int len, read; 362 int64_t ret; 363 364 if ((hdr->in_offset + hdr->in_length) > p->metadata_size) 365 return -EINVAL; 366 367 for (len = hdr->in_length; len; len -= read) { 368 369 data_offset = hdr->in_length - len; 370 offset = hdr->in_offset + data_offset; 371 372 if (len >= 8) 373 read = 8; 374 else if (len >= 4) 375 read = 4; 376 else if (len >= 2) 377 read = 2; 378 else 379 read = 1; 380 381 ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index, 382 offset, read); 383 384 if (ret == H_PARAMETER) /* bad DRC index */ 385 return -ENODEV; 386 if (ret) 387 return -EINVAL; /* other invalid parameter */ 388 389 switch (read) { 390 case 8: 391 *(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]); 392 break; 393 case 4: 394 *(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff); 395 break; 396 397 case 2: 398 *(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff); 399 break; 400 401 case 1: 402 *(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff); 403 break; 404 } 405 } 406 return 0; 407} 408 409static int papr_scm_meta_set(struct papr_scm_priv *p, 410 struct nd_cmd_set_config_hdr *hdr) 411{ 412 unsigned long offset, data_offset; 413 int len, wrote; 414 unsigned long data; 415 __be64 data_be; 416 int64_t ret; 417 418 if ((hdr->in_offset + hdr->in_length) > p->metadata_size) 419 return -EINVAL; 420 421 for (len = hdr->in_length; len; len -= wrote) { 422 423 data_offset = hdr->in_length - len; 424 offset = hdr->in_offset + data_offset; 425 426 if (len >= 8) { 427 data = *(uint64_t *)(hdr->in_buf + data_offset); 428 data_be = cpu_to_be64(data); 429 wrote = 8; 430 } else if (len >= 4) { 431 data = *(uint32_t *)(hdr->in_buf + data_offset); 432 data &= 0xffffffff; 433 data_be = cpu_to_be32(data); 434 wrote = 4; 435 } else if (len >= 2) { 436 data = *(uint16_t *)(hdr->in_buf + data_offset); 437 data &= 0xffff; 438 data_be = cpu_to_be16(data); 439 wrote = 2; 440 } else { 441 data_be = *(uint8_t *)(hdr->in_buf + data_offset); 442 data_be &= 0xff; 443 wrote = 1; 444 } 445 446 ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index, 447 offset, data_be, wrote); 448 if (ret == H_PARAMETER) /* bad DRC index */ 449 return -ENODEV; 450 if (ret) 451 return -EINVAL; /* other invalid parameter */ 452 } 453 454 return 0; 455} 456 457/* 458 * Do a sanity checks on the inputs args to dimm-control function and return 459 * '0' if valid. Validation of PDSM payloads happens later in 460 * papr_scm_service_pdsm. 461 */ 462static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf, 463 unsigned int buf_len) 464{ 465 unsigned long cmd_mask = PAPR_SCM_DIMM_CMD_MASK; 466 struct nd_cmd_pkg *nd_cmd; 467 struct papr_scm_priv *p; 468 enum papr_pdsm pdsm; 469 470 /* Only dimm-specific calls are supported atm */ 471 if (!nvdimm) 472 return -EINVAL; 473 474 /* get the provider data from struct nvdimm */ 475 p = nvdimm_provider_data(nvdimm); 476 477 if (!test_bit(cmd, &cmd_mask)) { 478 dev_dbg(&p->pdev->dev, "Unsupported cmd=%u\n", cmd); 479 return -EINVAL; 480 } 481 482 /* For CMD_CALL verify pdsm request */ 483 if (cmd == ND_CMD_CALL) { 484 /* Verify the envelope and envelop size */ 485 if (!buf || 486 buf_len < (sizeof(struct nd_cmd_pkg) + ND_PDSM_HDR_SIZE)) { 487 dev_dbg(&p->pdev->dev, "Invalid pkg size=%u\n", 488 buf_len); 489 return -EINVAL; 490 } 491 492 /* Verify that the nd_cmd_pkg.nd_family is correct */ 493 nd_cmd = (struct nd_cmd_pkg *)buf; 494 495 if (nd_cmd->nd_family != NVDIMM_FAMILY_PAPR) { 496 dev_dbg(&p->pdev->dev, "Invalid pkg family=0x%llx\n", 497 nd_cmd->nd_family); 498 return -EINVAL; 499 } 500 501 pdsm = (enum papr_pdsm)nd_cmd->nd_command; 502 503 /* Verify if the pdsm command is valid */ 504 if (pdsm <= PAPR_PDSM_MIN || pdsm >= PAPR_PDSM_MAX) { 505 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid PDSM\n", 506 pdsm); 507 return -EINVAL; 508 } 509 510 /* Have enough space to hold returned 'nd_pkg_pdsm' header */ 511 if (nd_cmd->nd_size_out < ND_PDSM_HDR_SIZE) { 512 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid payload\n", 513 pdsm); 514 return -EINVAL; 515 } 516 } 517 518 /* Let the command be further processed */ 519 return 0; 520} 521 522static int papr_pdsm_fuel_gauge(struct papr_scm_priv *p, 523 union nd_pdsm_payload *payload) 524{ 525 int rc, size; 526 u64 statval; 527 struct papr_scm_perf_stat *stat; 528 struct papr_scm_perf_stats *stats; 529 530 /* Silently fail if fetching performance metrics isn't supported */ 531 if (!p->stat_buffer_len) 532 return 0; 533 534 /* Allocate request buffer enough to hold single performance stat */ 535 size = sizeof(struct papr_scm_perf_stats) + 536 sizeof(struct papr_scm_perf_stat); 537 538 stats = kzalloc(size, GFP_KERNEL); 539 if (!stats) 540 return -ENOMEM; 541 542 stat = &stats->scm_statistic[0]; 543 memcpy(&stat->stat_id, "MemLife ", sizeof(stat->stat_id)); 544 stat->stat_val = 0; 545 546 /* Fetch the fuel gauge and populate it in payload */ 547 rc = drc_pmem_query_stats(p, stats, 1); 548 if (rc < 0) { 549 dev_dbg(&p->pdev->dev, "Err(%d) fetching fuel gauge\n", rc); 550 goto free_stats; 551 } 552 553 statval = be64_to_cpu(stat->stat_val); 554 dev_dbg(&p->pdev->dev, 555 "Fetched fuel-gauge %llu", statval); 556 payload->health.extension_flags |= 557 PDSM_DIMM_HEALTH_RUN_GAUGE_VALID; 558 payload->health.dimm_fuel_gauge = statval; 559 560 rc = sizeof(struct nd_papr_pdsm_health); 561 562free_stats: 563 kfree(stats); 564 return rc; 565} 566 567/* Fetch the DIMM health info and populate it in provided package. */ 568static int papr_pdsm_health(struct papr_scm_priv *p, 569 union nd_pdsm_payload *payload) 570{ 571 int rc; 572 573 /* Ensure dimm health mutex is taken preventing concurrent access */ 574 rc = mutex_lock_interruptible(&p->health_mutex); 575 if (rc) 576 goto out; 577 578 /* Always fetch upto date dimm health data ignoring cached values */ 579 rc = __drc_pmem_query_health(p); 580 if (rc) { 581 mutex_unlock(&p->health_mutex); 582 goto out; 583 } 584 585 /* update health struct with various flags derived from health bitmap */ 586 payload->health = (struct nd_papr_pdsm_health) { 587 .extension_flags = 0, 588 .dimm_unarmed = !!(p->health_bitmap & PAPR_PMEM_UNARMED_MASK), 589 .dimm_bad_shutdown = !!(p->health_bitmap & PAPR_PMEM_BAD_SHUTDOWN_MASK), 590 .dimm_bad_restore = !!(p->health_bitmap & PAPR_PMEM_BAD_RESTORE_MASK), 591 .dimm_scrubbed = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED), 592 .dimm_locked = !!(p->health_bitmap & PAPR_PMEM_SCRUBBED_AND_LOCKED), 593 .dimm_encrypted = !!(p->health_bitmap & PAPR_PMEM_ENCRYPTED), 594 .dimm_health = PAPR_PDSM_DIMM_HEALTHY, 595 }; 596 597 /* Update field dimm_health based on health_bitmap flags */ 598 if (p->health_bitmap & PAPR_PMEM_HEALTH_FATAL) 599 payload->health.dimm_health = PAPR_PDSM_DIMM_FATAL; 600 else if (p->health_bitmap & PAPR_PMEM_HEALTH_CRITICAL) 601 payload->health.dimm_health = PAPR_PDSM_DIMM_CRITICAL; 602 else if (p->health_bitmap & PAPR_PMEM_HEALTH_UNHEALTHY) 603 payload->health.dimm_health = PAPR_PDSM_DIMM_UNHEALTHY; 604 605 /* struct populated hence can release the mutex now */ 606 mutex_unlock(&p->health_mutex); 607 608 /* Populate the fuel gauge meter in the payload */ 609 papr_pdsm_fuel_gauge(p, payload); 610 611 rc = sizeof(struct nd_papr_pdsm_health); 612 613out: 614 return rc; 615} 616 617/* 618 * 'struct pdsm_cmd_desc' 619 * Identifies supported PDSMs' expected length of in/out payloads 620 * and pdsm service function. 621 * 622 * size_in : Size of input payload if any in the PDSM request. 623 * size_out : Size of output payload if any in the PDSM request. 624 * service : Service function for the PDSM request. Return semantics: 625 * rc < 0 : Error servicing PDSM and rc indicates the error. 626 * rc >=0 : Serviced successfully and 'rc' indicate number of 627 * bytes written to payload. 628 */ 629struct pdsm_cmd_desc { 630 u32 size_in; 631 u32 size_out; 632 int (*service)(struct papr_scm_priv *dimm, 633 union nd_pdsm_payload *payload); 634}; 635 636/* Holds all supported PDSMs' command descriptors */ 637static const struct pdsm_cmd_desc __pdsm_cmd_descriptors[] = { 638 [PAPR_PDSM_MIN] = { 639 .size_in = 0, 640 .size_out = 0, 641 .service = NULL, 642 }, 643 /* New PDSM command descriptors to be added below */ 644 645 [PAPR_PDSM_HEALTH] = { 646 .size_in = 0, 647 .size_out = sizeof(struct nd_papr_pdsm_health), 648 .service = papr_pdsm_health, 649 }, 650 /* Empty */ 651 [PAPR_PDSM_MAX] = { 652 .size_in = 0, 653 .size_out = 0, 654 .service = NULL, 655 }, 656}; 657 658/* Given a valid pdsm cmd return its command descriptor else return NULL */ 659static inline const struct pdsm_cmd_desc *pdsm_cmd_desc(enum papr_pdsm cmd) 660{ 661 if (cmd >= 0 || cmd < ARRAY_SIZE(__pdsm_cmd_descriptors)) 662 return &__pdsm_cmd_descriptors[cmd]; 663 664 return NULL; 665} 666 667/* 668 * For a given pdsm request call an appropriate service function. 669 * Returns errors if any while handling the pdsm command package. 670 */ 671static int papr_scm_service_pdsm(struct papr_scm_priv *p, 672 struct nd_cmd_pkg *pkg) 673{ 674 /* Get the PDSM header and PDSM command */ 675 struct nd_pkg_pdsm *pdsm_pkg = (struct nd_pkg_pdsm *)pkg->nd_payload; 676 enum papr_pdsm pdsm = (enum papr_pdsm)pkg->nd_command; 677 const struct pdsm_cmd_desc *pdsc; 678 int rc; 679 680 /* Fetch corresponding pdsm descriptor for validation and servicing */ 681 pdsc = pdsm_cmd_desc(pdsm); 682 683 /* Validate pdsm descriptor */ 684 /* Ensure that reserved fields are 0 */ 685 if (pdsm_pkg->reserved[0] || pdsm_pkg->reserved[1]) { 686 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Invalid reserved field\n", 687 pdsm); 688 return -EINVAL; 689 } 690 691 /* If pdsm expects some input, then ensure that the size_in matches */ 692 if (pdsc->size_in && 693 pkg->nd_size_in != (pdsc->size_in + ND_PDSM_HDR_SIZE)) { 694 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_in=%d\n", 695 pdsm, pkg->nd_size_in); 696 return -EINVAL; 697 } 698 699 /* If pdsm wants to return data, then ensure that size_out matches */ 700 if (pdsc->size_out && 701 pkg->nd_size_out != (pdsc->size_out + ND_PDSM_HDR_SIZE)) { 702 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Mismatched size_out=%d\n", 703 pdsm, pkg->nd_size_out); 704 return -EINVAL; 705 } 706 707 /* Service the pdsm */ 708 if (pdsc->service) { 709 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Servicing..\n", pdsm); 710 711 rc = pdsc->service(p, &pdsm_pkg->payload); 712 713 if (rc < 0) { 714 /* error encountered while servicing pdsm */ 715 pdsm_pkg->cmd_status = rc; 716 pkg->nd_fw_size = ND_PDSM_HDR_SIZE; 717 } else { 718 /* pdsm serviced and 'rc' bytes written to payload */ 719 pdsm_pkg->cmd_status = 0; 720 pkg->nd_fw_size = ND_PDSM_HDR_SIZE + rc; 721 } 722 } else { 723 dev_dbg(&p->pdev->dev, "PDSM[0x%x]: Unsupported PDSM request\n", 724 pdsm); 725 pdsm_pkg->cmd_status = -ENOENT; 726 pkg->nd_fw_size = ND_PDSM_HDR_SIZE; 727 } 728 729 return pdsm_pkg->cmd_status; 730} 731 732static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc, 733 struct nvdimm *nvdimm, unsigned int cmd, void *buf, 734 unsigned int buf_len, int *cmd_rc) 735{ 736 struct nd_cmd_get_config_size *get_size_hdr; 737 struct nd_cmd_pkg *call_pkg = NULL; 738 struct papr_scm_priv *p; 739 int rc; 740 741 rc = is_cmd_valid(nvdimm, cmd, buf, buf_len); 742 if (rc) { 743 pr_debug("Invalid cmd=0x%x. Err=%d\n", cmd, rc); 744 return rc; 745 } 746 747 /* Use a local variable in case cmd_rc pointer is NULL */ 748 if (!cmd_rc) 749 cmd_rc = &rc; 750 751 p = nvdimm_provider_data(nvdimm); 752 753 switch (cmd) { 754 case ND_CMD_GET_CONFIG_SIZE: 755 get_size_hdr = buf; 756 757 get_size_hdr->status = 0; 758 get_size_hdr->max_xfer = 8; 759 get_size_hdr->config_size = p->metadata_size; 760 *cmd_rc = 0; 761 break; 762 763 case ND_CMD_GET_CONFIG_DATA: 764 *cmd_rc = papr_scm_meta_get(p, buf); 765 break; 766 767 case ND_CMD_SET_CONFIG_DATA: 768 *cmd_rc = papr_scm_meta_set(p, buf); 769 break; 770 771 case ND_CMD_CALL: 772 call_pkg = (struct nd_cmd_pkg *)buf; 773 *cmd_rc = papr_scm_service_pdsm(p, call_pkg); 774 break; 775 776 default: 777 dev_dbg(&p->pdev->dev, "Unknown command = %d\n", cmd); 778 return -EINVAL; 779 } 780 781 dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc); 782 783 return 0; 784} 785 786static ssize_t perf_stats_show(struct device *dev, 787 struct device_attribute *attr, char *buf) 788{ 789 int index; 790 ssize_t rc; 791 struct seq_buf s; 792 struct papr_scm_perf_stat *stat; 793 struct papr_scm_perf_stats *stats; 794 struct nvdimm *dimm = to_nvdimm(dev); 795 struct papr_scm_priv *p = nvdimm_provider_data(dimm); 796 797 if (!p->stat_buffer_len) 798 return -ENOENT; 799 800 /* Allocate the buffer for phyp where stats are written */ 801 stats = kzalloc(p->stat_buffer_len, GFP_KERNEL); 802 if (!stats) 803 return -ENOMEM; 804 805 /* Ask phyp to return all dimm perf stats */ 806 rc = drc_pmem_query_stats(p, stats, 0); 807 if (rc) 808 goto free_stats; 809 /* 810 * Go through the returned output buffer and print stats and 811 * values. Since stat_id is essentially a char string of 812 * 8 bytes, simply use the string format specifier to print it. 813 */ 814 seq_buf_init(&s, buf, PAGE_SIZE); 815 for (index = 0, stat = stats->scm_statistic; 816 index < be32_to_cpu(stats->num_statistics); 817 ++index, ++stat) { 818 seq_buf_printf(&s, "%.8s = 0x%016llX\n", 819 stat->stat_id, 820 be64_to_cpu(stat->stat_val)); 821 } 822 823free_stats: 824 kfree(stats); 825 return rc ? rc : (ssize_t)seq_buf_used(&s); 826} 827static DEVICE_ATTR_ADMIN_RO(perf_stats); 828 829static ssize_t flags_show(struct device *dev, 830 struct device_attribute *attr, char *buf) 831{ 832 struct nvdimm *dimm = to_nvdimm(dev); 833 struct papr_scm_priv *p = nvdimm_provider_data(dimm); 834 struct seq_buf s; 835 u64 health; 836 int rc; 837 838 rc = drc_pmem_query_health(p); 839 if (rc) 840 return rc; 841 842 /* Copy health_bitmap locally, check masks & update out buffer */ 843 health = READ_ONCE(p->health_bitmap); 844 845 seq_buf_init(&s, buf, PAGE_SIZE); 846 if (health & PAPR_PMEM_UNARMED_MASK) 847 seq_buf_printf(&s, "not_armed "); 848 849 if (health & PAPR_PMEM_BAD_SHUTDOWN_MASK) 850 seq_buf_printf(&s, "flush_fail "); 851 852 if (health & PAPR_PMEM_BAD_RESTORE_MASK) 853 seq_buf_printf(&s, "restore_fail "); 854 855 if (health & PAPR_PMEM_ENCRYPTED) 856 seq_buf_printf(&s, "encrypted "); 857 858 if (health & PAPR_PMEM_SMART_EVENT_MASK) 859 seq_buf_printf(&s, "smart_notify "); 860 861 if (health & PAPR_PMEM_SCRUBBED_AND_LOCKED) 862 seq_buf_printf(&s, "scrubbed locked "); 863 864 if (seq_buf_used(&s)) 865 seq_buf_printf(&s, "\n"); 866 867 return seq_buf_used(&s); 868} 869DEVICE_ATTR_RO(flags); 870 871static umode_t papr_nd_attribute_visible(struct kobject *kobj, 872 struct attribute *attr, int n) 873{ 874 struct device *dev = kobj_to_dev(kobj); 875 struct nvdimm *nvdimm = to_nvdimm(dev); 876 struct papr_scm_priv *p = nvdimm_provider_data(nvdimm); 877 878 /* For if perf-stats not available remove perf_stats sysfs */ 879 if (attr == &dev_attr_perf_stats.attr && p->stat_buffer_len == 0) 880 return 0; 881 882 return attr->mode; 883} 884 885/* papr_scm specific dimm attributes */ 886static struct attribute *papr_nd_attributes[] = { 887 &dev_attr_flags.attr, 888 &dev_attr_perf_stats.attr, 889 NULL, 890}; 891 892static struct attribute_group papr_nd_attribute_group = { 893 .name = "papr", 894 .is_visible = papr_nd_attribute_visible, 895 .attrs = papr_nd_attributes, 896}; 897 898static const struct attribute_group *papr_nd_attr_groups[] = { 899 &papr_nd_attribute_group, 900 NULL, 901}; 902 903static int papr_scm_nvdimm_init(struct papr_scm_priv *p) 904{ 905 struct device *dev = &p->pdev->dev; 906 struct nd_mapping_desc mapping; 907 struct nd_region_desc ndr_desc; 908 unsigned long dimm_flags; 909 int target_nid, online_nid; 910 911 p->bus_desc.ndctl = papr_scm_ndctl; 912 p->bus_desc.module = THIS_MODULE; 913 p->bus_desc.of_node = p->pdev->dev.of_node; 914 p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL); 915 916 /* Set the dimm command family mask to accept PDSMs */ 917 set_bit(NVDIMM_FAMILY_PAPR, &p->bus_desc.dimm_family_mask); 918 919 if (!p->bus_desc.provider_name) 920 return -ENOMEM; 921 922 p->bus = nvdimm_bus_register(NULL, &p->bus_desc); 923 if (!p->bus) { 924 dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn); 925 kfree(p->bus_desc.provider_name); 926 return -ENXIO; 927 } 928 929 dimm_flags = 0; 930 set_bit(NDD_LABELING, &dimm_flags); 931 932 p->nvdimm = nvdimm_create(p->bus, p, papr_nd_attr_groups, 933 dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL); 934 if (!p->nvdimm) { 935 dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn); 936 goto err; 937 } 938 939 if (nvdimm_bus_check_dimm_count(p->bus, 1)) 940 goto err; 941 942 /* now add the region */ 943 944 memset(&mapping, 0, sizeof(mapping)); 945 mapping.nvdimm = p->nvdimm; 946 mapping.start = 0; 947 mapping.size = p->blocks * p->block_size; // XXX: potential overflow? 948 949 memset(&ndr_desc, 0, sizeof(ndr_desc)); 950 target_nid = dev_to_node(&p->pdev->dev); 951 online_nid = numa_map_to_online_node(target_nid); 952 ndr_desc.numa_node = online_nid; 953 ndr_desc.target_node = target_nid; 954 ndr_desc.res = &p->res; 955 ndr_desc.of_node = p->dn; 956 ndr_desc.provider_data = p; 957 ndr_desc.mapping = &mapping; 958 ndr_desc.num_mappings = 1; 959 ndr_desc.nd_set = &p->nd_set; 960 961 if (p->is_volatile) 962 p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc); 963 else { 964 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags); 965 p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc); 966 } 967 if (!p->region) { 968 dev_err(dev, "Error registering region %pR from %pOF\n", 969 ndr_desc.res, p->dn); 970 goto err; 971 } 972 if (target_nid != online_nid) 973 dev_info(dev, "Region registered with target node %d and online node %d", 974 target_nid, online_nid); 975 976 mutex_lock(&papr_ndr_lock); 977 list_add_tail(&p->region_list, &papr_nd_regions); 978 mutex_unlock(&papr_ndr_lock); 979 980 return 0; 981 982err: nvdimm_bus_unregister(p->bus); 983 kfree(p->bus_desc.provider_name); 984 return -ENXIO; 985} 986 987static void papr_scm_add_badblock(struct nd_region *region, 988 struct nvdimm_bus *bus, u64 phys_addr) 989{ 990 u64 aligned_addr = ALIGN_DOWN(phys_addr, L1_CACHE_BYTES); 991 992 if (nvdimm_bus_add_badrange(bus, aligned_addr, L1_CACHE_BYTES)) { 993 pr_err("Bad block registration for 0x%llx failed\n", phys_addr); 994 return; 995 } 996 997 pr_debug("Add memory range (0x%llx - 0x%llx) as bad range\n", 998 aligned_addr, aligned_addr + L1_CACHE_BYTES); 999 1000 nvdimm_region_notify(region, NVDIMM_REVALIDATE_POISON); 1001} 1002 1003static int handle_mce_ue(struct notifier_block *nb, unsigned long val, 1004 void *data) 1005{ 1006 struct machine_check_event *evt = data; 1007 struct papr_scm_priv *p; 1008 u64 phys_addr; 1009 bool found = false; 1010 1011 if (evt->error_type != MCE_ERROR_TYPE_UE) 1012 return NOTIFY_DONE; 1013 1014 if (list_empty(&papr_nd_regions)) 1015 return NOTIFY_DONE; 1016 1017 /* 1018 * The physical address obtained here is PAGE_SIZE aligned, so get the 1019 * exact address from the effective address 1020 */ 1021 phys_addr = evt->u.ue_error.physical_address + 1022 (evt->u.ue_error.effective_address & ~PAGE_MASK); 1023 1024 if (!evt->u.ue_error.physical_address_provided || 1025 !is_zone_device_page(pfn_to_page(phys_addr >> PAGE_SHIFT))) 1026 return NOTIFY_DONE; 1027 1028 /* mce notifier is called from a process context, so mutex is safe */ 1029 mutex_lock(&papr_ndr_lock); 1030 list_for_each_entry(p, &papr_nd_regions, region_list) { 1031 if (phys_addr >= p->res.start && phys_addr <= p->res.end) { 1032 found = true; 1033 break; 1034 } 1035 } 1036 1037 if (found) 1038 papr_scm_add_badblock(p->region, p->bus, phys_addr); 1039 1040 mutex_unlock(&papr_ndr_lock); 1041 1042 return found ? NOTIFY_OK : NOTIFY_DONE; 1043} 1044 1045static struct notifier_block mce_ue_nb = { 1046 .notifier_call = handle_mce_ue 1047}; 1048 1049static int papr_scm_probe(struct platform_device *pdev) 1050{ 1051 struct device_node *dn = pdev->dev.of_node; 1052 u32 drc_index, metadata_size; 1053 u64 blocks, block_size; 1054 struct papr_scm_priv *p; 1055 u8 uuid_raw[UUID_SIZE]; 1056 const char *uuid_str; 1057 ssize_t stat_size; 1058 uuid_t uuid; 1059 int rc; 1060 1061 /* check we have all the required DT properties */ 1062 if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) { 1063 dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn); 1064 return -ENODEV; 1065 } 1066 1067 if (of_property_read_u64(dn, "ibm,block-size", &block_size)) { 1068 dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn); 1069 return -ENODEV; 1070 } 1071 1072 if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) { 1073 dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn); 1074 return -ENODEV; 1075 } 1076 1077 if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) { 1078 dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn); 1079 return -ENODEV; 1080 } 1081 1082 /* 1083 * open firmware platform device create won't update the NUMA 1084 * distance table. For PAPR SCM devices we use numa_map_to_online_node() 1085 * to find the nearest online NUMA node and that requires correct 1086 * distance table information. 1087 */ 1088 update_numa_distance(dn); 1089 1090 p = kzalloc(sizeof(*p), GFP_KERNEL); 1091 if (!p) 1092 return -ENOMEM; 1093 1094 /* Initialize the dimm mutex */ 1095 mutex_init(&p->health_mutex); 1096 1097 /* optional DT properties */ 1098 of_property_read_u32(dn, "ibm,metadata-size", &metadata_size); 1099 1100 p->dn = dn; 1101 p->drc_index = drc_index; 1102 p->block_size = block_size; 1103 p->blocks = blocks; 1104 p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required"); 1105 1106 /* We just need to ensure that set cookies are unique across */ 1107 uuid_parse(uuid_str, &uuid); 1108 1109 /* 1110 * The cookie1 and cookie2 are not really little endian. 1111 * We store a raw buffer representation of the 1112 * uuid string so that we can compare this with the label 1113 * area cookie irrespective of the endian configuration 1114 * with which the kernel is built. 1115 * 1116 * Historically we stored the cookie in the below format. 1117 * for a uuid string 72511b67-0b3b-42fd-8d1d-5be3cae8bcaa 1118 * cookie1 was 0xfd423b0b671b5172 1119 * cookie2 was 0xaabce8cae35b1d8d 1120 */ 1121 export_uuid(uuid_raw, &uuid); 1122 p->nd_set.cookie1 = get_unaligned_le64(&uuid_raw[0]); 1123 p->nd_set.cookie2 = get_unaligned_le64(&uuid_raw[8]); 1124 1125 /* might be zero */ 1126 p->metadata_size = metadata_size; 1127 p->pdev = pdev; 1128 1129 /* request the hypervisor to bind this region to somewhere in memory */ 1130 rc = drc_pmem_bind(p); 1131 1132 /* If phyp says drc memory still bound then force unbound and retry */ 1133 if (rc == H_OVERLAP) 1134 rc = drc_pmem_query_n_bind(p); 1135 1136 if (rc != H_SUCCESS) { 1137 dev_err(&p->pdev->dev, "bind err: %d\n", rc); 1138 rc = -ENXIO; 1139 goto err; 1140 } 1141 1142 /* setup the resource for the newly bound range */ 1143 p->res.start = p->bound_addr; 1144 p->res.end = p->bound_addr + p->blocks * p->block_size - 1; 1145 p->res.name = pdev->name; 1146 p->res.flags = IORESOURCE_MEM; 1147 1148 /* Try retrieving the stat buffer and see if its supported */ 1149 stat_size = drc_pmem_query_stats(p, NULL, 0); 1150 if (stat_size > 0) { 1151 p->stat_buffer_len = stat_size; 1152 dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n", 1153 p->stat_buffer_len); 1154 } 1155 1156 rc = papr_scm_nvdimm_init(p); 1157 if (rc) 1158 goto err2; 1159 1160 platform_set_drvdata(pdev, p); 1161 1162 return 0; 1163 1164err2: drc_pmem_unbind(p); 1165err: kfree(p); 1166 return rc; 1167} 1168 1169static int papr_scm_remove(struct platform_device *pdev) 1170{ 1171 struct papr_scm_priv *p = platform_get_drvdata(pdev); 1172 1173 mutex_lock(&papr_ndr_lock); 1174 list_del(&p->region_list); 1175 mutex_unlock(&papr_ndr_lock); 1176 1177 nvdimm_bus_unregister(p->bus); 1178 drc_pmem_unbind(p); 1179 kfree(p->bus_desc.provider_name); 1180 kfree(p); 1181 1182 return 0; 1183} 1184 1185static const struct of_device_id papr_scm_match[] = { 1186 { .compatible = "ibm,pmemory" }, 1187 { .compatible = "ibm,pmemory-v2" }, 1188 { }, 1189}; 1190 1191static struct platform_driver papr_scm_driver = { 1192 .probe = papr_scm_probe, 1193 .remove = papr_scm_remove, 1194 .driver = { 1195 .name = "papr_scm", 1196 .of_match_table = papr_scm_match, 1197 }, 1198}; 1199 1200static int __init papr_scm_init(void) 1201{ 1202 int ret; 1203 1204 ret = platform_driver_register(&papr_scm_driver); 1205 if (!ret) 1206 mce_register_notifier(&mce_ue_nb); 1207 1208 return ret; 1209} 1210module_init(papr_scm_init); 1211 1212static void __exit papr_scm_exit(void) 1213{ 1214 mce_unregister_notifier(&mce_ue_nb); 1215 platform_driver_unregister(&papr_scm_driver); 1216} 1217module_exit(papr_scm_exit); 1218 1219MODULE_DEVICE_TABLE(of, papr_scm_match); 1220MODULE_LICENSE("GPL"); 1221MODULE_AUTHOR("IBM Corporation"); 1222