1 // SPDX-License-Identifier: GPL-2.0 2 /* mdesc.c: Sun4V machine description handling. 3 * 4 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net> 5 */ 6 #include <linux/kernel.h> 7 #include <linux/types.h> 8 #include <linux/memblock.h> 9 #include <linux/log2.h> 10 #include <linux/list.h> 11 #include <linux/slab.h> 12 #include <linux/mm.h> 13 #include <linux/miscdevice.h> 14 #include <linux/bootmem.h> 15 #include <linux/export.h> 16 17 #include <asm/cpudata.h> 18 #include <asm/hypervisor.h> 19 #include <asm/mdesc.h> 20 #include <asm/prom.h> 21 #include <linux/uaccess.h> 22 #include <asm/oplib.h> 23 #include <asm/smp.h> 24 25 /* Unlike the OBP device tree, the machine description is a full-on 26 * DAG. An arbitrary number of ARCs are possible from one 27 * node to other nodes and thus we can't use the OBP device_node 28 * data structure to represent these nodes inside of the kernel. 29 * 30 * Actually, it isn't even a DAG, because there are back pointers 31 * which create cycles in the graph. 32 * 33 * mdesc_hdr and mdesc_elem describe the layout of the data structure 34 * we get from the Hypervisor. 35 */ 36 struct mdesc_hdr { 37 u32 version; /* Transport version */ 38 u32 node_sz; /* node block size */ 39 u32 name_sz; /* name block size */ 40 u32 data_sz; /* data block size */ 41 } __attribute__((aligned(16))); 42 43 struct mdesc_elem { 44 u8 tag; 45 #define MD_LIST_END 0x00 46 #define MD_NODE 0x4e 47 #define MD_NODE_END 0x45 48 #define MD_NOOP 0x20 49 #define MD_PROP_ARC 0x61 50 #define MD_PROP_VAL 0x76 51 #define MD_PROP_STR 0x73 52 #define MD_PROP_DATA 0x64 53 u8 name_len; 54 u16 resv; 55 u32 name_offset; 56 union { 57 struct { 58 u32 data_len; 59 u32 data_offset; 60 } data; 61 u64 val; 62 } d; 63 }; 64 65 struct mdesc_mem_ops { 66 struct mdesc_handle *(*alloc)(unsigned int mdesc_size); 67 void (*free)(struct mdesc_handle *handle); 68 }; 69 70 struct mdesc_handle { 71 struct list_head list; 72 struct mdesc_mem_ops *mops; 73 void *self_base; 74 atomic_t refcnt; 75 unsigned int handle_size; 76 struct mdesc_hdr mdesc; 77 }; 78 79 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64, 80 union md_node_info *); 81 typedef void (*mdesc_node_info_rel_f)(union md_node_info *); 82 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *); 83 84 struct md_node_ops { 85 char *name; 86 mdesc_node_info_get_f get_info; 87 mdesc_node_info_rel_f rel_info; 88 mdesc_node_match_f node_match; 89 }; 90 91 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node, 92 union md_node_info *node_info); 93 static void rel_vdev_port_node_info(union md_node_info *node_info); 94 static bool vdev_port_node_match(union md_node_info *a_node_info, 95 union md_node_info *b_node_info); 96 97 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node, 98 union md_node_info *node_info); 99 static void rel_ds_port_node_info(union md_node_info *node_info); 100 static bool ds_port_node_match(union md_node_info *a_node_info, 101 union md_node_info *b_node_info); 102 103 /* supported node types which can be registered */ 104 static struct md_node_ops md_node_ops_table[] = { 105 {"virtual-device-port", get_vdev_port_node_info, 106 rel_vdev_port_node_info, vdev_port_node_match}, 107 {"domain-services-port", get_ds_port_node_info, 108 rel_ds_port_node_info, ds_port_node_match}, 109 {NULL, NULL, NULL, NULL} 110 }; 111 112 static void mdesc_get_node_ops(const char *node_name, 113 mdesc_node_info_get_f *get_info_f, 114 mdesc_node_info_rel_f *rel_info_f, 115 mdesc_node_match_f *match_f) 116 { 117 int i; 118 119 if (get_info_f) 120 *get_info_f = NULL; 121 122 if (rel_info_f) 123 *rel_info_f = NULL; 124 125 if (match_f) 126 *match_f = NULL; 127 128 if (!node_name) 129 return; 130 131 for (i = 0; md_node_ops_table[i].name != NULL; i++) { 132 if (strcmp(md_node_ops_table[i].name, node_name) == 0) { 133 if (get_info_f) 134 *get_info_f = md_node_ops_table[i].get_info; 135 136 if (rel_info_f) 137 *rel_info_f = md_node_ops_table[i].rel_info; 138 139 if (match_f) 140 *match_f = md_node_ops_table[i].node_match; 141 142 break; 143 } 144 } 145 } 146 147 static void mdesc_handle_init(struct mdesc_handle *hp, 148 unsigned int handle_size, 149 void *base) 150 { 151 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1)); 152 153 memset(hp, 0, handle_size); 154 INIT_LIST_HEAD(&hp->list); 155 hp->self_base = base; 156 atomic_set(&hp->refcnt, 1); 157 hp->handle_size = handle_size; 158 } 159 160 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size) 161 { 162 unsigned int handle_size, alloc_size; 163 struct mdesc_handle *hp; 164 unsigned long paddr; 165 166 handle_size = (sizeof(struct mdesc_handle) - 167 sizeof(struct mdesc_hdr) + 168 mdesc_size); 169 alloc_size = PAGE_ALIGN(handle_size); 170 171 paddr = memblock_alloc(alloc_size, PAGE_SIZE); 172 173 hp = NULL; 174 if (paddr) { 175 hp = __va(paddr); 176 mdesc_handle_init(hp, handle_size, hp); 177 } 178 return hp; 179 } 180 181 static void __init mdesc_memblock_free(struct mdesc_handle *hp) 182 { 183 unsigned int alloc_size; 184 unsigned long start; 185 186 BUG_ON(atomic_read(&hp->refcnt) != 0); 187 BUG_ON(!list_empty(&hp->list)); 188 189 alloc_size = PAGE_ALIGN(hp->handle_size); 190 start = __pa(hp); 191 free_bootmem_late(start, alloc_size); 192 } 193 194 static struct mdesc_mem_ops memblock_mdesc_ops = { 195 .alloc = mdesc_memblock_alloc, 196 .free = mdesc_memblock_free, 197 }; 198 199 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size) 200 { 201 unsigned int handle_size; 202 struct mdesc_handle *hp; 203 unsigned long addr; 204 void *base; 205 206 handle_size = (sizeof(struct mdesc_handle) - 207 sizeof(struct mdesc_hdr) + 208 mdesc_size); 209 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 210 if (!base) 211 return NULL; 212 213 addr = (unsigned long)base; 214 addr = (addr + 15UL) & ~15UL; 215 hp = (struct mdesc_handle *) addr; 216 217 mdesc_handle_init(hp, handle_size, base); 218 219 return hp; 220 } 221 222 static void mdesc_kfree(struct mdesc_handle *hp) 223 { 224 BUG_ON(atomic_read(&hp->refcnt) != 0); 225 BUG_ON(!list_empty(&hp->list)); 226 227 kfree(hp->self_base); 228 } 229 230 static struct mdesc_mem_ops kmalloc_mdesc_memops = { 231 .alloc = mdesc_kmalloc, 232 .free = mdesc_kfree, 233 }; 234 235 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size, 236 struct mdesc_mem_ops *mops) 237 { 238 struct mdesc_handle *hp = mops->alloc(mdesc_size); 239 240 if (hp) 241 hp->mops = mops; 242 243 return hp; 244 } 245 246 static void mdesc_free(struct mdesc_handle *hp) 247 { 248 hp->mops->free(hp); 249 } 250 251 static struct mdesc_handle *cur_mdesc; 252 static LIST_HEAD(mdesc_zombie_list); 253 static DEFINE_SPINLOCK(mdesc_lock); 254 255 struct mdesc_handle *mdesc_grab(void) 256 { 257 struct mdesc_handle *hp; 258 unsigned long flags; 259 260 spin_lock_irqsave(&mdesc_lock, flags); 261 hp = cur_mdesc; 262 if (hp) 263 atomic_inc(&hp->refcnt); 264 spin_unlock_irqrestore(&mdesc_lock, flags); 265 266 return hp; 267 } 268 EXPORT_SYMBOL(mdesc_grab); 269 270 void mdesc_release(struct mdesc_handle *hp) 271 { 272 unsigned long flags; 273 274 spin_lock_irqsave(&mdesc_lock, flags); 275 if (atomic_dec_and_test(&hp->refcnt)) { 276 list_del_init(&hp->list); 277 hp->mops->free(hp); 278 } 279 spin_unlock_irqrestore(&mdesc_lock, flags); 280 } 281 EXPORT_SYMBOL(mdesc_release); 282 283 static DEFINE_MUTEX(mdesc_mutex); 284 static struct mdesc_notifier_client *client_list; 285 286 void mdesc_register_notifier(struct mdesc_notifier_client *client) 287 { 288 bool supported = false; 289 u64 node; 290 int i; 291 292 mutex_lock(&mdesc_mutex); 293 294 /* check to see if the node is supported for registration */ 295 for (i = 0; md_node_ops_table[i].name != NULL; i++) { 296 if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) { 297 supported = true; 298 break; 299 } 300 } 301 302 if (!supported) { 303 pr_err("MD: %s node not supported\n", client->node_name); 304 mutex_unlock(&mdesc_mutex); 305 return; 306 } 307 308 client->next = client_list; 309 client_list = client; 310 311 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name) 312 client->add(cur_mdesc, node, client->node_name); 313 314 mutex_unlock(&mdesc_mutex); 315 } 316 317 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node) 318 { 319 const u64 *id; 320 u64 a; 321 322 id = NULL; 323 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) { 324 u64 target; 325 326 target = mdesc_arc_target(hp, a); 327 id = mdesc_get_property(hp, target, 328 "cfg-handle", NULL); 329 if (id) 330 break; 331 } 332 333 return id; 334 } 335 336 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node, 337 union md_node_info *node_info) 338 { 339 const u64 *parent_cfg_hdlp; 340 const char *name; 341 const u64 *idp; 342 343 /* 344 * Virtual device nodes are distinguished by: 345 * 1. "id" property 346 * 2. "name" property 347 * 3. parent node "cfg-handle" property 348 */ 349 idp = mdesc_get_property(md, node, "id", NULL); 350 name = mdesc_get_property(md, node, "name", NULL); 351 parent_cfg_hdlp = parent_cfg_handle(md, node); 352 353 if (!idp || !name || !parent_cfg_hdlp) 354 return -1; 355 356 node_info->vdev_port.id = *idp; 357 node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL); 358 node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp; 359 360 return 0; 361 } 362 363 static void rel_vdev_port_node_info(union md_node_info *node_info) 364 { 365 if (node_info && node_info->vdev_port.name) { 366 kfree_const(node_info->vdev_port.name); 367 node_info->vdev_port.name = NULL; 368 } 369 } 370 371 static bool vdev_port_node_match(union md_node_info *a_node_info, 372 union md_node_info *b_node_info) 373 { 374 if (a_node_info->vdev_port.id != b_node_info->vdev_port.id) 375 return false; 376 377 if (a_node_info->vdev_port.parent_cfg_hdl != 378 b_node_info->vdev_port.parent_cfg_hdl) 379 return false; 380 381 if (strncmp(a_node_info->vdev_port.name, 382 b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0) 383 return false; 384 385 return true; 386 } 387 388 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node, 389 union md_node_info *node_info) 390 { 391 const u64 *idp; 392 393 /* DS port nodes use the "id" property to distinguish them */ 394 idp = mdesc_get_property(md, node, "id", NULL); 395 if (!idp) 396 return -1; 397 398 node_info->ds_port.id = *idp; 399 400 return 0; 401 } 402 403 static void rel_ds_port_node_info(union md_node_info *node_info) 404 { 405 } 406 407 static bool ds_port_node_match(union md_node_info *a_node_info, 408 union md_node_info *b_node_info) 409 { 410 if (a_node_info->ds_port.id != b_node_info->ds_port.id) 411 return false; 412 413 return true; 414 } 415 416 /* Run 'func' on nodes which are in A but not in B. */ 417 static void invoke_on_missing(const char *name, 418 struct mdesc_handle *a, 419 struct mdesc_handle *b, 420 void (*func)(struct mdesc_handle *, u64, 421 const char *node_name)) 422 { 423 mdesc_node_info_get_f get_info_func; 424 mdesc_node_info_rel_f rel_info_func; 425 mdesc_node_match_f node_match_func; 426 union md_node_info a_node_info; 427 union md_node_info b_node_info; 428 bool found; 429 u64 a_node; 430 u64 b_node; 431 int rv; 432 433 /* 434 * Find the get_info, rel_info and node_match ops for the given 435 * node name 436 */ 437 mdesc_get_node_ops(name, &get_info_func, &rel_info_func, 438 &node_match_func); 439 440 /* If we didn't find a match, the node type is not supported */ 441 if (!get_info_func || !rel_info_func || !node_match_func) { 442 pr_err("MD: %s node type is not supported\n", name); 443 return; 444 } 445 446 mdesc_for_each_node_by_name(a, a_node, name) { 447 found = false; 448 449 rv = get_info_func(a, a_node, &a_node_info); 450 if (rv != 0) { 451 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n", 452 name); 453 continue; 454 } 455 456 /* Check each node in B for node matching a_node */ 457 mdesc_for_each_node_by_name(b, b_node, name) { 458 rv = get_info_func(b, b_node, &b_node_info); 459 if (rv != 0) 460 continue; 461 462 if (node_match_func(&a_node_info, &b_node_info)) { 463 found = true; 464 rel_info_func(&b_node_info); 465 break; 466 } 467 468 rel_info_func(&b_node_info); 469 } 470 471 rel_info_func(&a_node_info); 472 473 if (!found) 474 func(a, a_node, name); 475 } 476 } 477 478 static void notify_one(struct mdesc_notifier_client *p, 479 struct mdesc_handle *old_hp, 480 struct mdesc_handle *new_hp) 481 { 482 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove); 483 invoke_on_missing(p->node_name, new_hp, old_hp, p->add); 484 } 485 486 static void mdesc_notify_clients(struct mdesc_handle *old_hp, 487 struct mdesc_handle *new_hp) 488 { 489 struct mdesc_notifier_client *p = client_list; 490 491 while (p) { 492 notify_one(p, old_hp, new_hp); 493 p = p->next; 494 } 495 } 496 497 void mdesc_update(void) 498 { 499 unsigned long len, real_len, status; 500 struct mdesc_handle *hp, *orig_hp; 501 unsigned long flags; 502 503 mutex_lock(&mdesc_mutex); 504 505 (void) sun4v_mach_desc(0UL, 0UL, &len); 506 507 hp = mdesc_alloc(len, &kmalloc_mdesc_memops); 508 if (!hp) { 509 printk(KERN_ERR "MD: mdesc alloc fails\n"); 510 goto out; 511 } 512 513 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len); 514 if (status != HV_EOK || real_len > len) { 515 printk(KERN_ERR "MD: mdesc reread fails with %lu\n", 516 status); 517 atomic_dec(&hp->refcnt); 518 mdesc_free(hp); 519 goto out; 520 } 521 522 spin_lock_irqsave(&mdesc_lock, flags); 523 orig_hp = cur_mdesc; 524 cur_mdesc = hp; 525 spin_unlock_irqrestore(&mdesc_lock, flags); 526 527 mdesc_notify_clients(orig_hp, hp); 528 529 spin_lock_irqsave(&mdesc_lock, flags); 530 if (atomic_dec_and_test(&orig_hp->refcnt)) 531 mdesc_free(orig_hp); 532 else 533 list_add(&orig_hp->list, &mdesc_zombie_list); 534 spin_unlock_irqrestore(&mdesc_lock, flags); 535 536 out: 537 mutex_unlock(&mdesc_mutex); 538 } 539 540 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name, 541 union md_node_info *node_info) 542 { 543 mdesc_node_info_get_f get_info_func; 544 mdesc_node_info_rel_f rel_info_func; 545 mdesc_node_match_f node_match_func; 546 union md_node_info hp_node_info; 547 u64 hp_node; 548 int rv; 549 550 if (hp == NULL || node_name == NULL || node_info == NULL) 551 return MDESC_NODE_NULL; 552 553 /* Find the ops for the given node name */ 554 mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func, 555 &node_match_func); 556 557 /* If we didn't find ops for the given node name, it is not supported */ 558 if (!get_info_func || !rel_info_func || !node_match_func) { 559 pr_err("MD: %s node is not supported\n", node_name); 560 return -EINVAL; 561 } 562 563 mdesc_for_each_node_by_name(hp, hp_node, node_name) { 564 rv = get_info_func(hp, hp_node, &hp_node_info); 565 if (rv != 0) 566 continue; 567 568 if (node_match_func(node_info, &hp_node_info)) 569 break; 570 571 rel_info_func(&hp_node_info); 572 } 573 574 rel_info_func(&hp_node_info); 575 576 return hp_node; 577 } 578 EXPORT_SYMBOL(mdesc_get_node); 579 580 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node, 581 const char *node_name, union md_node_info *node_info) 582 { 583 mdesc_node_info_get_f get_info_func; 584 int rv; 585 586 if (hp == NULL || node == MDESC_NODE_NULL || 587 node_name == NULL || node_info == NULL) 588 return -EINVAL; 589 590 /* Find the get_info op for the given node name */ 591 mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL); 592 593 /* If we didn't find a get_info_func, the node name is not supported */ 594 if (get_info_func == NULL) { 595 pr_err("MD: %s node is not supported\n", node_name); 596 return -EINVAL; 597 } 598 599 rv = get_info_func(hp, node, node_info); 600 if (rv != 0) { 601 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n", 602 node_name); 603 return -1; 604 } 605 606 return 0; 607 } 608 EXPORT_SYMBOL(mdesc_get_node_info); 609 610 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc) 611 { 612 return (struct mdesc_elem *) (mdesc + 1); 613 } 614 615 static void *name_block(struct mdesc_hdr *mdesc) 616 { 617 return ((void *) node_block(mdesc)) + mdesc->node_sz; 618 } 619 620 static void *data_block(struct mdesc_hdr *mdesc) 621 { 622 return ((void *) name_block(mdesc)) + mdesc->name_sz; 623 } 624 625 u64 mdesc_node_by_name(struct mdesc_handle *hp, 626 u64 from_node, const char *name) 627 { 628 struct mdesc_elem *ep = node_block(&hp->mdesc); 629 const char *names = name_block(&hp->mdesc); 630 u64 last_node = hp->mdesc.node_sz / 16; 631 u64 ret; 632 633 if (from_node == MDESC_NODE_NULL) { 634 ret = from_node = 0; 635 } else if (from_node >= last_node) { 636 return MDESC_NODE_NULL; 637 } else { 638 ret = ep[from_node].d.val; 639 } 640 641 while (ret < last_node) { 642 if (ep[ret].tag != MD_NODE) 643 return MDESC_NODE_NULL; 644 if (!strcmp(names + ep[ret].name_offset, name)) 645 break; 646 ret = ep[ret].d.val; 647 } 648 if (ret >= last_node) 649 ret = MDESC_NODE_NULL; 650 return ret; 651 } 652 EXPORT_SYMBOL(mdesc_node_by_name); 653 654 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node, 655 const char *name, int *lenp) 656 { 657 const char *names = name_block(&hp->mdesc); 658 u64 last_node = hp->mdesc.node_sz / 16; 659 void *data = data_block(&hp->mdesc); 660 struct mdesc_elem *ep; 661 662 if (node == MDESC_NODE_NULL || node >= last_node) 663 return NULL; 664 665 ep = node_block(&hp->mdesc) + node; 666 ep++; 667 for (; ep->tag != MD_NODE_END; ep++) { 668 void *val = NULL; 669 int len = 0; 670 671 switch (ep->tag) { 672 case MD_PROP_VAL: 673 val = &ep->d.val; 674 len = 8; 675 break; 676 677 case MD_PROP_STR: 678 case MD_PROP_DATA: 679 val = data + ep->d.data.data_offset; 680 len = ep->d.data.data_len; 681 break; 682 683 default: 684 break; 685 } 686 if (!val) 687 continue; 688 689 if (!strcmp(names + ep->name_offset, name)) { 690 if (lenp) 691 *lenp = len; 692 return val; 693 } 694 } 695 696 return NULL; 697 } 698 EXPORT_SYMBOL(mdesc_get_property); 699 700 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type) 701 { 702 struct mdesc_elem *ep, *base = node_block(&hp->mdesc); 703 const char *names = name_block(&hp->mdesc); 704 u64 last_node = hp->mdesc.node_sz / 16; 705 706 if (from == MDESC_NODE_NULL || from >= last_node) 707 return MDESC_NODE_NULL; 708 709 ep = base + from; 710 711 ep++; 712 for (; ep->tag != MD_NODE_END; ep++) { 713 if (ep->tag != MD_PROP_ARC) 714 continue; 715 716 if (strcmp(names + ep->name_offset, arc_type)) 717 continue; 718 719 return ep - base; 720 } 721 722 return MDESC_NODE_NULL; 723 } 724 EXPORT_SYMBOL(mdesc_next_arc); 725 726 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc) 727 { 728 struct mdesc_elem *ep, *base = node_block(&hp->mdesc); 729 730 ep = base + arc; 731 732 return ep->d.val; 733 } 734 EXPORT_SYMBOL(mdesc_arc_target); 735 736 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node) 737 { 738 struct mdesc_elem *ep, *base = node_block(&hp->mdesc); 739 const char *names = name_block(&hp->mdesc); 740 u64 last_node = hp->mdesc.node_sz / 16; 741 742 if (node == MDESC_NODE_NULL || node >= last_node) 743 return NULL; 744 745 ep = base + node; 746 if (ep->tag != MD_NODE) 747 return NULL; 748 749 return names + ep->name_offset; 750 } 751 EXPORT_SYMBOL(mdesc_node_name); 752 753 static u64 max_cpus = 64; 754 755 static void __init report_platform_properties(void) 756 { 757 struct mdesc_handle *hp = mdesc_grab(); 758 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform"); 759 const char *s; 760 const u64 *v; 761 762 if (pn == MDESC_NODE_NULL) { 763 prom_printf("No platform node in machine-description.\n"); 764 prom_halt(); 765 } 766 767 s = mdesc_get_property(hp, pn, "banner-name", NULL); 768 printk("PLATFORM: banner-name [%s]\n", s); 769 s = mdesc_get_property(hp, pn, "name", NULL); 770 printk("PLATFORM: name [%s]\n", s); 771 772 v = mdesc_get_property(hp, pn, "hostid", NULL); 773 if (v) 774 printk("PLATFORM: hostid [%08llx]\n", *v); 775 v = mdesc_get_property(hp, pn, "serial#", NULL); 776 if (v) 777 printk("PLATFORM: serial# [%08llx]\n", *v); 778 v = mdesc_get_property(hp, pn, "stick-frequency", NULL); 779 printk("PLATFORM: stick-frequency [%08llx]\n", *v); 780 v = mdesc_get_property(hp, pn, "mac-address", NULL); 781 if (v) 782 printk("PLATFORM: mac-address [%llx]\n", *v); 783 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL); 784 if (v) 785 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v); 786 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL); 787 if (v) 788 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v); 789 v = mdesc_get_property(hp, pn, "max-cpus", NULL); 790 if (v) { 791 max_cpus = *v; 792 printk("PLATFORM: max-cpus [%llu]\n", max_cpus); 793 } 794 795 #ifdef CONFIG_SMP 796 { 797 int max_cpu, i; 798 799 if (v) { 800 max_cpu = *v; 801 if (max_cpu > NR_CPUS) 802 max_cpu = NR_CPUS; 803 } else { 804 max_cpu = NR_CPUS; 805 } 806 for (i = 0; i < max_cpu; i++) 807 set_cpu_possible(i, true); 808 } 809 #endif 810 811 mdesc_release(hp); 812 } 813 814 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp) 815 { 816 const u64 *level = mdesc_get_property(hp, mp, "level", NULL); 817 const u64 *size = mdesc_get_property(hp, mp, "size", NULL); 818 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL); 819 const char *type; 820 int type_len; 821 822 type = mdesc_get_property(hp, mp, "type", &type_len); 823 824 switch (*level) { 825 case 1: 826 if (of_find_in_proplist(type, "instn", type_len)) { 827 c->icache_size = *size; 828 c->icache_line_size = *line_size; 829 } else if (of_find_in_proplist(type, "data", type_len)) { 830 c->dcache_size = *size; 831 c->dcache_line_size = *line_size; 832 } 833 break; 834 835 case 2: 836 c->ecache_size = *size; 837 c->ecache_line_size = *line_size; 838 break; 839 840 default: 841 break; 842 } 843 844 if (*level == 1) { 845 u64 a; 846 847 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) { 848 u64 target = mdesc_arc_target(hp, a); 849 const char *name = mdesc_node_name(hp, target); 850 851 if (!strcmp(name, "cache")) 852 fill_in_one_cache(c, hp, target); 853 } 854 } 855 } 856 857 static void find_back_node_value(struct mdesc_handle *hp, u64 node, 858 char *srch_val, 859 void (*func)(struct mdesc_handle *, u64, int), 860 u64 val, int depth) 861 { 862 u64 arc; 863 864 /* Since we have an estimate of recursion depth, do a sanity check. */ 865 if (depth == 0) 866 return; 867 868 mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) { 869 u64 n = mdesc_arc_target(hp, arc); 870 const char *name = mdesc_node_name(hp, n); 871 872 if (!strcmp(srch_val, name)) 873 (*func)(hp, n, val); 874 875 find_back_node_value(hp, n, srch_val, func, val, depth-1); 876 } 877 } 878 879 static void __mark_core_id(struct mdesc_handle *hp, u64 node, 880 int core_id) 881 { 882 const u64 *id = mdesc_get_property(hp, node, "id", NULL); 883 884 if (*id < num_possible_cpus()) 885 cpu_data(*id).core_id = core_id; 886 } 887 888 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node, 889 int max_cache_id) 890 { 891 const u64 *id = mdesc_get_property(hp, node, "id", NULL); 892 893 if (*id < num_possible_cpus()) { 894 cpu_data(*id).max_cache_id = max_cache_id; 895 896 /** 897 * On systems without explicit socket descriptions socket 898 * is max_cache_id 899 */ 900 cpu_data(*id).sock_id = max_cache_id; 901 } 902 } 903 904 static void mark_core_ids(struct mdesc_handle *hp, u64 mp, 905 int core_id) 906 { 907 find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10); 908 } 909 910 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp, 911 int max_cache_id) 912 { 913 find_back_node_value(hp, mp, "cpu", __mark_max_cache_id, 914 max_cache_id, 10); 915 } 916 917 static void set_core_ids(struct mdesc_handle *hp) 918 { 919 int idx; 920 u64 mp; 921 922 idx = 1; 923 924 /* Identify unique cores by looking for cpus backpointed to by 925 * level 1 instruction caches. 926 */ 927 mdesc_for_each_node_by_name(hp, mp, "cache") { 928 const u64 *level; 929 const char *type; 930 int len; 931 932 level = mdesc_get_property(hp, mp, "level", NULL); 933 if (*level != 1) 934 continue; 935 936 type = mdesc_get_property(hp, mp, "type", &len); 937 if (!of_find_in_proplist(type, "instn", len)) 938 continue; 939 940 mark_core_ids(hp, mp, idx); 941 idx++; 942 } 943 } 944 945 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level) 946 { 947 u64 mp; 948 int idx = 1; 949 int fnd = 0; 950 951 /** 952 * Identify unique highest level of shared cache by looking for cpus 953 * backpointed to by shared level N caches. 954 */ 955 mdesc_for_each_node_by_name(hp, mp, "cache") { 956 const u64 *cur_lvl; 957 958 cur_lvl = mdesc_get_property(hp, mp, "level", NULL); 959 if (*cur_lvl != level) 960 continue; 961 mark_max_cache_ids(hp, mp, idx); 962 idx++; 963 fnd = 1; 964 } 965 return fnd; 966 } 967 968 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp) 969 { 970 int idx = 1; 971 972 mdesc_for_each_node_by_name(hp, mp, "socket") { 973 u64 a; 974 975 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) { 976 u64 t = mdesc_arc_target(hp, a); 977 const char *name; 978 const u64 *id; 979 980 name = mdesc_node_name(hp, t); 981 if (strcmp(name, "cpu")) 982 continue; 983 984 id = mdesc_get_property(hp, t, "id", NULL); 985 if (*id < num_possible_cpus()) 986 cpu_data(*id).sock_id = idx; 987 } 988 idx++; 989 } 990 } 991 992 static void set_sock_ids(struct mdesc_handle *hp) 993 { 994 u64 mp; 995 996 /** 997 * Find the highest level of shared cache which pre-T7 is also 998 * the socket. 999 */ 1000 if (!set_max_cache_ids_by_cache(hp, 3)) 1001 set_max_cache_ids_by_cache(hp, 2); 1002 1003 /* If machine description exposes sockets data use it.*/ 1004 mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets"); 1005 if (mp != MDESC_NODE_NULL) 1006 set_sock_ids_by_socket(hp, mp); 1007 } 1008 1009 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id) 1010 { 1011 u64 a; 1012 1013 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) { 1014 u64 t = mdesc_arc_target(hp, a); 1015 const char *name; 1016 const u64 *id; 1017 1018 name = mdesc_node_name(hp, t); 1019 if (strcmp(name, "cpu")) 1020 continue; 1021 1022 id = mdesc_get_property(hp, t, "id", NULL); 1023 if (*id < NR_CPUS) 1024 cpu_data(*id).proc_id = proc_id; 1025 } 1026 } 1027 1028 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name) 1029 { 1030 int idx; 1031 u64 mp; 1032 1033 idx = 0; 1034 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) { 1035 const char *type; 1036 int len; 1037 1038 type = mdesc_get_property(hp, mp, "type", &len); 1039 if (!of_find_in_proplist(type, "int", len) && 1040 !of_find_in_proplist(type, "integer", len)) 1041 continue; 1042 1043 mark_proc_ids(hp, mp, idx); 1044 idx++; 1045 } 1046 } 1047 1048 static void set_proc_ids(struct mdesc_handle *hp) 1049 { 1050 __set_proc_ids(hp, "exec_unit"); 1051 __set_proc_ids(hp, "exec-unit"); 1052 } 1053 1054 static void get_one_mondo_bits(const u64 *p, unsigned int *mask, 1055 unsigned long def, unsigned long max) 1056 { 1057 u64 val; 1058 1059 if (!p) 1060 goto use_default; 1061 val = *p; 1062 1063 if (!val || val >= 64) 1064 goto use_default; 1065 1066 if (val > max) 1067 val = max; 1068 1069 *mask = ((1U << val) * 64U) - 1U; 1070 return; 1071 1072 use_default: 1073 *mask = ((1U << def) * 64U) - 1U; 1074 } 1075 1076 static void get_mondo_data(struct mdesc_handle *hp, u64 mp, 1077 struct trap_per_cpu *tb) 1078 { 1079 static int printed; 1080 const u64 *val; 1081 1082 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL); 1083 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2)); 1084 1085 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL); 1086 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8); 1087 1088 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL); 1089 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7); 1090 1091 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL); 1092 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2); 1093 if (!printed++) { 1094 pr_info("SUN4V: Mondo queue sizes " 1095 "[cpu(%u) dev(%u) r(%u) nr(%u)]\n", 1096 tb->cpu_mondo_qmask + 1, 1097 tb->dev_mondo_qmask + 1, 1098 tb->resum_qmask + 1, 1099 tb->nonresum_qmask + 1); 1100 } 1101 } 1102 1103 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask) 1104 { 1105 struct mdesc_handle *hp = mdesc_grab(); 1106 void *ret = NULL; 1107 u64 mp; 1108 1109 mdesc_for_each_node_by_name(hp, mp, "cpu") { 1110 const u64 *id = mdesc_get_property(hp, mp, "id", NULL); 1111 int cpuid = *id; 1112 1113 #ifdef CONFIG_SMP 1114 if (cpuid >= NR_CPUS) { 1115 printk(KERN_WARNING "Ignoring CPU %d which is " 1116 ">= NR_CPUS (%d)\n", 1117 cpuid, NR_CPUS); 1118 continue; 1119 } 1120 if (!cpumask_test_cpu(cpuid, mask)) 1121 continue; 1122 #endif 1123 1124 ret = func(hp, mp, cpuid, arg); 1125 if (ret) 1126 goto out; 1127 } 1128 out: 1129 mdesc_release(hp); 1130 return ret; 1131 } 1132 1133 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, 1134 void *arg) 1135 { 1136 ncpus_probed++; 1137 #ifdef CONFIG_SMP 1138 set_cpu_present(cpuid, true); 1139 #endif 1140 return NULL; 1141 } 1142 1143 void mdesc_populate_present_mask(cpumask_t *mask) 1144 { 1145 if (tlb_type != hypervisor) 1146 return; 1147 1148 ncpus_probed = 0; 1149 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask); 1150 } 1151 1152 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg) 1153 { 1154 const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL); 1155 unsigned long *pgsz_mask = arg; 1156 u64 val; 1157 1158 val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K | 1159 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB); 1160 if (pgsz_prop) 1161 val = *pgsz_prop; 1162 1163 if (!*pgsz_mask) 1164 *pgsz_mask = val; 1165 else 1166 *pgsz_mask &= val; 1167 return NULL; 1168 } 1169 1170 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask) 1171 { 1172 *pgsz_mask = 0; 1173 mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask); 1174 } 1175 1176 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, 1177 void *arg) 1178 { 1179 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL); 1180 struct trap_per_cpu *tb; 1181 cpuinfo_sparc *c; 1182 u64 a; 1183 1184 #ifndef CONFIG_SMP 1185 /* On uniprocessor we only want the values for the 1186 * real physical cpu the kernel booted onto, however 1187 * cpu_data() only has one entry at index 0. 1188 */ 1189 if (cpuid != real_hard_smp_processor_id()) 1190 return NULL; 1191 cpuid = 0; 1192 #endif 1193 1194 c = &cpu_data(cpuid); 1195 c->clock_tick = *cfreq; 1196 1197 tb = &trap_block[cpuid]; 1198 get_mondo_data(hp, mp, tb); 1199 1200 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) { 1201 u64 j, t = mdesc_arc_target(hp, a); 1202 const char *t_name; 1203 1204 t_name = mdesc_node_name(hp, t); 1205 if (!strcmp(t_name, "cache")) { 1206 fill_in_one_cache(c, hp, t); 1207 continue; 1208 } 1209 1210 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) { 1211 u64 n = mdesc_arc_target(hp, j); 1212 const char *n_name; 1213 1214 n_name = mdesc_node_name(hp, n); 1215 if (!strcmp(n_name, "cache")) 1216 fill_in_one_cache(c, hp, n); 1217 } 1218 } 1219 1220 c->core_id = 0; 1221 c->proc_id = -1; 1222 1223 return NULL; 1224 } 1225 1226 void mdesc_fill_in_cpu_data(cpumask_t *mask) 1227 { 1228 struct mdesc_handle *hp; 1229 1230 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask); 1231 1232 hp = mdesc_grab(); 1233 1234 set_core_ids(hp); 1235 set_proc_ids(hp); 1236 set_sock_ids(hp); 1237 1238 mdesc_release(hp); 1239 1240 smp_fill_in_sib_core_maps(); 1241 } 1242 1243 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is 1244 * opened. Hold this reference until /dev/mdesc is closed to ensure 1245 * mdesc data structure is not released underneath us. Store the 1246 * pointer to mdesc structure in private_data for read and seek to use 1247 */ 1248 static int mdesc_open(struct inode *inode, struct file *file) 1249 { 1250 struct mdesc_handle *hp = mdesc_grab(); 1251 1252 if (!hp) 1253 return -ENODEV; 1254 1255 file->private_data = hp; 1256 1257 return 0; 1258 } 1259 1260 static ssize_t mdesc_read(struct file *file, char __user *buf, 1261 size_t len, loff_t *offp) 1262 { 1263 struct mdesc_handle *hp = file->private_data; 1264 unsigned char *mdesc; 1265 int bytes_left, count = len; 1266 1267 if (*offp >= hp->handle_size) 1268 return 0; 1269 1270 bytes_left = hp->handle_size - *offp; 1271 if (count > bytes_left) 1272 count = bytes_left; 1273 1274 mdesc = (unsigned char *)&hp->mdesc; 1275 mdesc += *offp; 1276 if (!copy_to_user(buf, mdesc, count)) { 1277 *offp += count; 1278 return count; 1279 } else { 1280 return -EFAULT; 1281 } 1282 } 1283 1284 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence) 1285 { 1286 struct mdesc_handle *hp = file->private_data; 1287 1288 return no_seek_end_llseek_size(file, offset, whence, hp->handle_size); 1289 } 1290 1291 /* mdesc_close() - /dev/mdesc is being closed, release the reference to 1292 * mdesc structure. 1293 */ 1294 static int mdesc_close(struct inode *inode, struct file *file) 1295 { 1296 mdesc_release(file->private_data); 1297 return 0; 1298 } 1299 1300 static const struct file_operations mdesc_fops = { 1301 .open = mdesc_open, 1302 .read = mdesc_read, 1303 .llseek = mdesc_llseek, 1304 .release = mdesc_close, 1305 .owner = THIS_MODULE, 1306 }; 1307 1308 static struct miscdevice mdesc_misc = { 1309 .minor = MISC_DYNAMIC_MINOR, 1310 .name = "mdesc", 1311 .fops = &mdesc_fops, 1312 }; 1313 1314 static int __init mdesc_misc_init(void) 1315 { 1316 return misc_register(&mdesc_misc); 1317 } 1318 1319 __initcall(mdesc_misc_init); 1320 1321 void __init sun4v_mdesc_init(void) 1322 { 1323 struct mdesc_handle *hp; 1324 unsigned long len, real_len, status; 1325 1326 (void) sun4v_mach_desc(0UL, 0UL, &len); 1327 1328 printk("MDESC: Size is %lu bytes.\n", len); 1329 1330 hp = mdesc_alloc(len, &memblock_mdesc_ops); 1331 if (hp == NULL) { 1332 prom_printf("MDESC: alloc of %lu bytes failed.\n", len); 1333 prom_halt(); 1334 } 1335 1336 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len); 1337 if (status != HV_EOK || real_len > len) { 1338 prom_printf("sun4v_mach_desc fails, err(%lu), " 1339 "len(%lu), real_len(%lu)\n", 1340 status, len, real_len); 1341 mdesc_free(hp); 1342 prom_halt(); 1343 } 1344 1345 cur_mdesc = hp; 1346 1347 report_platform_properties(); 1348 } 1349