1/* 2 * This file is subject to the terms and conditions of the GNU General Public 3 * License. See the file "COPYING" in the main directory of this archive 4 * for more details. 5 * 6 * Copyright (C) 2004-2006 Silicon Graphics, Inc. All rights reserved. 7 * 8 * SGI Altix topology and hardware performance monitoring API. 9 * Mark Goodwin <markgw@sgi.com>. 10 * 11 * Creates /proc/sgi_sn/sn_topology (read-only) to export 12 * info about Altix nodes, routers, CPUs and NumaLink 13 * interconnection/topology. 14 * 15 * Also creates a dynamic misc device named "sn_hwperf" 16 * that supports an ioctl interface to call down into SAL 17 * to discover hw objects, topology and to read/write 18 * memory mapped registers, e.g. for performance monitoring. 19 * The "sn_hwperf" device is registered only after the procfs 20 * file is first opened, i.e. only if/when it's needed. 21 * 22 * This API is used by SGI Performance Co-Pilot and other 23 * tools, see http://oss.sgi.com/projects/pcp 24 */ 25 26#include <linux/fs.h> 27#include <linux/slab.h> 28#include <linux/vmalloc.h> 29#include <linux/seq_file.h> 30#include <linux/miscdevice.h> 31#include <linux/utsname.h> 32#include <linux/cpumask.h> 33#include <linux/smp_lock.h> 34#include <linux/nodemask.h> 35#include <linux/smp.h> 36 37#include <asm/processor.h> 38#include <asm/topology.h> 39#include <asm/semaphore.h> 40#include <asm/uaccess.h> 41#include <asm/sal.h> 42#include <asm/sn/io.h> 43#include <asm/sn/sn_sal.h> 44#include <asm/sn/module.h> 45#include <asm/sn/geo.h> 46#include <asm/sn/sn2/sn_hwperf.h> 47#include <asm/sn/addrs.h> 48 49static void *sn_hwperf_salheap = NULL; 50static int sn_hwperf_obj_cnt = 0; 51static nasid_t sn_hwperf_master_nasid = INVALID_NASID; 52static int sn_hwperf_init(void); 53static DECLARE_MUTEX(sn_hwperf_init_mutex); 54 55#define cnode_possible(n) ((n) < num_cnodes) 56 57static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret) 58{ 59 int e; 60 u64 sz; 61 struct sn_hwperf_object_info *objbuf = NULL; 62 63 if ((e = sn_hwperf_init()) < 0) { 64 printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e); 65 goto out; 66 } 67 68 sz = sn_hwperf_obj_cnt * sizeof(struct sn_hwperf_object_info); 69 if ((objbuf = (struct sn_hwperf_object_info *) vmalloc(sz)) == NULL) { 70 printk("sn_hwperf_enum_objects: vmalloc(%d) failed\n", (int)sz); 71 e = -ENOMEM; 72 goto out; 73 } 74 75 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, SN_HWPERF_ENUM_OBJECTS, 76 0, sz, (u64) objbuf, 0, 0, NULL); 77 if (e != SN_HWPERF_OP_OK) { 78 e = -EINVAL; 79 vfree(objbuf); 80 } 81 82out: 83 *nobj = sn_hwperf_obj_cnt; 84 *ret = objbuf; 85 return e; 86} 87 88static int sn_hwperf_location_to_bpos(char *location, 89 int *rack, int *bay, int *slot, int *slab) 90{ 91 char type; 92 93 /* first scan for an old style geoid string */ 94 if (sscanf(location, "%03d%c%02d#%d", 95 rack, &type, bay, slab) == 4) 96 *slot = 0; 97 else /* scan for a new bladed geoid string */ 98 if (sscanf(location, "%03d%c%02d^%02d#%d", 99 rack, &type, bay, slot, slab) != 5) 100 return -1; 101 /* success */ 102 return 0; 103} 104 105static int sn_hwperf_geoid_to_cnode(char *location) 106{ 107 int cnode; 108 geoid_t geoid; 109 moduleid_t module_id; 110 int rack, bay, slot, slab; 111 int this_rack, this_bay, this_slot, this_slab; 112 113 if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab)) 114 return -1; 115 116 for (cnode = 0; cnode < num_cnodes; cnode++) { 117 geoid = cnodeid_get_geoid(cnode); 118 module_id = geo_module(geoid); 119 this_rack = MODULE_GET_RACK(module_id); 120 this_bay = MODULE_GET_BPOS(module_id); 121 this_slot = geo_slot(geoid); 122 this_slab = geo_slab(geoid); 123 if (rack == this_rack && bay == this_bay && 124 slot == this_slot && slab == this_slab) { 125 break; 126 } 127 } 128 129 return cnode_possible(cnode) ? cnode : -1; 130} 131 132static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj) 133{ 134 if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj)) 135 BUG(); 136 if (SN_HWPERF_FOREIGN(obj)) 137 return -1; 138 return sn_hwperf_geoid_to_cnode(obj->location); 139} 140 141static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj, 142 struct sn_hwperf_object_info *objs) 143{ 144 int ordinal; 145 struct sn_hwperf_object_info *p; 146 147 for (ordinal=0, p=objs; p != obj; p++) { 148 if (SN_HWPERF_FOREIGN(p)) 149 continue; 150 if (SN_HWPERF_SAME_OBJTYPE(p, obj)) 151 ordinal++; 152 } 153 154 return ordinal; 155} 156 157static const char *slabname_node = "node"; /* SHub asic */ 158static const char *slabname_ionode = "ionode"; /* TIO asic */ 159static const char *slabname_router = "router"; /* NL3R or NL4R */ 160static const char *slabname_other = "other"; /* unknown asic */ 161 162static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj, 163 struct sn_hwperf_object_info *objs, int *ordinal) 164{ 165 int isnode; 166 const char *slabname = slabname_other; 167 168 if ((isnode = SN_HWPERF_IS_NODE(obj)) || SN_HWPERF_IS_IONODE(obj)) { 169 slabname = isnode ? slabname_node : slabname_ionode; 170 *ordinal = sn_hwperf_obj_to_cnode(obj); 171 } 172 else { 173 *ordinal = sn_hwperf_generic_ordinal(obj, objs); 174 if (SN_HWPERF_IS_ROUTER(obj)) 175 slabname = slabname_router; 176 } 177 178 return slabname; 179} 180 181static void print_pci_topology(struct seq_file *s) 182{ 183 char *p; 184 size_t sz; 185 int e; 186 187 for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) { 188 if (!(p = kmalloc(sz, GFP_KERNEL))) 189 break; 190 e = ia64_sn_ioif_get_pci_topology(__pa(p), sz); 191 if (e == SALRET_OK) 192 seq_puts(s, p); 193 kfree(p); 194 if (e == SALRET_OK || e == SALRET_NOT_IMPLEMENTED) 195 break; 196 } 197} 198 199static inline int sn_hwperf_has_cpus(cnodeid_t node) 200{ 201 return node < MAX_NUMNODES && node_online(node) && nr_cpus_node(node); 202} 203 204static inline int sn_hwperf_has_mem(cnodeid_t node) 205{ 206 return node < MAX_NUMNODES && node_online(node) && NODE_DATA(node)->node_present_pages; 207} 208 209static struct sn_hwperf_object_info * 210sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf, 211 int nobj, int id) 212{ 213 int i; 214 struct sn_hwperf_object_info *p = objbuf; 215 216 for (i=0; i < nobj; i++, p++) { 217 if (p->id == id) 218 return p; 219 } 220 221 return NULL; 222 223} 224 225static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf, 226 int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node) 227{ 228 int e; 229 struct sn_hwperf_object_info *nodeobj = NULL; 230 struct sn_hwperf_object_info *op; 231 struct sn_hwperf_object_info *dest; 232 struct sn_hwperf_object_info *router; 233 struct sn_hwperf_port_info ptdata[16]; 234 int sz, i, j; 235 cnodeid_t c; 236 int found_mem = 0; 237 int found_cpu = 0; 238 239 if (!cnode_possible(node)) 240 return -EINVAL; 241 242 if (sn_hwperf_has_cpus(node)) { 243 if (near_cpu_node) 244 *near_cpu_node = node; 245 found_cpu++; 246 } 247 248 if (sn_hwperf_has_mem(node)) { 249 if (near_mem_node) 250 *near_mem_node = node; 251 found_mem++; 252 } 253 254 if (found_cpu && found_mem) 255 return 0; /* trivially successful */ 256 257 /* find the argument node object */ 258 for (i=0, op=objbuf; i < nobj; i++, op++) { 259 if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op)) 260 continue; 261 if (node == sn_hwperf_obj_to_cnode(op)) { 262 nodeobj = op; 263 break; 264 } 265 } 266 if (!nodeobj) { 267 e = -ENOENT; 268 goto err; 269 } 270 271 /* get it's interconnect topology */ 272 sz = op->ports * sizeof(struct sn_hwperf_port_info); 273 if (sz > sizeof(ptdata)) 274 BUG(); 275 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 276 SN_HWPERF_ENUM_PORTS, nodeobj->id, sz, 277 (u64)&ptdata, 0, 0, NULL); 278 if (e != SN_HWPERF_OP_OK) { 279 e = -EINVAL; 280 goto err; 281 } 282 283 /* find nearest node with cpus and nearest memory */ 284 for (router=NULL, j=0; j < op->ports; j++) { 285 dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id); 286 if (dest && SN_HWPERF_IS_ROUTER(dest)) 287 router = dest; 288 if (!dest || SN_HWPERF_FOREIGN(dest) || 289 !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) { 290 continue; 291 } 292 c = sn_hwperf_obj_to_cnode(dest); 293 if (!found_cpu && sn_hwperf_has_cpus(c)) { 294 if (near_cpu_node) 295 *near_cpu_node = c; 296 found_cpu++; 297 } 298 if (!found_mem && sn_hwperf_has_mem(c)) { 299 if (near_mem_node) 300 *near_mem_node = c; 301 found_mem++; 302 } 303 } 304 305 if (router && (!found_cpu || !found_mem)) { 306 /* search for a node connected to the same router */ 307 sz = router->ports * sizeof(struct sn_hwperf_port_info); 308 if (sz > sizeof(ptdata)) 309 BUG(); 310 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 311 SN_HWPERF_ENUM_PORTS, router->id, sz, 312 (u64)&ptdata, 0, 0, NULL); 313 if (e != SN_HWPERF_OP_OK) { 314 e = -EINVAL; 315 goto err; 316 } 317 for (j=0; j < router->ports; j++) { 318 dest = sn_hwperf_findobj_id(objbuf, nobj, 319 ptdata[j].conn_id); 320 if (!dest || dest->id == node || 321 SN_HWPERF_FOREIGN(dest) || 322 !SN_HWPERF_IS_NODE(dest) || 323 SN_HWPERF_IS_IONODE(dest)) { 324 continue; 325 } 326 c = sn_hwperf_obj_to_cnode(dest); 327 if (!found_cpu && sn_hwperf_has_cpus(c)) { 328 if (near_cpu_node) 329 *near_cpu_node = c; 330 found_cpu++; 331 } 332 if (!found_mem && sn_hwperf_has_mem(c)) { 333 if (near_mem_node) 334 *near_mem_node = c; 335 found_mem++; 336 } 337 if (found_cpu && found_mem) 338 break; 339 } 340 } 341 342 if (!found_cpu || !found_mem) { 343 /* resort to _any_ node with CPUs and memory */ 344 for (i=0, op=objbuf; i < nobj; i++, op++) { 345 if (SN_HWPERF_FOREIGN(op) || 346 SN_HWPERF_IS_IONODE(op) || 347 !SN_HWPERF_IS_NODE(op)) { 348 continue; 349 } 350 c = sn_hwperf_obj_to_cnode(op); 351 if (!found_cpu && sn_hwperf_has_cpus(c)) { 352 if (near_cpu_node) 353 *near_cpu_node = c; 354 found_cpu++; 355 } 356 if (!found_mem && sn_hwperf_has_mem(c)) { 357 if (near_mem_node) 358 *near_mem_node = c; 359 found_mem++; 360 } 361 if (found_cpu && found_mem) 362 break; 363 } 364 } 365 366 if (!found_cpu || !found_mem) 367 e = -ENODATA; 368 369err: 370 return e; 371} 372 373 374static int sn_topology_show(struct seq_file *s, void *d) 375{ 376 int sz; 377 int pt; 378 int e = 0; 379 int i; 380 int j; 381 const char *slabname; 382 int ordinal; 383 cpumask_t cpumask; 384 char slice; 385 struct cpuinfo_ia64 *c; 386 struct sn_hwperf_port_info *ptdata; 387 struct sn_hwperf_object_info *p; 388 struct sn_hwperf_object_info *obj = d; /* this object */ 389 struct sn_hwperf_object_info *objs = s->private; /* all objects */ 390 u8 shubtype; 391 u8 system_size; 392 u8 sharing_size; 393 u8 partid; 394 u8 coher; 395 u8 nasid_shift; 396 u8 region_size; 397 u16 nasid_mask; 398 int nasid_msb; 399 400 if (obj == objs) { 401 seq_printf(s, "# sn_topology version 2\n"); 402 seq_printf(s, "# objtype ordinal location partition" 403 " [attribute value [, ...]]\n"); 404 405 if (ia64_sn_get_sn_info(0, 406 &shubtype, &nasid_mask, &nasid_shift, &system_size, 407 &sharing_size, &partid, &coher, ®ion_size)) 408 BUG(); 409 for (nasid_msb=63; nasid_msb > 0; nasid_msb--) { 410 if (((u64)nasid_mask << nasid_shift) & (1ULL << nasid_msb)) 411 break; 412 } 413 seq_printf(s, "partition %u %s local " 414 "shubtype %s, " 415 "nasid_mask 0x%016lx, " 416 "nasid_bits %d:%d, " 417 "system_size %d, " 418 "sharing_size %d, " 419 "coherency_domain %d, " 420 "region_size %d\n", 421 422 partid, utsname()->nodename, 423 shubtype ? "shub2" : "shub1", 424 (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift, 425 system_size, sharing_size, coher, region_size); 426 427 print_pci_topology(s); 428 } 429 430 if (SN_HWPERF_FOREIGN(obj)) { 431 /* private in another partition: not interesting */ 432 return 0; 433 } 434 435 for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) { 436 if (obj->name[i] == ' ') 437 obj->name[i] = '_'; 438 } 439 440 slabname = sn_hwperf_get_slabname(obj, objs, &ordinal); 441 seq_printf(s, "%s %d %s %s asic %s", slabname, ordinal, obj->location, 442 obj->sn_hwp_this_part ? "local" : "shared", obj->name); 443 444 if (ordinal < 0 || (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))) 445 seq_putc(s, '\n'); 446 else { 447 cnodeid_t near_mem = -1; 448 cnodeid_t near_cpu = -1; 449 450 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal)); 451 452 if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt, 453 ordinal, &near_mem, &near_cpu) == 0) { 454 seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d", 455 near_mem, near_cpu); 456 } 457 458 if (!SN_HWPERF_IS_IONODE(obj)) { 459 for_each_online_node(i) { 460 seq_printf(s, i ? ":%d" : ", dist %d", 461 node_distance(ordinal, i)); 462 } 463 } 464 465 seq_putc(s, '\n'); 466 467 /* 468 * CPUs on this node, if any 469 */ 470 if (!SN_HWPERF_IS_IONODE(obj)) { 471 cpumask = node_to_cpumask(ordinal); 472 for_each_online_cpu(i) { 473 if (cpu_isset(i, cpumask)) { 474 slice = 'a' + cpuid_to_slice(i); 475 c = cpu_data(i); 476 seq_printf(s, "cpu %d %s%c local" 477 " freq %luMHz, arch ia64", 478 i, obj->location, slice, 479 c->proc_freq / 1000000); 480 for_each_online_cpu(j) { 481 seq_printf(s, j ? ":%d" : ", dist %d", 482 node_distance( 483 cpu_to_node(i), 484 cpu_to_node(j))); 485 } 486 seq_putc(s, '\n'); 487 } 488 } 489 } 490 } 491 492 if (obj->ports) { 493 /* 494 * numalink ports 495 */ 496 sz = obj->ports * sizeof(struct sn_hwperf_port_info); 497 if ((ptdata = kmalloc(sz, GFP_KERNEL)) == NULL) 498 return -ENOMEM; 499 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 500 SN_HWPERF_ENUM_PORTS, obj->id, sz, 501 (u64) ptdata, 0, 0, NULL); 502 if (e != SN_HWPERF_OP_OK) 503 return -EINVAL; 504 for (ordinal=0, p=objs; p != obj; p++) { 505 if (!SN_HWPERF_FOREIGN(p)) 506 ordinal += p->ports; 507 } 508 for (pt = 0; pt < obj->ports; pt++) { 509 for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) { 510 if (ptdata[pt].conn_id == p->id) { 511 break; 512 } 513 } 514 seq_printf(s, "numalink %d %s-%d", 515 ordinal+pt, obj->location, ptdata[pt].port); 516 517 if (i >= sn_hwperf_obj_cnt) { 518 /* no connection */ 519 seq_puts(s, " local endpoint disconnected" 520 ", protocol unknown\n"); 521 continue; 522 } 523 524 if (obj->sn_hwp_this_part && p->sn_hwp_this_part) 525 /* both ends local to this partition */ 526 seq_puts(s, " local"); 527 else if (SN_HWPERF_FOREIGN(p)) 528 /* both ends of the link in foreign partiton */ 529 seq_puts(s, " foreign"); 530 else 531 /* link straddles a partition */ 532 seq_puts(s, " shared"); 533 534 /* 535 * Unlikely, but strictly should query the LLP config 536 * registers because an NL4R can be configured to run 537 * NL3 protocol, even when not talking to an NL3 router. 538 * Ditto for node-node. 539 */ 540 seq_printf(s, " endpoint %s-%d, protocol %s\n", 541 p->location, ptdata[pt].conn_port, 542 (SN_HWPERF_IS_NL3ROUTER(obj) || 543 SN_HWPERF_IS_NL3ROUTER(p)) ? "LLP3" : "LLP4"); 544 } 545 kfree(ptdata); 546 } 547 548 return 0; 549} 550 551static void *sn_topology_start(struct seq_file *s, loff_t * pos) 552{ 553 struct sn_hwperf_object_info *objs = s->private; 554 555 if (*pos < sn_hwperf_obj_cnt) 556 return (void *)(objs + *pos); 557 558 return NULL; 559} 560 561static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos) 562{ 563 ++*pos; 564 return sn_topology_start(s, pos); 565} 566 567static void sn_topology_stop(struct seq_file *m, void *v) 568{ 569 return; 570} 571 572/* 573 * /proc/sgi_sn/sn_topology, read-only using seq_file 574 */ 575static struct seq_operations sn_topology_seq_ops = { 576 .start = sn_topology_start, 577 .next = sn_topology_next, 578 .stop = sn_topology_stop, 579 .show = sn_topology_show 580}; 581 582struct sn_hwperf_op_info { 583 u64 op; 584 struct sn_hwperf_ioctl_args *a; 585 void *p; 586 int *v0; 587 int ret; 588}; 589 590static void sn_hwperf_call_sal(void *info) 591{ 592 struct sn_hwperf_op_info *op_info = info; 593 int r; 594 595 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op, 596 op_info->a->arg, op_info->a->sz, 597 (u64) op_info->p, 0, 0, op_info->v0); 598 op_info->ret = r; 599} 600 601static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info) 602{ 603 u32 cpu; 604 u32 use_ipi; 605 int r = 0; 606 cpumask_t save_allowed; 607 608 cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32; 609 use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK; 610 op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK; 611 612 if (cpu != SN_HWPERF_ARG_ANY_CPU) { 613 if (cpu >= NR_CPUS || !cpu_online(cpu)) { 614 r = -EINVAL; 615 goto out; 616 } 617 } 618 619 if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) { 620 /* don't care, or already on correct cpu */ 621 sn_hwperf_call_sal(op_info); 622 } 623 else { 624 if (use_ipi) { 625 /* use an interprocessor interrupt to call SAL */ 626 smp_call_function_single(cpu, sn_hwperf_call_sal, 627 op_info, 1, 1); 628 } 629 else { 630 /* migrate the task before calling SAL */ 631 save_allowed = current->cpus_allowed; 632 set_cpus_allowed(current, cpumask_of_cpu(cpu)); 633 sn_hwperf_call_sal(op_info); 634 set_cpus_allowed(current, save_allowed); 635 } 636 } 637 r = op_info->ret; 638 639out: 640 return r; 641} 642 643/* map SAL hwperf error code to system error code */ 644static int sn_hwperf_map_err(int hwperf_err) 645{ 646 int e; 647 648 switch(hwperf_err) { 649 case SN_HWPERF_OP_OK: 650 e = 0; 651 break; 652 653 case SN_HWPERF_OP_NOMEM: 654 e = -ENOMEM; 655 break; 656 657 case SN_HWPERF_OP_NO_PERM: 658 e = -EPERM; 659 break; 660 661 case SN_HWPERF_OP_IO_ERROR: 662 e = -EIO; 663 break; 664 665 case SN_HWPERF_OP_BUSY: 666 e = -EBUSY; 667 break; 668 669 case SN_HWPERF_OP_RECONFIGURE: 670 e = -EAGAIN; 671 break; 672 673 case SN_HWPERF_OP_INVAL: 674 default: 675 e = -EINVAL; 676 break; 677 } 678 679 return e; 680} 681 682/* 683 * ioctl for "sn_hwperf" misc device 684 */ 685static int 686sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg) 687{ 688 struct sn_hwperf_ioctl_args a; 689 struct cpuinfo_ia64 *cdata; 690 struct sn_hwperf_object_info *objs; 691 struct sn_hwperf_object_info *cpuobj; 692 struct sn_hwperf_op_info op_info; 693 void *p = NULL; 694 int nobj; 695 char slice; 696 int node; 697 int r; 698 int v0; 699 int i; 700 int j; 701 702 unlock_kernel(); 703 704 /* only user requests are allowed here */ 705 if ((op & SN_HWPERF_OP_MASK) < 10) { 706 r = -EINVAL; 707 goto error; 708 } 709 r = copy_from_user(&a, (const void __user *)arg, 710 sizeof(struct sn_hwperf_ioctl_args)); 711 if (r != 0) { 712 r = -EFAULT; 713 goto error; 714 } 715 716 /* 717 * Allocate memory to hold a kernel copy of the user buffer. The 718 * buffer contents are either copied in or out (or both) of user 719 * space depending on the flags encoded in the requested operation. 720 */ 721 if (a.ptr) { 722 p = vmalloc(a.sz); 723 if (!p) { 724 r = -ENOMEM; 725 goto error; 726 } 727 } 728 729 if (op & SN_HWPERF_OP_MEM_COPYIN) { 730 r = copy_from_user(p, (const void __user *)a.ptr, a.sz); 731 if (r != 0) { 732 r = -EFAULT; 733 goto error; 734 } 735 } 736 737 switch (op) { 738 case SN_HWPERF_GET_CPU_INFO: 739 if (a.sz == sizeof(u64)) { 740 /* special case to get size needed */ 741 *(u64 *) p = (u64) num_online_cpus() * 742 sizeof(struct sn_hwperf_object_info); 743 } else 744 if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) { 745 r = -ENOMEM; 746 goto error; 747 } else 748 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) { 749 memset(p, 0, a.sz); 750 for (i = 0; i < nobj; i++) { 751 int cpuobj_index = 0; 752 if (!SN_HWPERF_IS_NODE(objs + i)) 753 continue; 754 node = sn_hwperf_obj_to_cnode(objs + i); 755 for_each_online_cpu(j) { 756 if (node != cpu_to_node(j)) 757 continue; 758 cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++; 759 slice = 'a' + cpuid_to_slice(j); 760 cdata = cpu_data(j); 761 cpuobj->id = j; 762 snprintf(cpuobj->name, 763 sizeof(cpuobj->name), 764 "CPU %luMHz %s", 765 cdata->proc_freq / 1000000, 766 cdata->vendor); 767 snprintf(cpuobj->location, 768 sizeof(cpuobj->location), 769 "%s%c", objs[i].location, 770 slice); 771 } 772 } 773 774 vfree(objs); 775 } 776 break; 777 778 case SN_HWPERF_GET_NODE_NASID: 779 if (a.sz != sizeof(u64) || 780 (node = a.arg) < 0 || !cnode_possible(node)) { 781 r = -EINVAL; 782 goto error; 783 } 784 *(u64 *)p = (u64)cnodeid_to_nasid(node); 785 break; 786 787 case SN_HWPERF_GET_OBJ_NODE: 788 if (a.sz != sizeof(u64) || a.arg < 0) { 789 r = -EINVAL; 790 goto error; 791 } 792 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) { 793 if (a.arg >= nobj) { 794 r = -EINVAL; 795 vfree(objs); 796 goto error; 797 } 798 if (objs[(i = a.arg)].id != a.arg) { 799 for (i = 0; i < nobj; i++) { 800 if (objs[i].id == a.arg) 801 break; 802 } 803 } 804 if (i == nobj) { 805 r = -EINVAL; 806 vfree(objs); 807 goto error; 808 } 809 810 if (!SN_HWPERF_IS_NODE(objs + i) && 811 !SN_HWPERF_IS_IONODE(objs + i)) { 812 r = -ENOENT; 813 vfree(objs); 814 goto error; 815 } 816 817 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i); 818 vfree(objs); 819 } 820 break; 821 822 case SN_HWPERF_GET_MMRS: 823 case SN_HWPERF_SET_MMRS: 824 case SN_HWPERF_OBJECT_DISTANCE: 825 op_info.p = p; 826 op_info.a = &a; 827 op_info.v0 = &v0; 828 op_info.op = op; 829 r = sn_hwperf_op_cpu(&op_info); 830 if (r) { 831 r = sn_hwperf_map_err(r); 832 a.v0 = v0; 833 goto error; 834 } 835 break; 836 837 default: 838 /* all other ops are a direct SAL call */ 839 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op, 840 a.arg, a.sz, (u64) p, 0, 0, &v0); 841 if (r) { 842 r = sn_hwperf_map_err(r); 843 goto error; 844 } 845 a.v0 = v0; 846 break; 847 } 848 849 if (op & SN_HWPERF_OP_MEM_COPYOUT) { 850 r = copy_to_user((void __user *)a.ptr, p, a.sz); 851 if (r != 0) { 852 r = -EFAULT; 853 goto error; 854 } 855 } 856 857error: 858 vfree(p); 859 860 lock_kernel(); 861 return r; 862} 863 864static const struct file_operations sn_hwperf_fops = { 865 .ioctl = sn_hwperf_ioctl, 866}; 867 868static struct miscdevice sn_hwperf_dev = { 869 MISC_DYNAMIC_MINOR, 870 "sn_hwperf", 871 &sn_hwperf_fops 872}; 873 874static int sn_hwperf_init(void) 875{ 876 u64 v; 877 int salr; 878 int e = 0; 879 880 /* single threaded, once-only initialization */ 881 down(&sn_hwperf_init_mutex); 882 883 if (sn_hwperf_salheap) { 884 up(&sn_hwperf_init_mutex); 885 return e; 886 } 887 888 /* 889 * The PROM code needs a fixed reference node. For convenience the 890 * same node as the console I/O is used. 891 */ 892 sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid(); 893 894 /* 895 * Request the needed size and install the PROM scratch area. 896 * The PROM keeps various tracking bits in this memory area. 897 */ 898 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 899 (u64) SN_HWPERF_GET_HEAPSIZE, 0, 900 (u64) sizeof(u64), (u64) &v, 0, 0, NULL); 901 if (salr != SN_HWPERF_OP_OK) { 902 e = -EINVAL; 903 goto out; 904 } 905 906 if ((sn_hwperf_salheap = vmalloc(v)) == NULL) { 907 e = -ENOMEM; 908 goto out; 909 } 910 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 911 SN_HWPERF_INSTALL_HEAP, 0, v, 912 (u64) sn_hwperf_salheap, 0, 0, NULL); 913 if (salr != SN_HWPERF_OP_OK) { 914 e = -EINVAL; 915 goto out; 916 } 917 918 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid, 919 SN_HWPERF_OBJECT_COUNT, 0, 920 sizeof(u64), (u64) &v, 0, 0, NULL); 921 if (salr != SN_HWPERF_OP_OK) { 922 e = -EINVAL; 923 goto out; 924 } 925 sn_hwperf_obj_cnt = (int)v; 926 927out: 928 if (e < 0 && sn_hwperf_salheap) { 929 vfree(sn_hwperf_salheap); 930 sn_hwperf_salheap = NULL; 931 sn_hwperf_obj_cnt = 0; 932 } 933 up(&sn_hwperf_init_mutex); 934 return e; 935} 936 937int sn_topology_open(struct inode *inode, struct file *file) 938{ 939 int e; 940 struct seq_file *seq; 941 struct sn_hwperf_object_info *objbuf; 942 int nobj; 943 944 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) { 945 e = seq_open(file, &sn_topology_seq_ops); 946 seq = file->private_data; 947 seq->private = objbuf; 948 } 949 950 return e; 951} 952 953int sn_topology_release(struct inode *inode, struct file *file) 954{ 955 struct seq_file *seq = file->private_data; 956 957 vfree(seq->private); 958 return seq_release(inode, file); 959} 960 961int sn_hwperf_get_nearest_node(cnodeid_t node, 962 cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node) 963{ 964 int e; 965 int nobj; 966 struct sn_hwperf_object_info *objbuf; 967 968 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) { 969 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj, 970 node, near_mem_node, near_cpu_node); 971 vfree(objbuf); 972 } 973 974 return e; 975} 976 977static int __devinit sn_hwperf_misc_register_init(void) 978{ 979 int e; 980 981 if (!ia64_platform_is("sn2")) 982 return 0; 983 984 sn_hwperf_init(); 985 986 /* 987 * Register a dynamic misc device for hwperf ioctls. Platforms 988 * supporting hotplug will create /dev/sn_hwperf, else user 989 * can to look up the minor number in /proc/misc. 990 */ 991 if ((e = misc_register(&sn_hwperf_dev)) != 0) { 992 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to " 993 "register misc device for \"%s\"\n", sn_hwperf_dev.name); 994 } 995 996 return e; 997} 998 999device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */ 1000EXPORT_SYMBOL(sn_hwperf_get_nearest_node); 1001