arch/ia64/sn/kernel/xpc_partition.c

   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) 2004-2005 Silicon Graphics, Inc.  All Rights Reserved.
   7  */
   8
   9
  10 /*
  11  * Cross Partition Communication (XPC) partition support.
  12  *
  13  *      This is the part of XPC that detects the presence/absence of
  14  *      other partitions. It provides a heartbeat and monitors the
  15  *      heartbeats of other partitions.
  16  *
  17  */
  18
  19
  20 #include <linux/kernel.h>
  21 #include <linux/sysctl.h>
  22 #include <linux/cache.h>
  23 #include <linux/mmzone.h>
  24 #include <linux/nodemask.h>
  25 #include <asm/uncached.h>
  26 #include <asm/sn/bte.h>
  27 #include <asm/sn/intr.h>
  28 #include <asm/sn/sn_sal.h>
  29 #include <asm/sn/nodepda.h>
  30 #include <asm/sn/addrs.h>
  31 #include "xpc.h"
  32
  33
  34 /* XPC is exiting flag */
  35 int xpc_exiting;
  36
  37
  38 /* SH_IPI_ACCESS shub register value on startup */
  39 static u64 xpc_sh1_IPI_access;
  40 static u64 xpc_sh2_IPI_access0;
  41 static u64 xpc_sh2_IPI_access1;
  42 static u64 xpc_sh2_IPI_access2;
  43 static u64 xpc_sh2_IPI_access3;
  44
  45
  46 /* original protection values for each node */
  47 u64 xpc_prot_vec[MAX_COMPACT_NODES];
  48
  49
  50 /* this partition's reserved page */
  51 struct xpc_rsvd_page *xpc_rsvd_page;
  52
  53 /* this partition's XPC variables (within the reserved page) */
  54 struct xpc_vars *xpc_vars;
  55 struct xpc_vars_part *xpc_vars_part;
  56
  57
  58 /*
  59  * For performance reasons, each entry of xpc_partitions[] is cacheline
  60  * aligned. And xpc_partitions[] is padded with an additional entry at the
  61  * end so that the last legitimate entry doesn't share its cacheline with
  62  * another variable.
  63  */
  64 struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
  65
  66
  67 /*
  68  * Generic buffer used to store a local copy of the remote partitions
  69  * reserved page or XPC variables.
  70  *
  71  * xpc_discovery runs only once and is a seperate thread that is
  72  * very likely going to be processing in parallel with receiving
  73  * interrupts.
  74  */
  75 char ____cacheline_aligned
  76                 xpc_remote_copy_buffer[XPC_RSVD_PAGE_ALIGNED_SIZE];
  77
  78
  79 /*
  80  * Given a nasid, get the physical address of the  partition's reserved page
  81  * for that nasid. This function returns 0 on any error.
  82  */
  83 static u64
  84 xpc_get_rsvd_page_pa(int nasid, u64 buf, u64 buf_size)
  85 {
  86         bte_result_t bte_res;
  87         s64 status;
  88         u64 cookie = 0;
  89         u64 rp_pa = nasid;      /* seed with nasid */
  90         u64 len = 0;
  91
  92
  93         while (1) {
  94
  95                 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
  96                                                                 &len);
  97
  98                 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
  99                         "0x%016lx, address=0x%016lx, len=0x%016lx\n",
 100                         status, cookie, rp_pa, len);
 101
 102                 if (status != SALRET_MORE_PASSES) {
 103                         break;
 104                 }
 105
 106                 if (len > buf_size) {
 107                         dev_err(xpc_part, "len (=0x%016lx) > buf_size\n", len);
 108                         status = SALRET_ERROR;
 109                         break;
 110                 }
 111
 112                 bte_res = xp_bte_copy(rp_pa, ia64_tpa(buf), buf_size,
 113                                         (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 114                 if (bte_res != BTE_SUCCESS) {
 115                         dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
 116                         status = SALRET_ERROR;
 117                         break;
 118                 }
 119         }
 120
 121         if (status != SALRET_OK) {
 122                 rp_pa = 0;
 123         }
 124         dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
 125         return rp_pa;
 126 }
 127
 128
 129 /*
 130  * Fill the partition reserved page with the information needed by
 131  * other partitions to discover we are alive and establish initial
 132  * communications.
 133  */
 134 struct xpc_rsvd_page *
 135 xpc_rsvd_page_init(void)
 136 {
 137         struct xpc_rsvd_page *rp;
 138         AMO_t *amos_page;
 139         u64 rp_pa, next_cl, nasid_array = 0;
 140         int i, ret;
 141
 142
 143         /* get the local reserved page's address */
 144
 145         rp_pa = xpc_get_rsvd_page_pa(cnodeid_to_nasid(0),
 146                                         (u64) xpc_remote_copy_buffer,
 147                                                 XPC_RSVD_PAGE_ALIGNED_SIZE);
 148         if (rp_pa == 0) {
 149                 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
 150                 return NULL;
 151         }
 152         rp = (struct xpc_rsvd_page *) __va(rp_pa);
 153
 154         if (rp->partid != sn_partition_id) {
 155                 dev_err(xpc_part, "the reserved page's partid of %d should be "
 156                         "%d\n", rp->partid, sn_partition_id);
 157                 return NULL;
 158         }
 159
 160         rp->version = XPC_RP_VERSION;
 161
 162         /*
 163          * Place the XPC variables on the cache line following the
 164          * reserved page structure.
 165          */
 166         next_cl = (u64) rp + XPC_RSVD_PAGE_ALIGNED_SIZE;
 167         xpc_vars = (struct xpc_vars *) next_cl;
 168
 169         /*
 170          * Before clearing xpc_vars, see if a page of AMOs had been previously
 171          * allocated. If not we'll need to allocate one and set permissions
 172          * so that cross-partition AMOs are allowed.
 173          *
 174          * The allocated AMO page needs MCA reporting to remain disabled after
 175          * XPC has unloaded.  To make this work, we keep a copy of the pointer
 176          * to this page (i.e., amos_page) in the struct xpc_vars structure,
 177          * which is pointed to by the reserved page, and re-use that saved copy
 178          * on subsequent loads of XPC. This AMO page is never freed, and its
 179          * memory protections are never restricted.
 180          */
 181         if ((amos_page = xpc_vars->amos_page) == NULL) {
 182                 amos_page = (AMO_t *) TO_AMO(uncached_alloc_page(0));
 183                 if (amos_page == NULL) {
 184                         dev_err(xpc_part, "can't allocate page of AMOs\n");
 185                         return NULL;
 186                 }
 187
 188                 /*
 189                  * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
 190                  * when xpc_allow_IPI_ops() is called via xpc_hb_init().
 191                  */
 192                 if (!enable_shub_wars_1_1()) {
 193                         ret = sn_change_memprotect(ia64_tpa((u64) amos_page),
 194                                         PAGE_SIZE, SN_MEMPROT_ACCESS_CLASS_1,
 195                                         &nasid_array);
 196                         if (ret != 0) {
 197                                 dev_err(xpc_part, "can't change memory "
 198                                         "protections\n");
 199                                 uncached_free_page(__IA64_UNCACHED_OFFSET |
 200                                                    TO_PHYS((u64) amos_page));
 201                                 return NULL;
 202                         }
 203                 }
 204         } else if (!IS_AMO_ADDRESS((u64) amos_page)) {
 205                 /*
 206                  * EFI's XPBOOT can also set amos_page in the reserved page,
 207                  * but it happens to leave it as an uncached physical address
 208                  * and we need it to be an uncached virtual, so we'll have to
 209                  * convert it.
 210                  */
 211                 if (!IS_AMO_PHYS_ADDRESS((u64) amos_page)) {
 212                         dev_err(xpc_part, "previously used amos_page address "
 213                                 "is bad = 0x%p\n", (void *) amos_page);
 214                         return NULL;
 215                 }
 216                 amos_page = (AMO_t *) TO_AMO((u64) amos_page);
 217         }
 218
 219         memset(xpc_vars, 0, sizeof(struct xpc_vars));
 220
 221         /*
 222          * Place the XPC per partition specific variables on the cache line
 223          * following the XPC variables structure.
 224          */
 225         next_cl += XPC_VARS_ALIGNED_SIZE;
 226         memset((u64 *) next_cl, 0, sizeof(struct xpc_vars_part) *
 227                                                         XP_MAX_PARTITIONS);
 228         xpc_vars_part = (struct xpc_vars_part *) next_cl;
 229         xpc_vars->vars_part_pa = __pa(next_cl);
 230
 231         xpc_vars->version = XPC_V_VERSION;
 232         xpc_vars->act_nasid = cpuid_to_nasid(0);
 233         xpc_vars->act_phys_cpuid = cpu_physical_id(0);
 234         xpc_vars->amos_page = amos_page;  /* save for next load of XPC */
 235
 236
 237         /* initialize the activate IRQ related AMO variables */
 238         for (i = 0; i < XP_NASID_MASK_WORDS; i++) {
 239                 (void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
 240         }
 241
 242         /* initialize the engaged remote partitions related AMO variables */
 243         (void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
 244         (void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
 245
 246         /* export AMO page's physical address to other partitions */
 247         xpc_vars->amos_page_pa = ia64_tpa((u64) xpc_vars->amos_page);
 248
 249         /* timestamp of when reserved page was initialized */
 250         rp->stamp = CURRENT_TIME;
 251
 252         /*
 253          * This signifies to the remote partition that our reserved
 254          * page is initialized.
 255          */
 256         rp->vars_pa = __pa(xpc_vars);
 257
 258         return rp;
 259 }
 260
 261
 262 /*
 263  * Change protections to allow IPI operations (and AMO operations on
 264  * Shub 1.1 systems).
 265  */
 266 void
 267 xpc_allow_IPI_ops(void)
 268 {
 269         int node;
 270         int nasid;
 271
 272
 273         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
 274
 275         if (is_shub2()) {
 276                 xpc_sh2_IPI_access0 =
 277                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
 278                 xpc_sh2_IPI_access1 =
 279                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
 280                 xpc_sh2_IPI_access2 =
 281                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
 282                 xpc_sh2_IPI_access3 =
 283                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
 284
 285                 for_each_online_node(node) {
 286                         nasid = cnodeid_to_nasid(node);
 287                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 288                                                                 -1UL);
 289                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 290                                                                 -1UL);
 291                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 292                                                                 -1UL);
 293                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 294                                                                 -1UL);
 295                 }
 296
 297         } else {
 298                 xpc_sh1_IPI_access =
 299                         (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
 300
 301                 for_each_online_node(node) {
 302                         nasid = cnodeid_to_nasid(node);
 303                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 304                                                                 -1UL);
 305
 306                         /*
 307                          * Since the BIST collides with memory operations on
 308                          * SHUB 1.1 sn_change_memprotect() cannot be used.
 309                          */
 310                         if (enable_shub_wars_1_1()) {
 311                                 /* open up everything */
 312                                 xpc_prot_vec[node] = (u64) HUB_L((u64 *)
 313                                                 GLOBAL_MMR_ADDR(nasid,
 314                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0));
 315                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
 316                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 317                                                                 -1UL);
 318                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
 319                                                 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 320                                                                 -1UL);
 321                         }
 322                 }
 323         }
 324 }
 325
 326
 327 /*
 328  * Restrict protections to disallow IPI operations (and AMO operations on
 329  * Shub 1.1 systems).
 330  */
 331 void
 332 xpc_restrict_IPI_ops(void)
 333 {
 334         int node;
 335         int nasid;
 336
 337
 338         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
 339
 340         if (is_shub2()) {
 341
 342                 for_each_online_node(node) {
 343                         nasid = cnodeid_to_nasid(node);
 344                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 345                                                         xpc_sh2_IPI_access0);
 346                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 347                                                         xpc_sh2_IPI_access1);
 348                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 349                                                         xpc_sh2_IPI_access2);
 350                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 351                                                         xpc_sh2_IPI_access3);
 352                 }
 353
 354         } else {
 355
 356                 for_each_online_node(node) {
 357                         nasid = cnodeid_to_nasid(node);
 358                         HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 359                                                         xpc_sh1_IPI_access);
 360
 361                         if (enable_shub_wars_1_1()) {
 362                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
 363                                                 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 364                                                         xpc_prot_vec[node]);
 365                                 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
 366                                                 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 367                                                         xpc_prot_vec[node]);
 368                         }
 369                 }
 370         }
 371 }
 372
 373
 374 /*
 375  * At periodic intervals, scan through all active partitions and ensure
 376  * their heartbeat is still active.  If not, the partition is deactivated.
 377  */
 378 void
 379 xpc_check_remote_hb(void)
 380 {
 381         struct xpc_vars *remote_vars;
 382         struct xpc_partition *part;
 383         partid_t partid;
 384         bte_result_t bres;
 385
 386
 387         remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
 388
 389         for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
 390
 391                 if (xpc_exiting) {
 392                         break;
 393                 }
 394
 395                 if (partid == sn_partition_id) {
 396                         continue;
 397                 }
 398
 399                 part = &xpc_partitions[partid];
 400
 401                 if (part->act_state == XPC_P_INACTIVE ||
 402                                 part->act_state == XPC_P_DEACTIVATING) {
 403                         continue;
 404                 }
 405
 406                 /* pull the remote_hb cache line */
 407                 bres = xp_bte_copy(part->remote_vars_pa,
 408                                         ia64_tpa((u64) remote_vars),
 409                                         XPC_VARS_ALIGNED_SIZE,
 410                                         (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 411                 if (bres != BTE_SUCCESS) {
 412                         XPC_DEACTIVATE_PARTITION(part,
 413                                                 xpc_map_bte_errors(bres));
 414                         continue;
 415                 }
 416
 417                 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
 418                         " = %ld, kdb_status = %ld, HB_mask = 0x%lx\n", partid,
 419                         remote_vars->heartbeat, part->last_heartbeat,
 420                         remote_vars->kdb_status,
 421                         remote_vars->heartbeating_to_mask);
 422
 423                 if (((remote_vars->heartbeat == part->last_heartbeat) &&
 424                         (remote_vars->kdb_status == 0)) ||
 425                              !xpc_hb_allowed(sn_partition_id, remote_vars)) {
 426
 427                         XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
 428                         continue;
 429                 }
 430
 431                 part->last_heartbeat = remote_vars->heartbeat;
 432         }
 433 }
 434
 435
 436 /*
 437  * Get a copy of the remote partition's rsvd page.
 438  *
 439  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
 440  * assumed to be of size XPC_RSVD_PAGE_ALIGNED_SIZE.
 441  */
 442 static enum xpc_retval
 443 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
 444                 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
 445 {
 446         int bres, i;
 447
 448
 449         /* get the reserved page's physical address */
 450
 451         *remote_rp_pa = xpc_get_rsvd_page_pa(nasid, (u64) remote_rp,
 452                                                 XPC_RSVD_PAGE_ALIGNED_SIZE);
 453         if (*remote_rp_pa == 0) {
 454                 return xpcNoRsvdPageAddr;
 455         }
 456
 457
 458         /* pull over the reserved page structure */
 459
 460         bres = xp_bte_copy(*remote_rp_pa, ia64_tpa((u64) remote_rp),
 461                                 XPC_RSVD_PAGE_ALIGNED_SIZE,
 462                                 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 463         if (bres != BTE_SUCCESS) {
 464                 return xpc_map_bte_errors(bres);
 465         }
 466
 467
 468         if (discovered_nasids != NULL) {
 469                 for (i = 0; i < XP_NASID_MASK_WORDS; i++) {
 470                         discovered_nasids[i] |= remote_rp->part_nasids[i];
 471                 }
 472         }
 473
 474
 475         /* check that the partid is for another partition */
 476
 477         if (remote_rp->partid < 1 ||
 478                                 remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
 479                 return xpcInvalidPartid;
 480         }
 481
 482         if (remote_rp->partid == sn_partition_id) {
 483                 return xpcLocalPartid;
 484         }
 485
 486
 487         if (XPC_VERSION_MAJOR(remote_rp->version) !=
 488                                         XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
 489                 return xpcBadVersion;
 490         }
 491
 492         return xpcSuccess;
 493 }
 494
 495
 496 /*
 497  * Get a copy of the remote partition's XPC variables.
 498  *
 499  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
 500  * assumed to be of size XPC_VARS_ALIGNED_SIZE.
 501  */
 502 static enum xpc_retval
 503 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
 504 {
 505         int bres;
 506
 507
 508         if (remote_vars_pa == 0) {
 509                 return xpcVarsNotSet;
 510         }
 511
 512
 513         /* pull over the cross partition variables */
 514
 515         bres = xp_bte_copy(remote_vars_pa, ia64_tpa((u64) remote_vars),
 516                                 XPC_VARS_ALIGNED_SIZE,
 517                                 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 518         if (bres != BTE_SUCCESS) {
 519                 return xpc_map_bte_errors(bres);
 520         }
 521
 522         if (XPC_VERSION_MAJOR(remote_vars->version) !=
 523                                         XPC_VERSION_MAJOR(XPC_V_VERSION)) {
 524                 return xpcBadVersion;
 525         }
 526
 527         return xpcSuccess;
 528 }
 529
 530
 531 /*
 532  * Update the remote partition's info.
 533  */
 534 static void
 535 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
 536                 struct timespec *remote_rp_stamp, u64 remote_rp_pa,
 537                 u64 remote_vars_pa, struct xpc_vars *remote_vars)
 538 {
 539         part->remote_rp_version = remote_rp_version;
 540         dev_dbg(xpc_part, "  remote_rp_version = 0x%016lx\n",
 541                 part->remote_rp_version);
 542
 543         part->remote_rp_stamp = *remote_rp_stamp;
 544         dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
 545                 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
 546
 547         part->remote_rp_pa = remote_rp_pa;
 548         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
 549
 550         part->remote_vars_pa = remote_vars_pa;
 551         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
 552                 part->remote_vars_pa);
 553
 554         part->last_heartbeat = remote_vars->heartbeat;
 555         dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
 556                 part->last_heartbeat);
 557
 558         part->remote_vars_part_pa = remote_vars->vars_part_pa;
 559         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
 560                 part->remote_vars_part_pa);
 561
 562         part->remote_act_nasid = remote_vars->act_nasid;
 563         dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
 564                 part->remote_act_nasid);
 565
 566         part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
 567         dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
 568                 part->remote_act_phys_cpuid);
 569
 570         part->remote_amos_page_pa = remote_vars->amos_page_pa;
 571         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
 572                 part->remote_amos_page_pa);
 573
 574         part->remote_vars_version = remote_vars->version;
 575         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
 576                 part->remote_vars_version);
 577 }
 578
 579
 580 /*
 581  * Prior code has determined the nasid which generated an IPI.  Inspect
 582  * that nasid to determine if its partition needs to be activated or
 583  * deactivated.
 584  *
 585  * A partition is consider "awaiting activation" if our partition
 586  * flags indicate it is not active and it has a heartbeat.  A
 587  * partition is considered "awaiting deactivation" if our partition
 588  * flags indicate it is active but it has no heartbeat or it is not
 589  * sending its heartbeat to us.
 590  *
 591  * To determine the heartbeat, the remote nasid must have a properly
 592  * initialized reserved page.
 593  */
 594 static void
 595 xpc_identify_act_IRQ_req(int nasid)
 596 {
 597         struct xpc_rsvd_page *remote_rp;
 598         struct xpc_vars *remote_vars;
 599         u64 remote_rp_pa;
 600         u64 remote_vars_pa;
 601         int remote_rp_version;
 602         int reactivate = 0;
 603         int stamp_diff;
 604         struct timespec remote_rp_stamp = { 0, 0 };
 605         partid_t partid;
 606         struct xpc_partition *part;
 607         enum xpc_retval ret;
 608
 609
 610         /* pull over the reserved page structure */
 611
 612         remote_rp = (struct xpc_rsvd_page *) xpc_remote_copy_buffer;
 613
 614         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
 615         if (ret != xpcSuccess) {
 616                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
 617                         "which sent interrupt, reason=%d\n", nasid, ret);
 618                 return;
 619         }
 620
 621         remote_vars_pa = remote_rp->vars_pa;
 622         remote_rp_version = remote_rp->version;
 623         if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 624                 remote_rp_stamp = remote_rp->stamp;
 625         }
 626         partid = remote_rp->partid;
 627         part = &xpc_partitions[partid];
 628
 629
 630         /* pull over the cross partition variables */
 631
 632         remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
 633
 634         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
 635         if (ret != xpcSuccess) {
 636
 637                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
 638                         "which sent interrupt, reason=%d\n", nasid, ret);
 639
 640                 XPC_DEACTIVATE_PARTITION(part, ret);
 641                 return;
 642         }
 643
 644
 645         part->act_IRQ_rcvd++;
 646
 647         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
 648                 "%ld:0x%lx\n", (int) nasid, (int) partid, part->act_IRQ_rcvd,
 649                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
 650
 651         if (xpc_partition_disengaged(part) &&
 652                                         part->act_state == XPC_P_INACTIVE) {
 653
 654                 xpc_update_partition_info(part, remote_rp_version,
 655                                         &remote_rp_stamp, remote_rp_pa,
 656                                         remote_vars_pa, remote_vars);
 657
 658                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 659                         if (xpc_partition_disengage_requested(1UL << partid)) {
 660                                 /*
 661                                  * Other side is waiting on us to disengage,
 662                                  * even though we already have.
 663                                  */
 664                                 return;
 665                         }
 666                 } else {
 667                         /* other side doesn't support disengage requests */
 668                         xpc_clear_partition_disengage_request(1UL << partid);
 669                 }
 670
 671                 xpc_activate_partition(part);
 672                 return;
 673         }
 674
 675         DBUG_ON(part->remote_rp_version == 0);
 676         DBUG_ON(part->remote_vars_version == 0);
 677
 678         if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
 679                 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
 680                                                         remote_vars_version));
 681
 682                 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 683                         DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
 684                                                                 version));
 685                         /* see if the other side rebooted */
 686                         if (part->remote_amos_page_pa ==
 687                                 remote_vars->amos_page_pa &&
 688                                         xpc_hb_allowed(sn_partition_id,
 689                                                                 remote_vars)) {
 690                                 /* doesn't look that way, so ignore the IPI */
 691                                 return;
 692                         }
 693                 }
 694
 695                 /*
 696                  * Other side rebooted and previous XPC didn't support the
 697                  * disengage request, so we don't need to do anything special.
 698                  */
 699
 700                 xpc_update_partition_info(part, remote_rp_version,
 701                                                 &remote_rp_stamp, remote_rp_pa,
 702                                                 remote_vars_pa, remote_vars);
 703                 part->reactivate_nasid = nasid;
 704                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
 705                 return;
 706         }
 707
 708         DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
 709
 710         if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 711                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 712
 713                 /*
 714                  * Other side rebooted and previous XPC did support the
 715                  * disengage request, but the new one doesn't.
 716                  */
 717
 718                 xpc_clear_partition_engaged(1UL << partid);
 719                 xpc_clear_partition_disengage_request(1UL << partid);
 720
 721                 xpc_update_partition_info(part, remote_rp_version,
 722                                                 &remote_rp_stamp, remote_rp_pa,
 723                                                 remote_vars_pa, remote_vars);
 724                 reactivate = 1;
 725
 726         } else {
 727                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 728
 729                 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
 730                                                         &remote_rp_stamp);
 731                 if (stamp_diff != 0) {
 732                         DBUG_ON(stamp_diff >= 0);
 733
 734                         /*
 735                          * Other side rebooted and the previous XPC did support
 736                          * the disengage request, as does the new one.
 737                          */
 738
 739                         DBUG_ON(xpc_partition_engaged(1UL << partid));
 740                         DBUG_ON(xpc_partition_disengage_requested(1UL <<
 741                                                                 partid));
 742
 743                         xpc_update_partition_info(part, remote_rp_version,
 744                                                 &remote_rp_stamp, remote_rp_pa,
 745                                                 remote_vars_pa, remote_vars);
 746                         reactivate = 1;
 747                 }
 748         }
 749
 750         if (!xpc_partition_disengaged(part)) {
 751                 /* still waiting on other side to disengage from us */
 752                 return;
 753         }
 754
 755         if (reactivate) {
 756                 part->reactivate_nasid = nasid;
 757                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
 758
 759         } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
 760                         xpc_partition_disengage_requested(1UL << partid)) {
 761                 XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
 762         }
 763 }
 764
 765
 766 /*
 767  * Loop through the activation AMO variables and process any bits
 768  * which are set.  Each bit indicates a nasid sending a partition
 769  * activation or deactivation request.
 770  *
 771  * Return #of IRQs detected.
 772  */
 773 int
 774 xpc_identify_act_IRQ_sender(void)
 775 {
 776         int word, bit;
 777         u64 nasid_mask;
 778         u64 nasid;                      /* remote nasid */
 779         int n_IRQs_detected = 0;
 780         AMO_t *act_amos;
 781         struct xpc_rsvd_page *rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
 782
 783
 784         act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
 785
 786
 787         /* scan through act AMO variable looking for non-zero entries */
 788         for (word = 0; word < XP_NASID_MASK_WORDS; word++) {
 789
 790                 if (xpc_exiting) {
 791                         break;
 792                 }
 793
 794                 nasid_mask = xpc_IPI_receive(&act_amos[word]);
 795                 if (nasid_mask == 0) {
 796                         /* no IRQs from nasids in this variable */
 797                         continue;
 798                 }
 799
 800                 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
 801                         nasid_mask);
 802
 803
 804                 /*
 805                  * If this nasid has been added to the machine since
 806                  * our partition was reset, this will retain the
 807                  * remote nasid in our reserved pages machine mask.
 808                  * This is used in the event of module reload.
 809                  */
 810                 rp->mach_nasids[word] |= nasid_mask;
 811
 812
 813                 /* locate the nasid(s) which sent interrupts */
 814
 815                 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
 816                         if (nasid_mask & (1UL << bit)) {
 817                                 n_IRQs_detected++;
 818                                 nasid = XPC_NASID_FROM_W_B(word, bit);
 819                                 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
 820                                         nasid);
 821                                 xpc_identify_act_IRQ_req(nasid);
 822                         }
 823                 }
 824         }
 825         return n_IRQs_detected;
 826 }
 827
 828
 829 /*
 830  * See if the other side has responded to a partition disengage request
 831  * from us.
 832  */
 833 int
 834 xpc_partition_disengaged(struct xpc_partition *part)
 835 {
 836         partid_t partid = XPC_PARTID(part);
 837         int disengaged;
 838
 839
 840         disengaged = (xpc_partition_engaged(1UL << partid) == 0);
 841         if (part->disengage_request_timeout) {
 842                 if (!disengaged) {
 843                         if (jiffies < part->disengage_request_timeout) {
 844                                 /* timelimit hasn't been reached yet */
 845                                 return 0;
 846                         }
 847
 848                         /*
 849                          * Other side hasn't responded to our disengage
 850                          * request in a timely fashion, so assume it's dead.
 851                          */
 852
 853                         xpc_clear_partition_engaged(1UL << partid);
 854                         disengaged = 1;
 855                 }
 856                 part->disengage_request_timeout = 0;
 857
 858                 /* cancel the timer function, provided it's not us */
 859                 if (!in_interrupt()) {
 860                         del_singleshot_timer_sync(&part->
 861                                                       disengage_request_timer);
 862                 }
 863
 864                 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
 865                                         part->act_state != XPC_P_INACTIVE);
 866                 if (part->act_state != XPC_P_INACTIVE) {
 867                         xpc_wakeup_channel_mgr(part);
 868                 }
 869
 870                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 871                         xpc_cancel_partition_disengage_request(part);
 872                 }
 873         }
 874         return disengaged;
 875 }
 876
 877
 878 /*
 879  * Mark specified partition as active.
 880  */
 881 enum xpc_retval
 882 xpc_mark_partition_active(struct xpc_partition *part)
 883 {
 884         unsigned long irq_flags;
 885         enum xpc_retval ret;
 886
 887
 888         dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
 889
 890         spin_lock_irqsave(&part->act_lock, irq_flags);
 891         if (part->act_state == XPC_P_ACTIVATING) {
 892                 part->act_state = XPC_P_ACTIVE;
 893                 ret = xpcSuccess;
 894         } else {
 895                 DBUG_ON(part->reason == xpcSuccess);
 896                 ret = part->reason;
 897         }
 898         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 899
 900         return ret;
 901 }
 902
 903
 904 /*
 905  * Notify XPC that the partition is down.
 906  */
 907 void
 908 xpc_deactivate_partition(const int line, struct xpc_partition *part,
 909                                 enum xpc_retval reason)
 910 {
 911         unsigned long irq_flags;
 912
 913
 914         spin_lock_irqsave(&part->act_lock, irq_flags);
 915
 916         if (part->act_state == XPC_P_INACTIVE) {
 917                 XPC_SET_REASON(part, reason, line);
 918                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 919                 if (reason == xpcReactivating) {
 920                         /* we interrupt ourselves to reactivate partition */
 921                         xpc_IPI_send_reactivate(part);
 922                 }
 923                 return;
 924         }
 925         if (part->act_state == XPC_P_DEACTIVATING) {
 926                 if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
 927                                         reason == xpcReactivating) {
 928                         XPC_SET_REASON(part, reason, line);
 929                 }
 930                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 931                 return;
 932         }
 933
 934         part->act_state = XPC_P_DEACTIVATING;
 935         XPC_SET_REASON(part, reason, line);
 936
 937         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 938
 939         if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 940                 xpc_request_partition_disengage(part);
 941                 xpc_IPI_send_disengage(part);
 942
 943                 /* set a timelimit on the disengage request */
 944                 part->disengage_request_timeout = jiffies +
 945                                         (xpc_disengage_request_timelimit * HZ);
 946                 part->disengage_request_timer.expires =
 947                                         part->disengage_request_timeout;
 948                 add_timer(&part->disengage_request_timer);
 949         }
 950
 951         dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
 952                 XPC_PARTID(part), reason);
 953
 954         xpc_partition_going_down(part, reason);
 955 }
 956
 957
 958 /*
 959  * Mark specified partition as inactive.
 960  */
 961 void
 962 xpc_mark_partition_inactive(struct xpc_partition *part)
 963 {
 964         unsigned long irq_flags;
 965
 966
 967         dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
 968                 XPC_PARTID(part));
 969
 970         spin_lock_irqsave(&part->act_lock, irq_flags);
 971         part->act_state = XPC_P_INACTIVE;
 972         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 973         part->remote_rp_pa = 0;
 974 }
 975
 976
 977 /*
 978  * SAL has provided a partition and machine mask.  The partition mask
 979  * contains a bit for each even nasid in our partition.  The machine
 980  * mask contains a bit for each even nasid in the entire machine.
 981  *
 982  * Using those two bit arrays, we can determine which nasids are
 983  * known in the machine.  Each should also have a reserved page
 984  * initialized if they are available for partitioning.
 985  */
 986 void
 987 xpc_discovery(void)
 988 {
 989         void *remote_rp_base;
 990         struct xpc_rsvd_page *remote_rp;
 991         struct xpc_vars *remote_vars;
 992         u64 remote_rp_pa;
 993         u64 remote_vars_pa;
 994         int region;
 995         int max_regions;
 996         int nasid;
 997         struct xpc_rsvd_page *rp;
 998         partid_t partid;
 999         struct xpc_partition *part;
1000         u64 *discovered_nasids;
1001         enum xpc_retval ret;
1002
1003
1004         remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RSVD_PAGE_ALIGNED_SIZE,
1005                                                 GFP_KERNEL, &remote_rp_base);
1006         if (remote_rp == NULL) {
1007                 return;
1008         }
1009         remote_vars = (struct xpc_vars *) remote_rp;
1010
1011
1012         discovered_nasids = kmalloc(sizeof(u64) * XP_NASID_MASK_WORDS,
1013                                                         GFP_KERNEL);
1014         if (discovered_nasids == NULL) {
1015                 kfree(remote_rp_base);
1016                 return;
1017         }
1018         memset(discovered_nasids, 0, sizeof(u64) * XP_NASID_MASK_WORDS);
1019
1020         rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
1021
1022         /*
1023          * The term 'region' in this context refers to the minimum number of
1024          * nodes that can comprise an access protection grouping. The access
1025          * protection is in regards to memory, IOI and IPI.
1026          */
1027 //>>> move the next two #defines into either include/asm-ia64/sn/arch.h or
1028 //>>> include/asm-ia64/sn/addrs.h
1029 #define SH1_MAX_REGIONS         64
1030 #define SH2_MAX_REGIONS         256
1031         max_regions = is_shub2() ? SH2_MAX_REGIONS : SH1_MAX_REGIONS;
1032
1033         for (region = 0; region < max_regions; region++) {
1034
1035                 if ((volatile int) xpc_exiting) {
1036                         break;
1037                 }
1038
1039                 dev_dbg(xpc_part, "searching region %d\n", region);
1040
1041                 for (nasid = (region * sn_region_size * 2);
1042                      nasid < ((region + 1) * sn_region_size * 2);
1043                      nasid += 2) {
1044
1045                         if ((volatile int) xpc_exiting) {
1046                                 break;
1047                         }
1048
1049                         dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1050
1051
1052                         if (XPC_NASID_IN_ARRAY(nasid, rp->part_nasids)) {
1053                                 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1054                                         "part of the local partition; skipping "
1055                                         "region\n", nasid);
1056                                 break;
1057                         }
1058
1059                         if (!(XPC_NASID_IN_ARRAY(nasid, rp->mach_nasids))) {
1060                                 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1061                                         "not on Numa-Link network at reset\n",
1062                                         nasid);
1063                                 continue;
1064                         }
1065
1066                         if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1067                                 dev_dbg(xpc_part, "Nasid %d is part of a "
1068                                         "partition which was previously "
1069                                         "discovered\n", nasid);
1070                                 continue;
1071                         }
1072
1073
1074                         /* pull over the reserved page structure */
1075
1076                         ret = xpc_get_remote_rp(nasid, discovered_nasids,
1077                                               remote_rp, &remote_rp_pa);
1078                         if (ret != xpcSuccess) {
1079                                 dev_dbg(xpc_part, "unable to get reserved page "
1080                                         "from nasid %d, reason=%d\n", nasid,
1081                                         ret);
1082
1083                                 if (ret == xpcLocalPartid) {
1084                                         break;
1085                                 }
1086                                 continue;
1087                         }
1088
1089                         remote_vars_pa = remote_rp->vars_pa;
1090
1091                         partid = remote_rp->partid;
1092                         part = &xpc_partitions[partid];
1093
1094
1095                         /* pull over the cross partition variables */
1096
1097                         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1098                         if (ret != xpcSuccess) {
1099                                 dev_dbg(xpc_part, "unable to get XPC variables "
1100                                         "from nasid %d, reason=%d\n", nasid,
1101                                         ret);
1102
1103                                 XPC_DEACTIVATE_PARTITION(part, ret);
1104                                 continue;
1105                         }
1106
1107                         if (part->act_state != XPC_P_INACTIVE) {
1108                                 dev_dbg(xpc_part, "partition %d on nasid %d is "
1109                                         "already activating\n", partid, nasid);
1110                                 break;
1111                         }
1112
1113                         /*
1114                          * Register the remote partition's AMOs with SAL so it
1115                          * can handle and cleanup errors within that address
1116                          * range should the remote partition go down. We don't
1117                          * unregister this range because it is difficult to
1118                          * tell when outstanding writes to the remote partition
1119                          * are finished and thus when it is thus safe to
1120                          * unregister. This should not result in wasted space
1121                          * in the SAL xp_addr_region table because we should
1122                          * get the same page for remote_act_amos_pa after
1123                          * module reloads and system reboots.
1124                          */
1125                         if (sn_register_xp_addr_region(
1126                                             remote_vars->amos_page_pa,
1127                                                         PAGE_SIZE, 1) < 0) {
1128                                 dev_dbg(xpc_part, "partition %d failed to "
1129                                         "register xp_addr region 0x%016lx\n",
1130                                         partid, remote_vars->amos_page_pa);
1131
1132                                 XPC_SET_REASON(part, xpcPhysAddrRegFailed,
1133                                                 __LINE__);
1134                                 break;
1135                         }
1136
1137                         /*
1138                          * The remote nasid is valid and available.
1139                          * Send an interrupt to that nasid to notify
1140                          * it that we are ready to begin activation.
1141                          */
1142                         dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1143                                 "nasid %d, phys_cpuid 0x%x\n",
1144                                 remote_vars->amos_page_pa,
1145                                 remote_vars->act_nasid,
1146                                 remote_vars->act_phys_cpuid);
1147
1148                         if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1149                                                                 version)) {
1150                                 part->remote_amos_page_pa =
1151                                                 remote_vars->amos_page_pa;
1152                                 xpc_mark_partition_disengaged(part);
1153                                 xpc_cancel_partition_disengage_request(part);
1154                         }
1155                         xpc_IPI_send_activate(remote_vars);
1156                 }
1157         }
1158
1159         kfree(discovered_nasids);
1160         kfree(remote_rp_base);
1161 }
1162
1163
1164 /*
1165  * Given a partid, get the nasids owned by that partition from the
1166  * remote partition's reserved page.
1167  */
1168 enum xpc_retval
1169 xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
1170 {
1171         struct xpc_partition *part;
1172         u64 part_nasid_pa;
1173         int bte_res;
1174
1175
1176         part = &xpc_partitions[partid];
1177         if (part->remote_rp_pa == 0) {
1178                 return xpcPartitionDown;
1179         }
1180
1181         part_nasid_pa = part->remote_rp_pa +
1182                 (u64) &((struct xpc_rsvd_page *) 0)->part_nasids;
1183
1184         bte_res = xp_bte_copy(part_nasid_pa, ia64_tpa((u64) nasid_mask),
1185                                 L1_CACHE_ALIGN(XP_NASID_MASK_BYTES),
1186                                 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
1187
1188         return xpc_map_bte_errors(bte_res);
1189 }
1190