SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/misc/sgi-xp/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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
   7  */
   8
   9 /*
  10  * Cross Partition Communication (XPC) partition support.
  11  *
  12  *      This is the part of XPC that detects the presence/absence of
  13  *      other partitions. It provides a heartbeat and monitors the
  14  *      heartbeats of other partitions.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/cache.h>
  21 #include <linux/mmzone.h>
  22 #include <linux/nodemask.h>
  23 #include <asm/uncached.h>
  24 #include <asm/sn/bte.h>
  25 #include <asm/sn/intr.h>
  26 #include <asm/sn/sn_sal.h>
  27 #include <asm/sn/nodepda.h>
  28 #include <asm/sn/addrs.h>
  29 #include "xpc.h"
  30
  31 /* XPC is exiting flag */
  32 int xpc_exiting;
  33
  34 /* SH_IPI_ACCESS shub register value on startup */
  35 static u64 xpc_sh1_IPI_access;
  36 static u64 xpc_sh2_IPI_access0;
  37 static u64 xpc_sh2_IPI_access1;
  38 static u64 xpc_sh2_IPI_access2;
  39 static u64 xpc_sh2_IPI_access3;
  40
  41 /* original protection values for each node */
  42 u64 xpc_prot_vec[MAX_NUMNODES];
  43
  44 /* this partition's reserved page pointers */
  45 struct xpc_rsvd_page *xpc_rsvd_page;
  46 static u64 *xpc_part_nasids;
  47 static u64 *xpc_mach_nasids;
  48 struct xpc_vars *xpc_vars;
  49 struct xpc_vars_part *xpc_vars_part;
  50
  51 static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
  52 static int xp_nasid_mask_words; /* actual size in words of nasid mask */
  53
  54 struct xpc_partition *xpc_partitions;
  55
  56 /*
  57  * Generic buffer used to store a local copy of portions of a remote
  58  * partition's reserved page (either its header and part_nasids mask,
  59  * or its vars).
  60  */
  61 char *xpc_remote_copy_buffer;
  62 void *xpc_remote_copy_buffer_base;
  63
  64 /*
  65  * Guarantee that the kmalloc'd memory is cacheline aligned.
  66  */
  67 void *
  68 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
  69 {
  70         /* see if kmalloc will give us cachline aligned memory by default */
  71         *base = kmalloc(size, flags);
  72         if (*base == NULL)
  73                 return NULL;
  74
  75         if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
  76                 return *base;
  77
  78         kfree(*base);
  79
  80         /* nope, we'll have to do it ourselves */
  81         *base = kmalloc(size + L1_CACHE_BYTES, flags);
  82         if (*base == NULL)
  83                 return NULL;
  84
  85         return (void *)L1_CACHE_ALIGN((u64)*base);
  86 }
  87
  88 /*
  89  * Given a nasid, get the physical address of the  partition's reserved page
  90  * for that nasid. This function returns 0 on any error.
  91  */
  92 static u64
  93 xpc_get_rsvd_page_pa(int nasid)
  94 {
  95         bte_result_t bte_res;
  96         s64 status;
  97         u64 cookie = 0;
  98         u64 rp_pa = nasid;      /* seed with nasid */
  99         u64 len = 0;
 100         u64 buf = buf;
 101         u64 buf_len = 0;
 102         void *buf_base = NULL;
 103
 104         while (1) {
 105
 106                 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
 107                                                        &len);
 108
 109                 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
 110                         "0x%016lx, address=0x%016lx, len=0x%016lx\n",
 111                         status, cookie, rp_pa, len);
 112
 113                 if (status != SALRET_MORE_PASSES)
 114                         break;
 115
 116                 if (L1_CACHE_ALIGN(len) > buf_len) {
 117                         kfree(buf_base);
 118                         buf_len = L1_CACHE_ALIGN(len);
 119                         buf = (u64)xpc_kmalloc_cacheline_aligned(buf_len,
 120                                                                  GFP_KERNEL,
 121                                                                  &buf_base);
 122                         if (buf_base == NULL) {
 123                                 dev_err(xpc_part, "unable to kmalloc "
 124                                         "len=0x%016lx\n", buf_len);
 125                                 status = SALRET_ERROR;
 126                                 break;
 127                         }
 128                 }
 129
 130                 bte_res = xp_bte_copy(rp_pa, buf, buf_len,
 131                                       (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 132                 if (bte_res != BTE_SUCCESS) {
 133                         dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
 134                         status = SALRET_ERROR;
 135                         break;
 136                 }
 137         }
 138
 139         kfree(buf_base);
 140
 141         if (status != SALRET_OK)
 142                 rp_pa = 0;
 143
 144         dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
 145         return rp_pa;
 146 }
 147
 148 /*
 149  * Fill the partition reserved page with the information needed by
 150  * other partitions to discover we are alive and establish initial
 151  * communications.
 152  */
 153 struct xpc_rsvd_page *
 154 xpc_rsvd_page_init(void)
 155 {
 156         struct xpc_rsvd_page *rp;
 157         AMO_t *amos_page;
 158         u64 rp_pa, nasid_array = 0;
 159         int i, ret;
 160
 161         /* get the local reserved page's address */
 162
 163         preempt_disable();
 164         rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
 165         preempt_enable();
 166         if (rp_pa == 0) {
 167                 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
 168                 return NULL;
 169         }
 170         rp = (struct xpc_rsvd_page *)__va(rp_pa);
 171
 172         if (rp->partid != sn_partition_id) {
 173                 dev_err(xpc_part, "the reserved page's partid of %d should be "
 174                         "%d\n", rp->partid, sn_partition_id);
 175                 return NULL;
 176         }
 177
 178         rp->version = XPC_RP_VERSION;
 179
 180         /* establish the actual sizes of the nasid masks */
 181         if (rp->SAL_version == 1) {
 182                 /* SAL_version 1 didn't set the nasids_size field */
 183                 rp->nasids_size = 128;
 184         }
 185         xp_nasid_mask_bytes = rp->nasids_size;
 186         xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
 187
 188         /* setup the pointers to the various items in the reserved page */
 189         xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
 190         xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
 191         xpc_vars = XPC_RP_VARS(rp);
 192         xpc_vars_part = XPC_RP_VARS_PART(rp);
 193
 194         /*
 195          * Before clearing xpc_vars, see if a page of AMOs had been previously
 196          * allocated. If not we'll need to allocate one and set permissions
 197          * so that cross-partition AMOs are allowed.
 198          *
 199          * The allocated AMO page needs MCA reporting to remain disabled after
 200          * XPC has unloaded.  To make this work, we keep a copy of the pointer
 201          * to this page (i.e., amos_page) in the struct xpc_vars structure,
 202          * which is pointed to by the reserved page, and re-use that saved copy
 203          * on subsequent loads of XPC. This AMO page is never freed, and its
 204          * memory protections are never restricted.
 205          */
 206         amos_page = xpc_vars->amos_page;
 207         if (amos_page == NULL) {
 208                 amos_page = (AMO_t *)TO_AMO(uncached_alloc_page(0, 1));
 209                 if (amos_page == NULL) {
 210                         dev_err(xpc_part, "can't allocate page of AMOs\n");
 211                         return NULL;
 212                 }
 213
 214                 /*
 215                  * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
 216                  * when xpc_allow_IPI_ops() is called via xpc_hb_init().
 217                  */
 218                 if (!enable_shub_wars_1_1()) {
 219                         ret = sn_change_memprotect(ia64_tpa((u64)amos_page),
 220                                                    PAGE_SIZE,
 221                                                    SN_MEMPROT_ACCESS_CLASS_1,
 222                                                    &nasid_array);
 223                         if (ret != 0) {
 224                                 dev_err(xpc_part, "can't change memory "
 225                                         "protections\n");
 226                                 uncached_free_page(__IA64_UNCACHED_OFFSET |
 227                                                    TO_PHYS((u64)amos_page), 1);
 228                                 return NULL;
 229                         }
 230                 }
 231         } else if (!IS_AMO_ADDRESS((u64)amos_page)) {
 232                 /*
 233                  * EFI's XPBOOT can also set amos_page in the reserved page,
 234                  * but it happens to leave it as an uncached physical address
 235                  * and we need it to be an uncached virtual, so we'll have to
 236                  * convert it.
 237                  */
 238                 if (!IS_AMO_PHYS_ADDRESS((u64)amos_page)) {
 239                         dev_err(xpc_part, "previously used amos_page address "
 240                                 "is bad = 0x%p\n", (void *)amos_page);
 241                         return NULL;
 242                 }
 243                 amos_page = (AMO_t *)TO_AMO((u64)amos_page);
 244         }
 245
 246         /* clear xpc_vars */
 247         memset(xpc_vars, 0, sizeof(struct xpc_vars));
 248
 249         xpc_vars->version = XPC_V_VERSION;
 250         xpc_vars->act_nasid = cpuid_to_nasid(0);
 251         xpc_vars->act_phys_cpuid = cpu_physical_id(0);
 252         xpc_vars->vars_part_pa = __pa(xpc_vars_part);
 253         xpc_vars->amos_page_pa = ia64_tpa((u64)amos_page);
 254         xpc_vars->amos_page = amos_page;        /* save for next load of XPC */
 255
 256         /* clear xpc_vars_part */
 257         memset((u64 *)xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
 258                xp_max_npartitions);
 259
 260         /* initialize the activate IRQ related AMO variables */
 261         for (i = 0; i < xp_nasid_mask_words; i++)
 262                 (void)xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
 263
 264         /* initialize the engaged remote partitions related AMO variables */
 265         (void)xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
 266         (void)xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
 267
 268         /* timestamp of when reserved page was setup by XPC */
 269         rp->stamp = CURRENT_TIME;
 270
 271         /*
 272          * This signifies to the remote partition that our reserved
 273          * page is initialized.
 274          */
 275         rp->vars_pa = __pa(xpc_vars);
 276
 277         return rp;
 278 }
 279
 280 /*
 281  * Change protections to allow IPI operations (and AMO operations on
 282  * Shub 1.1 systems).
 283  */
 284 void
 285 xpc_allow_IPI_ops(void)
 286 {
 287         int node;
 288         int nasid;
 289
 290         /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
 291
 292         if (is_shub2()) {
 293                 xpc_sh2_IPI_access0 =
 294                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
 295                 xpc_sh2_IPI_access1 =
 296                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
 297                 xpc_sh2_IPI_access2 =
 298                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
 299                 xpc_sh2_IPI_access3 =
 300                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
 301
 302                 for_each_online_node(node) {
 303                         nasid = cnodeid_to_nasid(node);
 304                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 305                               -1UL);
 306                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 307                               -1UL);
 308                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 309                               -1UL);
 310                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 311                               -1UL);
 312                 }
 313
 314         } else {
 315                 xpc_sh1_IPI_access =
 316                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
 317
 318                 for_each_online_node(node) {
 319                         nasid = cnodeid_to_nasid(node);
 320                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 321                               -1UL);
 322
 323                         /*
 324                          * Since the BIST collides with memory operations on
 325                          * SHUB 1.1 sn_change_memprotect() cannot be used.
 326                          */
 327                         if (enable_shub_wars_1_1()) {
 328                                 /* open up everything */
 329                                 xpc_prot_vec[node] = (u64)HUB_L((u64 *)
 330                                                                 GLOBAL_MMR_ADDR
 331                                                                 (nasid,
 332                                                   SH1_MD_DQLP_MMR_DIR_PRIVEC0));
 333                                 HUB_S((u64 *)
 334                                       GLOBAL_MMR_ADDR(nasid,
 335                                                    SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 336                                       -1UL);
 337                                 HUB_S((u64 *)
 338                                       GLOBAL_MMR_ADDR(nasid,
 339                                                    SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 340                                       -1UL);
 341                         }
 342                 }
 343         }
 344 }
 345
 346 /*
 347  * Restrict protections to disallow IPI operations (and AMO operations on
 348  * Shub 1.1 systems).
 349  */
 350 void
 351 xpc_restrict_IPI_ops(void)
 352 {
 353         int node;
 354         int nasid;
 355
 356         /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
 357
 358         if (is_shub2()) {
 359
 360                 for_each_online_node(node) {
 361                         nasid = cnodeid_to_nasid(node);
 362                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 363                               xpc_sh2_IPI_access0);
 364                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 365                               xpc_sh2_IPI_access1);
 366                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 367                               xpc_sh2_IPI_access2);
 368                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 369                               xpc_sh2_IPI_access3);
 370                 }
 371
 372         } else {
 373
 374                 for_each_online_node(node) {
 375                         nasid = cnodeid_to_nasid(node);
 376                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 377                               xpc_sh1_IPI_access);
 378
 379                         if (enable_shub_wars_1_1()) {
 380                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 381                                                    SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 382                                       xpc_prot_vec[node]);
 383                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 384                                                    SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 385                                       xpc_prot_vec[node]);
 386                         }
 387                 }
 388         }
 389 }
 390
 391 /*
 392  * At periodic intervals, scan through all active partitions and ensure
 393  * their heartbeat is still active.  If not, the partition is deactivated.
 394  */
 395 void
 396 xpc_check_remote_hb(void)
 397 {
 398         struct xpc_vars *remote_vars;
 399         struct xpc_partition *part;
 400         short partid;
 401         bte_result_t bres;
 402
 403         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
 404
 405         for (partid = 0; partid < xp_max_npartitions; partid++) {
 406
 407                 if (xpc_exiting)
 408                         break;
 409
 410                 if (partid == sn_partition_id)
 411                         continue;
 412
 413                 part = &xpc_partitions[partid];
 414
 415                 if (part->act_state == XPC_P_INACTIVE ||
 416                     part->act_state == XPC_P_DEACTIVATING) {
 417                         continue;
 418                 }
 419
 420                 /* pull the remote_hb cache line */
 421                 bres = xp_bte_copy(part->remote_vars_pa,
 422                                    (u64)remote_vars,
 423                                    XPC_RP_VARS_SIZE,
 424                                    (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 425                 if (bres != BTE_SUCCESS) {
 426                         XPC_DEACTIVATE_PARTITION(part,
 427                                                  xpc_map_bte_errors(bres));
 428                         continue;
 429                 }
 430
 431                 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
 432                         " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
 433                         partid, remote_vars->heartbeat, part->last_heartbeat,
 434                         remote_vars->heartbeat_offline,
 435                         remote_vars->heartbeating_to_mask);
 436
 437                 if (((remote_vars->heartbeat == part->last_heartbeat) &&
 438                      (remote_vars->heartbeat_offline == 0)) ||
 439                     !xpc_hb_allowed(sn_partition_id, remote_vars)) {
 440
 441                         XPC_DEACTIVATE_PARTITION(part, xpNoHeartbeat);
 442                         continue;
 443                 }
 444
 445                 part->last_heartbeat = remote_vars->heartbeat;
 446         }
 447 }
 448
 449 /*
 450  * Get a copy of a portion of the remote partition's rsvd page.
 451  *
 452  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
 453  * is large enough to contain a copy of their reserved page header and
 454  * part_nasids mask.
 455  */
 456 static enum xp_retval
 457 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
 458                   struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
 459 {
 460         int bres, i;
 461
 462         /* get the reserved page's physical address */
 463
 464         *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
 465         if (*remote_rp_pa == 0)
 466                 return xpNoRsvdPageAddr;
 467
 468         /* pull over the reserved page header and part_nasids mask */
 469         bres = xp_bte_copy(*remote_rp_pa, (u64)remote_rp,
 470                            XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
 471                            (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 472         if (bres != BTE_SUCCESS)
 473                 return xpc_map_bte_errors(bres);
 474
 475         if (discovered_nasids != NULL) {
 476                 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
 477
 478                 for (i = 0; i < xp_nasid_mask_words; i++)
 479                         discovered_nasids[i] |= remote_part_nasids[i];
 480         }
 481
 482         /* check that the partid is for another partition */
 483
 484         if (remote_rp->partid < 0 || remote_rp->partid >= xp_max_npartitions)
 485                 return xpInvalidPartid;
 486
 487         if (remote_rp->partid == sn_partition_id)
 488                 return xpLocalPartid;
 489
 490         if (XPC_VERSION_MAJOR(remote_rp->version) !=
 491             XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
 492                 return xpBadVersion;
 493         }
 494
 495         return xpSuccess;
 496 }
 497
 498 /*
 499  * Get a copy of the remote partition's XPC variables from the reserved page.
 500  *
 501  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
 502  * assumed to be of size XPC_RP_VARS_SIZE.
 503  */
 504 static enum xp_retval
 505 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
 506 {
 507         int bres;
 508
 509         if (remote_vars_pa == 0)
 510                 return xpVarsNotSet;
 511
 512         /* pull over the cross partition variables */
 513         bres = xp_bte_copy(remote_vars_pa, (u64)remote_vars, XPC_RP_VARS_SIZE,
 514                            (BTE_NOTIFY | BTE_WACQUIRE), NULL);
 515         if (bres != BTE_SUCCESS)
 516                 return xpc_map_bte_errors(bres);
 517
 518         if (XPC_VERSION_MAJOR(remote_vars->version) !=
 519             XPC_VERSION_MAJOR(XPC_V_VERSION)) {
 520                 return xpBadVersion;
 521         }
 522
 523         return xpSuccess;
 524 }
 525
 526 /*
 527  * Update the remote partition's info.
 528  */
 529 static void
 530 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
 531                           struct timespec *remote_rp_stamp, u64 remote_rp_pa,
 532                           u64 remote_vars_pa, struct xpc_vars *remote_vars)
 533 {
 534         part->remote_rp_version = remote_rp_version;
 535         dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
 536                 part->remote_rp_version);
 537
 538         part->remote_rp_stamp = *remote_rp_stamp;
 539         dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
 540                 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
 541
 542         part->remote_rp_pa = remote_rp_pa;
 543         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
 544
 545         part->remote_vars_pa = remote_vars_pa;
 546         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
 547                 part->remote_vars_pa);
 548
 549         part->last_heartbeat = remote_vars->heartbeat;
 550         dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
 551                 part->last_heartbeat);
 552
 553         part->remote_vars_part_pa = remote_vars->vars_part_pa;
 554         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
 555                 part->remote_vars_part_pa);
 556
 557         part->remote_act_nasid = remote_vars->act_nasid;
 558         dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
 559                 part->remote_act_nasid);
 560
 561         part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
 562         dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
 563                 part->remote_act_phys_cpuid);
 564
 565         part->remote_amos_page_pa = remote_vars->amos_page_pa;
 566         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
 567                 part->remote_amos_page_pa);
 568
 569         part->remote_vars_version = remote_vars->version;
 570         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
 571                 part->remote_vars_version);
 572 }
 573
 574 /*
 575  * Prior code has determined the nasid which generated an IPI.  Inspect
 576  * that nasid to determine if its partition needs to be activated or
 577  * deactivated.
 578  *
 579  * A partition is consider "awaiting activation" if our partition
 580  * flags indicate it is not active and it has a heartbeat.  A
 581  * partition is considered "awaiting deactivation" if our partition
 582  * flags indicate it is active but it has no heartbeat or it is not
 583  * sending its heartbeat to us.
 584  *
 585  * To determine the heartbeat, the remote nasid must have a properly
 586  * initialized reserved page.
 587  */
 588 static void
 589 xpc_identify_act_IRQ_req(int nasid)
 590 {
 591         struct xpc_rsvd_page *remote_rp;
 592         struct xpc_vars *remote_vars;
 593         u64 remote_rp_pa;
 594         u64 remote_vars_pa;
 595         int remote_rp_version;
 596         int reactivate = 0;
 597         int stamp_diff;
 598         struct timespec remote_rp_stamp = { 0, 0 };
 599         short partid;
 600         struct xpc_partition *part;
 601         enum xp_retval ret;
 602
 603         /* pull over the reserved page structure */
 604
 605         remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer;
 606
 607         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
 608         if (ret != xpSuccess) {
 609                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
 610                          "which sent interrupt, reason=%d\n", nasid, ret);
 611                 return;
 612         }
 613
 614         remote_vars_pa = remote_rp->vars_pa;
 615         remote_rp_version = remote_rp->version;
 616         if (XPC_SUPPORTS_RP_STAMP(remote_rp_version))
 617                 remote_rp_stamp = remote_rp->stamp;
 618
 619         partid = remote_rp->partid;
 620         part = &xpc_partitions[partid];
 621
 622         /* pull over the cross partition variables */
 623
 624         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
 625
 626         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
 627         if (ret != xpSuccess) {
 628
 629                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
 630                          "which sent interrupt, reason=%d\n", nasid, ret);
 631
 632                 XPC_DEACTIVATE_PARTITION(part, ret);
 633                 return;
 634         }
 635
 636         part->act_IRQ_rcvd++;
 637
 638         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
 639                 "%ld:0x%lx\n", (int)nasid, (int)partid, part->act_IRQ_rcvd,
 640                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
 641
 642         if (xpc_partition_disengaged(part) &&
 643             part->act_state == XPC_P_INACTIVE) {
 644
 645                 xpc_update_partition_info(part, remote_rp_version,
 646                                           &remote_rp_stamp, remote_rp_pa,
 647                                           remote_vars_pa, remote_vars);
 648
 649                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 650                         if (xpc_partition_disengage_requested(1UL << partid)) {
 651                                 /*
 652                                  * Other side is waiting on us to disengage,
 653                                  * even though we already have.
 654                                  */
 655                                 return;
 656                         }
 657                 } else {
 658                         /* other side doesn't support disengage requests */
 659                         xpc_clear_partition_disengage_request(1UL << partid);
 660                 }
 661
 662                 xpc_activate_partition(part);
 663                 return;
 664         }
 665
 666         DBUG_ON(part->remote_rp_version == 0);
 667         DBUG_ON(part->remote_vars_version == 0);
 668
 669         if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
 670                 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
 671                                                        remote_vars_version));
 672
 673                 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 674                         DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
 675                                                                version));
 676                         /* see if the other side rebooted */
 677                         if (part->remote_amos_page_pa ==
 678                             remote_vars->amos_page_pa &&
 679                             xpc_hb_allowed(sn_partition_id, remote_vars)) {
 680                                 /* doesn't look that way, so ignore the IPI */
 681                                 return;
 682                         }
 683                 }
 684
 685                 /*
 686                  * Other side rebooted and previous XPC didn't support the
 687                  * disengage request, so we don't need to do anything special.
 688                  */
 689
 690                 xpc_update_partition_info(part, remote_rp_version,
 691                                           &remote_rp_stamp, remote_rp_pa,
 692                                           remote_vars_pa, remote_vars);
 693                 part->reactivate_nasid = nasid;
 694                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 695                 return;
 696         }
 697
 698         DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
 699
 700         if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 701                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 702
 703                 /*
 704                  * Other side rebooted and previous XPC did support the
 705                  * disengage request, but the new one doesn't.
 706                  */
 707
 708                 xpc_clear_partition_engaged(1UL << partid);
 709                 xpc_clear_partition_disengage_request(1UL << partid);
 710
 711                 xpc_update_partition_info(part, remote_rp_version,
 712                                           &remote_rp_stamp, remote_rp_pa,
 713                                           remote_vars_pa, remote_vars);
 714                 reactivate = 1;
 715
 716         } else {
 717                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 718
 719                 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
 720                                                 &remote_rp_stamp);
 721                 if (stamp_diff != 0) {
 722                         DBUG_ON(stamp_diff >= 0);
 723
 724                         /*
 725                          * Other side rebooted and the previous XPC did support
 726                          * the disengage request, as does the new one.
 727                          */
 728
 729                         DBUG_ON(xpc_partition_engaged(1UL << partid));
 730                         DBUG_ON(xpc_partition_disengage_requested(1UL <<
 731                                                                   partid));
 732
 733                         xpc_update_partition_info(part, remote_rp_version,
 734                                                   &remote_rp_stamp,
 735                                                   remote_rp_pa, remote_vars_pa,
 736                                                   remote_vars);
 737                         reactivate = 1;
 738                 }
 739         }
 740
 741         if (part->disengage_request_timeout > 0 &&
 742             !xpc_partition_disengaged(part)) {
 743                 /* still waiting on other side to disengage from us */
 744                 return;
 745         }
 746
 747         if (reactivate) {
 748                 part->reactivate_nasid = nasid;
 749                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 750
 751         } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
 752                    xpc_partition_disengage_requested(1UL << partid)) {
 753                 XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
 754         }
 755 }
 756
 757 /*
 758  * Loop through the activation AMO variables and process any bits
 759  * which are set.  Each bit indicates a nasid sending a partition
 760  * activation or deactivation request.
 761  *
 762  * Return #of IRQs detected.
 763  */
 764 int
 765 xpc_identify_act_IRQ_sender(void)
 766 {
 767         int word, bit;
 768         u64 nasid_mask;
 769         u64 nasid;              /* remote nasid */
 770         int n_IRQs_detected = 0;
 771         AMO_t *act_amos;
 772
 773         act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
 774
 775         /* scan through act AMO variable looking for non-zero entries */
 776         for (word = 0; word < xp_nasid_mask_words; word++) {
 777
 778                 if (xpc_exiting)
 779                         break;
 780
 781                 nasid_mask = xpc_IPI_receive(&act_amos[word]);
 782                 if (nasid_mask == 0) {
 783                         /* no IRQs from nasids in this variable */
 784                         continue;
 785                 }
 786
 787                 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
 788                         nasid_mask);
 789
 790                 /*
 791                  * If this nasid has been added to the machine since
 792                  * our partition was reset, this will retain the
 793                  * remote nasid in our reserved pages machine mask.
 794                  * This is used in the event of module reload.
 795                  */
 796                 xpc_mach_nasids[word] |= nasid_mask;
 797
 798                 /* locate the nasid(s) which sent interrupts */
 799
 800                 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
 801                         if (nasid_mask & (1UL << bit)) {
 802                                 n_IRQs_detected++;
 803                                 nasid = XPC_NASID_FROM_W_B(word, bit);
 804                                 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
 805                                         nasid);
 806                                 xpc_identify_act_IRQ_req(nasid);
 807                         }
 808                 }
 809         }
 810         return n_IRQs_detected;
 811 }
 812
 813 /*
 814  * See if the other side has responded to a partition disengage request
 815  * from us.
 816  */
 817 int
 818 xpc_partition_disengaged(struct xpc_partition *part)
 819 {
 820         short partid = XPC_PARTID(part);
 821         int disengaged;
 822
 823         disengaged = (xpc_partition_engaged(1UL << partid) == 0);
 824         if (part->disengage_request_timeout) {
 825                 if (!disengaged) {
 826                         if (time_before(jiffies,
 827                             part->disengage_request_timeout)) {
 828                                 /* timelimit hasn't been reached yet */
 829                                 return 0;
 830                         }
 831
 832                         /*
 833                          * Other side hasn't responded to our disengage
 834                          * request in a timely fashion, so assume it's dead.
 835                          */
 836
 837                         dev_info(xpc_part, "disengage from remote partition %d "
 838                                  "timed out\n", partid);
 839                         xpc_disengage_request_timedout = 1;
 840                         xpc_clear_partition_engaged(1UL << partid);
 841                         disengaged = 1;
 842                 }
 843                 part->disengage_request_timeout = 0;
 844
 845                 /* cancel the timer function, provided it's not us */
 846                 if (!in_interrupt()) {
 847                         del_singleshot_timer_sync(&part->
 848                                                   disengage_request_timer);
 849                 }
 850
 851                 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
 852                         part->act_state != XPC_P_INACTIVE);
 853                 if (part->act_state != XPC_P_INACTIVE)
 854                         xpc_wakeup_channel_mgr(part);
 855
 856                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version))
 857                         xpc_cancel_partition_disengage_request(part);
 858         }
 859         return disengaged;
 860 }
 861
 862 /*
 863  * Mark specified partition as active.
 864  */
 865 enum xp_retval
 866 xpc_mark_partition_active(struct xpc_partition *part)
 867 {
 868         unsigned long irq_flags;
 869         enum xp_retval ret;
 870
 871         dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
 872
 873         spin_lock_irqsave(&part->act_lock, irq_flags);
 874         if (part->act_state == XPC_P_ACTIVATING) {
 875                 part->act_state = XPC_P_ACTIVE;
 876                 ret = xpSuccess;
 877         } else {
 878                 DBUG_ON(part->reason == xpSuccess);
 879                 ret = part->reason;
 880         }
 881         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 882
 883         return ret;
 884 }
 885
 886 /*
 887  * Notify XPC that the partition is down.
 888  */
 889 void
 890 xpc_deactivate_partition(const int line, struct xpc_partition *part,
 891                          enum xp_retval reason)
 892 {
 893         unsigned long irq_flags;
 894
 895         spin_lock_irqsave(&part->act_lock, irq_flags);
 896
 897         if (part->act_state == XPC_P_INACTIVE) {
 898                 XPC_SET_REASON(part, reason, line);
 899                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 900                 if (reason == xpReactivating) {
 901                         /* we interrupt ourselves to reactivate partition */
 902                         xpc_IPI_send_reactivate(part);
 903                 }
 904                 return;
 905         }
 906         if (part->act_state == XPC_P_DEACTIVATING) {
 907                 if ((part->reason == xpUnloading && reason != xpUnloading) ||
 908                     reason == xpReactivating) {
 909                         XPC_SET_REASON(part, reason, line);
 910                 }
 911                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 912                 return;
 913         }
 914
 915         part->act_state = XPC_P_DEACTIVATING;
 916         XPC_SET_REASON(part, reason, line);
 917
 918         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 919
 920         if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 921                 xpc_request_partition_disengage(part);
 922                 xpc_IPI_send_disengage(part);
 923
 924                 /* set a timelimit on the disengage request */
 925                 part->disengage_request_timeout = jiffies +
 926                     (xpc_disengage_request_timelimit * HZ);
 927                 part->disengage_request_timer.expires =
 928                     part->disengage_request_timeout;
 929                 add_timer(&part->disengage_request_timer);
 930         }
 931
 932         dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
 933                 XPC_PARTID(part), reason);
 934
 935         xpc_partition_going_down(part, reason);
 936 }
 937
 938 /*
 939  * Mark specified partition as inactive.
 940  */
 941 void
 942 xpc_mark_partition_inactive(struct xpc_partition *part)
 943 {
 944         unsigned long irq_flags;
 945
 946         dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
 947                 XPC_PARTID(part));
 948
 949         spin_lock_irqsave(&part->act_lock, irq_flags);
 950         part->act_state = XPC_P_INACTIVE;
 951         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 952         part->remote_rp_pa = 0;
 953 }
 954
 955 /*
 956  * SAL has provided a partition and machine mask.  The partition mask
 957  * contains a bit for each even nasid in our partition.  The machine
 958  * mask contains a bit for each even nasid in the entire machine.
 959  *
 960  * Using those two bit arrays, we can determine which nasids are
 961  * known in the machine.  Each should also have a reserved page
 962  * initialized if they are available for partitioning.
 963  */
 964 void
 965 xpc_discovery(void)
 966 {
 967         void *remote_rp_base;
 968         struct xpc_rsvd_page *remote_rp;
 969         struct xpc_vars *remote_vars;
 970         u64 remote_rp_pa;
 971         u64 remote_vars_pa;
 972         int region;
 973         int region_size;
 974         int max_regions;
 975         int nasid;
 976         struct xpc_rsvd_page *rp;
 977         short partid;
 978         struct xpc_partition *part;
 979         u64 *discovered_nasids;
 980         enum xp_retval ret;
 981
 982         remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
 983                                                   xp_nasid_mask_bytes,
 984                                                   GFP_KERNEL, &remote_rp_base);
 985         if (remote_rp == NULL)
 986                 return;
 987
 988         remote_vars = (struct xpc_vars *)remote_rp;
 989
 990         discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
 991                                     GFP_KERNEL);
 992         if (discovered_nasids == NULL) {
 993                 kfree(remote_rp_base);
 994                 return;
 995         }
 996
 997         rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
 998
 999         /*
1000          * The term 'region' in this context refers to the minimum number of
1001          * nodes that can comprise an access protection grouping. The access
1002          * protection is in regards to memory, IOI and IPI.
1003          */
1004         max_regions = 64;
1005         region_size = sn_region_size;
1006
1007         switch (region_size) {
1008         case 128:
1009                 max_regions *= 2;
1010         case 64:
1011                 max_regions *= 2;
1012         case 32:
1013                 max_regions *= 2;
1014                 region_size = 16;
1015                 DBUG_ON(!is_shub2());
1016         }
1017
1018         for (region = 0; region < max_regions; region++) {
1019
1020                 if (xpc_exiting)
1021                         break;
1022
1023                 dev_dbg(xpc_part, "searching region %d\n", region);
1024
1025                 for (nasid = (region * region_size * 2);
1026                      nasid < ((region + 1) * region_size * 2); nasid += 2) {
1027
1028                         if (xpc_exiting)
1029                                 break;
1030
1031                         dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1032
1033                         if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
1034                                 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1035                                         "part of the local partition; skipping "
1036                                         "region\n", nasid);
1037                                 break;
1038                         }
1039
1040                         if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
1041                                 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1042                                         "not on Numa-Link network at reset\n",
1043                                         nasid);
1044                                 continue;
1045                         }
1046
1047                         if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1048                                 dev_dbg(xpc_part, "Nasid %d is part of a "
1049                                         "partition which was previously "
1050                                         "discovered\n", nasid);
1051                                 continue;
1052                         }
1053
1054                         /* pull over the reserved page structure */
1055
1056                         ret = xpc_get_remote_rp(nasid, discovered_nasids,
1057                                                 remote_rp, &remote_rp_pa);
1058                         if (ret != xpSuccess) {
1059                                 dev_dbg(xpc_part, "unable to get reserved page "
1060                                         "from nasid %d, reason=%d\n", nasid,
1061                                         ret);
1062
1063                                 if (ret == xpLocalPartid)
1064                                         break;
1065
1066                                 continue;
1067                         }
1068
1069                         remote_vars_pa = remote_rp->vars_pa;
1070
1071                         partid = remote_rp->partid;
1072                         part = &xpc_partitions[partid];
1073
1074                         /* pull over the cross partition variables */
1075
1076                         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1077                         if (ret != xpSuccess) {
1078                                 dev_dbg(xpc_part, "unable to get XPC variables "
1079                                         "from nasid %d, reason=%d\n", nasid,
1080                                         ret);
1081
1082                                 XPC_DEACTIVATE_PARTITION(part, ret);
1083                                 continue;
1084                         }
1085
1086                         if (part->act_state != XPC_P_INACTIVE) {
1087                                 dev_dbg(xpc_part, "partition %d on nasid %d is "
1088                                         "already activating\n", partid, nasid);
1089                                 break;
1090                         }
1091
1092                         /*
1093                          * Register the remote partition's AMOs with SAL so it
1094                          * can handle and cleanup errors within that address
1095                          * range should the remote partition go down. We don't
1096                          * unregister this range because it is difficult to
1097                          * tell when outstanding writes to the remote partition
1098                          * are finished and thus when it is thus safe to
1099                          * unregister. This should not result in wasted space
1100                          * in the SAL xp_addr_region table because we should
1101                          * get the same page for remote_act_amos_pa after
1102                          * module reloads and system reboots.
1103                          */
1104                         if (sn_register_xp_addr_region
1105                             (remote_vars->amos_page_pa, PAGE_SIZE, 1) < 0) {
1106                                 dev_dbg(xpc_part,
1107                                         "partition %d failed to "
1108                                         "register xp_addr region 0x%016lx\n",
1109                                         partid, remote_vars->amos_page_pa);
1110
1111                                 XPC_SET_REASON(part, xpPhysAddrRegFailed,
1112                                                __LINE__);
1113                                 break;
1114                         }
1115
1116                         /*
1117                          * The remote nasid is valid and available.
1118                          * Send an interrupt to that nasid to notify
1119                          * it that we are ready to begin activation.
1120                          */
1121                         dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1122                                 "nasid %d, phys_cpuid 0x%x\n",
1123                                 remote_vars->amos_page_pa,
1124                                 remote_vars->act_nasid,
1125                                 remote_vars->act_phys_cpuid);
1126
1127                         if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1128                                                            version)) {
1129                                 part->remote_amos_page_pa =
1130                                     remote_vars->amos_page_pa;
1131                                 xpc_mark_partition_disengaged(part);
1132                                 xpc_cancel_partition_disengage_request(part);
1133                         }
1134                         xpc_IPI_send_activate(remote_vars);
1135                 }
1136         }
1137
1138         kfree(discovered_nasids);
1139         kfree(remote_rp_base);
1140 }
1141
1142 /*
1143  * Given a partid, get the nasids owned by that partition from the
1144  * remote partition's reserved page.
1145  */
1146 enum xp_retval
1147 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
1148 {
1149         struct xpc_partition *part;
1150         u64 part_nasid_pa;
1151         int bte_res;
1152
1153         part = &xpc_partitions[partid];
1154         if (part->remote_rp_pa == 0)
1155                 return xpPartitionDown;
1156
1157         memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1158
1159         part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
1160
1161         bte_res = xp_bte_copy(part_nasid_pa, (u64)nasid_mask,
1162                               xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE),
1163                               NULL);
1164
1165         return xpc_map_bte_errors(bte_res);
1166 }