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/intr.h>
  25 #include <asm/sn/sn_sal.h>
  26 #include <asm/sn/nodepda.h>
  27 #include <asm/sn/addrs.h>
  28 #include "xpc.h"
  29
  30 /* XPC is exiting flag */
  31 int xpc_exiting;
  32
  33 /* SH_IPI_ACCESS shub register value on startup */
  34 static u64 xpc_sh1_IPI_access;
  35 static u64 xpc_sh2_IPI_access0;
  36 static u64 xpc_sh2_IPI_access1;
  37 static u64 xpc_sh2_IPI_access2;
  38 static u64 xpc_sh2_IPI_access3;
  39
  40 /* original protection values for each node */
  41 u64 xpc_prot_vec[MAX_NUMNODES];
  42
  43 /* this partition's reserved page pointers */
  44 struct xpc_rsvd_page *xpc_rsvd_page;
  45 static u64 *xpc_part_nasids;
  46 static u64 *xpc_mach_nasids;
  47 struct xpc_vars *xpc_vars;
  48 struct xpc_vars_part *xpc_vars_part;
  49
  50 static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
  51 static int xp_nasid_mask_words; /* actual size in words of nasid mask */
  52
  53 struct xpc_partition *xpc_partitions;
  54
  55 /*
  56  * Generic buffer used to store a local copy of portions of a remote
  57  * partition's reserved page (either its header and part_nasids mask,
  58  * or its vars).
  59  */
  60 char *xpc_remote_copy_buffer;
  61 void *xpc_remote_copy_buffer_base;
  62
  63 /*
  64  * Guarantee that the kmalloc'd memory is cacheline aligned.
  65  */
  66 void *
  67 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
  68 {
  69         /* see if kmalloc will give us cachline aligned memory by default */
  70         *base = kmalloc(size, flags);
  71         if (*base == NULL)
  72                 return NULL;
  73
  74         if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
  75                 return *base;
  76
  77         kfree(*base);
  78
  79         /* nope, we'll have to do it ourselves */
  80         *base = kmalloc(size + L1_CACHE_BYTES, flags);
  81         if (*base == NULL)
  82                 return NULL;
  83
  84         return (void *)L1_CACHE_ALIGN((u64)*base);
  85 }
  86
  87 /*
  88  * Given a nasid, get the physical address of the  partition's reserved page
  89  * for that nasid. This function returns 0 on any error.
  90  */
  91 static u64
  92 xpc_get_rsvd_page_pa(int nasid)
  93 {
  94         enum xp_retval ret;
  95         s64 status;
  96         u64 cookie = 0;
  97         u64 rp_pa = nasid;      /* seed with nasid */
  98         u64 len = 0;
  99         u64 buf = buf;
 100         u64 buf_len = 0;
 101         void *buf_base = NULL;
 102
 103         while (1) {
 104
 105                 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
 106                                                        &len);
 107
 108                 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
 109                         "0x%016lx, address=0x%016lx, len=0x%016lx\n",
 110                         status, cookie, rp_pa, len);
 111
 112                 if (status != SALRET_MORE_PASSES)
 113                         break;
 114
 115                 /* >>> L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
 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                 ret = xp_remote_memcpy((void *)buf, (void *)rp_pa, buf_len);
 131                 if (ret != xpSuccess) {
 132                         dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
 133                         status = SALRET_ERROR;
 134                         break;
 135                 }
 136         }
 137
 138         kfree(buf_base);
 139
 140         if (status != SALRET_OK)
 141                 rp_pa = 0;
 142
 143         dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
 144         return rp_pa;
 145 }
 146
 147 /*
 148  * Fill the partition reserved page with the information needed by
 149  * other partitions to discover we are alive and establish initial
 150  * communications.
 151  */
 152 struct xpc_rsvd_page *
 153 xpc_rsvd_page_init(void)
 154 {
 155         struct xpc_rsvd_page *rp;
 156         AMO_t *amos_page;
 157         u64 rp_pa, nasid_array = 0;
 158         int i, ret;
 159
 160         /* get the local reserved page's address */
 161
 162         preempt_disable();
 163         rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
 164         preempt_enable();
 165         if (rp_pa == 0) {
 166                 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
 167                 return NULL;
 168         }
 169         rp = (struct xpc_rsvd_page *)__va(rp_pa);
 170
 171         if (rp->partid != sn_partition_id) {
 172                 dev_err(xpc_part, "the reserved page's partid of %d should be "
 173                         "%d\n", rp->partid, sn_partition_id);
 174                 return NULL;
 175         }
 176
 177         rp->version = XPC_RP_VERSION;
 178
 179         /* establish the actual sizes of the nasid masks */
 180         if (rp->SAL_version == 1) {
 181                 /* SAL_version 1 didn't set the nasids_size field */
 182                 rp->nasids_size = 128;
 183         }
 184         xp_nasid_mask_bytes = rp->nasids_size;
 185         xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
 186
 187         /* setup the pointers to the various items in the reserved page */
 188         xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
 189         xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
 190         xpc_vars = XPC_RP_VARS(rp);
 191         xpc_vars_part = XPC_RP_VARS_PART(rp);
 192
 193         /*
 194          * Before clearing xpc_vars, see if a page of AMOs had been previously
 195          * allocated. If not we'll need to allocate one and set permissions
 196          * so that cross-partition AMOs are allowed.
 197          *
 198          * The allocated AMO page needs MCA reporting to remain disabled after
 199          * XPC has unloaded.  To make this work, we keep a copy of the pointer
 200          * to this page (i.e., amos_page) in the struct xpc_vars structure,
 201          * which is pointed to by the reserved page, and re-use that saved copy
 202          * on subsequent loads of XPC. This AMO page is never freed, and its
 203          * memory protections are never restricted.
 204          */
 205         amos_page = xpc_vars->amos_page;
 206         if (amos_page == NULL) {
 207                 amos_page = (AMO_t *)TO_AMO(uncached_alloc_page(0, 1));
 208                 if (amos_page == NULL) {
 209                         dev_err(xpc_part, "can't allocate page of AMOs\n");
 210                         return NULL;
 211                 }
 212
 213                 /*
 214                  * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
 215                  * when xpc_allow_IPI_ops() is called via xpc_hb_init().
 216                  */
 217                 if (!enable_shub_wars_1_1()) {
 218                         ret = sn_change_memprotect(ia64_tpa((u64)amos_page),
 219                                                    PAGE_SIZE,
 220                                                    SN_MEMPROT_ACCESS_CLASS_1,
 221                                                    &nasid_array);
 222                         if (ret != 0) {
 223                                 dev_err(xpc_part, "can't change memory "
 224                                         "protections\n");
 225                                 uncached_free_page(__IA64_UNCACHED_OFFSET |
 226                                                    TO_PHYS((u64)amos_page), 1);
 227                                 return NULL;
 228                         }
 229                 }
 230         } else if (!IS_AMO_ADDRESS((u64)amos_page)) {
 231                 /*
 232                  * EFI's XPBOOT can also set amos_page in the reserved page,
 233                  * but it happens to leave it as an uncached physical address
 234                  * and we need it to be an uncached virtual, so we'll have to
 235                  * convert it.
 236                  */
 237                 if (!IS_AMO_PHYS_ADDRESS((u64)amos_page)) {
 238                         dev_err(xpc_part, "previously used amos_page address "
 239                                 "is bad = 0x%p\n", (void *)amos_page);
 240                         return NULL;
 241                 }
 242                 amos_page = (AMO_t *)TO_AMO((u64)amos_page);
 243         }
 244
 245         /* clear xpc_vars */
 246         memset(xpc_vars, 0, sizeof(struct xpc_vars));
 247
 248         xpc_vars->version = XPC_V_VERSION;
 249         xpc_vars->act_nasid = cpuid_to_nasid(0);
 250         xpc_vars->act_phys_cpuid = cpu_physical_id(0);
 251         xpc_vars->vars_part_pa = __pa(xpc_vars_part);
 252         xpc_vars->amos_page_pa = ia64_tpa((u64)amos_page);
 253         xpc_vars->amos_page = amos_page;        /* save for next load of XPC */
 254
 255         /* clear xpc_vars_part */
 256         memset((u64 *)xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
 257                xp_max_npartitions);
 258
 259         /* initialize the activate IRQ related AMO variables */
 260         for (i = 0; i < xp_nasid_mask_words; i++)
 261                 (void)xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
 262
 263         /* initialize the engaged remote partitions related AMO variables */
 264         (void)xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
 265         (void)xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
 266
 267         /* timestamp of when reserved page was setup by XPC */
 268         rp->stamp = CURRENT_TIME;
 269
 270         /*
 271          * This signifies to the remote partition that our reserved
 272          * page is initialized.
 273          */
 274         rp->vars_pa = __pa(xpc_vars);
 275
 276         return rp;
 277 }
 278
 279 /*
 280  * Change protections to allow IPI operations (and AMO operations on
 281  * Shub 1.1 systems).
 282  */
 283 void
 284 xpc_allow_IPI_ops(void)
 285 {
 286         int node;
 287         int nasid;
 288
 289         /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
 290
 291         if (is_shub2()) {
 292                 xpc_sh2_IPI_access0 =
 293                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
 294                 xpc_sh2_IPI_access1 =
 295                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
 296                 xpc_sh2_IPI_access2 =
 297                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
 298                 xpc_sh2_IPI_access3 =
 299                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
 300
 301                 for_each_online_node(node) {
 302                         nasid = cnodeid_to_nasid(node);
 303                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 304                               -1UL);
 305                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 306                               -1UL);
 307                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 308                               -1UL);
 309                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 310                               -1UL);
 311                 }
 312
 313         } else {
 314                 xpc_sh1_IPI_access =
 315                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
 316
 317                 for_each_online_node(node) {
 318                         nasid = cnodeid_to_nasid(node);
 319                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 320                               -1UL);
 321
 322                         /*
 323                          * Since the BIST collides with memory operations on
 324                          * SHUB 1.1 sn_change_memprotect() cannot be used.
 325                          */
 326                         if (enable_shub_wars_1_1()) {
 327                                 /* open up everything */
 328                                 xpc_prot_vec[node] = (u64)HUB_L((u64 *)
 329                                                                 GLOBAL_MMR_ADDR
 330                                                                 (nasid,
 331                                                   SH1_MD_DQLP_MMR_DIR_PRIVEC0));
 332                                 HUB_S((u64 *)
 333                                       GLOBAL_MMR_ADDR(nasid,
 334                                                    SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 335                                       -1UL);
 336                                 HUB_S((u64 *)
 337                                       GLOBAL_MMR_ADDR(nasid,
 338                                                    SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 339                                       -1UL);
 340                         }
 341                 }
 342         }
 343 }
 344
 345 /*
 346  * Restrict protections to disallow IPI operations (and AMO operations on
 347  * Shub 1.1 systems).
 348  */
 349 void
 350 xpc_restrict_IPI_ops(void)
 351 {
 352         int node;
 353         int nasid;
 354
 355         /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
 356
 357         if (is_shub2()) {
 358
 359                 for_each_online_node(node) {
 360                         nasid = cnodeid_to_nasid(node);
 361                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 362                               xpc_sh2_IPI_access0);
 363                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 364                               xpc_sh2_IPI_access1);
 365                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 366                               xpc_sh2_IPI_access2);
 367                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 368                               xpc_sh2_IPI_access3);
 369                 }
 370
 371         } else {
 372
 373                 for_each_online_node(node) {
 374                         nasid = cnodeid_to_nasid(node);
 375                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 376                               xpc_sh1_IPI_access);
 377
 378                         if (enable_shub_wars_1_1()) {
 379                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 380                                                    SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 381                                       xpc_prot_vec[node]);
 382                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 383                                                    SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 384                                       xpc_prot_vec[node]);
 385                         }
 386                 }
 387         }
 388 }
 389
 390 /*
 391  * At periodic intervals, scan through all active partitions and ensure
 392  * their heartbeat is still active.  If not, the partition is deactivated.
 393  */
 394 void
 395 xpc_check_remote_hb(void)
 396 {
 397         struct xpc_vars *remote_vars;
 398         struct xpc_partition *part;
 399         short partid;
 400         enum xp_retval ret;
 401
 402         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
 403
 404         for (partid = 0; partid < xp_max_npartitions; partid++) {
 405
 406                 if (xpc_exiting)
 407                         break;
 408
 409                 if (partid == sn_partition_id)
 410                         continue;
 411
 412                 part = &xpc_partitions[partid];
 413
 414                 if (part->act_state == XPC_P_INACTIVE ||
 415                     part->act_state == XPC_P_DEACTIVATING) {
 416                         continue;
 417                 }
 418
 419                 /* pull the remote_hb cache line */
 420                 ret = xp_remote_memcpy(remote_vars,
 421                                        (void *)part->remote_vars_pa,
 422                                        XPC_RP_VARS_SIZE);
 423                 if (ret != xpSuccess) {
 424                         XPC_DEACTIVATE_PARTITION(part, ret);
 425                         continue;
 426                 }
 427
 428                 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
 429                         " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
 430                         partid, remote_vars->heartbeat, part->last_heartbeat,
 431                         remote_vars->heartbeat_offline,
 432                         remote_vars->heartbeating_to_mask);
 433
 434                 if (((remote_vars->heartbeat == part->last_heartbeat) &&
 435                      (remote_vars->heartbeat_offline == 0)) ||
 436                     !xpc_hb_allowed(sn_partition_id, remote_vars)) {
 437
 438                         XPC_DEACTIVATE_PARTITION(part, xpNoHeartbeat);
 439                         continue;
 440                 }
 441
 442                 part->last_heartbeat = remote_vars->heartbeat;
 443         }
 444 }
 445
 446 /*
 447  * Get a copy of a portion of the remote partition's rsvd page.
 448  *
 449  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
 450  * is large enough to contain a copy of their reserved page header and
 451  * part_nasids mask.
 452  */
 453 static enum xp_retval
 454 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
 455                   struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
 456 {
 457         int i;
 458         enum xp_retval ret;
 459
 460         /* get the reserved page's physical address */
 461
 462         *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
 463         if (*remote_rp_pa == 0)
 464                 return xpNoRsvdPageAddr;
 465
 466         /* pull over the reserved page header and part_nasids mask */
 467         ret = xp_remote_memcpy(remote_rp, (void *)*remote_rp_pa,
 468                                XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes);
 469         if (ret != xpSuccess)
 470                 return ret;
 471
 472         if (discovered_nasids != NULL) {
 473                 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
 474
 475                 for (i = 0; i < xp_nasid_mask_words; i++)
 476                         discovered_nasids[i] |= remote_part_nasids[i];
 477         }
 478
 479         /* check that the partid is for another partition */
 480
 481         if (remote_rp->partid < 0 || remote_rp->partid >= xp_max_npartitions)
 482                 return xpInvalidPartid;
 483
 484         if (remote_rp->partid == sn_partition_id)
 485                 return xpLocalPartid;
 486
 487         if (XPC_VERSION_MAJOR(remote_rp->version) !=
 488             XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
 489                 return xpBadVersion;
 490         }
 491
 492         return xpSuccess;
 493 }
 494
 495 /*
 496  * Get a copy of the remote partition's XPC variables from the reserved page.
 497  *
 498  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
 499  * assumed to be of size XPC_RP_VARS_SIZE.
 500  */
 501 static enum xp_retval
 502 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
 503 {
 504         enum xp_retval ret;
 505
 506         if (remote_vars_pa == 0)
 507                 return xpVarsNotSet;
 508
 509         /* pull over the cross partition variables */
 510         ret = xp_remote_memcpy(remote_vars, (void *)remote_vars_pa,
 511                                XPC_RP_VARS_SIZE);
 512         if (ret != xpSuccess)
 513                 return ret;
 514
 515         if (XPC_VERSION_MAJOR(remote_vars->version) !=
 516             XPC_VERSION_MAJOR(XPC_V_VERSION)) {
 517                 return xpBadVersion;
 518         }
 519
 520         return xpSuccess;
 521 }
 522
 523 /*
 524  * Update the remote partition's info.
 525  */
 526 static void
 527 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
 528                           struct timespec *remote_rp_stamp, u64 remote_rp_pa,
 529                           u64 remote_vars_pa, struct xpc_vars *remote_vars)
 530 {
 531         part->remote_rp_version = remote_rp_version;
 532         dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
 533                 part->remote_rp_version);
 534
 535         part->remote_rp_stamp = *remote_rp_stamp;
 536         dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
 537                 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
 538
 539         part->remote_rp_pa = remote_rp_pa;
 540         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
 541
 542         part->remote_vars_pa = remote_vars_pa;
 543         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
 544                 part->remote_vars_pa);
 545
 546         part->last_heartbeat = remote_vars->heartbeat;
 547         dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
 548                 part->last_heartbeat);
 549
 550         part->remote_vars_part_pa = remote_vars->vars_part_pa;
 551         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
 552                 part->remote_vars_part_pa);
 553
 554         part->remote_act_nasid = remote_vars->act_nasid;
 555         dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
 556                 part->remote_act_nasid);
 557
 558         part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
 559         dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
 560                 part->remote_act_phys_cpuid);
 561
 562         part->remote_amos_page_pa = remote_vars->amos_page_pa;
 563         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
 564                 part->remote_amos_page_pa);
 565
 566         part->remote_vars_version = remote_vars->version;
 567         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
 568                 part->remote_vars_version);
 569 }
 570
 571 /*
 572  * Prior code has determined the nasid which generated an IPI.  Inspect
 573  * that nasid to determine if its partition needs to be activated or
 574  * deactivated.
 575  *
 576  * A partition is consider "awaiting activation" if our partition
 577  * flags indicate it is not active and it has a heartbeat.  A
 578  * partition is considered "awaiting deactivation" if our partition
 579  * flags indicate it is active but it has no heartbeat or it is not
 580  * sending its heartbeat to us.
 581  *
 582  * To determine the heartbeat, the remote nasid must have a properly
 583  * initialized reserved page.
 584  */
 585 static void
 586 xpc_identify_act_IRQ_req(int nasid)
 587 {
 588         struct xpc_rsvd_page *remote_rp;
 589         struct xpc_vars *remote_vars;
 590         u64 remote_rp_pa;
 591         u64 remote_vars_pa;
 592         int remote_rp_version;
 593         int reactivate = 0;
 594         int stamp_diff;
 595         struct timespec remote_rp_stamp = { 0, 0 };
 596         short partid;
 597         struct xpc_partition *part;
 598         enum xp_retval ret;
 599
 600         /* pull over the reserved page structure */
 601
 602         remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer;
 603
 604         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
 605         if (ret != xpSuccess) {
 606                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
 607                          "which sent interrupt, reason=%d\n", nasid, ret);
 608                 return;
 609         }
 610
 611         remote_vars_pa = remote_rp->vars_pa;
 612         remote_rp_version = remote_rp->version;
 613         if (XPC_SUPPORTS_RP_STAMP(remote_rp_version))
 614                 remote_rp_stamp = remote_rp->stamp;
 615
 616         partid = remote_rp->partid;
 617         part = &xpc_partitions[partid];
 618
 619         /* pull over the cross partition variables */
 620
 621         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
 622
 623         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
 624         if (ret != xpSuccess) {
 625
 626                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
 627                          "which sent interrupt, reason=%d\n", nasid, ret);
 628
 629                 XPC_DEACTIVATE_PARTITION(part, ret);
 630                 return;
 631         }
 632
 633         part->act_IRQ_rcvd++;
 634
 635         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
 636                 "%ld:0x%lx\n", (int)nasid, (int)partid, part->act_IRQ_rcvd,
 637                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
 638
 639         if (xpc_partition_disengaged(part) &&
 640             part->act_state == XPC_P_INACTIVE) {
 641
 642                 xpc_update_partition_info(part, remote_rp_version,
 643                                           &remote_rp_stamp, remote_rp_pa,
 644                                           remote_vars_pa, remote_vars);
 645
 646                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 647                         if (xpc_partition_disengage_requested(1UL << partid)) {
 648                                 /*
 649                                  * Other side is waiting on us to disengage,
 650                                  * even though we already have.
 651                                  */
 652                                 return;
 653                         }
 654                 } else {
 655                         /* other side doesn't support disengage requests */
 656                         xpc_clear_partition_disengage_request(1UL << partid);
 657                 }
 658
 659                 xpc_activate_partition(part);
 660                 return;
 661         }
 662
 663         DBUG_ON(part->remote_rp_version == 0);
 664         DBUG_ON(part->remote_vars_version == 0);
 665
 666         if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
 667                 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
 668                                                        remote_vars_version));
 669
 670                 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 671                         DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
 672                                                                version));
 673                         /* see if the other side rebooted */
 674                         if (part->remote_amos_page_pa ==
 675                             remote_vars->amos_page_pa &&
 676                             xpc_hb_allowed(sn_partition_id, remote_vars)) {
 677                                 /* doesn't look that way, so ignore the IPI */
 678                                 return;
 679                         }
 680                 }
 681
 682                 /*
 683                  * Other side rebooted and previous XPC didn't support the
 684                  * disengage request, so we don't need to do anything special.
 685                  */
 686
 687                 xpc_update_partition_info(part, remote_rp_version,
 688                                           &remote_rp_stamp, remote_rp_pa,
 689                                           remote_vars_pa, remote_vars);
 690                 part->reactivate_nasid = nasid;
 691                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 692                 return;
 693         }
 694
 695         DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
 696
 697         if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
 698                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 699
 700                 /*
 701                  * Other side rebooted and previous XPC did support the
 702                  * disengage request, but the new one doesn't.
 703                  */
 704
 705                 xpc_clear_partition_engaged(1UL << partid);
 706                 xpc_clear_partition_disengage_request(1UL << partid);
 707
 708                 xpc_update_partition_info(part, remote_rp_version,
 709                                           &remote_rp_stamp, remote_rp_pa,
 710                                           remote_vars_pa, remote_vars);
 711                 reactivate = 1;
 712
 713         } else {
 714                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
 715
 716                 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
 717                                                 &remote_rp_stamp);
 718                 if (stamp_diff != 0) {
 719                         DBUG_ON(stamp_diff >= 0);
 720
 721                         /*
 722                          * Other side rebooted and the previous XPC did support
 723                          * the disengage request, as does the new one.
 724                          */
 725
 726                         DBUG_ON(xpc_partition_engaged(1UL << partid));
 727                         DBUG_ON(xpc_partition_disengage_requested(1UL <<
 728                                                                   partid));
 729
 730                         xpc_update_partition_info(part, remote_rp_version,
 731                                                   &remote_rp_stamp,
 732                                                   remote_rp_pa, remote_vars_pa,
 733                                                   remote_vars);
 734                         reactivate = 1;
 735                 }
 736         }
 737
 738         if (part->disengage_request_timeout > 0 &&
 739             !xpc_partition_disengaged(part)) {
 740                 /* still waiting on other side to disengage from us */
 741                 return;
 742         }
 743
 744         if (reactivate) {
 745                 part->reactivate_nasid = nasid;
 746                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 747
 748         } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
 749                    xpc_partition_disengage_requested(1UL << partid)) {
 750                 XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
 751         }
 752 }
 753
 754 /*
 755  * Loop through the activation AMO variables and process any bits
 756  * which are set.  Each bit indicates a nasid sending a partition
 757  * activation or deactivation request.
 758  *
 759  * Return #of IRQs detected.
 760  */
 761 int
 762 xpc_identify_act_IRQ_sender(void)
 763 {
 764         int word, bit;
 765         u64 nasid_mask;
 766         u64 nasid;              /* remote nasid */
 767         int n_IRQs_detected = 0;
 768         AMO_t *act_amos;
 769
 770         act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
 771
 772         /* scan through act AMO variable looking for non-zero entries */
 773         for (word = 0; word < xp_nasid_mask_words; word++) {
 774
 775                 if (xpc_exiting)
 776                         break;
 777
 778                 nasid_mask = xpc_IPI_receive(&act_amos[word]);
 779                 if (nasid_mask == 0) {
 780                         /* no IRQs from nasids in this variable */
 781                         continue;
 782                 }
 783
 784                 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
 785                         nasid_mask);
 786
 787                 /*
 788                  * If this nasid has been added to the machine since
 789                  * our partition was reset, this will retain the
 790                  * remote nasid in our reserved pages machine mask.
 791                  * This is used in the event of module reload.
 792                  */
 793                 xpc_mach_nasids[word] |= nasid_mask;
 794
 795                 /* locate the nasid(s) which sent interrupts */
 796
 797                 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
 798                         if (nasid_mask & (1UL << bit)) {
 799                                 n_IRQs_detected++;
 800                                 nasid = XPC_NASID_FROM_W_B(word, bit);
 801                                 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
 802                                         nasid);
 803                                 xpc_identify_act_IRQ_req(nasid);
 804                         }
 805                 }
 806         }
 807         return n_IRQs_detected;
 808 }
 809
 810 /*
 811  * See if the other side has responded to a partition disengage request
 812  * from us.
 813  */
 814 int
 815 xpc_partition_disengaged(struct xpc_partition *part)
 816 {
 817         short partid = XPC_PARTID(part);
 818         int disengaged;
 819
 820         disengaged = (xpc_partition_engaged(1UL << partid) == 0);
 821         if (part->disengage_request_timeout) {
 822                 if (!disengaged) {
 823                         if (time_before(jiffies,
 824                             part->disengage_request_timeout)) {
 825                                 /* timelimit hasn't been reached yet */
 826                                 return 0;
 827                         }
 828
 829                         /*
 830                          * Other side hasn't responded to our disengage
 831                          * request in a timely fashion, so assume it's dead.
 832                          */
 833
 834                         dev_info(xpc_part, "disengage from remote partition %d "
 835                                  "timed out\n", partid);
 836                         xpc_disengage_request_timedout = 1;
 837                         xpc_clear_partition_engaged(1UL << partid);
 838                         disengaged = 1;
 839                 }
 840                 part->disengage_request_timeout = 0;
 841
 842                 /* cancel the timer function, provided it's not us */
 843                 if (!in_interrupt()) {
 844                         del_singleshot_timer_sync(&part->
 845                                                   disengage_request_timer);
 846                 }
 847
 848                 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
 849                         part->act_state != XPC_P_INACTIVE);
 850                 if (part->act_state != XPC_P_INACTIVE)
 851                         xpc_wakeup_channel_mgr(part);
 852
 853                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version))
 854                         xpc_cancel_partition_disengage_request(part);
 855         }
 856         return disengaged;
 857 }
 858
 859 /*
 860  * Mark specified partition as active.
 861  */
 862 enum xp_retval
 863 xpc_mark_partition_active(struct xpc_partition *part)
 864 {
 865         unsigned long irq_flags;
 866         enum xp_retval ret;
 867
 868         dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
 869
 870         spin_lock_irqsave(&part->act_lock, irq_flags);
 871         if (part->act_state == XPC_P_ACTIVATING) {
 872                 part->act_state = XPC_P_ACTIVE;
 873                 ret = xpSuccess;
 874         } else {
 875                 DBUG_ON(part->reason == xpSuccess);
 876                 ret = part->reason;
 877         }
 878         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 879
 880         return ret;
 881 }
 882
 883 /*
 884  * Notify XPC that the partition is down.
 885  */
 886 void
 887 xpc_deactivate_partition(const int line, struct xpc_partition *part,
 888                          enum xp_retval reason)
 889 {
 890         unsigned long irq_flags;
 891
 892         spin_lock_irqsave(&part->act_lock, irq_flags);
 893
 894         if (part->act_state == XPC_P_INACTIVE) {
 895                 XPC_SET_REASON(part, reason, line);
 896                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 897                 if (reason == xpReactivating) {
 898                         /* we interrupt ourselves to reactivate partition */
 899                         xpc_IPI_send_reactivate(part);
 900                 }
 901                 return;
 902         }
 903         if (part->act_state == XPC_P_DEACTIVATING) {
 904                 if ((part->reason == xpUnloading && reason != xpUnloading) ||
 905                     reason == xpReactivating) {
 906                         XPC_SET_REASON(part, reason, line);
 907                 }
 908                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 909                 return;
 910         }
 911
 912         part->act_state = XPC_P_DEACTIVATING;
 913         XPC_SET_REASON(part, reason, line);
 914
 915         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 916
 917         if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
 918                 xpc_request_partition_disengage(part);
 919                 xpc_IPI_send_disengage(part);
 920
 921                 /* set a timelimit on the disengage request */
 922                 part->disengage_request_timeout = jiffies +
 923                     (xpc_disengage_request_timelimit * HZ);
 924                 part->disengage_request_timer.expires =
 925                     part->disengage_request_timeout;
 926                 add_timer(&part->disengage_request_timer);
 927         }
 928
 929         dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
 930                 XPC_PARTID(part), reason);
 931
 932         xpc_partition_going_down(part, reason);
 933 }
 934
 935 /*
 936  * Mark specified partition as inactive.
 937  */
 938 void
 939 xpc_mark_partition_inactive(struct xpc_partition *part)
 940 {
 941         unsigned long irq_flags;
 942
 943         dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
 944                 XPC_PARTID(part));
 945
 946         spin_lock_irqsave(&part->act_lock, irq_flags);
 947         part->act_state = XPC_P_INACTIVE;
 948         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 949         part->remote_rp_pa = 0;
 950 }
 951
 952 /*
 953  * SAL has provided a partition and machine mask.  The partition mask
 954  * contains a bit for each even nasid in our partition.  The machine
 955  * mask contains a bit for each even nasid in the entire machine.
 956  *
 957  * Using those two bit arrays, we can determine which nasids are
 958  * known in the machine.  Each should also have a reserved page
 959  * initialized if they are available for partitioning.
 960  */
 961 void
 962 xpc_discovery(void)
 963 {
 964         void *remote_rp_base;
 965         struct xpc_rsvd_page *remote_rp;
 966         struct xpc_vars *remote_vars;
 967         u64 remote_rp_pa;
 968         u64 remote_vars_pa;
 969         int region;
 970         int region_size;
 971         int max_regions;
 972         int nasid;
 973         struct xpc_rsvd_page *rp;
 974         short partid;
 975         struct xpc_partition *part;
 976         u64 *discovered_nasids;
 977         enum xp_retval ret;
 978
 979         remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
 980                                                   xp_nasid_mask_bytes,
 981                                                   GFP_KERNEL, &remote_rp_base);
 982         if (remote_rp == NULL)
 983                 return;
 984
 985         remote_vars = (struct xpc_vars *)remote_rp;
 986
 987         discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
 988                                     GFP_KERNEL);
 989         if (discovered_nasids == NULL) {
 990                 kfree(remote_rp_base);
 991                 return;
 992         }
 993
 994         rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
 995
 996         /*
 997          * The term 'region' in this context refers to the minimum number of
 998          * nodes that can comprise an access protection grouping. The access
 999          * protection is in regards to memory, IOI and IPI.
1000          */
1001         max_regions = 64;
1002         region_size = sn_region_size;
1003
1004         switch (region_size) {
1005         case 128:
1006                 max_regions *= 2;
1007         case 64:
1008                 max_regions *= 2;
1009         case 32:
1010                 max_regions *= 2;
1011                 region_size = 16;
1012                 DBUG_ON(!is_shub2());
1013         }
1014
1015         for (region = 0; region < max_regions; region++) {
1016
1017                 if (xpc_exiting)
1018                         break;
1019
1020                 dev_dbg(xpc_part, "searching region %d\n", region);
1021
1022                 for (nasid = (region * region_size * 2);
1023                      nasid < ((region + 1) * region_size * 2); nasid += 2) {
1024
1025                         if (xpc_exiting)
1026                                 break;
1027
1028                         dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1029
1030                         if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
1031                                 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1032                                         "part of the local partition; skipping "
1033                                         "region\n", nasid);
1034                                 break;
1035                         }
1036
1037                         if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
1038                                 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1039                                         "not on Numa-Link network at reset\n",
1040                                         nasid);
1041                                 continue;
1042                         }
1043
1044                         if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1045                                 dev_dbg(xpc_part, "Nasid %d is part of a "
1046                                         "partition which was previously "
1047                                         "discovered\n", nasid);
1048                                 continue;
1049                         }
1050
1051                         /* pull over the reserved page structure */
1052
1053                         ret = xpc_get_remote_rp(nasid, discovered_nasids,
1054                                                 remote_rp, &remote_rp_pa);
1055                         if (ret != xpSuccess) {
1056                                 dev_dbg(xpc_part, "unable to get reserved page "
1057                                         "from nasid %d, reason=%d\n", nasid,
1058                                         ret);
1059
1060                                 if (ret == xpLocalPartid)
1061                                         break;
1062
1063                                 continue;
1064                         }
1065
1066                         remote_vars_pa = remote_rp->vars_pa;
1067
1068                         partid = remote_rp->partid;
1069                         part = &xpc_partitions[partid];
1070
1071                         /* pull over the cross partition variables */
1072
1073                         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1074                         if (ret != xpSuccess) {
1075                                 dev_dbg(xpc_part, "unable to get XPC variables "
1076                                         "from nasid %d, reason=%d\n", nasid,
1077                                         ret);
1078
1079                                 XPC_DEACTIVATE_PARTITION(part, ret);
1080                                 continue;
1081                         }
1082
1083                         if (part->act_state != XPC_P_INACTIVE) {
1084                                 dev_dbg(xpc_part, "partition %d on nasid %d is "
1085                                         "already activating\n", partid, nasid);
1086                                 break;
1087                         }
1088
1089                         /*
1090                          * Register the remote partition's AMOs with SAL so it
1091                          * can handle and cleanup errors within that address
1092                          * range should the remote partition go down. We don't
1093                          * unregister this range because it is difficult to
1094                          * tell when outstanding writes to the remote partition
1095                          * are finished and thus when it is thus safe to
1096                          * unregister. This should not result in wasted space
1097                          * in the SAL xp_addr_region table because we should
1098                          * get the same page for remote_act_amos_pa after
1099                          * module reloads and system reboots.
1100                          */
1101                         if (sn_register_xp_addr_region
1102                             (remote_vars->amos_page_pa, PAGE_SIZE, 1) < 0) {
1103                                 dev_dbg(xpc_part,
1104                                         "partition %d failed to "
1105                                         "register xp_addr region 0x%016lx\n",
1106                                         partid, remote_vars->amos_page_pa);
1107
1108                                 XPC_SET_REASON(part, xpPhysAddrRegFailed,
1109                                                __LINE__);
1110                                 break;
1111                         }
1112
1113                         /*
1114                          * The remote nasid is valid and available.
1115                          * Send an interrupt to that nasid to notify
1116                          * it that we are ready to begin activation.
1117                          */
1118                         dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1119                                 "nasid %d, phys_cpuid 0x%x\n",
1120                                 remote_vars->amos_page_pa,
1121                                 remote_vars->act_nasid,
1122                                 remote_vars->act_phys_cpuid);
1123
1124                         if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1125                                                            version)) {
1126                                 part->remote_amos_page_pa =
1127                                     remote_vars->amos_page_pa;
1128                                 xpc_mark_partition_disengaged(part);
1129                                 xpc_cancel_partition_disengage_request(part);
1130                         }
1131                         xpc_IPI_send_activate(remote_vars);
1132                 }
1133         }
1134
1135         kfree(discovered_nasids);
1136         kfree(remote_rp_base);
1137 }
1138
1139 /*
1140  * Given a partid, get the nasids owned by that partition from the
1141  * remote partition's reserved page.
1142  */
1143 enum xp_retval
1144 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
1145 {
1146         struct xpc_partition *part;
1147         u64 part_nasid_pa;
1148
1149         part = &xpc_partitions[partid];
1150         if (part->remote_rp_pa == 0)
1151                 return xpPartitionDown;
1152
1153         memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1154
1155         part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
1156
1157         return xp_remote_memcpy(nasid_mask, (void *)part_nasid_pa,
1158                                 xp_nasid_mask_bytes);
1159 }