drivers/misc/sgi-xp/xpc_uv.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) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
   7  */
   8
   9 /*
  10  * Cross Partition Communication (XPC) uv-based functions.
  11  *
  12  *     Architecture specific implementation of common functions.
  13  *
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/mm.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/delay.h>
  20 #include <linux/device.h>
  21 #include <linux/cpu.h>
  22 #include <linux/module.h>
  23 #include <linux/err.h>
  24 #include <linux/slab.h>
  25 #include <linux/numa.h>
  26 #include <asm/uv/uv_hub.h>
  27 #include <asm/uv/bios.h>
  28 #include <asm/uv/uv_irq.h>
  29 #include "../sgi-gru/gru.h"
  30 #include "../sgi-gru/grukservices.h"
  31 #include "xpc.h"
  32
  33 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
  34
  35 #define XPC_ACTIVATE_MSG_SIZE_UV        (1 * GRU_CACHE_LINE_BYTES)
  36 #define XPC_ACTIVATE_MQ_SIZE_UV         (4 * XP_MAX_NPARTITIONS_UV * \
  37                                          XPC_ACTIVATE_MSG_SIZE_UV)
  38 #define XPC_ACTIVATE_IRQ_NAME           "xpc_activate"
  39
  40 #define XPC_NOTIFY_MSG_SIZE_UV          (2 * GRU_CACHE_LINE_BYTES)
  41 #define XPC_NOTIFY_MQ_SIZE_UV           (4 * XP_MAX_NPARTITIONS_UV * \
  42                                          XPC_NOTIFY_MSG_SIZE_UV)
  43 #define XPC_NOTIFY_IRQ_NAME             "xpc_notify"
  44
  45 static int xpc_mq_node = NUMA_NO_NODE;
  46
  47 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
  48 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
  49
  50 static int
  51 xpc_setup_partitions_uv(void)
  52 {
  53         short partid;
  54         struct xpc_partition_uv *part_uv;
  55
  56         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
  57                 part_uv = &xpc_partitions[partid].sn.uv;
  58
  59                 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
  60                 spin_lock_init(&part_uv->flags_lock);
  61                 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
  62         }
  63         return 0;
  64 }
  65
  66 static void
  67 xpc_teardown_partitions_uv(void)
  68 {
  69         short partid;
  70         struct xpc_partition_uv *part_uv;
  71         unsigned long irq_flags;
  72
  73         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
  74                 part_uv = &xpc_partitions[partid].sn.uv;
  75
  76                 if (part_uv->cached_activate_gru_mq_desc != NULL) {
  77                         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
  78                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
  79                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
  80                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
  81                         kfree(part_uv->cached_activate_gru_mq_desc);
  82                         part_uv->cached_activate_gru_mq_desc = NULL;
  83                         mutex_unlock(&part_uv->
  84                                      cached_activate_gru_mq_desc_mutex);
  85                 }
  86         }
  87 }
  88
  89 static int
  90 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
  91 {
  92         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
  93
  94         mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
  95                         UV_AFFINITY_CPU);
  96         if (mq->irq < 0)
  97                 return mq->irq;
  98
  99         mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
 100
 101         return 0;
 102 }
 103
 104 static void
 105 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
 106 {
 107         uv_teardown_irq(mq->irq);
 108 }
 109
 110 static int
 111 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
 112 {
 113         int ret;
 114
 115         ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
 116                                          mq->order, &mq->mmr_offset);
 117         if (ret < 0) {
 118                 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
 119                         "ret=%d\n", ret);
 120                 return ret;
 121         }
 122
 123         mq->watchlist_num = ret;
 124         return 0;
 125 }
 126
 127 static void
 128 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
 129 {
 130         int ret;
 131         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
 132
 133         ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
 134         BUG_ON(ret != BIOS_STATUS_SUCCESS);
 135 }
 136
 137 static struct xpc_gru_mq_uv *
 138 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
 139                      irq_handler_t irq_handler)
 140 {
 141         enum xp_retval xp_ret;
 142         int ret;
 143         int nid;
 144         int nasid;
 145         int pg_order;
 146         struct page *page;
 147         struct xpc_gru_mq_uv *mq;
 148         struct uv_IO_APIC_route_entry *mmr_value;
 149
 150         mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
 151         if (mq == NULL) {
 152                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
 153                         "a xpc_gru_mq_uv structure\n");
 154                 ret = -ENOMEM;
 155                 goto out_0;
 156         }
 157
 158         mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
 159                                   GFP_KERNEL);
 160         if (mq->gru_mq_desc == NULL) {
 161                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
 162                         "a gru_message_queue_desc structure\n");
 163                 ret = -ENOMEM;
 164                 goto out_1;
 165         }
 166
 167         pg_order = get_order(mq_size);
 168         mq->order = pg_order + PAGE_SHIFT;
 169         mq_size = 1UL << mq->order;
 170
 171         mq->mmr_blade = uv_cpu_to_blade_id(cpu);
 172
 173         nid = cpu_to_node(cpu);
 174         page = __alloc_pages_node(nid,
 175                                       GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
 176                                       pg_order);
 177         if (page == NULL) {
 178                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
 179                         "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
 180                 ret = -ENOMEM;
 181                 goto out_2;
 182         }
 183         mq->address = page_address(page);
 184
 185         /* enable generation of irq when GRU mq operation occurs to this mq */
 186         ret = xpc_gru_mq_watchlist_alloc_uv(mq);
 187         if (ret != 0)
 188                 goto out_3;
 189
 190         ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
 191         if (ret != 0)
 192                 goto out_4;
 193
 194         ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
 195         if (ret != 0) {
 196                 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
 197                         mq->irq, -ret);
 198                 goto out_5;
 199         }
 200
 201         nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
 202
 203         mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
 204         ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
 205                                      nasid, mmr_value->vector, mmr_value->dest);
 206         if (ret != 0) {
 207                 dev_err(xpc_part, "gru_create_message_queue() returned "
 208                         "error=%d\n", ret);
 209                 ret = -EINVAL;
 210                 goto out_6;
 211         }
 212
 213         /* allow other partitions to access this GRU mq */
 214         xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
 215         if (xp_ret != xpSuccess) {
 216                 ret = -EACCES;
 217                 goto out_6;
 218         }
 219
 220         return mq;
 221
 222         /* something went wrong */
 223 out_6:
 224         free_irq(mq->irq, NULL);
 225 out_5:
 226         xpc_release_gru_mq_irq_uv(mq);
 227 out_4:
 228         xpc_gru_mq_watchlist_free_uv(mq);
 229 out_3:
 230         free_pages((unsigned long)mq->address, pg_order);
 231 out_2:
 232         kfree(mq->gru_mq_desc);
 233 out_1:
 234         kfree(mq);
 235 out_0:
 236         return ERR_PTR(ret);
 237 }
 238
 239 static void
 240 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
 241 {
 242         unsigned int mq_size;
 243         int pg_order;
 244         int ret;
 245
 246         /* disallow other partitions to access GRU mq */
 247         mq_size = 1UL << mq->order;
 248         ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
 249         BUG_ON(ret != xpSuccess);
 250
 251         /* unregister irq handler and release mq irq/vector mapping */
 252         free_irq(mq->irq, NULL);
 253         xpc_release_gru_mq_irq_uv(mq);
 254
 255         /* disable generation of irq when GRU mq op occurs to this mq */
 256         xpc_gru_mq_watchlist_free_uv(mq);
 257
 258         pg_order = mq->order - PAGE_SHIFT;
 259         free_pages((unsigned long)mq->address, pg_order);
 260
 261         kfree(mq);
 262 }
 263
 264 static enum xp_retval
 265 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
 266                  size_t msg_size)
 267 {
 268         enum xp_retval xp_ret;
 269         int ret;
 270
 271         while (1) {
 272                 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
 273                 if (ret == MQE_OK) {
 274                         xp_ret = xpSuccess;
 275                         break;
 276                 }
 277
 278                 if (ret == MQE_QUEUE_FULL) {
 279                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
 280                                 "error=MQE_QUEUE_FULL\n");
 281                         /* !!! handle QLimit reached; delay & try again */
 282                         /* ??? Do we add a limit to the number of retries? */
 283                         (void)msleep_interruptible(10);
 284                 } else if (ret == MQE_CONGESTION) {
 285                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
 286                                 "error=MQE_CONGESTION\n");
 287                         /* !!! handle LB Overflow; simply try again */
 288                         /* ??? Do we add a limit to the number of retries? */
 289                 } else {
 290                         /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
 291                         dev_err(xpc_chan, "gru_send_message_gpa() returned "
 292                                 "error=%d\n", ret);
 293                         xp_ret = xpGruSendMqError;
 294                         break;
 295                 }
 296         }
 297         return xp_ret;
 298 }
 299
 300 static void
 301 xpc_process_activate_IRQ_rcvd_uv(void)
 302 {
 303         unsigned long irq_flags;
 304         short partid;
 305         struct xpc_partition *part;
 306         u8 act_state_req;
 307
 308         DBUG_ON(xpc_activate_IRQ_rcvd == 0);
 309
 310         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 311         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
 312                 part = &xpc_partitions[partid];
 313
 314                 if (part->sn.uv.act_state_req == 0)
 315                         continue;
 316
 317                 xpc_activate_IRQ_rcvd--;
 318                 BUG_ON(xpc_activate_IRQ_rcvd < 0);
 319
 320                 act_state_req = part->sn.uv.act_state_req;
 321                 part->sn.uv.act_state_req = 0;
 322                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 323
 324                 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
 325                         if (part->act_state == XPC_P_AS_INACTIVE)
 326                                 xpc_activate_partition(part);
 327                         else if (part->act_state == XPC_P_AS_DEACTIVATING)
 328                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 329
 330                 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
 331                         if (part->act_state == XPC_P_AS_INACTIVE)
 332                                 xpc_activate_partition(part);
 333                         else
 334                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 335
 336                 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
 337                         XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
 338
 339                 } else {
 340                         BUG();
 341                 }
 342
 343                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 344                 if (xpc_activate_IRQ_rcvd == 0)
 345                         break;
 346         }
 347         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 348
 349 }
 350
 351 static void
 352 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
 353                               struct xpc_activate_mq_msghdr_uv *msg_hdr,
 354                               int part_setup,
 355                               int *wakeup_hb_checker)
 356 {
 357         unsigned long irq_flags;
 358         struct xpc_partition_uv *part_uv = &part->sn.uv;
 359         struct xpc_openclose_args *args;
 360
 361         part_uv->remote_act_state = msg_hdr->act_state;
 362
 363         switch (msg_hdr->type) {
 364         case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
 365                 /* syncing of remote_act_state was just done above */
 366                 break;
 367
 368         case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
 369                 struct xpc_activate_mq_msg_activate_req_uv *msg;
 370
 371                 /*
 372                  * ??? Do we deal here with ts_jiffies being different
 373                  * ??? if act_state != XPC_P_AS_INACTIVE instead of
 374                  * ??? below?
 375                  */
 376                 msg = container_of(msg_hdr, struct
 377                                    xpc_activate_mq_msg_activate_req_uv, hdr);
 378
 379                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 380                 if (part_uv->act_state_req == 0)
 381                         xpc_activate_IRQ_rcvd++;
 382                 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
 383                 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
 384                 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
 385                 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
 386
 387                 if (msg->activate_gru_mq_desc_gpa !=
 388                     part_uv->activate_gru_mq_desc_gpa) {
 389                         spin_lock(&part_uv->flags_lock);
 390                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
 391                         spin_unlock(&part_uv->flags_lock);
 392                         part_uv->activate_gru_mq_desc_gpa =
 393                             msg->activate_gru_mq_desc_gpa;
 394                 }
 395                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 396
 397                 (*wakeup_hb_checker)++;
 398                 break;
 399         }
 400         case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
 401                 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
 402
 403                 msg = container_of(msg_hdr, struct
 404                                    xpc_activate_mq_msg_deactivate_req_uv, hdr);
 405
 406                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 407                 if (part_uv->act_state_req == 0)
 408                         xpc_activate_IRQ_rcvd++;
 409                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
 410                 part_uv->reason = msg->reason;
 411                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 412
 413                 (*wakeup_hb_checker)++;
 414                 return;
 415         }
 416         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
 417                 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
 418
 419                 if (!part_setup)
 420                         break;
 421
 422                 msg = container_of(msg_hdr, struct
 423                                    xpc_activate_mq_msg_chctl_closerequest_uv,
 424                                    hdr);
 425                 args = &part->remote_openclose_args[msg->ch_number];
 426                 args->reason = msg->reason;
 427
 428                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 429                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
 430                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 431
 432                 xpc_wakeup_channel_mgr(part);
 433                 break;
 434         }
 435         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
 436                 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
 437
 438                 if (!part_setup)
 439                         break;
 440
 441                 msg = container_of(msg_hdr, struct
 442                                    xpc_activate_mq_msg_chctl_closereply_uv,
 443                                    hdr);
 444
 445                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 446                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
 447                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 448
 449                 xpc_wakeup_channel_mgr(part);
 450                 break;
 451         }
 452         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
 453                 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
 454
 455                 if (!part_setup)
 456                         break;
 457
 458                 msg = container_of(msg_hdr, struct
 459                                    xpc_activate_mq_msg_chctl_openrequest_uv,
 460                                    hdr);
 461                 args = &part->remote_openclose_args[msg->ch_number];
 462                 args->entry_size = msg->entry_size;
 463                 args->local_nentries = msg->local_nentries;
 464
 465                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 466                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
 467                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 468
 469                 xpc_wakeup_channel_mgr(part);
 470                 break;
 471         }
 472         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
 473                 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
 474
 475                 if (!part_setup)
 476                         break;
 477
 478                 msg = container_of(msg_hdr, struct
 479                                    xpc_activate_mq_msg_chctl_openreply_uv, hdr);
 480                 args = &part->remote_openclose_args[msg->ch_number];
 481                 args->remote_nentries = msg->remote_nentries;
 482                 args->local_nentries = msg->local_nentries;
 483                 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
 484
 485                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 486                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
 487                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 488
 489                 xpc_wakeup_channel_mgr(part);
 490                 break;
 491         }
 492         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
 493                 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
 494
 495                 if (!part_setup)
 496                         break;
 497
 498                 msg = container_of(msg_hdr, struct
 499                                 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
 500                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 501                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
 502                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 503
 504                 xpc_wakeup_channel_mgr(part);
 505         }
 506                 fallthrough;
 507         case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
 508                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 509                 part_uv->flags |= XPC_P_ENGAGED_UV;
 510                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 511                 break;
 512
 513         case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
 514                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 515                 part_uv->flags &= ~XPC_P_ENGAGED_UV;
 516                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 517                 break;
 518
 519         default:
 520                 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
 521                         "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
 522
 523                 /* get hb checker to deactivate from the remote partition */
 524                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 525                 if (part_uv->act_state_req == 0)
 526                         xpc_activate_IRQ_rcvd++;
 527                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
 528                 part_uv->reason = xpBadMsgType;
 529                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 530
 531                 (*wakeup_hb_checker)++;
 532                 return;
 533         }
 534
 535         if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
 536             part->remote_rp_ts_jiffies != 0) {
 537                 /*
 538                  * ??? Does what we do here need to be sensitive to
 539                  * ??? act_state or remote_act_state?
 540                  */
 541                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 542                 if (part_uv->act_state_req == 0)
 543                         xpc_activate_IRQ_rcvd++;
 544                 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
 545                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 546
 547                 (*wakeup_hb_checker)++;
 548         }
 549 }
 550
 551 static irqreturn_t
 552 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
 553 {
 554         struct xpc_activate_mq_msghdr_uv *msg_hdr;
 555         short partid;
 556         struct xpc_partition *part;
 557         int wakeup_hb_checker = 0;
 558         int part_referenced;
 559
 560         while (1) {
 561                 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
 562                 if (msg_hdr == NULL)
 563                         break;
 564
 565                 partid = msg_hdr->partid;
 566                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
 567                         dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
 568                                 "received invalid partid=0x%x in message\n",
 569                                 partid);
 570                 } else {
 571                         part = &xpc_partitions[partid];
 572
 573                         part_referenced = xpc_part_ref(part);
 574                         xpc_handle_activate_mq_msg_uv(part, msg_hdr,
 575                                                       part_referenced,
 576                                                       &wakeup_hb_checker);
 577                         if (part_referenced)
 578                                 xpc_part_deref(part);
 579                 }
 580
 581                 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
 582         }
 583
 584         if (wakeup_hb_checker)
 585                 wake_up_interruptible(&xpc_activate_IRQ_wq);
 586
 587         return IRQ_HANDLED;
 588 }
 589
 590 static enum xp_retval
 591 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
 592                                 unsigned long gru_mq_desc_gpa)
 593 {
 594         enum xp_retval ret;
 595
 596         ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
 597                                sizeof(struct gru_message_queue_desc));
 598         if (ret == xpSuccess)
 599                 gru_mq_desc->mq = NULL;
 600
 601         return ret;
 602 }
 603
 604 static enum xp_retval
 605 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
 606                          int msg_type)
 607 {
 608         struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
 609         struct xpc_partition_uv *part_uv = &part->sn.uv;
 610         struct gru_message_queue_desc *gru_mq_desc;
 611         unsigned long irq_flags;
 612         enum xp_retval ret;
 613
 614         DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
 615
 616         msg_hdr->type = msg_type;
 617         msg_hdr->partid = xp_partition_id;
 618         msg_hdr->act_state = part->act_state;
 619         msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
 620
 621         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
 622 again:
 623         if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
 624                 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
 625                 if (gru_mq_desc == NULL) {
 626                         gru_mq_desc = kmalloc(sizeof(struct
 627                                               gru_message_queue_desc),
 628                                               GFP_ATOMIC);
 629                         if (gru_mq_desc == NULL) {
 630                                 ret = xpNoMemory;
 631                                 goto done;
 632                         }
 633                         part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
 634                 }
 635
 636                 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
 637                                                       part_uv->
 638                                                       activate_gru_mq_desc_gpa);
 639                 if (ret != xpSuccess)
 640                         goto done;
 641
 642                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 643                 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
 644                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 645         }
 646
 647         /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
 648         ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
 649                                msg_size);
 650         if (ret != xpSuccess) {
 651                 smp_rmb();      /* ensure a fresh copy of part_uv->flags */
 652                 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
 653                         goto again;
 654         }
 655 done:
 656         mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
 657         return ret;
 658 }
 659
 660 static void
 661 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
 662                               size_t msg_size, int msg_type)
 663 {
 664         enum xp_retval ret;
 665
 666         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
 667         if (unlikely(ret != xpSuccess))
 668                 XPC_DEACTIVATE_PARTITION(part, ret);
 669 }
 670
 671 static void
 672 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
 673                          void *msg, size_t msg_size, int msg_type)
 674 {
 675         struct xpc_partition *part = &xpc_partitions[ch->partid];
 676         enum xp_retval ret;
 677
 678         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
 679         if (unlikely(ret != xpSuccess)) {
 680                 if (irq_flags != NULL)
 681                         spin_unlock_irqrestore(&ch->lock, *irq_flags);
 682
 683                 XPC_DEACTIVATE_PARTITION(part, ret);
 684
 685                 if (irq_flags != NULL)
 686                         spin_lock_irqsave(&ch->lock, *irq_flags);
 687         }
 688 }
 689
 690 static void
 691 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
 692 {
 693         unsigned long irq_flags;
 694         struct xpc_partition_uv *part_uv = &part->sn.uv;
 695
 696         /*
 697          * !!! Make our side think that the remote partition sent an activate
 698          * !!! mq message our way by doing what the activate IRQ handler would
 699          * !!! do had one really been sent.
 700          */
 701
 702         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 703         if (part_uv->act_state_req == 0)
 704                 xpc_activate_IRQ_rcvd++;
 705         part_uv->act_state_req = act_state_req;
 706         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 707
 708         wake_up_interruptible(&xpc_activate_IRQ_wq);
 709 }
 710
 711 static enum xp_retval
 712 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
 713                                   size_t *len)
 714 {
 715         s64 status;
 716         enum xp_retval ret;
 717
 718         status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
 719                                           (u64 *)len);
 720         if (status == BIOS_STATUS_SUCCESS)
 721                 ret = xpSuccess;
 722         else if (status == BIOS_STATUS_MORE_PASSES)
 723                 ret = xpNeedMoreInfo;
 724         else
 725                 ret = xpBiosError;
 726
 727         return ret;
 728 }
 729
 730 static int
 731 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
 732 {
 733         xpc_heartbeat_uv =
 734             &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
 735         rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
 736         rp->sn.uv.activate_gru_mq_desc_gpa =
 737             uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
 738         return 0;
 739 }
 740
 741 static void
 742 xpc_allow_hb_uv(short partid)
 743 {
 744 }
 745
 746 static void
 747 xpc_disallow_hb_uv(short partid)
 748 {
 749 }
 750
 751 static void
 752 xpc_disallow_all_hbs_uv(void)
 753 {
 754 }
 755
 756 static void
 757 xpc_increment_heartbeat_uv(void)
 758 {
 759         xpc_heartbeat_uv->value++;
 760 }
 761
 762 static void
 763 xpc_offline_heartbeat_uv(void)
 764 {
 765         xpc_increment_heartbeat_uv();
 766         xpc_heartbeat_uv->offline = 1;
 767 }
 768
 769 static void
 770 xpc_online_heartbeat_uv(void)
 771 {
 772         xpc_increment_heartbeat_uv();
 773         xpc_heartbeat_uv->offline = 0;
 774 }
 775
 776 static void
 777 xpc_heartbeat_init_uv(void)
 778 {
 779         xpc_heartbeat_uv->value = 1;
 780         xpc_heartbeat_uv->offline = 0;
 781 }
 782
 783 static void
 784 xpc_heartbeat_exit_uv(void)
 785 {
 786         xpc_offline_heartbeat_uv();
 787 }
 788
 789 static enum xp_retval
 790 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
 791 {
 792         struct xpc_partition_uv *part_uv = &part->sn.uv;
 793         enum xp_retval ret;
 794
 795         ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
 796                                part_uv->heartbeat_gpa,
 797                                sizeof(struct xpc_heartbeat_uv));
 798         if (ret != xpSuccess)
 799                 return ret;
 800
 801         if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
 802             !part_uv->cached_heartbeat.offline) {
 803
 804                 ret = xpNoHeartbeat;
 805         } else {
 806                 part->last_heartbeat = part_uv->cached_heartbeat.value;
 807         }
 808         return ret;
 809 }
 810
 811 static void
 812 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
 813                                     unsigned long remote_rp_gpa, int nasid)
 814 {
 815         short partid = remote_rp->SAL_partid;
 816         struct xpc_partition *part = &xpc_partitions[partid];
 817         struct xpc_activate_mq_msg_activate_req_uv msg;
 818
 819         part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
 820         part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
 821         part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
 822         part->sn.uv.activate_gru_mq_desc_gpa =
 823             remote_rp->sn.uv.activate_gru_mq_desc_gpa;
 824
 825         /*
 826          * ??? Is it a good idea to make this conditional on what is
 827          * ??? potentially stale state information?
 828          */
 829         if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
 830                 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
 831                 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
 832                 msg.activate_gru_mq_desc_gpa =
 833                     xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
 834                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
 835                                            XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
 836         }
 837
 838         if (part->act_state == XPC_P_AS_INACTIVE)
 839                 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
 840 }
 841
 842 static void
 843 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
 844 {
 845         xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
 846 }
 847
 848 static void
 849 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
 850 {
 851         struct xpc_activate_mq_msg_deactivate_req_uv msg;
 852
 853         /*
 854          * ??? Is it a good idea to make this conditional on what is
 855          * ??? potentially stale state information?
 856          */
 857         if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
 858             part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
 859
 860                 msg.reason = part->reason;
 861                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
 862                                          XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
 863         }
 864 }
 865
 866 static void
 867 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
 868 {
 869         /* nothing needs to be done */
 870         return;
 871 }
 872
 873 static void
 874 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
 875 {
 876         head->first = NULL;
 877         head->last = NULL;
 878         spin_lock_init(&head->lock);
 879         head->n_entries = 0;
 880 }
 881
 882 static void *
 883 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
 884 {
 885         unsigned long irq_flags;
 886         struct xpc_fifo_entry_uv *first;
 887
 888         spin_lock_irqsave(&head->lock, irq_flags);
 889         first = head->first;
 890         if (head->first != NULL) {
 891                 head->first = first->next;
 892                 if (head->first == NULL)
 893                         head->last = NULL;
 894
 895                 head->n_entries--;
 896                 BUG_ON(head->n_entries < 0);
 897
 898                 first->next = NULL;
 899         }
 900         spin_unlock_irqrestore(&head->lock, irq_flags);
 901         return first;
 902 }
 903
 904 static void
 905 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
 906                       struct xpc_fifo_entry_uv *last)
 907 {
 908         unsigned long irq_flags;
 909
 910         last->next = NULL;
 911         spin_lock_irqsave(&head->lock, irq_flags);
 912         if (head->last != NULL)
 913                 head->last->next = last;
 914         else
 915                 head->first = last;
 916         head->last = last;
 917         head->n_entries++;
 918         spin_unlock_irqrestore(&head->lock, irq_flags);
 919 }
 920
 921 static int
 922 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
 923 {
 924         return head->n_entries;
 925 }
 926
 927 /*
 928  * Setup the channel structures that are uv specific.
 929  */
 930 static enum xp_retval
 931 xpc_setup_ch_structures_uv(struct xpc_partition *part)
 932 {
 933         struct xpc_channel_uv *ch_uv;
 934         int ch_number;
 935
 936         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
 937                 ch_uv = &part->channels[ch_number].sn.uv;
 938
 939                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
 940                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
 941         }
 942
 943         return xpSuccess;
 944 }
 945
 946 /*
 947  * Teardown the channel structures that are uv specific.
 948  */
 949 static void
 950 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
 951 {
 952         /* nothing needs to be done */
 953         return;
 954 }
 955
 956 static enum xp_retval
 957 xpc_make_first_contact_uv(struct xpc_partition *part)
 958 {
 959         struct xpc_activate_mq_msg_uv msg;
 960
 961         /*
 962          * We send a sync msg to get the remote partition's remote_act_state
 963          * updated to our current act_state which at this point should
 964          * be XPC_P_AS_ACTIVATING.
 965          */
 966         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
 967                                       XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
 968
 969         while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
 970                  (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
 971
 972                 dev_dbg(xpc_part, "waiting to make first contact with "
 973                         "partition %d\n", XPC_PARTID(part));
 974
 975                 /* wait a 1/4 of a second or so */
 976                 (void)msleep_interruptible(250);
 977
 978                 if (part->act_state == XPC_P_AS_DEACTIVATING)
 979                         return part->reason;
 980         }
 981
 982         return xpSuccess;
 983 }
 984
 985 static u64
 986 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
 987 {
 988         unsigned long irq_flags;
 989         union xpc_channel_ctl_flags chctl;
 990
 991         spin_lock_irqsave(&part->chctl_lock, irq_flags);
 992         chctl = part->chctl;
 993         if (chctl.all_flags != 0)
 994                 part->chctl.all_flags = 0;
 995
 996         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 997         return chctl.all_flags;
 998 }
 999
1000 static enum xp_retval
1001 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1002 {
1003         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1004         struct xpc_send_msg_slot_uv *msg_slot;
1005         unsigned long irq_flags;
1006         int nentries;
1007         int entry;
1008         size_t nbytes;
1009
1010         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1011                 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1012                 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1013                 if (ch_uv->send_msg_slots == NULL)
1014                         continue;
1015
1016                 for (entry = 0; entry < nentries; entry++) {
1017                         msg_slot = &ch_uv->send_msg_slots[entry];
1018
1019                         msg_slot->msg_slot_number = entry;
1020                         xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1021                                               &msg_slot->next);
1022                 }
1023
1024                 spin_lock_irqsave(&ch->lock, irq_flags);
1025                 if (nentries < ch->local_nentries)
1026                         ch->local_nentries = nentries;
1027                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1028                 return xpSuccess;
1029         }
1030
1031         return xpNoMemory;
1032 }
1033
1034 static enum xp_retval
1035 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1036 {
1037         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1038         struct xpc_notify_mq_msg_uv *msg_slot;
1039         unsigned long irq_flags;
1040         int nentries;
1041         int entry;
1042         size_t nbytes;
1043
1044         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1045                 nbytes = nentries * ch->entry_size;
1046                 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1047                 if (ch_uv->recv_msg_slots == NULL)
1048                         continue;
1049
1050                 for (entry = 0; entry < nentries; entry++) {
1051                         msg_slot = ch_uv->recv_msg_slots +
1052                             entry * ch->entry_size;
1053
1054                         msg_slot->hdr.msg_slot_number = entry;
1055                 }
1056
1057                 spin_lock_irqsave(&ch->lock, irq_flags);
1058                 if (nentries < ch->remote_nentries)
1059                         ch->remote_nentries = nentries;
1060                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1061                 return xpSuccess;
1062         }
1063
1064         return xpNoMemory;
1065 }
1066
1067 /*
1068  * Allocate msg_slots associated with the channel.
1069  */
1070 static enum xp_retval
1071 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1072 {
1073         static enum xp_retval ret;
1074         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1075
1076         DBUG_ON(ch->flags & XPC_C_SETUP);
1077
1078         ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1079                                                    gru_message_queue_desc),
1080                                                    GFP_KERNEL);
1081         if (ch_uv->cached_notify_gru_mq_desc == NULL)
1082                 return xpNoMemory;
1083
1084         ret = xpc_allocate_send_msg_slot_uv(ch);
1085         if (ret == xpSuccess) {
1086
1087                 ret = xpc_allocate_recv_msg_slot_uv(ch);
1088                 if (ret != xpSuccess) {
1089                         kfree(ch_uv->send_msg_slots);
1090                         xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1091                 }
1092         }
1093         return ret;
1094 }
1095
1096 /*
1097  * Free up msg_slots and clear other stuff that were setup for the specified
1098  * channel.
1099  */
1100 static void
1101 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1102 {
1103         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1104
1105         lockdep_assert_held(&ch->lock);
1106
1107         kfree(ch_uv->cached_notify_gru_mq_desc);
1108         ch_uv->cached_notify_gru_mq_desc = NULL;
1109
1110         if (ch->flags & XPC_C_SETUP) {
1111                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1112                 kfree(ch_uv->send_msg_slots);
1113                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1114                 kfree(ch_uv->recv_msg_slots);
1115         }
1116 }
1117
1118 static void
1119 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1120 {
1121         struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1122
1123         msg.ch_number = ch->number;
1124         msg.reason = ch->reason;
1125         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1126                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1127 }
1128
1129 static void
1130 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1131 {
1132         struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1133
1134         msg.ch_number = ch->number;
1135         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1136                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1137 }
1138
1139 static void
1140 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1141 {
1142         struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1143
1144         msg.ch_number = ch->number;
1145         msg.entry_size = ch->entry_size;
1146         msg.local_nentries = ch->local_nentries;
1147         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1148                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1149 }
1150
1151 static void
1152 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1153 {
1154         struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1155
1156         msg.ch_number = ch->number;
1157         msg.local_nentries = ch->local_nentries;
1158         msg.remote_nentries = ch->remote_nentries;
1159         msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1160         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1161                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1162 }
1163
1164 static void
1165 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1166 {
1167         struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1168
1169         msg.ch_number = ch->number;
1170         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1171                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1172 }
1173
1174 static void
1175 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1176 {
1177         unsigned long irq_flags;
1178
1179         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1180         part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1181         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1182
1183         xpc_wakeup_channel_mgr(part);
1184 }
1185
1186 static enum xp_retval
1187 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1188                                unsigned long gru_mq_desc_gpa)
1189 {
1190         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1191
1192         DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1193         return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1194                                                gru_mq_desc_gpa);
1195 }
1196
1197 static void
1198 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1199 {
1200         struct xpc_activate_mq_msg_uv msg;
1201
1202         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1203                                       XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1204 }
1205
1206 static void
1207 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1208 {
1209         struct xpc_activate_mq_msg_uv msg;
1210
1211         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1212                                       XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1213 }
1214
1215 static void
1216 xpc_assume_partition_disengaged_uv(short partid)
1217 {
1218         struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1219         unsigned long irq_flags;
1220
1221         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1222         part_uv->flags &= ~XPC_P_ENGAGED_UV;
1223         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1224 }
1225
1226 static int
1227 xpc_partition_engaged_uv(short partid)
1228 {
1229         return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1230 }
1231
1232 static int
1233 xpc_any_partition_engaged_uv(void)
1234 {
1235         struct xpc_partition_uv *part_uv;
1236         short partid;
1237
1238         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1239                 part_uv = &xpc_partitions[partid].sn.uv;
1240                 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1241                         return 1;
1242         }
1243         return 0;
1244 }
1245
1246 static enum xp_retval
1247 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1248                          struct xpc_send_msg_slot_uv **address_of_msg_slot)
1249 {
1250         enum xp_retval ret;
1251         struct xpc_send_msg_slot_uv *msg_slot;
1252         struct xpc_fifo_entry_uv *entry;
1253
1254         while (1) {
1255                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1256                 if (entry != NULL)
1257                         break;
1258
1259                 if (flags & XPC_NOWAIT)
1260                         return xpNoWait;
1261
1262                 ret = xpc_allocate_msg_wait(ch);
1263                 if (ret != xpInterrupted && ret != xpTimeout)
1264                         return ret;
1265         }
1266
1267         msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1268         *address_of_msg_slot = msg_slot;
1269         return xpSuccess;
1270 }
1271
1272 static void
1273 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1274                      struct xpc_send_msg_slot_uv *msg_slot)
1275 {
1276         xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1277
1278         /* wakeup anyone waiting for a free msg slot */
1279         if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1280                 wake_up(&ch->msg_allocate_wq);
1281 }
1282
1283 static void
1284 xpc_notify_sender_uv(struct xpc_channel *ch,
1285                      struct xpc_send_msg_slot_uv *msg_slot,
1286                      enum xp_retval reason)
1287 {
1288         xpc_notify_func func = msg_slot->func;
1289
1290         if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1291
1292                 atomic_dec(&ch->n_to_notify);
1293
1294                 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1295                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1296                         msg_slot->msg_slot_number, ch->partid, ch->number);
1297
1298                 func(reason, ch->partid, ch->number, msg_slot->key);
1299
1300                 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1301                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1302                         msg_slot->msg_slot_number, ch->partid, ch->number);
1303         }
1304 }
1305
1306 static void
1307 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1308                             struct xpc_notify_mq_msg_uv *msg)
1309 {
1310         struct xpc_send_msg_slot_uv *msg_slot;
1311         int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1312
1313         msg_slot = &ch->sn.uv.send_msg_slots[entry];
1314
1315         BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1316         msg_slot->msg_slot_number += ch->local_nentries;
1317
1318         if (msg_slot->func != NULL)
1319                 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1320
1321         xpc_free_msg_slot_uv(ch, msg_slot);
1322 }
1323
1324 static void
1325 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1326                             struct xpc_notify_mq_msg_uv *msg)
1327 {
1328         struct xpc_partition_uv *part_uv = &part->sn.uv;
1329         struct xpc_channel *ch;
1330         struct xpc_channel_uv *ch_uv;
1331         struct xpc_notify_mq_msg_uv *msg_slot;
1332         unsigned long irq_flags;
1333         int ch_number = msg->hdr.ch_number;
1334
1335         if (unlikely(ch_number >= part->nchannels)) {
1336                 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1337                         "channel number=0x%x in message from partid=%d\n",
1338                         ch_number, XPC_PARTID(part));
1339
1340                 /* get hb checker to deactivate from the remote partition */
1341                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1342                 if (part_uv->act_state_req == 0)
1343                         xpc_activate_IRQ_rcvd++;
1344                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1345                 part_uv->reason = xpBadChannelNumber;
1346                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1347
1348                 wake_up_interruptible(&xpc_activate_IRQ_wq);
1349                 return;
1350         }
1351
1352         ch = &part->channels[ch_number];
1353         xpc_msgqueue_ref(ch);
1354
1355         if (!(ch->flags & XPC_C_CONNECTED)) {
1356                 xpc_msgqueue_deref(ch);
1357                 return;
1358         }
1359
1360         /* see if we're really dealing with an ACK for a previously sent msg */
1361         if (msg->hdr.size == 0) {
1362                 xpc_handle_notify_mq_ack_uv(ch, msg);
1363                 xpc_msgqueue_deref(ch);
1364                 return;
1365         }
1366
1367         /* we're dealing with a normal message sent via the notify_mq */
1368         ch_uv = &ch->sn.uv;
1369
1370         msg_slot = ch_uv->recv_msg_slots +
1371             (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1372
1373         BUG_ON(msg_slot->hdr.size != 0);
1374
1375         memcpy(msg_slot, msg, msg->hdr.size);
1376
1377         xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1378
1379         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1380                 /*
1381                  * If there is an existing idle kthread get it to deliver
1382                  * the payload, otherwise we'll have to get the channel mgr
1383                  * for this partition to create a kthread to do the delivery.
1384                  */
1385                 if (atomic_read(&ch->kthreads_idle) > 0)
1386                         wake_up_nr(&ch->idle_wq, 1);
1387                 else
1388                         xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1389         }
1390         xpc_msgqueue_deref(ch);
1391 }
1392
1393 static irqreturn_t
1394 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1395 {
1396         struct xpc_notify_mq_msg_uv *msg;
1397         short partid;
1398         struct xpc_partition *part;
1399
1400         while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1401                NULL) {
1402
1403                 partid = msg->hdr.partid;
1404                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1405                         dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1406                                 "invalid partid=0x%x in message\n", partid);
1407                 } else {
1408                         part = &xpc_partitions[partid];
1409
1410                         if (xpc_part_ref(part)) {
1411                                 xpc_handle_notify_mq_msg_uv(part, msg);
1412                                 xpc_part_deref(part);
1413                         }
1414                 }
1415
1416                 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1417         }
1418
1419         return IRQ_HANDLED;
1420 }
1421
1422 static int
1423 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1424 {
1425         return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1426 }
1427
1428 static void
1429 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1430 {
1431         struct xpc_channel *ch = &part->channels[ch_number];
1432         int ndeliverable_payloads;
1433
1434         xpc_msgqueue_ref(ch);
1435
1436         ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1437
1438         if (ndeliverable_payloads > 0 &&
1439             (ch->flags & XPC_C_CONNECTED) &&
1440             (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1441
1442                 xpc_activate_kthreads(ch, ndeliverable_payloads);
1443         }
1444
1445         xpc_msgqueue_deref(ch);
1446 }
1447
1448 static enum xp_retval
1449 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1450                     u16 payload_size, u8 notify_type, xpc_notify_func func,
1451                     void *key)
1452 {
1453         enum xp_retval ret = xpSuccess;
1454         struct xpc_send_msg_slot_uv *msg_slot = NULL;
1455         struct xpc_notify_mq_msg_uv *msg;
1456         u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1457         size_t msg_size;
1458
1459         DBUG_ON(notify_type != XPC_N_CALL);
1460
1461         msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1462         if (msg_size > ch->entry_size)
1463                 return xpPayloadTooBig;
1464
1465         xpc_msgqueue_ref(ch);
1466
1467         if (ch->flags & XPC_C_DISCONNECTING) {
1468                 ret = ch->reason;
1469                 goto out_1;
1470         }
1471         if (!(ch->flags & XPC_C_CONNECTED)) {
1472                 ret = xpNotConnected;
1473                 goto out_1;
1474         }
1475
1476         ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1477         if (ret != xpSuccess)
1478                 goto out_1;
1479
1480         if (func != NULL) {
1481                 atomic_inc(&ch->n_to_notify);
1482
1483                 msg_slot->key = key;
1484                 smp_wmb(); /* a non-NULL func must hit memory after the key */
1485                 msg_slot->func = func;
1486
1487                 if (ch->flags & XPC_C_DISCONNECTING) {
1488                         ret = ch->reason;
1489                         goto out_2;
1490                 }
1491         }
1492
1493         msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1494         msg->hdr.partid = xp_partition_id;
1495         msg->hdr.ch_number = ch->number;
1496         msg->hdr.size = msg_size;
1497         msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1498         memcpy(&msg->payload, payload, payload_size);
1499
1500         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1501                                msg_size);
1502         if (ret == xpSuccess)
1503                 goto out_1;
1504
1505         XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1506 out_2:
1507         if (func != NULL) {
1508                 /*
1509                  * Try to NULL the msg_slot's func field. If we fail, then
1510                  * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1511                  * case we need to pretend we succeeded to send the message
1512                  * since the user will get a callout for the disconnect error
1513                  * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1514                  * error returned here will confuse them. Additionally, since
1515                  * in this case the channel is being disconnected we don't need
1516                  * to put the msg_slot back on the free list.
1517                  */
1518                 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1519                         ret = xpSuccess;
1520                         goto out_1;
1521                 }
1522
1523                 msg_slot->key = NULL;
1524                 atomic_dec(&ch->n_to_notify);
1525         }
1526         xpc_free_msg_slot_uv(ch, msg_slot);
1527 out_1:
1528         xpc_msgqueue_deref(ch);
1529         return ret;
1530 }
1531
1532 /*
1533  * Tell the callers of xpc_send_notify() that the status of their payloads
1534  * is unknown because the channel is now disconnecting.
1535  *
1536  * We don't worry about putting these msg_slots on the free list since the
1537  * msg_slots themselves are about to be kfree'd.
1538  */
1539 static void
1540 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1541 {
1542         struct xpc_send_msg_slot_uv *msg_slot;
1543         int entry;
1544
1545         DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1546
1547         for (entry = 0; entry < ch->local_nentries; entry++) {
1548
1549                 if (atomic_read(&ch->n_to_notify) == 0)
1550                         break;
1551
1552                 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1553                 if (msg_slot->func != NULL)
1554                         xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1555         }
1556 }
1557
1558 /*
1559  * Get the next deliverable message's payload.
1560  */
1561 static void *
1562 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1563 {
1564         struct xpc_fifo_entry_uv *entry;
1565         struct xpc_notify_mq_msg_uv *msg;
1566         void *payload = NULL;
1567
1568         if (!(ch->flags & XPC_C_DISCONNECTING)) {
1569                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1570                 if (entry != NULL) {
1571                         msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1572                                            hdr.u.next);
1573                         payload = &msg->payload;
1574                 }
1575         }
1576         return payload;
1577 }
1578
1579 static void
1580 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1581 {
1582         struct xpc_notify_mq_msg_uv *msg;
1583         enum xp_retval ret;
1584
1585         msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1586
1587         /* return an ACK to the sender of this message */
1588
1589         msg->hdr.partid = xp_partition_id;
1590         msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
1591
1592         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1593                                sizeof(struct xpc_notify_mq_msghdr_uv));
1594         if (ret != xpSuccess)
1595                 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1596 }
1597
1598 static const struct xpc_arch_operations xpc_arch_ops_uv = {
1599         .setup_partitions = xpc_setup_partitions_uv,
1600         .teardown_partitions = xpc_teardown_partitions_uv,
1601         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1602         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1603         .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1604
1605         .allow_hb = xpc_allow_hb_uv,
1606         .disallow_hb = xpc_disallow_hb_uv,
1607         .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1608         .increment_heartbeat = xpc_increment_heartbeat_uv,
1609         .offline_heartbeat = xpc_offline_heartbeat_uv,
1610         .online_heartbeat = xpc_online_heartbeat_uv,
1611         .heartbeat_init = xpc_heartbeat_init_uv,
1612         .heartbeat_exit = xpc_heartbeat_exit_uv,
1613         .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1614
1615         .request_partition_activation =
1616                 xpc_request_partition_activation_uv,
1617         .request_partition_reactivation =
1618                 xpc_request_partition_reactivation_uv,
1619         .request_partition_deactivation =
1620                 xpc_request_partition_deactivation_uv,
1621         .cancel_partition_deactivation_request =
1622                 xpc_cancel_partition_deactivation_request_uv,
1623
1624         .setup_ch_structures = xpc_setup_ch_structures_uv,
1625         .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1626
1627         .make_first_contact = xpc_make_first_contact_uv,
1628
1629         .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1630         .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1631         .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1632         .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1633         .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1634         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1635         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1636
1637         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1638
1639         .setup_msg_structures = xpc_setup_msg_structures_uv,
1640         .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1641
1642         .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1643         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1644         .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1645         .partition_engaged = xpc_partition_engaged_uv,
1646         .any_partition_engaged = xpc_any_partition_engaged_uv,
1647
1648         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1649         .send_payload = xpc_send_payload_uv,
1650         .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1651         .received_payload = xpc_received_payload_uv,
1652         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1653 };
1654
1655 static int
1656 xpc_init_mq_node(int nid)
1657 {
1658         int cpu;
1659
1660         cpus_read_lock();
1661
1662         for_each_cpu(cpu, cpumask_of_node(nid)) {
1663                 xpc_activate_mq_uv =
1664                         xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
1665                                              XPC_ACTIVATE_IRQ_NAME,
1666                                              xpc_handle_activate_IRQ_uv);
1667                 if (!IS_ERR(xpc_activate_mq_uv))
1668                         break;
1669         }
1670         if (IS_ERR(xpc_activate_mq_uv)) {
1671                 cpus_read_unlock();
1672                 return PTR_ERR(xpc_activate_mq_uv);
1673         }
1674
1675         for_each_cpu(cpu, cpumask_of_node(nid)) {
1676                 xpc_notify_mq_uv =
1677                         xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
1678                                              XPC_NOTIFY_IRQ_NAME,
1679                                              xpc_handle_notify_IRQ_uv);
1680                 if (!IS_ERR(xpc_notify_mq_uv))
1681                         break;
1682         }
1683         if (IS_ERR(xpc_notify_mq_uv)) {
1684                 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1685                 cpus_read_unlock();
1686                 return PTR_ERR(xpc_notify_mq_uv);
1687         }
1688
1689         cpus_read_unlock();
1690         return 0;
1691 }
1692
1693 int
1694 xpc_init_uv(void)
1695 {
1696         int nid;
1697         int ret = 0;
1698
1699         xpc_arch_ops = xpc_arch_ops_uv;
1700
1701         if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1702                 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1703                         XPC_MSG_HDR_MAX_SIZE);
1704                 return -E2BIG;
1705         }
1706
1707         if (xpc_mq_node < 0)
1708                 for_each_online_node(nid) {
1709                         ret = xpc_init_mq_node(nid);
1710
1711                         if (!ret)
1712                                 break;
1713                 }
1714         else
1715                 ret = xpc_init_mq_node(xpc_mq_node);
1716
1717         if (ret < 0)
1718                 dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1719                         -ret);
1720
1721         return ret;
1722 }
1723
1724 void
1725 xpc_exit_uv(void)
1726 {
1727         xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1728         xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1729 }
1730
1731 module_param(xpc_mq_node, int, 0);
1732 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");