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 <asm/uv/uv_hub.h> 26 #if defined CONFIG_X86_64 27 #include <asm/uv/bios.h> 28 #include <asm/uv/uv_irq.h> 29 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 30 #include <asm/sn/intr.h> 31 #include <asm/sn/sn_sal.h> 32 #endif 33 #include "../sgi-gru/gru.h" 34 #include "../sgi-gru/grukservices.h" 35 #include "xpc.h" 36 37 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 38 struct uv_IO_APIC_route_entry { 39 __u64 vector : 8, 40 delivery_mode : 3, 41 dest_mode : 1, 42 delivery_status : 1, 43 polarity : 1, 44 __reserved_1 : 1, 45 trigger : 1, 46 mask : 1, 47 __reserved_2 : 15, 48 dest : 32; 49 }; 50 #endif 51 52 static struct xpc_heartbeat_uv *xpc_heartbeat_uv; 53 54 #define XPC_ACTIVATE_MSG_SIZE_UV (1 * GRU_CACHE_LINE_BYTES) 55 #define XPC_ACTIVATE_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \ 56 XPC_ACTIVATE_MSG_SIZE_UV) 57 #define XPC_ACTIVATE_IRQ_NAME "xpc_activate" 58 59 #define XPC_NOTIFY_MSG_SIZE_UV (2 * GRU_CACHE_LINE_BYTES) 60 #define XPC_NOTIFY_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \ 61 XPC_NOTIFY_MSG_SIZE_UV) 62 #define XPC_NOTIFY_IRQ_NAME "xpc_notify" 63 64 static int xpc_mq_node = -1; 65 66 static struct xpc_gru_mq_uv *xpc_activate_mq_uv; 67 static struct xpc_gru_mq_uv *xpc_notify_mq_uv; 68 69 static int 70 xpc_setup_partitions_uv(void) 71 { 72 short partid; 73 struct xpc_partition_uv *part_uv; 74 75 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) { 76 part_uv = &xpc_partitions[partid].sn.uv; 77 78 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex); 79 spin_lock_init(&part_uv->flags_lock); 80 part_uv->remote_act_state = XPC_P_AS_INACTIVE; 81 } 82 return 0; 83 } 84 85 static void 86 xpc_teardown_partitions_uv(void) 87 { 88 short partid; 89 struct xpc_partition_uv *part_uv; 90 unsigned long irq_flags; 91 92 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) { 93 part_uv = &xpc_partitions[partid].sn.uv; 94 95 if (part_uv->cached_activate_gru_mq_desc != NULL) { 96 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex); 97 spin_lock_irqsave(&part_uv->flags_lock, irq_flags); 98 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV; 99 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags); 100 kfree(part_uv->cached_activate_gru_mq_desc); 101 part_uv->cached_activate_gru_mq_desc = NULL; 102 mutex_unlock(&part_uv-> 103 cached_activate_gru_mq_desc_mutex); 104 } 105 } 106 } 107 108 static int 109 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name) 110 { 111 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade); 112 113 #if defined CONFIG_X86_64 114 mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset, 115 UV_AFFINITY_CPU); 116 if (mq->irq < 0) 117 return mq->irq; 118 119 mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset); 120 121 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 122 if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0) 123 mq->irq = SGI_XPC_ACTIVATE; 124 else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0) 125 mq->irq = SGI_XPC_NOTIFY; 126 else 127 return -EINVAL; 128 129 mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq; 130 uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value); 131 #else 132 #error not a supported configuration 133 #endif 134 135 return 0; 136 } 137 138 static void 139 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq) 140 { 141 #if defined CONFIG_X86_64 142 uv_teardown_irq(mq->irq); 143 144 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 145 int mmr_pnode; 146 unsigned long mmr_value; 147 148 mmr_pnode = uv_blade_to_pnode(mq->mmr_blade); 149 mmr_value = 1UL << 16; 150 151 uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value); 152 #else 153 #error not a supported configuration 154 #endif 155 } 156 157 static int 158 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq) 159 { 160 int ret; 161 162 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 163 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade); 164 165 ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address), 166 mq->order, &mq->mmr_offset); 167 if (ret < 0) { 168 dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n", 169 ret); 170 return -EBUSY; 171 } 172 #elif defined CONFIG_X86_64 173 ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address), 174 mq->order, &mq->mmr_offset); 175 if (ret < 0) { 176 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, " 177 "ret=%d\n", ret); 178 return ret; 179 } 180 #else 181 #error not a supported configuration 182 #endif 183 184 mq->watchlist_num = ret; 185 return 0; 186 } 187 188 static void 189 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq) 190 { 191 int ret; 192 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade); 193 194 #if defined CONFIG_X86_64 195 ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num); 196 BUG_ON(ret != BIOS_STATUS_SUCCESS); 197 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 198 ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num); 199 BUG_ON(ret != SALRET_OK); 200 #else 201 #error not a supported configuration 202 #endif 203 } 204 205 static struct xpc_gru_mq_uv * 206 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name, 207 irq_handler_t irq_handler) 208 { 209 enum xp_retval xp_ret; 210 int ret; 211 int nid; 212 int nasid; 213 int pg_order; 214 struct page *page; 215 struct xpc_gru_mq_uv *mq; 216 struct uv_IO_APIC_route_entry *mmr_value; 217 218 mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL); 219 if (mq == NULL) { 220 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() " 221 "a xpc_gru_mq_uv structure\n"); 222 ret = -ENOMEM; 223 goto out_0; 224 } 225 226 mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc), 227 GFP_KERNEL); 228 if (mq->gru_mq_desc == NULL) { 229 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() " 230 "a gru_message_queue_desc structure\n"); 231 ret = -ENOMEM; 232 goto out_1; 233 } 234 235 pg_order = get_order(mq_size); 236 mq->order = pg_order + PAGE_SHIFT; 237 mq_size = 1UL << mq->order; 238 239 mq->mmr_blade = uv_cpu_to_blade_id(cpu); 240 241 nid = cpu_to_node(cpu); 242 page = __alloc_pages_node(nid, 243 GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE, 244 pg_order); 245 if (page == NULL) { 246 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d " 247 "bytes of memory on nid=%d for GRU mq\n", mq_size, nid); 248 ret = -ENOMEM; 249 goto out_2; 250 } 251 mq->address = page_address(page); 252 253 /* enable generation of irq when GRU mq operation occurs to this mq */ 254 ret = xpc_gru_mq_watchlist_alloc_uv(mq); 255 if (ret != 0) 256 goto out_3; 257 258 ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name); 259 if (ret != 0) 260 goto out_4; 261 262 ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL); 263 if (ret != 0) { 264 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n", 265 mq->irq, -ret); 266 goto out_5; 267 } 268 269 nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu)); 270 271 mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value; 272 ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size, 273 nasid, mmr_value->vector, mmr_value->dest); 274 if (ret != 0) { 275 dev_err(xpc_part, "gru_create_message_queue() returned " 276 "error=%d\n", ret); 277 ret = -EINVAL; 278 goto out_6; 279 } 280 281 /* allow other partitions to access this GRU mq */ 282 xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size); 283 if (xp_ret != xpSuccess) { 284 ret = -EACCES; 285 goto out_6; 286 } 287 288 return mq; 289 290 /* something went wrong */ 291 out_6: 292 free_irq(mq->irq, NULL); 293 out_5: 294 xpc_release_gru_mq_irq_uv(mq); 295 out_4: 296 xpc_gru_mq_watchlist_free_uv(mq); 297 out_3: 298 free_pages((unsigned long)mq->address, pg_order); 299 out_2: 300 kfree(mq->gru_mq_desc); 301 out_1: 302 kfree(mq); 303 out_0: 304 return ERR_PTR(ret); 305 } 306 307 static void 308 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq) 309 { 310 unsigned int mq_size; 311 int pg_order; 312 int ret; 313 314 /* disallow other partitions to access GRU mq */ 315 mq_size = 1UL << mq->order; 316 ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size); 317 BUG_ON(ret != xpSuccess); 318 319 /* unregister irq handler and release mq irq/vector mapping */ 320 free_irq(mq->irq, NULL); 321 xpc_release_gru_mq_irq_uv(mq); 322 323 /* disable generation of irq when GRU mq op occurs to this mq */ 324 xpc_gru_mq_watchlist_free_uv(mq); 325 326 pg_order = mq->order - PAGE_SHIFT; 327 free_pages((unsigned long)mq->address, pg_order); 328 329 kfree(mq); 330 } 331 332 static enum xp_retval 333 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg, 334 size_t msg_size) 335 { 336 enum xp_retval xp_ret; 337 int ret; 338 339 while (1) { 340 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size); 341 if (ret == MQE_OK) { 342 xp_ret = xpSuccess; 343 break; 344 } 345 346 if (ret == MQE_QUEUE_FULL) { 347 dev_dbg(xpc_chan, "gru_send_message_gpa() returned " 348 "error=MQE_QUEUE_FULL\n"); 349 /* !!! handle QLimit reached; delay & try again */ 350 /* ??? Do we add a limit to the number of retries? */ 351 (void)msleep_interruptible(10); 352 } else if (ret == MQE_CONGESTION) { 353 dev_dbg(xpc_chan, "gru_send_message_gpa() returned " 354 "error=MQE_CONGESTION\n"); 355 /* !!! handle LB Overflow; simply try again */ 356 /* ??? Do we add a limit to the number of retries? */ 357 } else { 358 /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */ 359 dev_err(xpc_chan, "gru_send_message_gpa() returned " 360 "error=%d\n", ret); 361 xp_ret = xpGruSendMqError; 362 break; 363 } 364 } 365 return xp_ret; 366 } 367 368 static void 369 xpc_process_activate_IRQ_rcvd_uv(void) 370 { 371 unsigned long irq_flags; 372 short partid; 373 struct xpc_partition *part; 374 u8 act_state_req; 375 376 DBUG_ON(xpc_activate_IRQ_rcvd == 0); 377 378 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 379 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) { 380 part = &xpc_partitions[partid]; 381 382 if (part->sn.uv.act_state_req == 0) 383 continue; 384 385 xpc_activate_IRQ_rcvd--; 386 BUG_ON(xpc_activate_IRQ_rcvd < 0); 387 388 act_state_req = part->sn.uv.act_state_req; 389 part->sn.uv.act_state_req = 0; 390 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 391 392 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) { 393 if (part->act_state == XPC_P_AS_INACTIVE) 394 xpc_activate_partition(part); 395 else if (part->act_state == XPC_P_AS_DEACTIVATING) 396 XPC_DEACTIVATE_PARTITION(part, xpReactivating); 397 398 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) { 399 if (part->act_state == XPC_P_AS_INACTIVE) 400 xpc_activate_partition(part); 401 else 402 XPC_DEACTIVATE_PARTITION(part, xpReactivating); 403 404 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) { 405 XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason); 406 407 } else { 408 BUG(); 409 } 410 411 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 412 if (xpc_activate_IRQ_rcvd == 0) 413 break; 414 } 415 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 416 417 } 418 419 static void 420 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part, 421 struct xpc_activate_mq_msghdr_uv *msg_hdr, 422 int part_setup, 423 int *wakeup_hb_checker) 424 { 425 unsigned long irq_flags; 426 struct xpc_partition_uv *part_uv = &part->sn.uv; 427 struct xpc_openclose_args *args; 428 429 part_uv->remote_act_state = msg_hdr->act_state; 430 431 switch (msg_hdr->type) { 432 case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV: 433 /* syncing of remote_act_state was just done above */ 434 break; 435 436 case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: { 437 struct xpc_activate_mq_msg_activate_req_uv *msg; 438 439 /* 440 * ??? Do we deal here with ts_jiffies being different 441 * ??? if act_state != XPC_P_AS_INACTIVE instead of 442 * ??? below? 443 */ 444 msg = container_of(msg_hdr, struct 445 xpc_activate_mq_msg_activate_req_uv, hdr); 446 447 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 448 if (part_uv->act_state_req == 0) 449 xpc_activate_IRQ_rcvd++; 450 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV; 451 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */ 452 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies; 453 part_uv->heartbeat_gpa = msg->heartbeat_gpa; 454 455 if (msg->activate_gru_mq_desc_gpa != 456 part_uv->activate_gru_mq_desc_gpa) { 457 spin_lock(&part_uv->flags_lock); 458 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV; 459 spin_unlock(&part_uv->flags_lock); 460 part_uv->activate_gru_mq_desc_gpa = 461 msg->activate_gru_mq_desc_gpa; 462 } 463 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 464 465 (*wakeup_hb_checker)++; 466 break; 467 } 468 case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: { 469 struct xpc_activate_mq_msg_deactivate_req_uv *msg; 470 471 msg = container_of(msg_hdr, struct 472 xpc_activate_mq_msg_deactivate_req_uv, hdr); 473 474 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 475 if (part_uv->act_state_req == 0) 476 xpc_activate_IRQ_rcvd++; 477 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV; 478 part_uv->reason = msg->reason; 479 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 480 481 (*wakeup_hb_checker)++; 482 return; 483 } 484 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: { 485 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg; 486 487 if (!part_setup) 488 break; 489 490 msg = container_of(msg_hdr, struct 491 xpc_activate_mq_msg_chctl_closerequest_uv, 492 hdr); 493 args = &part->remote_openclose_args[msg->ch_number]; 494 args->reason = msg->reason; 495 496 spin_lock_irqsave(&part->chctl_lock, irq_flags); 497 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST; 498 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 499 500 xpc_wakeup_channel_mgr(part); 501 break; 502 } 503 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: { 504 struct xpc_activate_mq_msg_chctl_closereply_uv *msg; 505 506 if (!part_setup) 507 break; 508 509 msg = container_of(msg_hdr, struct 510 xpc_activate_mq_msg_chctl_closereply_uv, 511 hdr); 512 513 spin_lock_irqsave(&part->chctl_lock, irq_flags); 514 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY; 515 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 516 517 xpc_wakeup_channel_mgr(part); 518 break; 519 } 520 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: { 521 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg; 522 523 if (!part_setup) 524 break; 525 526 msg = container_of(msg_hdr, struct 527 xpc_activate_mq_msg_chctl_openrequest_uv, 528 hdr); 529 args = &part->remote_openclose_args[msg->ch_number]; 530 args->entry_size = msg->entry_size; 531 args->local_nentries = msg->local_nentries; 532 533 spin_lock_irqsave(&part->chctl_lock, irq_flags); 534 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST; 535 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 536 537 xpc_wakeup_channel_mgr(part); 538 break; 539 } 540 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: { 541 struct xpc_activate_mq_msg_chctl_openreply_uv *msg; 542 543 if (!part_setup) 544 break; 545 546 msg = container_of(msg_hdr, struct 547 xpc_activate_mq_msg_chctl_openreply_uv, hdr); 548 args = &part->remote_openclose_args[msg->ch_number]; 549 args->remote_nentries = msg->remote_nentries; 550 args->local_nentries = msg->local_nentries; 551 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa; 552 553 spin_lock_irqsave(&part->chctl_lock, irq_flags); 554 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY; 555 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 556 557 xpc_wakeup_channel_mgr(part); 558 break; 559 } 560 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: { 561 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg; 562 563 if (!part_setup) 564 break; 565 566 msg = container_of(msg_hdr, struct 567 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr); 568 spin_lock_irqsave(&part->chctl_lock, irq_flags); 569 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE; 570 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 571 572 xpc_wakeup_channel_mgr(part); 573 } 574 case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV: 575 spin_lock_irqsave(&part_uv->flags_lock, irq_flags); 576 part_uv->flags |= XPC_P_ENGAGED_UV; 577 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags); 578 break; 579 580 case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV: 581 spin_lock_irqsave(&part_uv->flags_lock, irq_flags); 582 part_uv->flags &= ~XPC_P_ENGAGED_UV; 583 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags); 584 break; 585 586 default: 587 dev_err(xpc_part, "received unknown activate_mq msg type=%d " 588 "from partition=%d\n", msg_hdr->type, XPC_PARTID(part)); 589 590 /* get hb checker to deactivate from the remote partition */ 591 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 592 if (part_uv->act_state_req == 0) 593 xpc_activate_IRQ_rcvd++; 594 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV; 595 part_uv->reason = xpBadMsgType; 596 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 597 598 (*wakeup_hb_checker)++; 599 return; 600 } 601 602 if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies && 603 part->remote_rp_ts_jiffies != 0) { 604 /* 605 * ??? Does what we do here need to be sensitive to 606 * ??? act_state or remote_act_state? 607 */ 608 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 609 if (part_uv->act_state_req == 0) 610 xpc_activate_IRQ_rcvd++; 611 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV; 612 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 613 614 (*wakeup_hb_checker)++; 615 } 616 } 617 618 static irqreturn_t 619 xpc_handle_activate_IRQ_uv(int irq, void *dev_id) 620 { 621 struct xpc_activate_mq_msghdr_uv *msg_hdr; 622 short partid; 623 struct xpc_partition *part; 624 int wakeup_hb_checker = 0; 625 int part_referenced; 626 627 while (1) { 628 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc); 629 if (msg_hdr == NULL) 630 break; 631 632 partid = msg_hdr->partid; 633 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) { 634 dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() " 635 "received invalid partid=0x%x in message\n", 636 partid); 637 } else { 638 part = &xpc_partitions[partid]; 639 640 part_referenced = xpc_part_ref(part); 641 xpc_handle_activate_mq_msg_uv(part, msg_hdr, 642 part_referenced, 643 &wakeup_hb_checker); 644 if (part_referenced) 645 xpc_part_deref(part); 646 } 647 648 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr); 649 } 650 651 if (wakeup_hb_checker) 652 wake_up_interruptible(&xpc_activate_IRQ_wq); 653 654 return IRQ_HANDLED; 655 } 656 657 static enum xp_retval 658 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc, 659 unsigned long gru_mq_desc_gpa) 660 { 661 enum xp_retval ret; 662 663 ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa, 664 sizeof(struct gru_message_queue_desc)); 665 if (ret == xpSuccess) 666 gru_mq_desc->mq = NULL; 667 668 return ret; 669 } 670 671 static enum xp_retval 672 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size, 673 int msg_type) 674 { 675 struct xpc_activate_mq_msghdr_uv *msg_hdr = msg; 676 struct xpc_partition_uv *part_uv = &part->sn.uv; 677 struct gru_message_queue_desc *gru_mq_desc; 678 unsigned long irq_flags; 679 enum xp_retval ret; 680 681 DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV); 682 683 msg_hdr->type = msg_type; 684 msg_hdr->partid = xp_partition_id; 685 msg_hdr->act_state = part->act_state; 686 msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies; 687 688 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex); 689 again: 690 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) { 691 gru_mq_desc = part_uv->cached_activate_gru_mq_desc; 692 if (gru_mq_desc == NULL) { 693 gru_mq_desc = kmalloc(sizeof(struct 694 gru_message_queue_desc), 695 GFP_KERNEL); 696 if (gru_mq_desc == NULL) { 697 ret = xpNoMemory; 698 goto done; 699 } 700 part_uv->cached_activate_gru_mq_desc = gru_mq_desc; 701 } 702 703 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc, 704 part_uv-> 705 activate_gru_mq_desc_gpa); 706 if (ret != xpSuccess) 707 goto done; 708 709 spin_lock_irqsave(&part_uv->flags_lock, irq_flags); 710 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV; 711 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags); 712 } 713 714 /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */ 715 ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg, 716 msg_size); 717 if (ret != xpSuccess) { 718 smp_rmb(); /* ensure a fresh copy of part_uv->flags */ 719 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) 720 goto again; 721 } 722 done: 723 mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex); 724 return ret; 725 } 726 727 static void 728 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg, 729 size_t msg_size, int msg_type) 730 { 731 enum xp_retval ret; 732 733 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type); 734 if (unlikely(ret != xpSuccess)) 735 XPC_DEACTIVATE_PARTITION(part, ret); 736 } 737 738 static void 739 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags, 740 void *msg, size_t msg_size, int msg_type) 741 { 742 struct xpc_partition *part = &xpc_partitions[ch->partid]; 743 enum xp_retval ret; 744 745 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type); 746 if (unlikely(ret != xpSuccess)) { 747 if (irq_flags != NULL) 748 spin_unlock_irqrestore(&ch->lock, *irq_flags); 749 750 XPC_DEACTIVATE_PARTITION(part, ret); 751 752 if (irq_flags != NULL) 753 spin_lock_irqsave(&ch->lock, *irq_flags); 754 } 755 } 756 757 static void 758 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req) 759 { 760 unsigned long irq_flags; 761 struct xpc_partition_uv *part_uv = &part->sn.uv; 762 763 /* 764 * !!! Make our side think that the remote partition sent an activate 765 * !!! mq message our way by doing what the activate IRQ handler would 766 * !!! do had one really been sent. 767 */ 768 769 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 770 if (part_uv->act_state_req == 0) 771 xpc_activate_IRQ_rcvd++; 772 part_uv->act_state_req = act_state_req; 773 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 774 775 wake_up_interruptible(&xpc_activate_IRQ_wq); 776 } 777 778 static enum xp_retval 779 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa, 780 size_t *len) 781 { 782 s64 status; 783 enum xp_retval ret; 784 785 #if defined CONFIG_X86_64 786 status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa, 787 (u64 *)len); 788 if (status == BIOS_STATUS_SUCCESS) 789 ret = xpSuccess; 790 else if (status == BIOS_STATUS_MORE_PASSES) 791 ret = xpNeedMoreInfo; 792 else 793 ret = xpBiosError; 794 795 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV 796 status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len); 797 if (status == SALRET_OK) 798 ret = xpSuccess; 799 else if (status == SALRET_MORE_PASSES) 800 ret = xpNeedMoreInfo; 801 else 802 ret = xpSalError; 803 804 #else 805 #error not a supported configuration 806 #endif 807 808 return ret; 809 } 810 811 static int 812 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp) 813 { 814 xpc_heartbeat_uv = 815 &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat; 816 rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv); 817 rp->sn.uv.activate_gru_mq_desc_gpa = 818 uv_gpa(xpc_activate_mq_uv->gru_mq_desc); 819 return 0; 820 } 821 822 static void 823 xpc_allow_hb_uv(short partid) 824 { 825 } 826 827 static void 828 xpc_disallow_hb_uv(short partid) 829 { 830 } 831 832 static void 833 xpc_disallow_all_hbs_uv(void) 834 { 835 } 836 837 static void 838 xpc_increment_heartbeat_uv(void) 839 { 840 xpc_heartbeat_uv->value++; 841 } 842 843 static void 844 xpc_offline_heartbeat_uv(void) 845 { 846 xpc_increment_heartbeat_uv(); 847 xpc_heartbeat_uv->offline = 1; 848 } 849 850 static void 851 xpc_online_heartbeat_uv(void) 852 { 853 xpc_increment_heartbeat_uv(); 854 xpc_heartbeat_uv->offline = 0; 855 } 856 857 static void 858 xpc_heartbeat_init_uv(void) 859 { 860 xpc_heartbeat_uv->value = 1; 861 xpc_heartbeat_uv->offline = 0; 862 } 863 864 static void 865 xpc_heartbeat_exit_uv(void) 866 { 867 xpc_offline_heartbeat_uv(); 868 } 869 870 static enum xp_retval 871 xpc_get_remote_heartbeat_uv(struct xpc_partition *part) 872 { 873 struct xpc_partition_uv *part_uv = &part->sn.uv; 874 enum xp_retval ret; 875 876 ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat), 877 part_uv->heartbeat_gpa, 878 sizeof(struct xpc_heartbeat_uv)); 879 if (ret != xpSuccess) 880 return ret; 881 882 if (part_uv->cached_heartbeat.value == part->last_heartbeat && 883 !part_uv->cached_heartbeat.offline) { 884 885 ret = xpNoHeartbeat; 886 } else { 887 part->last_heartbeat = part_uv->cached_heartbeat.value; 888 } 889 return ret; 890 } 891 892 static void 893 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp, 894 unsigned long remote_rp_gpa, int nasid) 895 { 896 short partid = remote_rp->SAL_partid; 897 struct xpc_partition *part = &xpc_partitions[partid]; 898 struct xpc_activate_mq_msg_activate_req_uv msg; 899 900 part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */ 901 part->remote_rp_ts_jiffies = remote_rp->ts_jiffies; 902 part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa; 903 part->sn.uv.activate_gru_mq_desc_gpa = 904 remote_rp->sn.uv.activate_gru_mq_desc_gpa; 905 906 /* 907 * ??? Is it a good idea to make this conditional on what is 908 * ??? potentially stale state information? 909 */ 910 if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) { 911 msg.rp_gpa = uv_gpa(xpc_rsvd_page); 912 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa; 913 msg.activate_gru_mq_desc_gpa = 914 xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa; 915 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg), 916 XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV); 917 } 918 919 if (part->act_state == XPC_P_AS_INACTIVE) 920 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV); 921 } 922 923 static void 924 xpc_request_partition_reactivation_uv(struct xpc_partition *part) 925 { 926 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV); 927 } 928 929 static void 930 xpc_request_partition_deactivation_uv(struct xpc_partition *part) 931 { 932 struct xpc_activate_mq_msg_deactivate_req_uv msg; 933 934 /* 935 * ??? Is it a good idea to make this conditional on what is 936 * ??? potentially stale state information? 937 */ 938 if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING && 939 part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) { 940 941 msg.reason = part->reason; 942 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg), 943 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV); 944 } 945 } 946 947 static void 948 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part) 949 { 950 /* nothing needs to be done */ 951 return; 952 } 953 954 static void 955 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head) 956 { 957 head->first = NULL; 958 head->last = NULL; 959 spin_lock_init(&head->lock); 960 head->n_entries = 0; 961 } 962 963 static void * 964 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head) 965 { 966 unsigned long irq_flags; 967 struct xpc_fifo_entry_uv *first; 968 969 spin_lock_irqsave(&head->lock, irq_flags); 970 first = head->first; 971 if (head->first != NULL) { 972 head->first = first->next; 973 if (head->first == NULL) 974 head->last = NULL; 975 976 head->n_entries--; 977 BUG_ON(head->n_entries < 0); 978 979 first->next = NULL; 980 } 981 spin_unlock_irqrestore(&head->lock, irq_flags); 982 return first; 983 } 984 985 static void 986 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head, 987 struct xpc_fifo_entry_uv *last) 988 { 989 unsigned long irq_flags; 990 991 last->next = NULL; 992 spin_lock_irqsave(&head->lock, irq_flags); 993 if (head->last != NULL) 994 head->last->next = last; 995 else 996 head->first = last; 997 head->last = last; 998 head->n_entries++; 999 spin_unlock_irqrestore(&head->lock, irq_flags); 1000 } 1001 1002 static int 1003 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head) 1004 { 1005 return head->n_entries; 1006 } 1007 1008 /* 1009 * Setup the channel structures that are uv specific. 1010 */ 1011 static enum xp_retval 1012 xpc_setup_ch_structures_uv(struct xpc_partition *part) 1013 { 1014 struct xpc_channel_uv *ch_uv; 1015 int ch_number; 1016 1017 for (ch_number = 0; ch_number < part->nchannels; ch_number++) { 1018 ch_uv = &part->channels[ch_number].sn.uv; 1019 1020 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list); 1021 xpc_init_fifo_uv(&ch_uv->recv_msg_list); 1022 } 1023 1024 return xpSuccess; 1025 } 1026 1027 /* 1028 * Teardown the channel structures that are uv specific. 1029 */ 1030 static void 1031 xpc_teardown_ch_structures_uv(struct xpc_partition *part) 1032 { 1033 /* nothing needs to be done */ 1034 return; 1035 } 1036 1037 static enum xp_retval 1038 xpc_make_first_contact_uv(struct xpc_partition *part) 1039 { 1040 struct xpc_activate_mq_msg_uv msg; 1041 1042 /* 1043 * We send a sync msg to get the remote partition's remote_act_state 1044 * updated to our current act_state which at this point should 1045 * be XPC_P_AS_ACTIVATING. 1046 */ 1047 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg), 1048 XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV); 1049 1050 while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) || 1051 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) { 1052 1053 dev_dbg(xpc_part, "waiting to make first contact with " 1054 "partition %d\n", XPC_PARTID(part)); 1055 1056 /* wait a 1/4 of a second or so */ 1057 (void)msleep_interruptible(250); 1058 1059 if (part->act_state == XPC_P_AS_DEACTIVATING) 1060 return part->reason; 1061 } 1062 1063 return xpSuccess; 1064 } 1065 1066 static u64 1067 xpc_get_chctl_all_flags_uv(struct xpc_partition *part) 1068 { 1069 unsigned long irq_flags; 1070 union xpc_channel_ctl_flags chctl; 1071 1072 spin_lock_irqsave(&part->chctl_lock, irq_flags); 1073 chctl = part->chctl; 1074 if (chctl.all_flags != 0) 1075 part->chctl.all_flags = 0; 1076 1077 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 1078 return chctl.all_flags; 1079 } 1080 1081 static enum xp_retval 1082 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch) 1083 { 1084 struct xpc_channel_uv *ch_uv = &ch->sn.uv; 1085 struct xpc_send_msg_slot_uv *msg_slot; 1086 unsigned long irq_flags; 1087 int nentries; 1088 int entry; 1089 size_t nbytes; 1090 1091 for (nentries = ch->local_nentries; nentries > 0; nentries--) { 1092 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv); 1093 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL); 1094 if (ch_uv->send_msg_slots == NULL) 1095 continue; 1096 1097 for (entry = 0; entry < nentries; entry++) { 1098 msg_slot = &ch_uv->send_msg_slots[entry]; 1099 1100 msg_slot->msg_slot_number = entry; 1101 xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list, 1102 &msg_slot->next); 1103 } 1104 1105 spin_lock_irqsave(&ch->lock, irq_flags); 1106 if (nentries < ch->local_nentries) 1107 ch->local_nentries = nentries; 1108 spin_unlock_irqrestore(&ch->lock, irq_flags); 1109 return xpSuccess; 1110 } 1111 1112 return xpNoMemory; 1113 } 1114 1115 static enum xp_retval 1116 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch) 1117 { 1118 struct xpc_channel_uv *ch_uv = &ch->sn.uv; 1119 struct xpc_notify_mq_msg_uv *msg_slot; 1120 unsigned long irq_flags; 1121 int nentries; 1122 int entry; 1123 size_t nbytes; 1124 1125 for (nentries = ch->remote_nentries; nentries > 0; nentries--) { 1126 nbytes = nentries * ch->entry_size; 1127 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL); 1128 if (ch_uv->recv_msg_slots == NULL) 1129 continue; 1130 1131 for (entry = 0; entry < nentries; entry++) { 1132 msg_slot = ch_uv->recv_msg_slots + 1133 entry * ch->entry_size; 1134 1135 msg_slot->hdr.msg_slot_number = entry; 1136 } 1137 1138 spin_lock_irqsave(&ch->lock, irq_flags); 1139 if (nentries < ch->remote_nentries) 1140 ch->remote_nentries = nentries; 1141 spin_unlock_irqrestore(&ch->lock, irq_flags); 1142 return xpSuccess; 1143 } 1144 1145 return xpNoMemory; 1146 } 1147 1148 /* 1149 * Allocate msg_slots associated with the channel. 1150 */ 1151 static enum xp_retval 1152 xpc_setup_msg_structures_uv(struct xpc_channel *ch) 1153 { 1154 static enum xp_retval ret; 1155 struct xpc_channel_uv *ch_uv = &ch->sn.uv; 1156 1157 DBUG_ON(ch->flags & XPC_C_SETUP); 1158 1159 ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct 1160 gru_message_queue_desc), 1161 GFP_KERNEL); 1162 if (ch_uv->cached_notify_gru_mq_desc == NULL) 1163 return xpNoMemory; 1164 1165 ret = xpc_allocate_send_msg_slot_uv(ch); 1166 if (ret == xpSuccess) { 1167 1168 ret = xpc_allocate_recv_msg_slot_uv(ch); 1169 if (ret != xpSuccess) { 1170 kfree(ch_uv->send_msg_slots); 1171 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list); 1172 } 1173 } 1174 return ret; 1175 } 1176 1177 /* 1178 * Free up msg_slots and clear other stuff that were setup for the specified 1179 * channel. 1180 */ 1181 static void 1182 xpc_teardown_msg_structures_uv(struct xpc_channel *ch) 1183 { 1184 struct xpc_channel_uv *ch_uv = &ch->sn.uv; 1185 1186 DBUG_ON(!spin_is_locked(&ch->lock)); 1187 1188 kfree(ch_uv->cached_notify_gru_mq_desc); 1189 ch_uv->cached_notify_gru_mq_desc = NULL; 1190 1191 if (ch->flags & XPC_C_SETUP) { 1192 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list); 1193 kfree(ch_uv->send_msg_slots); 1194 xpc_init_fifo_uv(&ch_uv->recv_msg_list); 1195 kfree(ch_uv->recv_msg_slots); 1196 } 1197 } 1198 1199 static void 1200 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags) 1201 { 1202 struct xpc_activate_mq_msg_chctl_closerequest_uv msg; 1203 1204 msg.ch_number = ch->number; 1205 msg.reason = ch->reason; 1206 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg), 1207 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV); 1208 } 1209 1210 static void 1211 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags) 1212 { 1213 struct xpc_activate_mq_msg_chctl_closereply_uv msg; 1214 1215 msg.ch_number = ch->number; 1216 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg), 1217 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV); 1218 } 1219 1220 static void 1221 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags) 1222 { 1223 struct xpc_activate_mq_msg_chctl_openrequest_uv msg; 1224 1225 msg.ch_number = ch->number; 1226 msg.entry_size = ch->entry_size; 1227 msg.local_nentries = ch->local_nentries; 1228 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg), 1229 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV); 1230 } 1231 1232 static void 1233 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags) 1234 { 1235 struct xpc_activate_mq_msg_chctl_openreply_uv msg; 1236 1237 msg.ch_number = ch->number; 1238 msg.local_nentries = ch->local_nentries; 1239 msg.remote_nentries = ch->remote_nentries; 1240 msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc); 1241 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg), 1242 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV); 1243 } 1244 1245 static void 1246 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags) 1247 { 1248 struct xpc_activate_mq_msg_chctl_opencomplete_uv msg; 1249 1250 msg.ch_number = ch->number; 1251 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg), 1252 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV); 1253 } 1254 1255 static void 1256 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number) 1257 { 1258 unsigned long irq_flags; 1259 1260 spin_lock_irqsave(&part->chctl_lock, irq_flags); 1261 part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST; 1262 spin_unlock_irqrestore(&part->chctl_lock, irq_flags); 1263 1264 xpc_wakeup_channel_mgr(part); 1265 } 1266 1267 static enum xp_retval 1268 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch, 1269 unsigned long gru_mq_desc_gpa) 1270 { 1271 struct xpc_channel_uv *ch_uv = &ch->sn.uv; 1272 1273 DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL); 1274 return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc, 1275 gru_mq_desc_gpa); 1276 } 1277 1278 static void 1279 xpc_indicate_partition_engaged_uv(struct xpc_partition *part) 1280 { 1281 struct xpc_activate_mq_msg_uv msg; 1282 1283 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg), 1284 XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV); 1285 } 1286 1287 static void 1288 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part) 1289 { 1290 struct xpc_activate_mq_msg_uv msg; 1291 1292 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg), 1293 XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV); 1294 } 1295 1296 static void 1297 xpc_assume_partition_disengaged_uv(short partid) 1298 { 1299 struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv; 1300 unsigned long irq_flags; 1301 1302 spin_lock_irqsave(&part_uv->flags_lock, irq_flags); 1303 part_uv->flags &= ~XPC_P_ENGAGED_UV; 1304 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags); 1305 } 1306 1307 static int 1308 xpc_partition_engaged_uv(short partid) 1309 { 1310 return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0; 1311 } 1312 1313 static int 1314 xpc_any_partition_engaged_uv(void) 1315 { 1316 struct xpc_partition_uv *part_uv; 1317 short partid; 1318 1319 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) { 1320 part_uv = &xpc_partitions[partid].sn.uv; 1321 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0) 1322 return 1; 1323 } 1324 return 0; 1325 } 1326 1327 static enum xp_retval 1328 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags, 1329 struct xpc_send_msg_slot_uv **address_of_msg_slot) 1330 { 1331 enum xp_retval ret; 1332 struct xpc_send_msg_slot_uv *msg_slot; 1333 struct xpc_fifo_entry_uv *entry; 1334 1335 while (1) { 1336 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list); 1337 if (entry != NULL) 1338 break; 1339 1340 if (flags & XPC_NOWAIT) 1341 return xpNoWait; 1342 1343 ret = xpc_allocate_msg_wait(ch); 1344 if (ret != xpInterrupted && ret != xpTimeout) 1345 return ret; 1346 } 1347 1348 msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next); 1349 *address_of_msg_slot = msg_slot; 1350 return xpSuccess; 1351 } 1352 1353 static void 1354 xpc_free_msg_slot_uv(struct xpc_channel *ch, 1355 struct xpc_send_msg_slot_uv *msg_slot) 1356 { 1357 xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next); 1358 1359 /* wakeup anyone waiting for a free msg slot */ 1360 if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) 1361 wake_up(&ch->msg_allocate_wq); 1362 } 1363 1364 static void 1365 xpc_notify_sender_uv(struct xpc_channel *ch, 1366 struct xpc_send_msg_slot_uv *msg_slot, 1367 enum xp_retval reason) 1368 { 1369 xpc_notify_func func = msg_slot->func; 1370 1371 if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) { 1372 1373 atomic_dec(&ch->n_to_notify); 1374 1375 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p " 1376 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot, 1377 msg_slot->msg_slot_number, ch->partid, ch->number); 1378 1379 func(reason, ch->partid, ch->number, msg_slot->key); 1380 1381 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p " 1382 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot, 1383 msg_slot->msg_slot_number, ch->partid, ch->number); 1384 } 1385 } 1386 1387 static void 1388 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch, 1389 struct xpc_notify_mq_msg_uv *msg) 1390 { 1391 struct xpc_send_msg_slot_uv *msg_slot; 1392 int entry = msg->hdr.msg_slot_number % ch->local_nentries; 1393 1394 msg_slot = &ch->sn.uv.send_msg_slots[entry]; 1395 1396 BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number); 1397 msg_slot->msg_slot_number += ch->local_nentries; 1398 1399 if (msg_slot->func != NULL) 1400 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered); 1401 1402 xpc_free_msg_slot_uv(ch, msg_slot); 1403 } 1404 1405 static void 1406 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part, 1407 struct xpc_notify_mq_msg_uv *msg) 1408 { 1409 struct xpc_partition_uv *part_uv = &part->sn.uv; 1410 struct xpc_channel *ch; 1411 struct xpc_channel_uv *ch_uv; 1412 struct xpc_notify_mq_msg_uv *msg_slot; 1413 unsigned long irq_flags; 1414 int ch_number = msg->hdr.ch_number; 1415 1416 if (unlikely(ch_number >= part->nchannels)) { 1417 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid " 1418 "channel number=0x%x in message from partid=%d\n", 1419 ch_number, XPC_PARTID(part)); 1420 1421 /* get hb checker to deactivate from the remote partition */ 1422 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags); 1423 if (part_uv->act_state_req == 0) 1424 xpc_activate_IRQ_rcvd++; 1425 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV; 1426 part_uv->reason = xpBadChannelNumber; 1427 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags); 1428 1429 wake_up_interruptible(&xpc_activate_IRQ_wq); 1430 return; 1431 } 1432 1433 ch = &part->channels[ch_number]; 1434 xpc_msgqueue_ref(ch); 1435 1436 if (!(ch->flags & XPC_C_CONNECTED)) { 1437 xpc_msgqueue_deref(ch); 1438 return; 1439 } 1440 1441 /* see if we're really dealing with an ACK for a previously sent msg */ 1442 if (msg->hdr.size == 0) { 1443 xpc_handle_notify_mq_ack_uv(ch, msg); 1444 xpc_msgqueue_deref(ch); 1445 return; 1446 } 1447 1448 /* we're dealing with a normal message sent via the notify_mq */ 1449 ch_uv = &ch->sn.uv; 1450 1451 msg_slot = ch_uv->recv_msg_slots + 1452 (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size; 1453 1454 BUG_ON(msg_slot->hdr.size != 0); 1455 1456 memcpy(msg_slot, msg, msg->hdr.size); 1457 1458 xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next); 1459 1460 if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) { 1461 /* 1462 * If there is an existing idle kthread get it to deliver 1463 * the payload, otherwise we'll have to get the channel mgr 1464 * for this partition to create a kthread to do the delivery. 1465 */ 1466 if (atomic_read(&ch->kthreads_idle) > 0) 1467 wake_up_nr(&ch->idle_wq, 1); 1468 else 1469 xpc_send_chctl_local_msgrequest_uv(part, ch->number); 1470 } 1471 xpc_msgqueue_deref(ch); 1472 } 1473 1474 static irqreturn_t 1475 xpc_handle_notify_IRQ_uv(int irq, void *dev_id) 1476 { 1477 struct xpc_notify_mq_msg_uv *msg; 1478 short partid; 1479 struct xpc_partition *part; 1480 1481 while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) != 1482 NULL) { 1483 1484 partid = msg->hdr.partid; 1485 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) { 1486 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received " 1487 "invalid partid=0x%x in message\n", partid); 1488 } else { 1489 part = &xpc_partitions[partid]; 1490 1491 if (xpc_part_ref(part)) { 1492 xpc_handle_notify_mq_msg_uv(part, msg); 1493 xpc_part_deref(part); 1494 } 1495 } 1496 1497 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg); 1498 } 1499 1500 return IRQ_HANDLED; 1501 } 1502 1503 static int 1504 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch) 1505 { 1506 return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list); 1507 } 1508 1509 static void 1510 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number) 1511 { 1512 struct xpc_channel *ch = &part->channels[ch_number]; 1513 int ndeliverable_payloads; 1514 1515 xpc_msgqueue_ref(ch); 1516 1517 ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch); 1518 1519 if (ndeliverable_payloads > 0 && 1520 (ch->flags & XPC_C_CONNECTED) && 1521 (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) { 1522 1523 xpc_activate_kthreads(ch, ndeliverable_payloads); 1524 } 1525 1526 xpc_msgqueue_deref(ch); 1527 } 1528 1529 static enum xp_retval 1530 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload, 1531 u16 payload_size, u8 notify_type, xpc_notify_func func, 1532 void *key) 1533 { 1534 enum xp_retval ret = xpSuccess; 1535 struct xpc_send_msg_slot_uv *msg_slot = NULL; 1536 struct xpc_notify_mq_msg_uv *msg; 1537 u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV]; 1538 size_t msg_size; 1539 1540 DBUG_ON(notify_type != XPC_N_CALL); 1541 1542 msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size; 1543 if (msg_size > ch->entry_size) 1544 return xpPayloadTooBig; 1545 1546 xpc_msgqueue_ref(ch); 1547 1548 if (ch->flags & XPC_C_DISCONNECTING) { 1549 ret = ch->reason; 1550 goto out_1; 1551 } 1552 if (!(ch->flags & XPC_C_CONNECTED)) { 1553 ret = xpNotConnected; 1554 goto out_1; 1555 } 1556 1557 ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot); 1558 if (ret != xpSuccess) 1559 goto out_1; 1560 1561 if (func != NULL) { 1562 atomic_inc(&ch->n_to_notify); 1563 1564 msg_slot->key = key; 1565 smp_wmb(); /* a non-NULL func must hit memory after the key */ 1566 msg_slot->func = func; 1567 1568 if (ch->flags & XPC_C_DISCONNECTING) { 1569 ret = ch->reason; 1570 goto out_2; 1571 } 1572 } 1573 1574 msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer; 1575 msg->hdr.partid = xp_partition_id; 1576 msg->hdr.ch_number = ch->number; 1577 msg->hdr.size = msg_size; 1578 msg->hdr.msg_slot_number = msg_slot->msg_slot_number; 1579 memcpy(&msg->payload, payload, payload_size); 1580 1581 ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg, 1582 msg_size); 1583 if (ret == xpSuccess) 1584 goto out_1; 1585 1586 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret); 1587 out_2: 1588 if (func != NULL) { 1589 /* 1590 * Try to NULL the msg_slot's func field. If we fail, then 1591 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which 1592 * case we need to pretend we succeeded to send the message 1593 * since the user will get a callout for the disconnect error 1594 * by xpc_notify_senders_of_disconnect_uv(), and to also get an 1595 * error returned here will confuse them. Additionally, since 1596 * in this case the channel is being disconnected we don't need 1597 * to put the the msg_slot back on the free list. 1598 */ 1599 if (cmpxchg(&msg_slot->func, func, NULL) != func) { 1600 ret = xpSuccess; 1601 goto out_1; 1602 } 1603 1604 msg_slot->key = NULL; 1605 atomic_dec(&ch->n_to_notify); 1606 } 1607 xpc_free_msg_slot_uv(ch, msg_slot); 1608 out_1: 1609 xpc_msgqueue_deref(ch); 1610 return ret; 1611 } 1612 1613 /* 1614 * Tell the callers of xpc_send_notify() that the status of their payloads 1615 * is unknown because the channel is now disconnecting. 1616 * 1617 * We don't worry about putting these msg_slots on the free list since the 1618 * msg_slots themselves are about to be kfree'd. 1619 */ 1620 static void 1621 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch) 1622 { 1623 struct xpc_send_msg_slot_uv *msg_slot; 1624 int entry; 1625 1626 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING)); 1627 1628 for (entry = 0; entry < ch->local_nentries; entry++) { 1629 1630 if (atomic_read(&ch->n_to_notify) == 0) 1631 break; 1632 1633 msg_slot = &ch->sn.uv.send_msg_slots[entry]; 1634 if (msg_slot->func != NULL) 1635 xpc_notify_sender_uv(ch, msg_slot, ch->reason); 1636 } 1637 } 1638 1639 /* 1640 * Get the next deliverable message's payload. 1641 */ 1642 static void * 1643 xpc_get_deliverable_payload_uv(struct xpc_channel *ch) 1644 { 1645 struct xpc_fifo_entry_uv *entry; 1646 struct xpc_notify_mq_msg_uv *msg; 1647 void *payload = NULL; 1648 1649 if (!(ch->flags & XPC_C_DISCONNECTING)) { 1650 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list); 1651 if (entry != NULL) { 1652 msg = container_of(entry, struct xpc_notify_mq_msg_uv, 1653 hdr.u.next); 1654 payload = &msg->payload; 1655 } 1656 } 1657 return payload; 1658 } 1659 1660 static void 1661 xpc_received_payload_uv(struct xpc_channel *ch, void *payload) 1662 { 1663 struct xpc_notify_mq_msg_uv *msg; 1664 enum xp_retval ret; 1665 1666 msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload); 1667 1668 /* return an ACK to the sender of this message */ 1669 1670 msg->hdr.partid = xp_partition_id; 1671 msg->hdr.size = 0; /* size of zero indicates this is an ACK */ 1672 1673 ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg, 1674 sizeof(struct xpc_notify_mq_msghdr_uv)); 1675 if (ret != xpSuccess) 1676 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret); 1677 } 1678 1679 static struct xpc_arch_operations xpc_arch_ops_uv = { 1680 .setup_partitions = xpc_setup_partitions_uv, 1681 .teardown_partitions = xpc_teardown_partitions_uv, 1682 .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv, 1683 .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv, 1684 .setup_rsvd_page = xpc_setup_rsvd_page_uv, 1685 1686 .allow_hb = xpc_allow_hb_uv, 1687 .disallow_hb = xpc_disallow_hb_uv, 1688 .disallow_all_hbs = xpc_disallow_all_hbs_uv, 1689 .increment_heartbeat = xpc_increment_heartbeat_uv, 1690 .offline_heartbeat = xpc_offline_heartbeat_uv, 1691 .online_heartbeat = xpc_online_heartbeat_uv, 1692 .heartbeat_init = xpc_heartbeat_init_uv, 1693 .heartbeat_exit = xpc_heartbeat_exit_uv, 1694 .get_remote_heartbeat = xpc_get_remote_heartbeat_uv, 1695 1696 .request_partition_activation = 1697 xpc_request_partition_activation_uv, 1698 .request_partition_reactivation = 1699 xpc_request_partition_reactivation_uv, 1700 .request_partition_deactivation = 1701 xpc_request_partition_deactivation_uv, 1702 .cancel_partition_deactivation_request = 1703 xpc_cancel_partition_deactivation_request_uv, 1704 1705 .setup_ch_structures = xpc_setup_ch_structures_uv, 1706 .teardown_ch_structures = xpc_teardown_ch_structures_uv, 1707 1708 .make_first_contact = xpc_make_first_contact_uv, 1709 1710 .get_chctl_all_flags = xpc_get_chctl_all_flags_uv, 1711 .send_chctl_closerequest = xpc_send_chctl_closerequest_uv, 1712 .send_chctl_closereply = xpc_send_chctl_closereply_uv, 1713 .send_chctl_openrequest = xpc_send_chctl_openrequest_uv, 1714 .send_chctl_openreply = xpc_send_chctl_openreply_uv, 1715 .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv, 1716 .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv, 1717 1718 .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv, 1719 1720 .setup_msg_structures = xpc_setup_msg_structures_uv, 1721 .teardown_msg_structures = xpc_teardown_msg_structures_uv, 1722 1723 .indicate_partition_engaged = xpc_indicate_partition_engaged_uv, 1724 .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv, 1725 .assume_partition_disengaged = xpc_assume_partition_disengaged_uv, 1726 .partition_engaged = xpc_partition_engaged_uv, 1727 .any_partition_engaged = xpc_any_partition_engaged_uv, 1728 1729 .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv, 1730 .send_payload = xpc_send_payload_uv, 1731 .get_deliverable_payload = xpc_get_deliverable_payload_uv, 1732 .received_payload = xpc_received_payload_uv, 1733 .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv, 1734 }; 1735 1736 static int 1737 xpc_init_mq_node(int nid) 1738 { 1739 int cpu; 1740 1741 get_online_cpus(); 1742 1743 for_each_cpu(cpu, cpumask_of_node(nid)) { 1744 xpc_activate_mq_uv = 1745 xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid, 1746 XPC_ACTIVATE_IRQ_NAME, 1747 xpc_handle_activate_IRQ_uv); 1748 if (!IS_ERR(xpc_activate_mq_uv)) 1749 break; 1750 } 1751 if (IS_ERR(xpc_activate_mq_uv)) { 1752 put_online_cpus(); 1753 return PTR_ERR(xpc_activate_mq_uv); 1754 } 1755 1756 for_each_cpu(cpu, cpumask_of_node(nid)) { 1757 xpc_notify_mq_uv = 1758 xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid, 1759 XPC_NOTIFY_IRQ_NAME, 1760 xpc_handle_notify_IRQ_uv); 1761 if (!IS_ERR(xpc_notify_mq_uv)) 1762 break; 1763 } 1764 if (IS_ERR(xpc_notify_mq_uv)) { 1765 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv); 1766 put_online_cpus(); 1767 return PTR_ERR(xpc_notify_mq_uv); 1768 } 1769 1770 put_online_cpus(); 1771 return 0; 1772 } 1773 1774 int 1775 xpc_init_uv(void) 1776 { 1777 int nid; 1778 int ret = 0; 1779 1780 xpc_arch_ops = xpc_arch_ops_uv; 1781 1782 if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) { 1783 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n", 1784 XPC_MSG_HDR_MAX_SIZE); 1785 return -E2BIG; 1786 } 1787 1788 if (xpc_mq_node < 0) 1789 for_each_online_node(nid) { 1790 ret = xpc_init_mq_node(nid); 1791 1792 if (!ret) 1793 break; 1794 } 1795 else 1796 ret = xpc_init_mq_node(xpc_mq_node); 1797 1798 if (ret < 0) 1799 dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n", 1800 -ret); 1801 1802 return ret; 1803 } 1804 1805 void 1806 xpc_exit_uv(void) 1807 { 1808 xpc_destroy_gru_mq_uv(xpc_notify_mq_uv); 1809 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv); 1810 } 1811 1812 module_param(xpc_mq_node, int, 0); 1813 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues."); 1814