git.oblomov.eu Git - linux-2.6/blob - drivers/infiniband/hw/ehca/ehca_irq.c

   1 /*
   2  *  IBM eServer eHCA Infiniband device driver for Linux on POWER
   3  *
   4  *  Functions for EQs, NEQs and interrupts
   5  *
   6  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
   7  *           Khadija Souissi <souissi@de.ibm.com>
   8  *
   9  *  Copyright (c) 2005 IBM Corporation
  10  *
  11  *  All rights reserved.
  12  *
  13  *  This source code is distributed under a dual license of GPL v2.0 and OpenIB
  14  *  BSD.
  15  *
  16  * OpenIB BSD License
  17  *
  18  * Redistribution and use in source and binary forms, with or without
  19  * modification, are permitted provided that the following conditions are met:
  20  *
  21  * Redistributions of source code must retain the above copyright notice, this
  22  * list of conditions and the following disclaimer.
  23  *
  24  * Redistributions in binary form must reproduce the above copyright notice,
  25  * this list of conditions and the following disclaimer in the documentation
  26  * and/or other materials
  27  * provided with the distribution.
  28  *
  29  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  30  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  32  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  33  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  36  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
  37  * IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  39  * POSSIBILITY OF SUCH DAMAGE.
  40  */
  41
  42 #include "ehca_classes.h"
  43 #include "ehca_irq.h"
  44 #include "ehca_iverbs.h"
  45 #include "ehca_tools.h"
  46 #include "hcp_if.h"
  47 #include "hipz_fns.h"
  48 #include "ipz_pt_fn.h"
  49
  50 #define EQE_COMPLETION_EVENT   EHCA_BMASK_IBM(1,1)
  51 #define EQE_CQ_QP_NUMBER       EHCA_BMASK_IBM(8,31)
  52 #define EQE_EE_IDENTIFIER      EHCA_BMASK_IBM(2,7)
  53 #define EQE_CQ_NUMBER          EHCA_BMASK_IBM(8,31)
  54 #define EQE_QP_NUMBER          EHCA_BMASK_IBM(8,31)
  55 #define EQE_QP_TOKEN           EHCA_BMASK_IBM(32,63)
  56 #define EQE_CQ_TOKEN           EHCA_BMASK_IBM(32,63)
  57
  58 #define NEQE_COMPLETION_EVENT  EHCA_BMASK_IBM(1,1)
  59 #define NEQE_EVENT_CODE        EHCA_BMASK_IBM(2,7)
  60 #define NEQE_PORT_NUMBER       EHCA_BMASK_IBM(8,15)
  61 #define NEQE_PORT_AVAILABILITY EHCA_BMASK_IBM(16,16)
  62
  63 #define ERROR_DATA_LENGTH      EHCA_BMASK_IBM(52,63)
  64 #define ERROR_DATA_TYPE        EHCA_BMASK_IBM(0,7)
  65
  66 static void queue_comp_task(struct ehca_cq *__cq);
  67
  68 static struct ehca_comp_pool* pool;
  69 #ifdef CONFIG_HOTPLUG_CPU
  70 static struct notifier_block comp_pool_callback_nb;
  71 #endif
  72
  73 static inline void comp_event_callback(struct ehca_cq *cq)
  74 {
  75         if (!cq->ib_cq.comp_handler)
  76                 return;
  77
  78         spin_lock(&cq->cb_lock);
  79         cq->ib_cq.comp_handler(&cq->ib_cq, cq->ib_cq.cq_context);
  80         spin_unlock(&cq->cb_lock);
  81
  82         return;
  83 }
  84
  85 static void print_error_data(struct ehca_shca * shca, void* data,
  86                              u64* rblock, int length)
  87 {
  88         u64 type = EHCA_BMASK_GET(ERROR_DATA_TYPE, rblock[2]);
  89         u64 resource = rblock[1];
  90
  91         switch (type) {
  92         case 0x1: /* Queue Pair */
  93         {
  94                 struct ehca_qp *qp = (struct ehca_qp*)data;
  95
  96                 /* only print error data if AER is set */
  97                 if (rblock[6] == 0)
  98                         return;
  99
 100                 ehca_err(&shca->ib_device,
 101                          "QP 0x%x (resource=%lx) has errors.",
 102                          qp->ib_qp.qp_num, resource);
 103                 break;
 104         }
 105         case 0x4: /* Completion Queue */
 106         {
 107                 struct ehca_cq *cq = (struct ehca_cq*)data;
 108
 109                 ehca_err(&shca->ib_device,
 110                          "CQ 0x%x (resource=%lx) has errors.",
 111                          cq->cq_number, resource);
 112                 break;
 113         }
 114         default:
 115                 ehca_err(&shca->ib_device,
 116                          "Unknown errror type: %lx on %s.",
 117                          type, shca->ib_device.name);
 118                 break;
 119         }
 120
 121         ehca_err(&shca->ib_device, "Error data is available: %lx.", resource);
 122         ehca_err(&shca->ib_device, "EHCA ----- error data begin "
 123                  "---------------------------------------------------");
 124         ehca_dmp(rblock, length, "resource=%lx", resource);
 125         ehca_err(&shca->ib_device, "EHCA ----- error data end "
 126                  "----------------------------------------------------");
 127
 128         return;
 129 }
 130
 131 int ehca_error_data(struct ehca_shca *shca, void *data,
 132                     u64 resource)
 133 {
 134
 135         unsigned long ret;
 136         u64 *rblock;
 137         unsigned long block_count;
 138
 139         rblock = ehca_alloc_fw_ctrlblock(GFP_ATOMIC);
 140         if (!rblock) {
 141                 ehca_err(&shca->ib_device, "Cannot allocate rblock memory.");
 142                 ret = -ENOMEM;
 143                 goto error_data1;
 144         }
 145
 146         /* rblock must be 4K aligned and should be 4K large */
 147         ret = hipz_h_error_data(shca->ipz_hca_handle,
 148                                 resource,
 149                                 rblock,
 150                                 &block_count);
 151
 152         if (ret == H_R_STATE)
 153                 ehca_err(&shca->ib_device,
 154                          "No error data is available: %lx.", resource);
 155         else if (ret == H_SUCCESS) {
 156                 int length;
 157
 158                 length = EHCA_BMASK_GET(ERROR_DATA_LENGTH, rblock[0]);
 159
 160                 if (length > EHCA_PAGESIZE)
 161                         length = EHCA_PAGESIZE;
 162
 163                 print_error_data(shca, data, rblock, length);
 164         } else
 165                 ehca_err(&shca->ib_device,
 166                          "Error data could not be fetched: %lx", resource);
 167
 168         ehca_free_fw_ctrlblock(rblock);
 169
 170 error_data1:
 171         return ret;
 172
 173 }
 174
 175 static void qp_event_callback(struct ehca_shca *shca,
 176                               u64 eqe,
 177                               enum ib_event_type event_type)
 178 {
 179         struct ib_event event;
 180         struct ehca_qp *qp;
 181         unsigned long flags;
 182         u32 token = EHCA_BMASK_GET(EQE_QP_TOKEN, eqe);
 183
 184         spin_lock_irqsave(&ehca_qp_idr_lock, flags);
 185         qp = idr_find(&ehca_qp_idr, token);
 186         spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
 187
 188
 189         if (!qp)
 190                 return;
 191
 192         ehca_error_data(shca, qp, qp->ipz_qp_handle.handle);
 193
 194         if (!qp->ib_qp.event_handler)
 195                 return;
 196
 197         event.device     = &shca->ib_device;
 198         event.event      = event_type;
 199         event.element.qp = &qp->ib_qp;
 200
 201         qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
 202
 203         return;
 204 }
 205
 206 static void cq_event_callback(struct ehca_shca *shca,
 207                               u64 eqe)
 208 {
 209         struct ehca_cq *cq;
 210         unsigned long flags;
 211         u32 token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe);
 212
 213         spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 214         cq = idr_find(&ehca_cq_idr, token);
 215         spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 216
 217         if (!cq)
 218                 return;
 219
 220         ehca_error_data(shca, cq, cq->ipz_cq_handle.handle);
 221
 222         return;
 223 }
 224
 225 static void parse_identifier(struct ehca_shca *shca, u64 eqe)
 226 {
 227         u8 identifier = EHCA_BMASK_GET(EQE_EE_IDENTIFIER, eqe);
 228
 229         switch (identifier) {
 230         case 0x02: /* path migrated */
 231                 qp_event_callback(shca, eqe, IB_EVENT_PATH_MIG);
 232                 break;
 233         case 0x03: /* communication established */
 234                 qp_event_callback(shca, eqe, IB_EVENT_COMM_EST);
 235                 break;
 236         case 0x04: /* send queue drained */
 237                 qp_event_callback(shca, eqe, IB_EVENT_SQ_DRAINED);
 238                 break;
 239         case 0x05: /* QP error */
 240         case 0x06: /* QP error */
 241                 qp_event_callback(shca, eqe, IB_EVENT_QP_FATAL);
 242                 break;
 243         case 0x07: /* CQ error */
 244         case 0x08: /* CQ error */
 245                 cq_event_callback(shca, eqe);
 246                 break;
 247         case 0x09: /* MRMWPTE error */
 248                 ehca_err(&shca->ib_device, "MRMWPTE error.");
 249                 break;
 250         case 0x0A: /* port event */
 251                 ehca_err(&shca->ib_device, "Port event.");
 252                 break;
 253         case 0x0B: /* MR access error */
 254                 ehca_err(&shca->ib_device, "MR access error.");
 255                 break;
 256         case 0x0C: /* EQ error */
 257                 ehca_err(&shca->ib_device, "EQ error.");
 258                 break;
 259         case 0x0D: /* P/Q_Key mismatch */
 260                 ehca_err(&shca->ib_device, "P/Q_Key mismatch.");
 261                 break;
 262         case 0x10: /* sampling complete */
 263                 ehca_err(&shca->ib_device, "Sampling complete.");
 264                 break;
 265         case 0x11: /* unaffiliated access error */
 266                 ehca_err(&shca->ib_device, "Unaffiliated access error.");
 267                 break;
 268         case 0x12: /* path migrating error */
 269                 ehca_err(&shca->ib_device, "Path migration error.");
 270                 break;
 271         case 0x13: /* interface trace stopped */
 272                 ehca_err(&shca->ib_device, "Interface trace stopped.");
 273                 break;
 274         case 0x14: /* first error capture info available */
 275         default:
 276                 ehca_err(&shca->ib_device, "Unknown identifier: %x on %s.",
 277                          identifier, shca->ib_device.name);
 278                 break;
 279         }
 280
 281         return;
 282 }
 283
 284 static void parse_ec(struct ehca_shca *shca, u64 eqe)
 285 {
 286         struct ib_event event;
 287         u8 ec   = EHCA_BMASK_GET(NEQE_EVENT_CODE, eqe);
 288         u8 port = EHCA_BMASK_GET(NEQE_PORT_NUMBER, eqe);
 289
 290         switch (ec) {
 291         case 0x30: /* port availability change */
 292                 if (EHCA_BMASK_GET(NEQE_PORT_AVAILABILITY, eqe)) {
 293                         ehca_info(&shca->ib_device,
 294                                   "port %x is active.", port);
 295                         event.device = &shca->ib_device;
 296                         event.event = IB_EVENT_PORT_ACTIVE;
 297                         event.element.port_num = port;
 298                         shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
 299                         ib_dispatch_event(&event);
 300                 } else {
 301                         ehca_info(&shca->ib_device,
 302                                   "port %x is inactive.", port);
 303                         event.device = &shca->ib_device;
 304                         event.event = IB_EVENT_PORT_ERR;
 305                         event.element.port_num = port;
 306                         shca->sport[port - 1].port_state = IB_PORT_DOWN;
 307                         ib_dispatch_event(&event);
 308                 }
 309                 break;
 310         case 0x31:
 311                 /* port configuration change
 312                  * disruptive change is caused by
 313                  * LID, PKEY or SM change
 314                  */
 315                 ehca_warn(&shca->ib_device,
 316                           "disruptive port %x configuration change", port);
 317
 318                 ehca_info(&shca->ib_device,
 319                           "port %x is inactive.", port);
 320                 event.device = &shca->ib_device;
 321                 event.event = IB_EVENT_PORT_ERR;
 322                 event.element.port_num = port;
 323                 shca->sport[port - 1].port_state = IB_PORT_DOWN;
 324                 ib_dispatch_event(&event);
 325
 326                 ehca_info(&shca->ib_device,
 327                           "port %x is active.", port);
 328                 event.device = &shca->ib_device;
 329                 event.event = IB_EVENT_PORT_ACTIVE;
 330                 event.element.port_num = port;
 331                 shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
 332                 ib_dispatch_event(&event);
 333                 break;
 334         case 0x32: /* adapter malfunction */
 335                 ehca_err(&shca->ib_device, "Adapter malfunction.");
 336                 break;
 337         case 0x33:  /* trace stopped */
 338                 ehca_err(&shca->ib_device, "Traced stopped.");
 339                 break;
 340         default:
 341                 ehca_err(&shca->ib_device, "Unknown event code: %x on %s.",
 342                          ec, shca->ib_device.name);
 343                 break;
 344         }
 345
 346         return;
 347 }
 348
 349 static inline void reset_eq_pending(struct ehca_cq *cq)
 350 {
 351         u64 CQx_EP;
 352         struct h_galpa gal = cq->galpas.kernel;
 353
 354         hipz_galpa_store_cq(gal, cqx_ep, 0x0);
 355         CQx_EP = hipz_galpa_load(gal, CQTEMM_OFFSET(cqx_ep));
 356
 357         return;
 358 }
 359
 360 irqreturn_t ehca_interrupt_neq(int irq, void *dev_id)
 361 {
 362         struct ehca_shca *shca = (struct ehca_shca*)dev_id;
 363
 364         tasklet_hi_schedule(&shca->neq.interrupt_task);
 365
 366         return IRQ_HANDLED;
 367 }
 368
 369 void ehca_tasklet_neq(unsigned long data)
 370 {
 371         struct ehca_shca *shca = (struct ehca_shca*)data;
 372         struct ehca_eqe *eqe;
 373         u64 ret;
 374
 375         eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
 376
 377         while (eqe) {
 378                 if (!EHCA_BMASK_GET(NEQE_COMPLETION_EVENT, eqe->entry))
 379                         parse_ec(shca, eqe->entry);
 380
 381                 eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
 382         }
 383
 384         ret = hipz_h_reset_event(shca->ipz_hca_handle,
 385                                  shca->neq.ipz_eq_handle, 0xFFFFFFFFFFFFFFFFL);
 386
 387         if (ret != H_SUCCESS)
 388                 ehca_err(&shca->ib_device, "Can't clear notification events.");
 389
 390         return;
 391 }
 392
 393 irqreturn_t ehca_interrupt_eq(int irq, void *dev_id)
 394 {
 395         struct ehca_shca *shca = (struct ehca_shca*)dev_id;
 396
 397         tasklet_hi_schedule(&shca->eq.interrupt_task);
 398
 399         return IRQ_HANDLED;
 400 }
 401
 402
 403 static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 404 {
 405         u64 eqe_value;
 406         u32 token;
 407         unsigned long flags;
 408         struct ehca_cq *cq;
 409
 410         eqe_value = eqe->entry;
 411         ehca_dbg(&shca->ib_device, "eqe_value=%lx", eqe_value);
 412         if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
 413                 ehca_dbg(&shca->ib_device, "Got completion event");
 414                 token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
 415                 spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 416                 cq = idr_find(&ehca_cq_idr, token);
 417                 if (cq == NULL) {
 418                         spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 419                         ehca_err(&shca->ib_device,
 420                                  "Invalid eqe for non-existing cq token=%x",
 421                                  token);
 422                         return;
 423                 }
 424                 reset_eq_pending(cq);
 425                 cq->nr_events++;
 426                 spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 427                 if (ehca_scaling_code)
 428                         queue_comp_task(cq);
 429                 else {
 430                         comp_event_callback(cq);
 431                         spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 432                         cq->nr_events--;
 433                         if (!cq->nr_events)
 434                                 wake_up(&cq->wait_completion);
 435                         spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 436                 }
 437         } else {
 438                 ehca_dbg(&shca->ib_device, "Got non completion event");
 439                 parse_identifier(shca, eqe_value);
 440         }
 441 }
 442
 443 void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 444 {
 445         struct ehca_eq *eq = &shca->eq;
 446         struct ehca_eqe_cache_entry *eqe_cache = eq->eqe_cache;
 447         u64 eqe_value;
 448         unsigned long flags;
 449         int eqe_cnt, i;
 450         int eq_empty = 0;
 451
 452         spin_lock_irqsave(&eq->irq_spinlock, flags);
 453         if (is_irq) {
 454                 const int max_query_cnt = 100;
 455                 int query_cnt = 0;
 456                 int int_state = 1;
 457                 do {
 458                         int_state = hipz_h_query_int_state(
 459                                 shca->ipz_hca_handle, eq->ist);
 460                         query_cnt++;
 461                         iosync();
 462                 } while (int_state && query_cnt < max_query_cnt);
 463                 if (unlikely((query_cnt == max_query_cnt)))
 464                         ehca_dbg(&shca->ib_device, "int_state=%x query_cnt=%x",
 465                                  int_state, query_cnt);
 466         }
 467
 468         /* read out all eqes */
 469         eqe_cnt = 0;
 470         do {
 471                 u32 token;
 472                 eqe_cache[eqe_cnt].eqe =
 473                         (struct ehca_eqe *)ehca_poll_eq(shca, eq);
 474                 if (!eqe_cache[eqe_cnt].eqe)
 475                         break;
 476                 eqe_value = eqe_cache[eqe_cnt].eqe->entry;
 477                 if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
 478                         token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
 479                         spin_lock(&ehca_cq_idr_lock);
 480                         eqe_cache[eqe_cnt].cq = idr_find(&ehca_cq_idr, token);
 481                         if (!eqe_cache[eqe_cnt].cq) {
 482                                 spin_unlock(&ehca_cq_idr_lock);
 483                                 ehca_err(&shca->ib_device,
 484                                          "Invalid eqe for non-existing cq "
 485                                          "token=%x", token);
 486                                 continue;
 487                         }
 488                         eqe_cache[eqe_cnt].cq->nr_events++;
 489                         spin_unlock(&ehca_cq_idr_lock);
 490                 } else
 491                         eqe_cache[eqe_cnt].cq = NULL;
 492                 eqe_cnt++;
 493         } while (eqe_cnt < EHCA_EQE_CACHE_SIZE);
 494         if (!eqe_cnt) {
 495                 if (is_irq)
 496                         ehca_dbg(&shca->ib_device,
 497                                  "No eqe found for irq event");
 498                 goto unlock_irq_spinlock;
 499         } else if (!is_irq)
 500                 ehca_dbg(&shca->ib_device, "deadman found %x eqe", eqe_cnt);
 501         if (unlikely(eqe_cnt == EHCA_EQE_CACHE_SIZE))
 502                 ehca_dbg(&shca->ib_device, "too many eqes for one irq event");
 503         /* enable irq for new packets */
 504         for (i = 0; i < eqe_cnt; i++) {
 505                 if (eq->eqe_cache[i].cq)
 506                         reset_eq_pending(eq->eqe_cache[i].cq);
 507         }
 508         /* check eq */
 509         spin_lock(&eq->spinlock);
 510         eq_empty = (!ipz_eqit_eq_peek_valid(&shca->eq.ipz_queue));
 511         spin_unlock(&eq->spinlock);
 512         /* call completion handler for cached eqes */
 513         for (i = 0; i < eqe_cnt; i++)
 514                 if (eq->eqe_cache[i].cq) {
 515                         if (ehca_scaling_code)
 516                                 queue_comp_task(eq->eqe_cache[i].cq);
 517                         else {
 518                                 struct ehca_cq *cq = eq->eqe_cache[i].cq;
 519                                 comp_event_callback(cq);
 520                                 spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 521                                 cq->nr_events--;
 522                                 if (!cq->nr_events)
 523                                         wake_up(&cq->wait_completion);
 524                                 spin_unlock_irqrestore(&ehca_cq_idr_lock,
 525                                                        flags);
 526                         }
 527                 } else {
 528                         ehca_dbg(&shca->ib_device, "Got non completion event");
 529                         parse_identifier(shca, eq->eqe_cache[i].eqe->entry);
 530                 }
 531         /* poll eq if not empty */
 532         if (eq_empty)
 533                 goto unlock_irq_spinlock;
 534         do {
 535                 struct ehca_eqe *eqe;
 536                 eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->eq);
 537                 if (!eqe)
 538                         break;
 539                 process_eqe(shca, eqe);
 540         } while (1);
 541
 542 unlock_irq_spinlock:
 543         spin_unlock_irqrestore(&eq->irq_spinlock, flags);
 544 }
 545
 546 void ehca_tasklet_eq(unsigned long data)
 547 {
 548         ehca_process_eq((struct ehca_shca*)data, 1);
 549 }
 550
 551 static inline int find_next_online_cpu(struct ehca_comp_pool* pool)
 552 {
 553         int cpu;
 554         unsigned long flags;
 555
 556         WARN_ON_ONCE(!in_interrupt());
 557         if (ehca_debug_level)
 558                 ehca_dmp(&cpu_online_map, sizeof(cpumask_t), "");
 559
 560         spin_lock_irqsave(&pool->last_cpu_lock, flags);
 561         cpu = next_cpu(pool->last_cpu, cpu_online_map);
 562         if (cpu == NR_CPUS)
 563                 cpu = first_cpu(cpu_online_map);
 564         pool->last_cpu = cpu;
 565         spin_unlock_irqrestore(&pool->last_cpu_lock, flags);
 566
 567         return cpu;
 568 }
 569
 570 static void __queue_comp_task(struct ehca_cq *__cq,
 571                               struct ehca_cpu_comp_task *cct)
 572 {
 573         unsigned long flags;
 574
 575         spin_lock_irqsave(&cct->task_lock, flags);
 576         spin_lock(&__cq->task_lock);
 577
 578         if (__cq->nr_callbacks == 0) {
 579                 __cq->nr_callbacks++;
 580                 list_add_tail(&__cq->entry, &cct->cq_list);
 581                 cct->cq_jobs++;
 582                 wake_up(&cct->wait_queue);
 583         } else
 584                 __cq->nr_callbacks++;
 585
 586         spin_unlock(&__cq->task_lock);
 587         spin_unlock_irqrestore(&cct->task_lock, flags);
 588 }
 589
 590 static void queue_comp_task(struct ehca_cq *__cq)
 591 {
 592         int cpu_id;
 593         struct ehca_cpu_comp_task *cct;
 594         int cq_jobs;
 595         unsigned long flags;
 596
 597         cpu_id = find_next_online_cpu(pool);
 598         BUG_ON(!cpu_online(cpu_id));
 599
 600         cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
 601         BUG_ON(!cct);
 602
 603         spin_lock_irqsave(&cct->task_lock, flags);
 604         cq_jobs = cct->cq_jobs;
 605         spin_unlock_irqrestore(&cct->task_lock, flags);
 606         if (cq_jobs > 0) {
 607                 cpu_id = find_next_online_cpu(pool);
 608                 cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
 609                 BUG_ON(!cct);
 610         }
 611
 612         __queue_comp_task(__cq, cct);
 613 }
 614
 615 static void run_comp_task(struct ehca_cpu_comp_task* cct)
 616 {
 617         struct ehca_cq *cq;
 618         unsigned long flags;
 619
 620         spin_lock_irqsave(&cct->task_lock, flags);
 621
 622         while (!list_empty(&cct->cq_list)) {
 623                 cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
 624                 spin_unlock_irqrestore(&cct->task_lock, flags);
 625                 comp_event_callback(cq);
 626
 627                 spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 628                 cq->nr_events--;
 629                 if (!cq->nr_events)
 630                         wake_up(&cq->wait_completion);
 631                 spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 632
 633                 spin_lock_irqsave(&cct->task_lock, flags);
 634                 spin_lock(&cq->task_lock);
 635                 cq->nr_callbacks--;
 636                 if (!cq->nr_callbacks) {
 637                         list_del_init(cct->cq_list.next);
 638                         cct->cq_jobs--;
 639                 }
 640                 spin_unlock(&cq->task_lock);
 641         }
 642
 643         spin_unlock_irqrestore(&cct->task_lock, flags);
 644 }
 645
 646 static int comp_task(void *__cct)
 647 {
 648         struct ehca_cpu_comp_task* cct = __cct;
 649         int cql_empty;
 650         DECLARE_WAITQUEUE(wait, current);
 651
 652         set_current_state(TASK_INTERRUPTIBLE);
 653         while(!kthread_should_stop()) {
 654                 add_wait_queue(&cct->wait_queue, &wait);
 655
 656                 spin_lock_irq(&cct->task_lock);
 657                 cql_empty = list_empty(&cct->cq_list);
 658                 spin_unlock_irq(&cct->task_lock);
 659                 if (cql_empty)
 660                         schedule();
 661                 else
 662                         __set_current_state(TASK_RUNNING);
 663
 664                 remove_wait_queue(&cct->wait_queue, &wait);
 665
 666                 spin_lock_irq(&cct->task_lock);
 667                 cql_empty = list_empty(&cct->cq_list);
 668                 spin_unlock_irq(&cct->task_lock);
 669                 if (!cql_empty)
 670                         run_comp_task(__cct);
 671
 672                 set_current_state(TASK_INTERRUPTIBLE);
 673         }
 674         __set_current_state(TASK_RUNNING);
 675
 676         return 0;
 677 }
 678
 679 static struct task_struct *create_comp_task(struct ehca_comp_pool *pool,
 680                                             int cpu)
 681 {
 682         struct ehca_cpu_comp_task *cct;
 683
 684         cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 685         spin_lock_init(&cct->task_lock);
 686         INIT_LIST_HEAD(&cct->cq_list);
 687         init_waitqueue_head(&cct->wait_queue);
 688         cct->task = kthread_create(comp_task, cct, "ehca_comp/%d", cpu);
 689
 690         return cct->task;
 691 }
 692
 693 static void destroy_comp_task(struct ehca_comp_pool *pool,
 694                               int cpu)
 695 {
 696         struct ehca_cpu_comp_task *cct;
 697         struct task_struct *task;
 698         unsigned long flags_cct;
 699
 700         cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 701
 702         spin_lock_irqsave(&cct->task_lock, flags_cct);
 703
 704         task = cct->task;
 705         cct->task = NULL;
 706         cct->cq_jobs = 0;
 707
 708         spin_unlock_irqrestore(&cct->task_lock, flags_cct);
 709
 710         if (task)
 711                 kthread_stop(task);
 712 }
 713
 714 static void take_over_work(struct ehca_comp_pool *pool,
 715                            int cpu)
 716 {
 717         struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 718         LIST_HEAD(list);
 719         struct ehca_cq *cq;
 720         unsigned long flags_cct;
 721
 722         spin_lock_irqsave(&cct->task_lock, flags_cct);
 723
 724         list_splice_init(&cct->cq_list, &list);
 725
 726         while(!list_empty(&list)) {
 727                 cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
 728
 729                 list_del(&cq->entry);
 730                 __queue_comp_task(cq, per_cpu_ptr(pool->cpu_comp_tasks,
 731                                                   smp_processor_id()));
 732         }
 733
 734         spin_unlock_irqrestore(&cct->task_lock, flags_cct);
 735
 736 }
 737
 738 #ifdef CONFIG_HOTPLUG_CPU
 739 static int comp_pool_callback(struct notifier_block *nfb,
 740                               unsigned long action,
 741                               void *hcpu)
 742 {
 743         unsigned int cpu = (unsigned long)hcpu;
 744         struct ehca_cpu_comp_task *cct;
 745
 746         switch (action) {
 747         case CPU_UP_PREPARE:
 748         case CPU_UP_PREPARE_FROZEN:
 749                 ehca_gen_dbg("CPU: %x (CPU_PREPARE)", cpu);
 750                 if(!create_comp_task(pool, cpu)) {
 751                         ehca_gen_err("Can't create comp_task for cpu: %x", cpu);
 752                         return NOTIFY_BAD;
 753                 }
 754                 break;
 755         case CPU_UP_CANCELED:
 756         case CPU_UP_CANCELED_FROZEN:
 757                 ehca_gen_dbg("CPU: %x (CPU_CANCELED)", cpu);
 758                 cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 759                 kthread_bind(cct->task, any_online_cpu(cpu_online_map));
 760                 destroy_comp_task(pool, cpu);
 761                 break;
 762         case CPU_ONLINE:
 763         case CPU_ONLINE_FROZEN:
 764                 ehca_gen_dbg("CPU: %x (CPU_ONLINE)", cpu);
 765                 cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 766                 kthread_bind(cct->task, cpu);
 767                 wake_up_process(cct->task);
 768                 break;
 769         case CPU_DOWN_PREPARE:
 770         case CPU_DOWN_PREPARE_FROZEN:
 771                 ehca_gen_dbg("CPU: %x (CPU_DOWN_PREPARE)", cpu);
 772                 break;
 773         case CPU_DOWN_FAILED:
 774         case CPU_DOWN_FAILED_FROZEN:
 775                 ehca_gen_dbg("CPU: %x (CPU_DOWN_FAILED)", cpu);
 776                 break;
 777         case CPU_DEAD:
 778         case CPU_DEAD_FROZEN:
 779                 ehca_gen_dbg("CPU: %x (CPU_DEAD)", cpu);
 780                 destroy_comp_task(pool, cpu);
 781                 take_over_work(pool, cpu);
 782                 break;
 783         }
 784
 785         return NOTIFY_OK;
 786 }
 787 #endif
 788
 789 int ehca_create_comp_pool(void)
 790 {
 791         int cpu;
 792         struct task_struct *task;
 793
 794         if (!ehca_scaling_code)
 795                 return 0;
 796
 797         pool = kzalloc(sizeof(struct ehca_comp_pool), GFP_KERNEL);
 798         if (pool == NULL)
 799                 return -ENOMEM;
 800
 801         spin_lock_init(&pool->last_cpu_lock);
 802         pool->last_cpu = any_online_cpu(cpu_online_map);
 803
 804         pool->cpu_comp_tasks = alloc_percpu(struct ehca_cpu_comp_task);
 805         if (pool->cpu_comp_tasks == NULL) {
 806                 kfree(pool);
 807                 return -EINVAL;
 808         }
 809
 810         for_each_online_cpu(cpu) {
 811                 task = create_comp_task(pool, cpu);
 812                 if (task) {
 813                         kthread_bind(task, cpu);
 814                         wake_up_process(task);
 815                 }
 816         }
 817
 818 #ifdef CONFIG_HOTPLUG_CPU
 819         comp_pool_callback_nb.notifier_call = comp_pool_callback;
 820         comp_pool_callback_nb.priority =0;
 821         register_cpu_notifier(&comp_pool_callback_nb);
 822 #endif
 823
 824         printk(KERN_INFO "eHCA scaling code enabled\n");
 825
 826         return 0;
 827 }
 828
 829 void ehca_destroy_comp_pool(void)
 830 {
 831         int i;
 832
 833         if (!ehca_scaling_code)
 834                 return;
 835
 836 #ifdef CONFIG_HOTPLUG_CPU
 837         unregister_cpu_notifier(&comp_pool_callback_nb);
 838 #endif
 839
 840         for (i = 0; i < NR_CPUS; i++) {
 841                 if (cpu_online(i))
 842                         destroy_comp_task(pool, i);
 843         }
 844         free_percpu(pool->cpu_comp_tasks);
 845         kfree(pool);
 846 }