lib/librte_eal/linux/eal/eal_interrupts.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdint.h>
   7 #include <stdlib.h>
   8 #include <pthread.h>
   9 #include <sys/queue.h>
  10 #include <stdarg.h>
  11 #include <unistd.h>
  12 #include <string.h>
  13 #include <errno.h>
  14 #include <inttypes.h>
  15 #include <sys/epoll.h>
  16 #include <sys/signalfd.h>
  17 #include <sys/ioctl.h>
  18 #include <sys/eventfd.h>
  19 #include <assert.h>
  20 #include <stdbool.h>
  21
  22 #include <rte_common.h>
  23 #include <rte_interrupts.h>
  24 #include <rte_memory.h>
  25 #include <rte_launch.h>
  26 #include <rte_eal.h>
  27 #include <rte_per_lcore.h>
  28 #include <rte_lcore.h>
  29 #include <rte_atomic.h>
  30 #include <rte_branch_prediction.h>
  31 #include <rte_debug.h>
  32 #include <rte_log.h>
  33 #include <rte_errno.h>
  34 #include <rte_spinlock.h>
  35 #include <rte_pause.h>
  36 #include <rte_vfio.h>
  37
  38 #include "eal_private.h"
  39 #include "eal_vfio.h"
  40 #include "eal_thread.h"
  41
  42 #define EAL_INTR_EPOLL_WAIT_FOREVER (-1)
  43 #define NB_OTHER_INTR               1
  44
  45 static RTE_DEFINE_PER_LCORE(int, _epfd) = -1; /**< epoll fd per thread */
  46
  47 /**
  48  * union for pipe fds.
  49  */
  50 union intr_pipefds{
  51         struct {
  52                 int pipefd[2];
  53         };
  54         struct {
  55                 int readfd;
  56                 int writefd;
  57         };
  58 };
  59
  60 /**
  61  * union buffer for reading on different devices
  62  */
  63 union rte_intr_read_buffer {
  64         int uio_intr_count;              /* for uio device */
  65 #ifdef VFIO_PRESENT
  66         uint64_t vfio_intr_count;        /* for vfio device */
  67 #endif
  68         uint64_t timerfd_num;            /* for timerfd */
  69         char charbuf[16];                /* for others */
  70 };
  71
  72 TAILQ_HEAD(rte_intr_cb_list, rte_intr_callback);
  73 TAILQ_HEAD(rte_intr_source_list, rte_intr_source);
  74
  75 struct rte_intr_callback {
  76         TAILQ_ENTRY(rte_intr_callback) next;
  77         rte_intr_callback_fn cb_fn;  /**< callback address */
  78         void *cb_arg;                /**< parameter for callback */
  79         uint8_t pending_delete;      /**< delete after callback is called */
  80         rte_intr_unregister_callback_fn ucb_fn; /**< fn to call before cb is deleted */
  81 };
  82
  83 struct rte_intr_source {
  84         TAILQ_ENTRY(rte_intr_source) next;
  85         struct rte_intr_handle intr_handle; /**< interrupt handle */
  86         struct rte_intr_cb_list callbacks;  /**< user callbacks */
  87         uint32_t active;
  88 };
  89
  90 /* global spinlock for interrupt data operation */
  91 static rte_spinlock_t intr_lock = RTE_SPINLOCK_INITIALIZER;
  92
  93 /* union buffer for pipe read/write */
  94 static union intr_pipefds intr_pipe;
  95
  96 /* interrupt sources list */
  97 static struct rte_intr_source_list intr_sources;
  98
  99 /* interrupt handling thread */
 100 static pthread_t intr_thread;
 101
 102 /* VFIO interrupts */
 103 #ifdef VFIO_PRESENT
 104
 105 #define IRQ_SET_BUF_LEN  (sizeof(struct vfio_irq_set) + sizeof(int))
 106 /* irq set buffer length for queue interrupts and LSC interrupt */
 107 #define MSIX_IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + \
 108                               sizeof(int) * (RTE_MAX_RXTX_INTR_VEC_ID + 1))
 109
 110 /* enable legacy (INTx) interrupts */
 111 static int
 112 vfio_enable_intx(const struct rte_intr_handle *intr_handle) {
 113         struct vfio_irq_set *irq_set;
 114         char irq_set_buf[IRQ_SET_BUF_LEN];
 115         int len, ret;
 116         int *fd_ptr;
 117
 118         len = sizeof(irq_set_buf);
 119
 120         /* enable INTx */
 121         irq_set = (struct vfio_irq_set *) irq_set_buf;
 122         irq_set->argsz = len;
 123         irq_set->count = 1;
 124         irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
 125         irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
 126         irq_set->start = 0;
 127         fd_ptr = (int *) &irq_set->data;
 128         *fd_ptr = intr_handle->fd;
 129
 130         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 131
 132         if (ret) {
 133                 RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
 134                                                 intr_handle->fd);
 135                 return -1;
 136         }
 137
 138         /* unmask INTx after enabling */
 139         memset(irq_set, 0, len);
 140         len = sizeof(struct vfio_irq_set);
 141         irq_set->argsz = len;
 142         irq_set->count = 1;
 143         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
 144         irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
 145         irq_set->start = 0;
 146
 147         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 148
 149         if (ret) {
 150                 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
 151                                                 intr_handle->fd);
 152                 return -1;
 153         }
 154         return 0;
 155 }
 156
 157 /* disable legacy (INTx) interrupts */
 158 static int
 159 vfio_disable_intx(const struct rte_intr_handle *intr_handle) {
 160         struct vfio_irq_set *irq_set;
 161         char irq_set_buf[IRQ_SET_BUF_LEN];
 162         int len, ret;
 163
 164         len = sizeof(struct vfio_irq_set);
 165
 166         /* mask interrupts before disabling */
 167         irq_set = (struct vfio_irq_set *) irq_set_buf;
 168         irq_set->argsz = len;
 169         irq_set->count = 1;
 170         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_MASK;
 171         irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
 172         irq_set->start = 0;
 173
 174         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 175
 176         if (ret) {
 177                 RTE_LOG(ERR, EAL, "Error masking INTx interrupts for fd %d\n",
 178                                                 intr_handle->fd);
 179                 return -1;
 180         }
 181
 182         /* disable INTx*/
 183         memset(irq_set, 0, len);
 184         irq_set->argsz = len;
 185         irq_set->count = 0;
 186         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
 187         irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
 188         irq_set->start = 0;
 189
 190         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 191
 192         if (ret) {
 193                 RTE_LOG(ERR, EAL,
 194                         "Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
 195                 return -1;
 196         }
 197         return 0;
 198 }
 199
 200 /* enable MSI interrupts */
 201 static int
 202 vfio_enable_msi(const struct rte_intr_handle *intr_handle) {
 203         int len, ret;
 204         char irq_set_buf[IRQ_SET_BUF_LEN];
 205         struct vfio_irq_set *irq_set;
 206         int *fd_ptr;
 207
 208         len = sizeof(irq_set_buf);
 209
 210         irq_set = (struct vfio_irq_set *) irq_set_buf;
 211         irq_set->argsz = len;
 212         irq_set->count = 1;
 213         irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
 214         irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
 215         irq_set->start = 0;
 216         fd_ptr = (int *) &irq_set->data;
 217         *fd_ptr = intr_handle->fd;
 218
 219         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 220
 221         if (ret) {
 222                 RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
 223                                                 intr_handle->fd);
 224                 return -1;
 225         }
 226         return 0;
 227 }
 228
 229 /* disable MSI interrupts */
 230 static int
 231 vfio_disable_msi(const struct rte_intr_handle *intr_handle) {
 232         struct vfio_irq_set *irq_set;
 233         char irq_set_buf[IRQ_SET_BUF_LEN];
 234         int len, ret;
 235
 236         len = sizeof(struct vfio_irq_set);
 237
 238         irq_set = (struct vfio_irq_set *) irq_set_buf;
 239         irq_set->argsz = len;
 240         irq_set->count = 0;
 241         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
 242         irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
 243         irq_set->start = 0;
 244
 245         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 246
 247         if (ret)
 248                 RTE_LOG(ERR, EAL,
 249                         "Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
 250
 251         return ret;
 252 }
 253
 254 /* enable MSI-X interrupts */
 255 static int
 256 vfio_enable_msix(const struct rte_intr_handle *intr_handle) {
 257         int len, ret;
 258         char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
 259         struct vfio_irq_set *irq_set;
 260         int *fd_ptr;
 261
 262         len = sizeof(irq_set_buf);
 263
 264         irq_set = (struct vfio_irq_set *) irq_set_buf;
 265         irq_set->argsz = len;
 266         /* 0 < irq_set->count < RTE_MAX_RXTX_INTR_VEC_ID + 1 */
 267         irq_set->count = intr_handle->max_intr ?
 268                 (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID + 1 ?
 269                 RTE_MAX_RXTX_INTR_VEC_ID + 1 : intr_handle->max_intr) : 1;
 270         irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
 271         irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
 272         irq_set->start = 0;
 273         fd_ptr = (int *) &irq_set->data;
 274         /* INTR vector offset 0 reserve for non-efds mapping */
 275         fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
 276         memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
 277                 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
 278
 279         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 280
 281         if (ret) {
 282                 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
 283                                                 intr_handle->fd);
 284                 return -1;
 285         }
 286
 287         return 0;
 288 }
 289
 290 /* disable MSI-X interrupts */
 291 static int
 292 vfio_disable_msix(const struct rte_intr_handle *intr_handle) {
 293         struct vfio_irq_set *irq_set;
 294         char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
 295         int len, ret;
 296
 297         len = sizeof(struct vfio_irq_set);
 298
 299         irq_set = (struct vfio_irq_set *) irq_set_buf;
 300         irq_set->argsz = len;
 301         irq_set->count = 0;
 302         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
 303         irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
 304         irq_set->start = 0;
 305
 306         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 307
 308         if (ret)
 309                 RTE_LOG(ERR, EAL,
 310                         "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
 311
 312         return ret;
 313 }
 314
 315 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
 316 /* enable req notifier */
 317 static int
 318 vfio_enable_req(const struct rte_intr_handle *intr_handle)
 319 {
 320         int len, ret;
 321         char irq_set_buf[IRQ_SET_BUF_LEN];
 322         struct vfio_irq_set *irq_set;
 323         int *fd_ptr;
 324
 325         len = sizeof(irq_set_buf);
 326
 327         irq_set = (struct vfio_irq_set *) irq_set_buf;
 328         irq_set->argsz = len;
 329         irq_set->count = 1;
 330         irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD |
 331                          VFIO_IRQ_SET_ACTION_TRIGGER;
 332         irq_set->index = VFIO_PCI_REQ_IRQ_INDEX;
 333         irq_set->start = 0;
 334         fd_ptr = (int *) &irq_set->data;
 335         *fd_ptr = intr_handle->fd;
 336
 337         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 338
 339         if (ret) {
 340                 RTE_LOG(ERR, EAL, "Error enabling req interrupts for fd %d\n",
 341                                                 intr_handle->fd);
 342                 return -1;
 343         }
 344
 345         return 0;
 346 }
 347
 348 /* disable req notifier */
 349 static int
 350 vfio_disable_req(const struct rte_intr_handle *intr_handle)
 351 {
 352         struct vfio_irq_set *irq_set;
 353         char irq_set_buf[IRQ_SET_BUF_LEN];
 354         int len, ret;
 355
 356         len = sizeof(struct vfio_irq_set);
 357
 358         irq_set = (struct vfio_irq_set *) irq_set_buf;
 359         irq_set->argsz = len;
 360         irq_set->count = 0;
 361         irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
 362         irq_set->index = VFIO_PCI_REQ_IRQ_INDEX;
 363         irq_set->start = 0;
 364
 365         ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
 366
 367         if (ret)
 368                 RTE_LOG(ERR, EAL, "Error disabling req interrupts for fd %d\n",
 369                         intr_handle->fd);
 370
 371         return ret;
 372 }
 373 #endif
 374 #endif
 375
 376 static int
 377 uio_intx_intr_disable(const struct rte_intr_handle *intr_handle)
 378 {
 379         unsigned char command_high;
 380
 381         /* use UIO config file descriptor for uio_pci_generic */
 382         if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
 383                 RTE_LOG(ERR, EAL,
 384                         "Error reading interrupts status for fd %d\n",
 385                         intr_handle->uio_cfg_fd);
 386                 return -1;
 387         }
 388         /* disable interrupts */
 389         command_high |= 0x4;
 390         if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
 391                 RTE_LOG(ERR, EAL,
 392                         "Error disabling interrupts for fd %d\n",
 393                         intr_handle->uio_cfg_fd);
 394                 return -1;
 395         }
 396
 397         return 0;
 398 }
 399
 400 static int
 401 uio_intx_intr_enable(const struct rte_intr_handle *intr_handle)
 402 {
 403         unsigned char command_high;
 404
 405         /* use UIO config file descriptor for uio_pci_generic */
 406         if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
 407                 RTE_LOG(ERR, EAL,
 408                         "Error reading interrupts status for fd %d\n",
 409                         intr_handle->uio_cfg_fd);
 410                 return -1;
 411         }
 412         /* enable interrupts */
 413         command_high &= ~0x4;
 414         if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
 415                 RTE_LOG(ERR, EAL,
 416                         "Error enabling interrupts for fd %d\n",
 417                         intr_handle->uio_cfg_fd);
 418                 return -1;
 419         }
 420
 421         return 0;
 422 }
 423
 424 static int
 425 uio_intr_disable(const struct rte_intr_handle *intr_handle)
 426 {
 427         const int value = 0;
 428
 429         if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
 430                 RTE_LOG(ERR, EAL,
 431                         "Error disabling interrupts for fd %d (%s)\n",
 432                         intr_handle->fd, strerror(errno));
 433                 return -1;
 434         }
 435         return 0;
 436 }
 437
 438 static int
 439 uio_intr_enable(const struct rte_intr_handle *intr_handle)
 440 {
 441         const int value = 1;
 442
 443         if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
 444                 RTE_LOG(ERR, EAL,
 445                         "Error enabling interrupts for fd %d (%s)\n",
 446                         intr_handle->fd, strerror(errno));
 447                 return -1;
 448         }
 449         return 0;
 450 }
 451
 452 int
 453 rte_intr_callback_register(const struct rte_intr_handle *intr_handle,
 454                         rte_intr_callback_fn cb, void *cb_arg)
 455 {
 456         int ret, wake_thread;
 457         struct rte_intr_source *src;
 458         struct rte_intr_callback *callback;
 459
 460         wake_thread = 0;
 461
 462         /* first do parameter checking */
 463         if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
 464                 RTE_LOG(ERR, EAL,
 465                         "Registering with invalid input parameter\n");
 466                 return -EINVAL;
 467         }
 468
 469         /* allocate a new interrupt callback entity */
 470         callback = calloc(1, sizeof(*callback));
 471         if (callback == NULL) {
 472                 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
 473                 return -ENOMEM;
 474         }
 475         callback->cb_fn = cb;
 476         callback->cb_arg = cb_arg;
 477         callback->pending_delete = 0;
 478         callback->ucb_fn = NULL;
 479
 480         rte_spinlock_lock(&intr_lock);
 481
 482         /* check if there is at least one callback registered for the fd */
 483         TAILQ_FOREACH(src, &intr_sources, next) {
 484                 if (src->intr_handle.fd == intr_handle->fd) {
 485                         /* we had no interrupts for this */
 486                         if (TAILQ_EMPTY(&src->callbacks))
 487                                 wake_thread = 1;
 488
 489                         TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
 490                         ret = 0;
 491                         break;
 492                 }
 493         }
 494
 495         /* no existing callbacks for this - add new source */
 496         if (src == NULL) {
 497                 src = calloc(1, sizeof(*src));
 498                 if (src == NULL) {
 499                         RTE_LOG(ERR, EAL, "Can not allocate memory\n");
 500                         free(callback);
 501                         ret = -ENOMEM;
 502                 } else {
 503                         src->intr_handle = *intr_handle;
 504                         TAILQ_INIT(&src->callbacks);
 505                         TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
 506                         TAILQ_INSERT_TAIL(&intr_sources, src, next);
 507                         wake_thread = 1;
 508                         ret = 0;
 509                 }
 510         }
 511
 512         rte_spinlock_unlock(&intr_lock);
 513
 514         /**
 515          * check if need to notify the pipe fd waited by epoll_wait to
 516          * rebuild the wait list.
 517          */
 518         if (wake_thread)
 519                 if (write(intr_pipe.writefd, "1", 1) < 0)
 520                         return -EPIPE;
 521
 522         return ret;
 523 }
 524
 525 int __rte_experimental
 526 rte_intr_callback_unregister_pending(const struct rte_intr_handle *intr_handle,
 527                                 rte_intr_callback_fn cb_fn, void *cb_arg,
 528                                 rte_intr_unregister_callback_fn ucb_fn)
 529 {
 530         int ret;
 531         struct rte_intr_source *src;
 532         struct rte_intr_callback *cb, *next;
 533
 534         /* do parameter checking first */
 535         if (intr_handle == NULL || intr_handle->fd < 0) {
 536                 RTE_LOG(ERR, EAL,
 537                 "Unregistering with invalid input parameter\n");
 538                 return -EINVAL;
 539         }
 540
 541         rte_spinlock_lock(&intr_lock);
 542
 543         /* check if the insterrupt source for the fd is existent */
 544         TAILQ_FOREACH(src, &intr_sources, next)
 545                 if (src->intr_handle.fd == intr_handle->fd)
 546                         break;
 547
 548         /* No interrupt source registered for the fd */
 549         if (src == NULL) {
 550                 ret = -ENOENT;
 551
 552         /* only usable if the source is active */
 553         } else if (src->active == 0) {
 554                 ret = -EAGAIN;
 555
 556         } else {
 557                 ret = 0;
 558
 559                 /* walk through the callbacks and mark all that match. */
 560                 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
 561                         next = TAILQ_NEXT(cb, next);
 562                         if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
 563                                         cb->cb_arg == cb_arg)) {
 564                                 cb->pending_delete = 1;
 565                                 cb->ucb_fn = ucb_fn;
 566                                 ret++;
 567                         }
 568                 }
 569         }
 570
 571         rte_spinlock_unlock(&intr_lock);
 572
 573         return ret;
 574 }
 575
 576 int
 577 rte_intr_callback_unregister(const struct rte_intr_handle *intr_handle,
 578                         rte_intr_callback_fn cb_fn, void *cb_arg)
 579 {
 580         int ret;
 581         struct rte_intr_source *src;
 582         struct rte_intr_callback *cb, *next;
 583
 584         /* do parameter checking first */
 585         if (intr_handle == NULL || intr_handle->fd < 0) {
 586                 RTE_LOG(ERR, EAL,
 587                 "Unregistering with invalid input parameter\n");
 588                 return -EINVAL;
 589         }
 590
 591         rte_spinlock_lock(&intr_lock);
 592
 593         /* check if the insterrupt source for the fd is existent */
 594         TAILQ_FOREACH(src, &intr_sources, next)
 595                 if (src->intr_handle.fd == intr_handle->fd)
 596                         break;
 597
 598         /* No interrupt source registered for the fd */
 599         if (src == NULL) {
 600                 ret = -ENOENT;
 601
 602         /* interrupt source has some active callbacks right now. */
 603         } else if (src->active != 0) {
 604                 ret = -EAGAIN;
 605
 606         /* ok to remove. */
 607         } else {
 608                 ret = 0;
 609
 610                 /*walk through the callbacks and remove all that match. */
 611                 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
 612
 613                         next = TAILQ_NEXT(cb, next);
 614
 615                         if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
 616                                         cb->cb_arg == cb_arg)) {
 617                                 TAILQ_REMOVE(&src->callbacks, cb, next);
 618                                 free(cb);
 619                                 ret++;
 620                         }
 621                 }
 622
 623                 /* all callbacks for that source are removed. */
 624                 if (TAILQ_EMPTY(&src->callbacks)) {
 625                         TAILQ_REMOVE(&intr_sources, src, next);
 626                         free(src);
 627                 }
 628         }
 629
 630         rte_spinlock_unlock(&intr_lock);
 631
 632         /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
 633         if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
 634                 ret = -EPIPE;
 635         }
 636
 637         return ret;
 638 }
 639
 640 int
 641 rte_intr_enable(const struct rte_intr_handle *intr_handle)
 642 {
 643         if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
 644                 return 0;
 645
 646         if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
 647                 return -1;
 648
 649         switch (intr_handle->type){
 650         /* write to the uio fd to enable the interrupt */
 651         case RTE_INTR_HANDLE_UIO:
 652                 if (uio_intr_enable(intr_handle))
 653                         return -1;
 654                 break;
 655         case RTE_INTR_HANDLE_UIO_INTX:
 656                 if (uio_intx_intr_enable(intr_handle))
 657                         return -1;
 658                 break;
 659         /* not used at this moment */
 660         case RTE_INTR_HANDLE_ALARM:
 661                 return -1;
 662 #ifdef VFIO_PRESENT
 663         case RTE_INTR_HANDLE_VFIO_MSIX:
 664                 if (vfio_enable_msix(intr_handle))
 665                         return -1;
 666                 break;
 667         case RTE_INTR_HANDLE_VFIO_MSI:
 668                 if (vfio_enable_msi(intr_handle))
 669                         return -1;
 670                 break;
 671         case RTE_INTR_HANDLE_VFIO_LEGACY:
 672                 if (vfio_enable_intx(intr_handle))
 673                         return -1;
 674                 break;
 675 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
 676         case RTE_INTR_HANDLE_VFIO_REQ:
 677                 if (vfio_enable_req(intr_handle))
 678                         return -1;
 679                 break;
 680 #endif
 681 #endif
 682         /* not used at this moment */
 683         case RTE_INTR_HANDLE_DEV_EVENT:
 684                 return -1;
 685         /* unknown handle type */
 686         default:
 687                 RTE_LOG(ERR, EAL,
 688                         "Unknown handle type of fd %d\n",
 689                                         intr_handle->fd);
 690                 return -1;
 691         }
 692
 693         return 0;
 694 }
 695
 696 int
 697 rte_intr_disable(const struct rte_intr_handle *intr_handle)
 698 {
 699         if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
 700                 return 0;
 701
 702         if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
 703                 return -1;
 704
 705         switch (intr_handle->type){
 706         /* write to the uio fd to disable the interrupt */
 707         case RTE_INTR_HANDLE_UIO:
 708                 if (uio_intr_disable(intr_handle))
 709                         return -1;
 710                 break;
 711         case RTE_INTR_HANDLE_UIO_INTX:
 712                 if (uio_intx_intr_disable(intr_handle))
 713                         return -1;
 714                 break;
 715         /* not used at this moment */
 716         case RTE_INTR_HANDLE_ALARM:
 717                 return -1;
 718 #ifdef VFIO_PRESENT
 719         case RTE_INTR_HANDLE_VFIO_MSIX:
 720                 if (vfio_disable_msix(intr_handle))
 721                         return -1;
 722                 break;
 723         case RTE_INTR_HANDLE_VFIO_MSI:
 724                 if (vfio_disable_msi(intr_handle))
 725                         return -1;
 726                 break;
 727         case RTE_INTR_HANDLE_VFIO_LEGACY:
 728                 if (vfio_disable_intx(intr_handle))
 729                         return -1;
 730                 break;
 731 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
 732         case RTE_INTR_HANDLE_VFIO_REQ:
 733                 if (vfio_disable_req(intr_handle))
 734                         return -1;
 735                 break;
 736 #endif
 737 #endif
 738         /* not used at this moment */
 739         case RTE_INTR_HANDLE_DEV_EVENT:
 740                 return -1;
 741         /* unknown handle type */
 742         default:
 743                 RTE_LOG(ERR, EAL,
 744                         "Unknown handle type of fd %d\n",
 745                                         intr_handle->fd);
 746                 return -1;
 747         }
 748
 749         return 0;
 750 }
 751
 752 static int
 753 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
 754 {
 755         bool call = false;
 756         int n, bytes_read, rv;
 757         struct rte_intr_source *src;
 758         struct rte_intr_callback *cb, *next;
 759         union rte_intr_read_buffer buf;
 760         struct rte_intr_callback active_cb;
 761
 762         for (n = 0; n < nfds; n++) {
 763
 764                 /**
 765                  * if the pipe fd is ready to read, return out to
 766                  * rebuild the wait list.
 767                  */
 768                 if (events[n].data.fd == intr_pipe.readfd){
 769                         int r = read(intr_pipe.readfd, buf.charbuf,
 770                                         sizeof(buf.charbuf));
 771                         RTE_SET_USED(r);
 772                         return -1;
 773                 }
 774                 rte_spinlock_lock(&intr_lock);
 775                 TAILQ_FOREACH(src, &intr_sources, next)
 776                         if (src->intr_handle.fd ==
 777                                         events[n].data.fd)
 778                                 break;
 779                 if (src == NULL){
 780                         rte_spinlock_unlock(&intr_lock);
 781                         continue;
 782                 }
 783
 784                 /* mark this interrupt source as active and release the lock. */
 785                 src->active = 1;
 786                 rte_spinlock_unlock(&intr_lock);
 787
 788                 /* set the length to be read dor different handle type */
 789                 switch (src->intr_handle.type) {
 790                 case RTE_INTR_HANDLE_UIO:
 791                 case RTE_INTR_HANDLE_UIO_INTX:
 792                         bytes_read = sizeof(buf.uio_intr_count);
 793                         break;
 794                 case RTE_INTR_HANDLE_ALARM:
 795                         bytes_read = sizeof(buf.timerfd_num);
 796                         break;
 797 #ifdef VFIO_PRESENT
 798                 case RTE_INTR_HANDLE_VFIO_MSIX:
 799                 case RTE_INTR_HANDLE_VFIO_MSI:
 800                 case RTE_INTR_HANDLE_VFIO_LEGACY:
 801                         bytes_read = sizeof(buf.vfio_intr_count);
 802                         break;
 803 #ifdef HAVE_VFIO_DEV_REQ_INTERFACE
 804                 case RTE_INTR_HANDLE_VFIO_REQ:
 805                         bytes_read = 0;
 806                         call = true;
 807                         break;
 808 #endif
 809 #endif
 810                 case RTE_INTR_HANDLE_VDEV:
 811                 case RTE_INTR_HANDLE_EXT:
 812                         bytes_read = 0;
 813                         call = true;
 814                         break;
 815                 case RTE_INTR_HANDLE_DEV_EVENT:
 816                         bytes_read = 0;
 817                         call = true;
 818                         break;
 819                 default:
 820                         bytes_read = 1;
 821                         break;
 822                 }
 823
 824                 if (bytes_read > 0) {
 825                         /**
 826                          * read out to clear the ready-to-be-read flag
 827                          * for epoll_wait.
 828                          */
 829                         bytes_read = read(events[n].data.fd, &buf, bytes_read);
 830                         if (bytes_read < 0) {
 831                                 if (errno == EINTR || errno == EWOULDBLOCK)
 832                                         continue;
 833
 834                                 RTE_LOG(ERR, EAL, "Error reading from file "
 835                                         "descriptor %d: %s\n",
 836                                         events[n].data.fd,
 837                                         strerror(errno));
 838                                 /*
 839                                  * The device is unplugged or buggy, remove
 840                                  * it as an interrupt source and return to
 841                                  * force the wait list to be rebuilt.
 842                                  */
 843                                 rte_spinlock_lock(&intr_lock);
 844                                 TAILQ_REMOVE(&intr_sources, src, next);
 845                                 rte_spinlock_unlock(&intr_lock);
 846
 847                                 for (cb = TAILQ_FIRST(&src->callbacks); cb;
 848                                                         cb = next) {
 849                                         next = TAILQ_NEXT(cb, next);
 850                                         TAILQ_REMOVE(&src->callbacks, cb, next);
 851                                         free(cb);
 852                                 }
 853                                 free(src);
 854                                 return -1;
 855                         } else if (bytes_read == 0)
 856                                 RTE_LOG(ERR, EAL, "Read nothing from file "
 857                                         "descriptor %d\n", events[n].data.fd);
 858                         else
 859                                 call = true;
 860                 }
 861
 862                 /* grab a lock, again to call callbacks and update status. */
 863                 rte_spinlock_lock(&intr_lock);
 864
 865                 if (call) {
 866
 867                         /* Finally, call all callbacks. */
 868                         TAILQ_FOREACH(cb, &src->callbacks, next) {
 869
 870                                 /* make a copy and unlock. */
 871                                 active_cb = *cb;
 872                                 rte_spinlock_unlock(&intr_lock);
 873
 874                                 /* call the actual callback */
 875                                 active_cb.cb_fn(active_cb.cb_arg);
 876
 877                                 /*get the lock back. */
 878                                 rte_spinlock_lock(&intr_lock);
 879                         }
 880                 }
 881                 /* we done with that interrupt source, release it. */
 882                 src->active = 0;
 883
 884                 rv = 0;
 885
 886                 /* check if any callback are supposed to be removed */
 887                 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
 888                         next = TAILQ_NEXT(cb, next);
 889                         if (cb->pending_delete) {
 890                                 TAILQ_REMOVE(&src->callbacks, cb, next);
 891                                 if (cb->ucb_fn)
 892                                         cb->ucb_fn(&src->intr_handle, cb->cb_arg);
 893                                 free(cb);
 894                                 rv++;
 895                         }
 896                 }
 897
 898                 /* all callbacks for that source are removed. */
 899                 if (TAILQ_EMPTY(&src->callbacks)) {
 900                         TAILQ_REMOVE(&intr_sources, src, next);
 901                         free(src);
 902                 }
 903
 904                 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
 905                 if (rv >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
 906                         rte_spinlock_unlock(&intr_lock);
 907                         return -EPIPE;
 908                 }
 909
 910                 rte_spinlock_unlock(&intr_lock);
 911         }
 912
 913         return 0;
 914 }
 915
 916 /**
 917  * It handles all the interrupts.
 918  *
 919  * @param pfd
 920  *  epoll file descriptor.
 921  * @param totalfds
 922  *  The number of file descriptors added in epoll.
 923  *
 924  * @return
 925  *  void
 926  */
 927 static void
 928 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
 929 {
 930         struct epoll_event events[totalfds];
 931         int nfds = 0;
 932
 933         for(;;) {
 934                 nfds = epoll_wait(pfd, events, totalfds,
 935                         EAL_INTR_EPOLL_WAIT_FOREVER);
 936                 /* epoll_wait fail */
 937                 if (nfds < 0) {
 938                         if (errno == EINTR)
 939                                 continue;
 940                         RTE_LOG(ERR, EAL,
 941                                 "epoll_wait returns with fail\n");
 942                         return;
 943                 }
 944                 /* epoll_wait timeout, will never happens here */
 945                 else if (nfds == 0)
 946                         continue;
 947                 /* epoll_wait has at least one fd ready to read */
 948                 if (eal_intr_process_interrupts(events, nfds) < 0)
 949                         return;
 950         }
 951 }
 952
 953 /**
 954  * It builds/rebuilds up the epoll file descriptor with all the
 955  * file descriptors being waited on. Then handles the interrupts.
 956  *
 957  * @param arg
 958  *  pointer. (unused)
 959  *
 960  * @return
 961  *  never return;
 962  */
 963 static __attribute__((noreturn)) void *
 964 eal_intr_thread_main(__rte_unused void *arg)
 965 {
 966         struct epoll_event ev;
 967
 968         /* host thread, never break out */
 969         for (;;) {
 970                 /* build up the epoll fd with all descriptors we are to
 971                  * wait on then pass it to the handle_interrupts function
 972                  */
 973                 static struct epoll_event pipe_event = {
 974                         .events = EPOLLIN | EPOLLPRI,
 975                 };
 976                 struct rte_intr_source *src;
 977                 unsigned numfds = 0;
 978
 979                 /* create epoll fd */
 980                 int pfd = epoll_create(1);
 981                 if (pfd < 0)
 982                         rte_panic("Cannot create epoll instance\n");
 983
 984                 pipe_event.data.fd = intr_pipe.readfd;
 985                 /**
 986                  * add pipe fd into wait list, this pipe is used to
 987                  * rebuild the wait list.
 988                  */
 989                 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
 990                                                 &pipe_event) < 0) {
 991                         rte_panic("Error adding fd to %d epoll_ctl, %s\n",
 992                                         intr_pipe.readfd, strerror(errno));
 993                 }
 994                 numfds++;
 995
 996                 rte_spinlock_lock(&intr_lock);
 997
 998                 TAILQ_FOREACH(src, &intr_sources, next) {
 999                         if (src->callbacks.tqh_first == NULL)
1000                                 continue; /* skip those with no callbacks */
1001                         ev.events = EPOLLIN | EPOLLPRI | EPOLLRDHUP | EPOLLHUP;
1002                         ev.data.fd = src->intr_handle.fd;
1003
1004                         /**
1005                          * add all the uio device file descriptor
1006                          * into wait list.
1007                          */
1008                         if (epoll_ctl(pfd, EPOLL_CTL_ADD,
1009                                         src->intr_handle.fd, &ev) < 0){
1010                                 rte_panic("Error adding fd %d epoll_ctl, %s\n",
1011                                         src->intr_handle.fd, strerror(errno));
1012                         }
1013                         else
1014                                 numfds++;
1015                 }
1016                 rte_spinlock_unlock(&intr_lock);
1017                 /* serve the interrupt */
1018                 eal_intr_handle_interrupts(pfd, numfds);
1019
1020                 /**
1021                  * when we return, we need to rebuild the
1022                  * list of fds to monitor.
1023                  */
1024                 close(pfd);
1025         }
1026 }
1027
1028 int
1029 rte_eal_intr_init(void)
1030 {
1031         int ret = 0;
1032
1033         /* init the global interrupt source head */
1034         TAILQ_INIT(&intr_sources);
1035
1036         /**
1037          * create a pipe which will be waited by epoll and notified to
1038          * rebuild the wait list of epoll.
1039          */
1040         if (pipe(intr_pipe.pipefd) < 0) {
1041                 rte_errno = errno;
1042                 return -1;
1043         }
1044
1045         /* create the host thread to wait/handle the interrupt */
1046         ret = rte_ctrl_thread_create(&intr_thread, "eal-intr-thread", NULL,
1047                         eal_intr_thread_main, NULL);
1048         if (ret != 0) {
1049                 rte_errno = -ret;
1050                 RTE_LOG(ERR, EAL,
1051                         "Failed to create thread for interrupt handling\n");
1052         }
1053
1054         return ret;
1055 }
1056
1057 static void
1058 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
1059 {
1060         union rte_intr_read_buffer buf;
1061         int bytes_read = 0;
1062         int nbytes;
1063
1064         switch (intr_handle->type) {
1065         case RTE_INTR_HANDLE_UIO:
1066         case RTE_INTR_HANDLE_UIO_INTX:
1067                 bytes_read = sizeof(buf.uio_intr_count);
1068                 break;
1069 #ifdef VFIO_PRESENT
1070         case RTE_INTR_HANDLE_VFIO_MSIX:
1071         case RTE_INTR_HANDLE_VFIO_MSI:
1072         case RTE_INTR_HANDLE_VFIO_LEGACY:
1073                 bytes_read = sizeof(buf.vfio_intr_count);
1074                 break;
1075 #endif
1076         case RTE_INTR_HANDLE_VDEV:
1077                 bytes_read = intr_handle->efd_counter_size;
1078                 /* For vdev, number of bytes to read is set by driver */
1079                 break;
1080         case RTE_INTR_HANDLE_EXT:
1081                 return;
1082         default:
1083                 bytes_read = 1;
1084                 RTE_LOG(INFO, EAL, "unexpected intr type\n");
1085                 break;
1086         }
1087
1088         /**
1089          * read out to clear the ready-to-be-read flag
1090          * for epoll_wait.
1091          */
1092         if (bytes_read == 0)
1093                 return;
1094         do {
1095                 nbytes = read(fd, &buf, bytes_read);
1096                 if (nbytes < 0) {
1097                         if (errno == EINTR || errno == EWOULDBLOCK ||
1098                             errno == EAGAIN)
1099                                 continue;
1100                         RTE_LOG(ERR, EAL,
1101                                 "Error reading from fd %d: %s\n",
1102                                 fd, strerror(errno));
1103                 } else if (nbytes == 0)
1104                         RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
1105                 return;
1106         } while (1);
1107 }
1108
1109 static int
1110 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
1111                         struct rte_epoll_event *events)
1112 {
1113         unsigned int i, count = 0;
1114         struct rte_epoll_event *rev;
1115
1116         for (i = 0; i < n; i++) {
1117                 rev = evs[i].data.ptr;
1118                 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
1119                                                  RTE_EPOLL_EXEC))
1120                         continue;
1121
1122                 events[count].status        = RTE_EPOLL_VALID;
1123                 events[count].fd            = rev->fd;
1124                 events[count].epfd          = rev->epfd;
1125                 events[count].epdata.event  = rev->epdata.event;
1126                 events[count].epdata.data   = rev->epdata.data;
1127                 if (rev->epdata.cb_fun)
1128                         rev->epdata.cb_fun(rev->fd,
1129                                            rev->epdata.cb_arg);
1130
1131                 rte_compiler_barrier();
1132                 rev->status = RTE_EPOLL_VALID;
1133                 count++;
1134         }
1135         return count;
1136 }
1137
1138 static inline int
1139 eal_init_tls_epfd(void)
1140 {
1141         int pfd = epoll_create(255);
1142
1143         if (pfd < 0) {
1144                 RTE_LOG(ERR, EAL,
1145                         "Cannot create epoll instance\n");
1146                 return -1;
1147         }
1148         return pfd;
1149 }
1150
1151 int
1152 rte_intr_tls_epfd(void)
1153 {
1154         if (RTE_PER_LCORE(_epfd) == -1)
1155                 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
1156
1157         return RTE_PER_LCORE(_epfd);
1158 }
1159
1160 int
1161 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
1162                int maxevents, int timeout)
1163 {
1164         struct epoll_event evs[maxevents];
1165         int rc;
1166
1167         if (!events) {
1168                 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1169                 return -1;
1170         }
1171
1172         /* using per thread epoll fd */
1173         if (epfd == RTE_EPOLL_PER_THREAD)
1174                 epfd = rte_intr_tls_epfd();
1175
1176         while (1) {
1177                 rc = epoll_wait(epfd, evs, maxevents, timeout);
1178                 if (likely(rc > 0)) {
1179                         /* epoll_wait has at least one fd ready to read */
1180                         rc = eal_epoll_process_event(evs, rc, events);
1181                         break;
1182                 } else if (rc < 0) {
1183                         if (errno == EINTR)
1184                                 continue;
1185                         /* epoll_wait fail */
1186                         RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1187                                 strerror(errno));
1188                         rc = -1;
1189                         break;
1190                 } else {
1191                         /* rc == 0, epoll_wait timed out */
1192                         break;
1193                 }
1194         }
1195
1196         return rc;
1197 }
1198
1199 static inline void
1200 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1201 {
1202         while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1203                                     RTE_EPOLL_INVALID))
1204                 while (ev->status != RTE_EPOLL_VALID)
1205                         rte_pause();
1206         memset(&ev->epdata, 0, sizeof(ev->epdata));
1207         ev->fd = -1;
1208         ev->epfd = -1;
1209 }
1210
1211 int
1212 rte_epoll_ctl(int epfd, int op, int fd,
1213               struct rte_epoll_event *event)
1214 {
1215         struct epoll_event ev;
1216
1217         if (!event) {
1218                 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1219                 return -1;
1220         }
1221
1222         /* using per thread epoll fd */
1223         if (epfd == RTE_EPOLL_PER_THREAD)
1224                 epfd = rte_intr_tls_epfd();
1225
1226         if (op == EPOLL_CTL_ADD) {
1227                 event->status = RTE_EPOLL_VALID;
1228                 event->fd = fd;  /* ignore fd in event */
1229                 event->epfd = epfd;
1230                 ev.data.ptr = (void *)event;
1231         }
1232
1233         ev.events = event->epdata.event;
1234         if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1235                 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1236                         op, fd, strerror(errno));
1237                 if (op == EPOLL_CTL_ADD)
1238                         /* rollback status when CTL_ADD fail */
1239                         event->status = RTE_EPOLL_INVALID;
1240                 return -1;
1241         }
1242
1243         if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1244                 eal_epoll_data_safe_free(event);
1245
1246         return 0;
1247 }
1248
1249 int
1250 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1251                 int op, unsigned int vec, void *data)
1252 {
1253         struct rte_epoll_event *rev;
1254         struct rte_epoll_data *epdata;
1255         int epfd_op;
1256         unsigned int efd_idx;
1257         int rc = 0;
1258
1259         efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1260                 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1261
1262         if (!intr_handle || intr_handle->nb_efd == 0 ||
1263             efd_idx >= intr_handle->nb_efd) {
1264                 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1265                 return -EPERM;
1266         }
1267
1268         switch (op) {
1269         case RTE_INTR_EVENT_ADD:
1270                 epfd_op = EPOLL_CTL_ADD;
1271                 rev = &intr_handle->elist[efd_idx];
1272                 if (rev->status != RTE_EPOLL_INVALID) {
1273                         RTE_LOG(INFO, EAL, "Event already been added.\n");
1274                         return -EEXIST;
1275                 }
1276
1277                 /* attach to intr vector fd */
1278                 epdata = &rev->epdata;
1279                 epdata->event  = EPOLLIN | EPOLLPRI | EPOLLET;
1280                 epdata->data   = data;
1281                 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1282                 epdata->cb_arg = (void *)intr_handle;
1283                 rc = rte_epoll_ctl(epfd, epfd_op,
1284                                    intr_handle->efds[efd_idx], rev);
1285                 if (!rc)
1286                         RTE_LOG(DEBUG, EAL,
1287                                 "efd %d associated with vec %d added on epfd %d"
1288                                 "\n", rev->fd, vec, epfd);
1289                 else
1290                         rc = -EPERM;
1291                 break;
1292         case RTE_INTR_EVENT_DEL:
1293                 epfd_op = EPOLL_CTL_DEL;
1294                 rev = &intr_handle->elist[efd_idx];
1295                 if (rev->status == RTE_EPOLL_INVALID) {
1296                         RTE_LOG(INFO, EAL, "Event does not exist.\n");
1297                         return -EPERM;
1298                 }
1299
1300                 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1301                 if (rc)
1302                         rc = -EPERM;
1303                 break;
1304         default:
1305                 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1306                 rc = -EPERM;
1307         }
1308
1309         return rc;
1310 }
1311
1312 void
1313 rte_intr_free_epoll_fd(struct rte_intr_handle *intr_handle)
1314 {
1315         uint32_t i;
1316         struct rte_epoll_event *rev;
1317
1318         for (i = 0; i < intr_handle->nb_efd; i++) {
1319                 rev = &intr_handle->elist[i];
1320                 if (rev->status == RTE_EPOLL_INVALID)
1321                         continue;
1322                 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1323                         /* force free if the entry valid */
1324                         eal_epoll_data_safe_free(rev);
1325                         rev->status = RTE_EPOLL_INVALID;
1326                 }
1327         }
1328 }
1329
1330 int
1331 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1332 {
1333         uint32_t i;
1334         int fd;
1335         uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1336
1337         assert(nb_efd != 0);
1338
1339         if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1340                 for (i = 0; i < n; i++) {
1341                         fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1342                         if (fd < 0) {
1343                                 RTE_LOG(ERR, EAL,
1344                                         "can't setup eventfd, error %i (%s)\n",
1345                                         errno, strerror(errno));
1346                                 return -errno;
1347                         }
1348                         intr_handle->efds[i] = fd;
1349                 }
1350                 intr_handle->nb_efd   = n;
1351                 intr_handle->max_intr = NB_OTHER_INTR + n;
1352         } else if (intr_handle->type == RTE_INTR_HANDLE_VDEV) {
1353                 /* only check, initialization would be done in vdev driver.*/
1354                 if (intr_handle->efd_counter_size >
1355                     sizeof(union rte_intr_read_buffer)) {
1356                         RTE_LOG(ERR, EAL, "the efd_counter_size is oversized");
1357                         return -EINVAL;
1358                 }
1359         } else {
1360                 intr_handle->efds[0]  = intr_handle->fd;
1361                 intr_handle->nb_efd   = RTE_MIN(nb_efd, 1U);
1362                 intr_handle->max_intr = NB_OTHER_INTR;
1363         }
1364
1365         return 0;
1366 }
1367
1368 void
1369 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1370 {
1371         uint32_t i;
1372
1373         rte_intr_free_epoll_fd(intr_handle);
1374         if (intr_handle->max_intr > intr_handle->nb_efd) {
1375                 for (i = 0; i < intr_handle->nb_efd; i++)
1376                         close(intr_handle->efds[i]);
1377         }
1378         intr_handle->nb_efd = 0;
1379         intr_handle->max_intr = 0;
1380 }
1381
1382 int
1383 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1384 {
1385         return !(!intr_handle->nb_efd);
1386 }
1387
1388 int
1389 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1390 {
1391         if (!rte_intr_dp_is_en(intr_handle))
1392                 return 1;
1393         else
1394                 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1395 }
1396
1397 int
1398 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1399 {
1400         if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)
1401                 return 1;
1402
1403         if (intr_handle->type == RTE_INTR_HANDLE_VDEV)
1404                 return 1;
1405
1406         return 0;
1407 }