linux/drivers/net/ethernet/mellanox/mlx5/core/health.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and/or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 */
  32
  33#include <linux/kernel.h>
  34#include <linux/module.h>
  35#include <linux/random.h>
  36#include <linux/vmalloc.h>
  37#include <linux/hardirq.h>
  38#include <linux/mlx5/driver.h>
  39#include "mlx5_core.h"
  40#include "lib/eq.h"
  41#include "lib/mlx5.h"
  42#include "lib/pci_vsc.h"
  43#include "diag/fw_tracer.h"
  44
  45enum {
  46        MLX5_HEALTH_POLL_INTERVAL       = 2 * HZ,
  47        MAX_MISSES                      = 3,
  48};
  49
  50enum {
  51        MLX5_HEALTH_SYNDR_FW_ERR                = 0x1,
  52        MLX5_HEALTH_SYNDR_IRISC_ERR             = 0x7,
  53        MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR  = 0x8,
  54        MLX5_HEALTH_SYNDR_CRC_ERR               = 0x9,
  55        MLX5_HEALTH_SYNDR_FETCH_PCI_ERR         = 0xa,
  56        MLX5_HEALTH_SYNDR_HW_FTL_ERR            = 0xb,
  57        MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR  = 0xc,
  58        MLX5_HEALTH_SYNDR_EQ_ERR                = 0xd,
  59        MLX5_HEALTH_SYNDR_EQ_INV                = 0xe,
  60        MLX5_HEALTH_SYNDR_FFSER_ERR             = 0xf,
  61        MLX5_HEALTH_SYNDR_HIGH_TEMP             = 0x10
  62};
  63
  64enum {
  65        MLX5_DROP_NEW_HEALTH_WORK,
  66};
  67
  68enum  {
  69        MLX5_SENSOR_NO_ERR              = 0,
  70        MLX5_SENSOR_PCI_COMM_ERR        = 1,
  71        MLX5_SENSOR_PCI_ERR             = 2,
  72        MLX5_SENSOR_NIC_DISABLED        = 3,
  73        MLX5_SENSOR_NIC_SW_RESET        = 4,
  74        MLX5_SENSOR_FW_SYND_RFR         = 5,
  75};
  76
  77u8 mlx5_get_nic_state(struct mlx5_core_dev *dev)
  78{
  79        return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
  80}
  81
  82void mlx5_set_nic_state(struct mlx5_core_dev *dev, u8 state)
  83{
  84        u32 cur_cmdq_addr_l_sz;
  85
  86        cur_cmdq_addr_l_sz = ioread32be(&dev->iseg->cmdq_addr_l_sz);
  87        iowrite32be((cur_cmdq_addr_l_sz & 0xFFFFF000) |
  88                    state << MLX5_NIC_IFC_OFFSET,
  89                    &dev->iseg->cmdq_addr_l_sz);
  90}
  91
  92static bool sensor_pci_not_working(struct mlx5_core_dev *dev)
  93{
  94        struct mlx5_core_health *health = &dev->priv.health;
  95        struct health_buffer __iomem *h = health->health;
  96
  97        /* Offline PCI reads return 0xffffffff */
  98        return (ioread32be(&h->fw_ver) == 0xffffffff);
  99}
 100
 101static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
 102{
 103        struct mlx5_core_health *health = &dev->priv.health;
 104        struct health_buffer __iomem *h = health->health;
 105        u32 rfr = ioread32be(&h->rfr) >> MLX5_RFR_OFFSET;
 106        u8 synd = ioread8(&h->synd);
 107
 108        if (rfr && synd)
 109                mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
 110        return rfr && synd;
 111}
 112
 113u32 mlx5_health_check_fatal_sensors(struct mlx5_core_dev *dev)
 114{
 115        if (sensor_pci_not_working(dev))
 116                return MLX5_SENSOR_PCI_COMM_ERR;
 117        if (pci_channel_offline(dev->pdev))
 118                return MLX5_SENSOR_PCI_ERR;
 119        if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
 120                return MLX5_SENSOR_NIC_DISABLED;
 121        if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_SW_RESET)
 122                return MLX5_SENSOR_NIC_SW_RESET;
 123        if (sensor_fw_synd_rfr(dev))
 124                return MLX5_SENSOR_FW_SYND_RFR;
 125
 126        return MLX5_SENSOR_NO_ERR;
 127}
 128
 129static int lock_sem_sw_reset(struct mlx5_core_dev *dev, bool lock)
 130{
 131        enum mlx5_vsc_state state;
 132        int ret;
 133
 134        if (!mlx5_core_is_pf(dev))
 135                return -EBUSY;
 136
 137        /* Try to lock GW access, this stage doesn't return
 138         * EBUSY because locked GW does not mean that other PF
 139         * already started the reset.
 140         */
 141        ret = mlx5_vsc_gw_lock(dev);
 142        if (ret == -EBUSY)
 143                return -EINVAL;
 144        if (ret)
 145                return ret;
 146
 147        state = lock ? MLX5_VSC_LOCK : MLX5_VSC_UNLOCK;
 148        /* At this stage, if the return status == EBUSY, then we know
 149         * for sure that another PF started the reset, so don't allow
 150         * another reset.
 151         */
 152        ret = mlx5_vsc_sem_set_space(dev, MLX5_SEMAPHORE_SW_RESET, state);
 153        if (ret)
 154                mlx5_core_warn(dev, "Failed to lock SW reset semaphore\n");
 155
 156        /* Unlock GW access */
 157        mlx5_vsc_gw_unlock(dev);
 158
 159        return ret;
 160}
 161
 162static bool reset_fw_if_needed(struct mlx5_core_dev *dev)
 163{
 164        bool supported = (ioread32be(&dev->iseg->initializing) >>
 165                          MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
 166        u32 fatal_error;
 167
 168        if (!supported)
 169                return false;
 170
 171        /* The reset only needs to be issued by one PF. The health buffer is
 172         * shared between all functions, and will be cleared during a reset.
 173         * Check again to avoid a redundant 2nd reset. If the fatal erros was
 174         * PCI related a reset won't help.
 175         */
 176        fatal_error = mlx5_health_check_fatal_sensors(dev);
 177        if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
 178            fatal_error == MLX5_SENSOR_NIC_DISABLED ||
 179            fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
 180                mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.");
 181                return false;
 182        }
 183
 184        mlx5_core_warn(dev, "Issuing FW Reset\n");
 185        /* Write the NIC interface field to initiate the reset, the command
 186         * interface address also resides here, don't overwrite it.
 187         */
 188        mlx5_set_nic_state(dev, MLX5_NIC_IFC_SW_RESET);
 189
 190        return true;
 191}
 192
 193void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
 194{
 195        bool err_detected = false;
 196
 197        /* Mark the device as fatal in order to abort FW commands */
 198        if ((mlx5_health_check_fatal_sensors(dev) || force) &&
 199            dev->state == MLX5_DEVICE_STATE_UP) {
 200                dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
 201                err_detected = true;
 202        }
 203        mutex_lock(&dev->intf_state_mutex);
 204        if (!err_detected && dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
 205                goto unlock;/* a previous error is still being handled */
 206        if (dev->state == MLX5_DEVICE_STATE_UNINITIALIZED) {
 207                dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
 208                goto unlock;
 209        }
 210
 211        if (mlx5_health_check_fatal_sensors(dev) || force) { /* protected state setting */
 212                dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
 213                mlx5_cmd_flush(dev);
 214        }
 215
 216        mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_SYS_ERROR, (void *)1);
 217unlock:
 218        mutex_unlock(&dev->intf_state_mutex);
 219}
 220
 221#define MLX5_CRDUMP_WAIT_MS     60000
 222#define MLX5_FW_RESET_WAIT_MS   1000
 223void mlx5_error_sw_reset(struct mlx5_core_dev *dev)
 224{
 225        unsigned long end, delay_ms = MLX5_FW_RESET_WAIT_MS;
 226        int lock = -EBUSY;
 227
 228        mutex_lock(&dev->intf_state_mutex);
 229        if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
 230                goto unlock;
 231
 232        mlx5_core_err(dev, "start\n");
 233
 234        if (mlx5_health_check_fatal_sensors(dev) == MLX5_SENSOR_FW_SYND_RFR) {
 235                /* Get cr-dump and reset FW semaphore */
 236                lock = lock_sem_sw_reset(dev, true);
 237
 238                if (lock == -EBUSY) {
 239                        delay_ms = MLX5_CRDUMP_WAIT_MS;
 240                        goto recover_from_sw_reset;
 241                }
 242                /* Execute SW reset */
 243                reset_fw_if_needed(dev);
 244        }
 245
 246recover_from_sw_reset:
 247        /* Recover from SW reset */
 248        end = jiffies + msecs_to_jiffies(delay_ms);
 249        do {
 250                if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
 251                        break;
 252
 253                msleep(20);
 254        } while (!time_after(jiffies, end));
 255
 256        if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
 257                dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
 258                        mlx5_get_nic_state(dev), delay_ms);
 259        }
 260
 261        /* Release FW semaphore if you are the lock owner */
 262        if (!lock)
 263                lock_sem_sw_reset(dev, false);
 264
 265        mlx5_core_err(dev, "end\n");
 266
 267unlock:
 268        mutex_unlock(&dev->intf_state_mutex);
 269}
 270
 271static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
 272{
 273        u8 nic_interface = mlx5_get_nic_state(dev);
 274
 275        switch (nic_interface) {
 276        case MLX5_NIC_IFC_FULL:
 277                mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
 278                break;
 279
 280        case MLX5_NIC_IFC_DISABLED:
 281                mlx5_core_warn(dev, "starting teardown\n");
 282                break;
 283
 284        case MLX5_NIC_IFC_NO_DRAM_NIC:
 285                mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
 286                break;
 287
 288        case MLX5_NIC_IFC_SW_RESET:
 289                /* The IFC mode field is 3 bits, so it will read 0x7 in 2 cases:
 290                 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
 291                 *    and this is a VF), this is not recoverable by SW reset.
 292                 *    Logging of this is handled elsewhere.
 293                 * 2. FW reset has been issued by another function, driver can
 294                 *    be reloaded to recover after the mode switches to
 295                 *    MLX5_NIC_IFC_DISABLED.
 296                 */
 297                if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
 298                        mlx5_core_warn(dev, "NIC SW reset in progress\n");
 299                break;
 300
 301        default:
 302                mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
 303                               nic_interface);
 304        }
 305
 306        mlx5_disable_device(dev);
 307}
 308
 309/* How much time to wait until health resetting the driver (in msecs) */
 310#define MLX5_RECOVERY_WAIT_MSECS 60000
 311int mlx5_health_wait_pci_up(struct mlx5_core_dev *dev)
 312{
 313        unsigned long end;
 314
 315        end = jiffies + msecs_to_jiffies(MLX5_RECOVERY_WAIT_MSECS);
 316        while (sensor_pci_not_working(dev)) {
 317                if (time_after(jiffies, end))
 318                        return -ETIMEDOUT;
 319                msleep(100);
 320        }
 321        return 0;
 322}
 323
 324static int mlx5_health_try_recover(struct mlx5_core_dev *dev)
 325{
 326        mlx5_core_warn(dev, "handling bad device here\n");
 327        mlx5_handle_bad_state(dev);
 328        if (mlx5_health_wait_pci_up(dev)) {
 329                mlx5_core_err(dev, "health recovery flow aborted, PCI reads still not working\n");
 330                return -EIO;
 331        }
 332        mlx5_core_err(dev, "starting health recovery flow\n");
 333        mlx5_recover_device(dev);
 334        if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state) ||
 335            mlx5_health_check_fatal_sensors(dev)) {
 336                mlx5_core_err(dev, "health recovery failed\n");
 337                return -EIO;
 338        }
 339        return 0;
 340}
 341
 342static const char *hsynd_str(u8 synd)
 343{
 344        switch (synd) {
 345        case MLX5_HEALTH_SYNDR_FW_ERR:
 346                return "firmware internal error";
 347        case MLX5_HEALTH_SYNDR_IRISC_ERR:
 348                return "irisc not responding";
 349        case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
 350                return "unrecoverable hardware error";
 351        case MLX5_HEALTH_SYNDR_CRC_ERR:
 352                return "firmware CRC error";
 353        case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
 354                return "ICM fetch PCI error";
 355        case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
 356                return "HW fatal error\n";
 357        case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
 358                return "async EQ buffer overrun";
 359        case MLX5_HEALTH_SYNDR_EQ_ERR:
 360                return "EQ error";
 361        case MLX5_HEALTH_SYNDR_EQ_INV:
 362                return "Invalid EQ referenced";
 363        case MLX5_HEALTH_SYNDR_FFSER_ERR:
 364                return "FFSER error";
 365        case MLX5_HEALTH_SYNDR_HIGH_TEMP:
 366                return "High temperature";
 367        default:
 368                return "unrecognized error";
 369        }
 370}
 371
 372static void print_health_info(struct mlx5_core_dev *dev)
 373{
 374        struct mlx5_core_health *health = &dev->priv.health;
 375        struct health_buffer __iomem *h = health->health;
 376        char fw_str[18];
 377        u32 fw;
 378        int i;
 379
 380        /* If the syndrome is 0, the device is OK and no need to print buffer */
 381        if (!ioread8(&h->synd))
 382                return;
 383
 384        for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
 385                mlx5_core_err(dev, "assert_var[%d] 0x%08x\n", i,
 386                              ioread32be(h->assert_var + i));
 387
 388        mlx5_core_err(dev, "assert_exit_ptr 0x%08x\n",
 389                      ioread32be(&h->assert_exit_ptr));
 390        mlx5_core_err(dev, "assert_callra 0x%08x\n",
 391                      ioread32be(&h->assert_callra));
 392        sprintf(fw_str, "%d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
 393        mlx5_core_err(dev, "fw_ver %s\n", fw_str);
 394        mlx5_core_err(dev, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
 395        mlx5_core_err(dev, "irisc_index %d\n", ioread8(&h->irisc_index));
 396        mlx5_core_err(dev, "synd 0x%x: %s\n", ioread8(&h->synd),
 397                      hsynd_str(ioread8(&h->synd)));
 398        mlx5_core_err(dev, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
 399        fw = ioread32be(&h->fw_ver);
 400        mlx5_core_err(dev, "raw fw_ver 0x%08x\n", fw);
 401}
 402
 403static int
 404mlx5_fw_reporter_diagnose(struct devlink_health_reporter *reporter,
 405                          struct devlink_fmsg *fmsg,
 406                          struct netlink_ext_ack *extack)
 407{
 408        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 409        struct mlx5_core_health *health = &dev->priv.health;
 410        struct health_buffer __iomem *h = health->health;
 411        u8 synd;
 412        int err;
 413
 414        synd = ioread8(&h->synd);
 415        err = devlink_fmsg_u8_pair_put(fmsg, "Syndrome", synd);
 416        if (err || !synd)
 417                return err;
 418        return devlink_fmsg_string_pair_put(fmsg, "Description", hsynd_str(synd));
 419}
 420
 421struct mlx5_fw_reporter_ctx {
 422        u8 err_synd;
 423        int miss_counter;
 424};
 425
 426static int
 427mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg *fmsg,
 428                               struct mlx5_fw_reporter_ctx *fw_reporter_ctx)
 429{
 430        int err;
 431
 432        err = devlink_fmsg_u8_pair_put(fmsg, "syndrome",
 433                                       fw_reporter_ctx->err_synd);
 434        if (err)
 435                return err;
 436        err = devlink_fmsg_u32_pair_put(fmsg, "fw_miss_counter",
 437                                        fw_reporter_ctx->miss_counter);
 438        if (err)
 439                return err;
 440        return 0;
 441}
 442
 443static int
 444mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev *dev,
 445                                       struct devlink_fmsg *fmsg)
 446{
 447        struct mlx5_core_health *health = &dev->priv.health;
 448        struct health_buffer __iomem *h = health->health;
 449        int err;
 450        int i;
 451
 452        if (!ioread8(&h->synd))
 453                return 0;
 454
 455        err = devlink_fmsg_pair_nest_start(fmsg, "health buffer");
 456        if (err)
 457                return err;
 458        err = devlink_fmsg_obj_nest_start(fmsg);
 459        if (err)
 460                return err;
 461        err = devlink_fmsg_arr_pair_nest_start(fmsg, "assert_var");
 462        if (err)
 463                return err;
 464
 465        for (i = 0; i < ARRAY_SIZE(h->assert_var); i++) {
 466                err = devlink_fmsg_u32_put(fmsg, ioread32be(h->assert_var + i));
 467                if (err)
 468                        return err;
 469        }
 470        err = devlink_fmsg_arr_pair_nest_end(fmsg);
 471        if (err)
 472                return err;
 473        err = devlink_fmsg_u32_pair_put(fmsg, "assert_exit_ptr",
 474                                        ioread32be(&h->assert_exit_ptr));
 475        if (err)
 476                return err;
 477        err = devlink_fmsg_u32_pair_put(fmsg, "assert_callra",
 478                                        ioread32be(&h->assert_callra));
 479        if (err)
 480                return err;
 481        err = devlink_fmsg_u32_pair_put(fmsg, "hw_id", ioread32be(&h->hw_id));
 482        if (err)
 483                return err;
 484        err = devlink_fmsg_u8_pair_put(fmsg, "irisc_index",
 485                                       ioread8(&h->irisc_index));
 486        if (err)
 487                return err;
 488        err = devlink_fmsg_u8_pair_put(fmsg, "synd", ioread8(&h->synd));
 489        if (err)
 490                return err;
 491        err = devlink_fmsg_u32_pair_put(fmsg, "ext_synd",
 492                                        ioread16be(&h->ext_synd));
 493        if (err)
 494                return err;
 495        err = devlink_fmsg_u32_pair_put(fmsg, "raw_fw_ver",
 496                                        ioread32be(&h->fw_ver));
 497        if (err)
 498                return err;
 499        err = devlink_fmsg_obj_nest_end(fmsg);
 500        if (err)
 501                return err;
 502        return devlink_fmsg_pair_nest_end(fmsg);
 503}
 504
 505static int
 506mlx5_fw_reporter_dump(struct devlink_health_reporter *reporter,
 507                      struct devlink_fmsg *fmsg, void *priv_ctx,
 508                      struct netlink_ext_ack *extack)
 509{
 510        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 511        int err;
 512
 513        err = mlx5_fw_tracer_trigger_core_dump_general(dev);
 514        if (err)
 515                return err;
 516
 517        if (priv_ctx) {
 518                struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
 519
 520                err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
 521                if (err)
 522                        return err;
 523        }
 524
 525        err = mlx5_fw_reporter_heath_buffer_data_put(dev, fmsg);
 526        if (err)
 527                return err;
 528        return mlx5_fw_tracer_get_saved_traces_objects(dev->tracer, fmsg);
 529}
 530
 531static void mlx5_fw_reporter_err_work(struct work_struct *work)
 532{
 533        struct mlx5_fw_reporter_ctx fw_reporter_ctx;
 534        struct mlx5_core_health *health;
 535
 536        health = container_of(work, struct mlx5_core_health, report_work);
 537
 538        if (IS_ERR_OR_NULL(health->fw_reporter))
 539                return;
 540
 541        fw_reporter_ctx.err_synd = health->synd;
 542        fw_reporter_ctx.miss_counter = health->miss_counter;
 543        if (fw_reporter_ctx.err_synd) {
 544                devlink_health_report(health->fw_reporter,
 545                                      "FW syndrom reported", &fw_reporter_ctx);
 546                return;
 547        }
 548        if (fw_reporter_ctx.miss_counter)
 549                devlink_health_report(health->fw_reporter,
 550                                      "FW miss counter reported",
 551                                      &fw_reporter_ctx);
 552}
 553
 554static const struct devlink_health_reporter_ops mlx5_fw_reporter_ops = {
 555                .name = "fw",
 556                .diagnose = mlx5_fw_reporter_diagnose,
 557                .dump = mlx5_fw_reporter_dump,
 558};
 559
 560static int
 561mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter *reporter,
 562                               void *priv_ctx,
 563                               struct netlink_ext_ack *extack)
 564{
 565        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 566
 567        return mlx5_health_try_recover(dev);
 568}
 569
 570static int
 571mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter *reporter,
 572                            struct devlink_fmsg *fmsg, void *priv_ctx,
 573                            struct netlink_ext_ack *extack)
 574{
 575        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 576        u32 crdump_size = dev->priv.health.crdump_size;
 577        u32 *cr_data;
 578        int err;
 579
 580        if (!mlx5_core_is_pf(dev))
 581                return -EPERM;
 582
 583        cr_data = kvmalloc(crdump_size, GFP_KERNEL);
 584        if (!cr_data)
 585                return -ENOMEM;
 586        err = mlx5_crdump_collect(dev, cr_data);
 587        if (err)
 588                goto free_data;
 589
 590        if (priv_ctx) {
 591                struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
 592
 593                err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
 594                if (err)
 595                        goto free_data;
 596        }
 597
 598        err = devlink_fmsg_binary_pair_put(fmsg, "crdump_data", cr_data, crdump_size);
 599
 600free_data:
 601        kvfree(cr_data);
 602        return err;
 603}
 604
 605static void mlx5_fw_fatal_reporter_err_work(struct work_struct *work)
 606{
 607        struct mlx5_fw_reporter_ctx fw_reporter_ctx;
 608        struct mlx5_core_health *health;
 609        struct mlx5_core_dev *dev;
 610        struct mlx5_priv *priv;
 611
 612        health = container_of(work, struct mlx5_core_health, fatal_report_work);
 613        priv = container_of(health, struct mlx5_priv, health);
 614        dev = container_of(priv, struct mlx5_core_dev, priv);
 615
 616        mlx5_enter_error_state(dev, false);
 617        if (IS_ERR_OR_NULL(health->fw_fatal_reporter)) {
 618                if (mlx5_health_try_recover(dev))
 619                        mlx5_core_err(dev, "health recovery failed\n");
 620                return;
 621        }
 622        fw_reporter_ctx.err_synd = health->synd;
 623        fw_reporter_ctx.miss_counter = health->miss_counter;
 624        devlink_health_report(health->fw_fatal_reporter,
 625                              "FW fatal error reported", &fw_reporter_ctx);
 626}
 627
 628static const struct devlink_health_reporter_ops mlx5_fw_fatal_reporter_ops = {
 629                .name = "fw_fatal",
 630                .recover = mlx5_fw_fatal_reporter_recover,
 631                .dump = mlx5_fw_fatal_reporter_dump,
 632};
 633
 634#define MLX5_REPORTER_FW_GRACEFUL_PERIOD 1200000
 635static void mlx5_fw_reporters_create(struct mlx5_core_dev *dev)
 636{
 637        struct mlx5_core_health *health = &dev->priv.health;
 638        struct devlink *devlink = priv_to_devlink(dev);
 639
 640        health->fw_reporter =
 641                devlink_health_reporter_create(devlink, &mlx5_fw_reporter_ops,
 642                                               0, dev);
 643        if (IS_ERR(health->fw_reporter))
 644                mlx5_core_warn(dev, "Failed to create fw reporter, err = %ld\n",
 645                               PTR_ERR(health->fw_reporter));
 646
 647        health->fw_fatal_reporter =
 648                devlink_health_reporter_create(devlink,
 649                                               &mlx5_fw_fatal_reporter_ops,
 650                                               MLX5_REPORTER_FW_GRACEFUL_PERIOD,
 651                                               dev);
 652        if (IS_ERR(health->fw_fatal_reporter))
 653                mlx5_core_warn(dev, "Failed to create fw fatal reporter, err = %ld\n",
 654                               PTR_ERR(health->fw_fatal_reporter));
 655}
 656
 657static void mlx5_fw_reporters_destroy(struct mlx5_core_dev *dev)
 658{
 659        struct mlx5_core_health *health = &dev->priv.health;
 660
 661        if (!IS_ERR_OR_NULL(health->fw_reporter))
 662                devlink_health_reporter_destroy(health->fw_reporter);
 663
 664        if (!IS_ERR_OR_NULL(health->fw_fatal_reporter))
 665                devlink_health_reporter_destroy(health->fw_fatal_reporter);
 666}
 667
 668static unsigned long get_next_poll_jiffies(void)
 669{
 670        unsigned long next;
 671
 672        get_random_bytes(&next, sizeof(next));
 673        next %= HZ;
 674        next += jiffies + MLX5_HEALTH_POLL_INTERVAL;
 675
 676        return next;
 677}
 678
 679void mlx5_trigger_health_work(struct mlx5_core_dev *dev)
 680{
 681        struct mlx5_core_health *health = &dev->priv.health;
 682        unsigned long flags;
 683
 684        spin_lock_irqsave(&health->wq_lock, flags);
 685        if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
 686                queue_work(health->wq, &health->fatal_report_work);
 687        else
 688                mlx5_core_err(dev, "new health works are not permitted at this stage\n");
 689        spin_unlock_irqrestore(&health->wq_lock, flags);
 690}
 691
 692static void poll_health(struct timer_list *t)
 693{
 694        struct mlx5_core_dev *dev = from_timer(dev, t, priv.health.timer);
 695        struct mlx5_core_health *health = &dev->priv.health;
 696        struct health_buffer __iomem *h = health->health;
 697        u32 fatal_error;
 698        u8 prev_synd;
 699        u32 count;
 700
 701        if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
 702                goto out;
 703
 704        fatal_error = mlx5_health_check_fatal_sensors(dev);
 705
 706        if (fatal_error && !health->fatal_error) {
 707                mlx5_core_err(dev, "Fatal error %u detected\n", fatal_error);
 708                dev->priv.health.fatal_error = fatal_error;
 709                print_health_info(dev);
 710                mlx5_trigger_health_work(dev);
 711                goto out;
 712        }
 713
 714        count = ioread32be(health->health_counter);
 715        if (count == health->prev)
 716                ++health->miss_counter;
 717        else
 718                health->miss_counter = 0;
 719
 720        health->prev = count;
 721        if (health->miss_counter == MAX_MISSES) {
 722                mlx5_core_err(dev, "device's health compromised - reached miss count\n");
 723                print_health_info(dev);
 724                queue_work(health->wq, &health->report_work);
 725        }
 726
 727        prev_synd = health->synd;
 728        health->synd = ioread8(&h->synd);
 729        if (health->synd && health->synd != prev_synd)
 730                queue_work(health->wq, &health->report_work);
 731
 732out:
 733        mod_timer(&health->timer, get_next_poll_jiffies());
 734}
 735
 736void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 737{
 738        struct mlx5_core_health *health = &dev->priv.health;
 739
 740        timer_setup(&health->timer, poll_health, 0);
 741        health->fatal_error = MLX5_SENSOR_NO_ERR;
 742        clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
 743        health->health = &dev->iseg->health;
 744        health->health_counter = &dev->iseg->health_counter;
 745
 746        health->timer.expires = round_jiffies(jiffies + MLX5_HEALTH_POLL_INTERVAL);
 747        add_timer(&health->timer);
 748}
 749
 750void mlx5_stop_health_poll(struct mlx5_core_dev *dev, bool disable_health)
 751{
 752        struct mlx5_core_health *health = &dev->priv.health;
 753        unsigned long flags;
 754
 755        if (disable_health) {
 756                spin_lock_irqsave(&health->wq_lock, flags);
 757                set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
 758                spin_unlock_irqrestore(&health->wq_lock, flags);
 759        }
 760
 761        del_timer_sync(&health->timer);
 762}
 763
 764void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
 765{
 766        struct mlx5_core_health *health = &dev->priv.health;
 767        unsigned long flags;
 768
 769        spin_lock_irqsave(&health->wq_lock, flags);
 770        set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
 771        spin_unlock_irqrestore(&health->wq_lock, flags);
 772        cancel_work_sync(&health->report_work);
 773        cancel_work_sync(&health->fatal_report_work);
 774}
 775
 776void mlx5_health_flush(struct mlx5_core_dev *dev)
 777{
 778        struct mlx5_core_health *health = &dev->priv.health;
 779
 780        flush_workqueue(health->wq);
 781}
 782
 783void mlx5_health_cleanup(struct mlx5_core_dev *dev)
 784{
 785        struct mlx5_core_health *health = &dev->priv.health;
 786
 787        destroy_workqueue(health->wq);
 788        mlx5_fw_reporters_destroy(dev);
 789}
 790
 791int mlx5_health_init(struct mlx5_core_dev *dev)
 792{
 793        struct mlx5_core_health *health;
 794        char *name;
 795
 796        mlx5_fw_reporters_create(dev);
 797
 798        health = &dev->priv.health;
 799        name = kmalloc(64, GFP_KERNEL);
 800        if (!name)
 801                goto out_err;
 802
 803        strcpy(name, "mlx5_health");
 804        strcat(name, dev_name(dev->device));
 805        health->wq = create_singlethread_workqueue(name);
 806        kfree(name);
 807        if (!health->wq)
 808                goto out_err;
 809        spin_lock_init(&health->wq_lock);
 810        INIT_WORK(&health->fatal_report_work, mlx5_fw_fatal_reporter_err_work);
 811        INIT_WORK(&health->report_work, mlx5_fw_reporter_err_work);
 812
 813        return 0;
 814
 815out_err:
 816        mlx5_fw_reporters_destroy(dev);
 817        return -ENOMEM;
 818}
 819