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 <linux/mlx5/cmd.h>
  40#include "mlx5_core.h"
  41#include "lib/eq.h"
  42#include "lib/mlx5.h"
  43#include "lib/pci_vsc.h"
  44#include "diag/fw_tracer.h"
  45
  46enum {
  47        MLX5_HEALTH_POLL_INTERVAL       = 2 * HZ,
  48        MAX_MISSES                      = 3,
  49};
  50
  51enum {
  52        MLX5_HEALTH_SYNDR_FW_ERR                = 0x1,
  53        MLX5_HEALTH_SYNDR_IRISC_ERR             = 0x7,
  54        MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR  = 0x8,
  55        MLX5_HEALTH_SYNDR_CRC_ERR               = 0x9,
  56        MLX5_HEALTH_SYNDR_FETCH_PCI_ERR         = 0xa,
  57        MLX5_HEALTH_SYNDR_HW_FTL_ERR            = 0xb,
  58        MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR  = 0xc,
  59        MLX5_HEALTH_SYNDR_EQ_ERR                = 0xd,
  60        MLX5_HEALTH_SYNDR_EQ_INV                = 0xe,
  61        MLX5_HEALTH_SYNDR_FFSER_ERR             = 0xf,
  62        MLX5_HEALTH_SYNDR_HIGH_TEMP             = 0x10
  63};
  64
  65enum {
  66        MLX5_DROP_NEW_HEALTH_WORK,
  67};
  68
  69enum  {
  70        MLX5_SENSOR_NO_ERR              = 0,
  71        MLX5_SENSOR_PCI_COMM_ERR        = 1,
  72        MLX5_SENSOR_PCI_ERR             = 2,
  73        MLX5_SENSOR_NIC_DISABLED        = 3,
  74        MLX5_SENSOR_NIC_SW_RESET        = 4,
  75        MLX5_SENSOR_FW_SYND_RFR         = 5,
  76};
  77
  78u8 mlx5_get_nic_state(struct mlx5_core_dev *dev)
  79{
  80        return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 7;
  81}
  82
  83void mlx5_set_nic_state(struct mlx5_core_dev *dev, u8 state)
  84{
  85        u32 cur_cmdq_addr_l_sz;
  86
  87        cur_cmdq_addr_l_sz = ioread32be(&dev->iseg->cmdq_addr_l_sz);
  88        iowrite32be((cur_cmdq_addr_l_sz & 0xFFFFF000) |
  89                    state << MLX5_NIC_IFC_OFFSET,
  90                    &dev->iseg->cmdq_addr_l_sz);
  91}
  92
  93static bool sensor_pci_not_working(struct mlx5_core_dev *dev)
  94{
  95        struct mlx5_core_health *health = &dev->priv.health;
  96        struct health_buffer __iomem *h = health->health;
  97
  98        /* Offline PCI reads return 0xffffffff */
  99        return (ioread32be(&h->fw_ver) == 0xffffffff);
 100}
 101
 102static bool sensor_fw_synd_rfr(struct mlx5_core_dev *dev)
 103{
 104        struct mlx5_core_health *health = &dev->priv.health;
 105        struct health_buffer __iomem *h = health->health;
 106        u32 rfr = ioread32be(&h->rfr) >> MLX5_RFR_OFFSET;
 107        u8 synd = ioread8(&h->synd);
 108
 109        if (rfr && synd)
 110                mlx5_core_dbg(dev, "FW requests reset, synd: %d\n", synd);
 111        return rfr && synd;
 112}
 113
 114static u32 check_fatal_sensors(struct mlx5_core_dev *dev)
 115{
 116        if (sensor_pci_not_working(dev))
 117                return MLX5_SENSOR_PCI_COMM_ERR;
 118        if (pci_channel_offline(dev->pdev))
 119                return MLX5_SENSOR_PCI_ERR;
 120        if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
 121                return MLX5_SENSOR_NIC_DISABLED;
 122        if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_SW_RESET)
 123                return MLX5_SENSOR_NIC_SW_RESET;
 124        if (sensor_fw_synd_rfr(dev))
 125                return MLX5_SENSOR_FW_SYND_RFR;
 126
 127        return MLX5_SENSOR_NO_ERR;
 128}
 129
 130static int lock_sem_sw_reset(struct mlx5_core_dev *dev, bool lock)
 131{
 132        enum mlx5_vsc_state state;
 133        int ret;
 134
 135        if (!mlx5_core_is_pf(dev))
 136                return -EBUSY;
 137
 138        /* Try to lock GW access, this stage doesn't return
 139         * EBUSY because locked GW does not mean that other PF
 140         * already started the reset.
 141         */
 142        ret = mlx5_vsc_gw_lock(dev);
 143        if (ret == -EBUSY)
 144                return -EINVAL;
 145        if (ret)
 146                return ret;
 147
 148        state = lock ? MLX5_VSC_LOCK : MLX5_VSC_UNLOCK;
 149        /* At this stage, if the return status == EBUSY, then we know
 150         * for sure that another PF started the reset, so don't allow
 151         * another reset.
 152         */
 153        ret = mlx5_vsc_sem_set_space(dev, MLX5_SEMAPHORE_SW_RESET, state);
 154        if (ret)
 155                mlx5_core_warn(dev, "Failed to lock SW reset semaphore\n");
 156
 157        /* Unlock GW access */
 158        mlx5_vsc_gw_unlock(dev);
 159
 160        return ret;
 161}
 162
 163static bool reset_fw_if_needed(struct mlx5_core_dev *dev)
 164{
 165        bool supported = (ioread32be(&dev->iseg->initializing) >>
 166                          MLX5_FW_RESET_SUPPORTED_OFFSET) & 1;
 167        u32 fatal_error;
 168
 169        if (!supported)
 170                return false;
 171
 172        /* The reset only needs to be issued by one PF. The health buffer is
 173         * shared between all functions, and will be cleared during a reset.
 174         * Check again to avoid a redundant 2nd reset. If the fatal erros was
 175         * PCI related a reset won't help.
 176         */
 177        fatal_error = check_fatal_sensors(dev);
 178        if (fatal_error == MLX5_SENSOR_PCI_COMM_ERR ||
 179            fatal_error == MLX5_SENSOR_NIC_DISABLED ||
 180            fatal_error == MLX5_SENSOR_NIC_SW_RESET) {
 181                mlx5_core_warn(dev, "Not issuing FW reset. Either it's already done or won't help.");
 182                return false;
 183        }
 184
 185        mlx5_core_warn(dev, "Issuing FW Reset\n");
 186        /* Write the NIC interface field to initiate the reset, the command
 187         * interface address also resides here, don't overwrite it.
 188         */
 189        mlx5_set_nic_state(dev, MLX5_NIC_IFC_SW_RESET);
 190
 191        return true;
 192}
 193
 194void mlx5_enter_error_state(struct mlx5_core_dev *dev, bool force)
 195{
 196        mutex_lock(&dev->intf_state_mutex);
 197        if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR)
 198                goto unlock;
 199        if (dev->state == MLX5_DEVICE_STATE_UNINITIALIZED) {
 200                dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
 201                goto unlock;
 202        }
 203
 204        if (check_fatal_sensors(dev) || force) {
 205                dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
 206                mlx5_cmd_flush(dev);
 207        }
 208
 209        mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_SYS_ERROR, (void *)1);
 210unlock:
 211        mutex_unlock(&dev->intf_state_mutex);
 212}
 213
 214#define MLX5_CRDUMP_WAIT_MS     60000
 215#define MLX5_FW_RESET_WAIT_MS   1000
 216void mlx5_error_sw_reset(struct mlx5_core_dev *dev)
 217{
 218        unsigned long end, delay_ms = MLX5_FW_RESET_WAIT_MS;
 219        int lock = -EBUSY;
 220
 221        mutex_lock(&dev->intf_state_mutex);
 222        if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
 223                goto unlock;
 224
 225        mlx5_core_err(dev, "start\n");
 226
 227        if (check_fatal_sensors(dev) == MLX5_SENSOR_FW_SYND_RFR) {
 228                /* Get cr-dump and reset FW semaphore */
 229                lock = lock_sem_sw_reset(dev, true);
 230
 231                if (lock == -EBUSY) {
 232                        delay_ms = MLX5_CRDUMP_WAIT_MS;
 233                        goto recover_from_sw_reset;
 234                }
 235                /* Execute SW reset */
 236                reset_fw_if_needed(dev);
 237        }
 238
 239recover_from_sw_reset:
 240        /* Recover from SW reset */
 241        end = jiffies + msecs_to_jiffies(delay_ms);
 242        do {
 243                if (mlx5_get_nic_state(dev) == MLX5_NIC_IFC_DISABLED)
 244                        break;
 245
 246                cond_resched();
 247        } while (!time_after(jiffies, end));
 248
 249        if (mlx5_get_nic_state(dev) != MLX5_NIC_IFC_DISABLED) {
 250                dev_err(&dev->pdev->dev, "NIC IFC still %d after %lums.\n",
 251                        mlx5_get_nic_state(dev), delay_ms);
 252        }
 253
 254        /* Release FW semaphore if you are the lock owner */
 255        if (!lock)
 256                lock_sem_sw_reset(dev, false);
 257
 258        mlx5_core_err(dev, "end\n");
 259
 260unlock:
 261        mutex_unlock(&dev->intf_state_mutex);
 262}
 263
 264static void mlx5_handle_bad_state(struct mlx5_core_dev *dev)
 265{
 266        u8 nic_interface = mlx5_get_nic_state(dev);
 267
 268        switch (nic_interface) {
 269        case MLX5_NIC_IFC_FULL:
 270                mlx5_core_warn(dev, "Expected to see disabled NIC but it is full driver\n");
 271                break;
 272
 273        case MLX5_NIC_IFC_DISABLED:
 274                mlx5_core_warn(dev, "starting teardown\n");
 275                break;
 276
 277        case MLX5_NIC_IFC_NO_DRAM_NIC:
 278                mlx5_core_warn(dev, "Expected to see disabled NIC but it is no dram nic\n");
 279                break;
 280
 281        case MLX5_NIC_IFC_SW_RESET:
 282                /* The IFC mode field is 3 bits, so it will read 0x7 in 2 cases:
 283                 * 1. PCI has been disabled (ie. PCI-AER, PF driver unloaded
 284                 *    and this is a VF), this is not recoverable by SW reset.
 285                 *    Logging of this is handled elsewhere.
 286                 * 2. FW reset has been issued by another function, driver can
 287                 *    be reloaded to recover after the mode switches to
 288                 *    MLX5_NIC_IFC_DISABLED.
 289                 */
 290                if (dev->priv.health.fatal_error != MLX5_SENSOR_PCI_COMM_ERR)
 291                        mlx5_core_warn(dev, "NIC SW reset in progress\n");
 292                break;
 293
 294        default:
 295                mlx5_core_warn(dev, "Expected to see disabled NIC but it is has invalid value %d\n",
 296                               nic_interface);
 297        }
 298
 299        mlx5_disable_device(dev);
 300}
 301
 302/* How much time to wait until health resetting the driver (in msecs) */
 303#define MLX5_RECOVERY_WAIT_MSECS 60000
 304static int mlx5_health_try_recover(struct mlx5_core_dev *dev)
 305{
 306        unsigned long end;
 307
 308        mlx5_core_warn(dev, "handling bad device here\n");
 309        mlx5_handle_bad_state(dev);
 310        end = jiffies + msecs_to_jiffies(MLX5_RECOVERY_WAIT_MSECS);
 311        while (sensor_pci_not_working(dev)) {
 312                if (time_after(jiffies, end)) {
 313                        mlx5_core_err(dev,
 314                                      "health recovery flow aborted, PCI reads still not working\n");
 315                        return -EIO;
 316                }
 317                msleep(100);
 318        }
 319
 320        mlx5_core_err(dev, "starting health recovery flow\n");
 321        mlx5_recover_device(dev);
 322        if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state) ||
 323            check_fatal_sensors(dev)) {
 324                mlx5_core_err(dev, "health recovery failed\n");
 325                return -EIO;
 326        }
 327        return 0;
 328}
 329
 330static const char *hsynd_str(u8 synd)
 331{
 332        switch (synd) {
 333        case MLX5_HEALTH_SYNDR_FW_ERR:
 334                return "firmware internal error";
 335        case MLX5_HEALTH_SYNDR_IRISC_ERR:
 336                return "irisc not responding";
 337        case MLX5_HEALTH_SYNDR_HW_UNRECOVERABLE_ERR:
 338                return "unrecoverable hardware error";
 339        case MLX5_HEALTH_SYNDR_CRC_ERR:
 340                return "firmware CRC error";
 341        case MLX5_HEALTH_SYNDR_FETCH_PCI_ERR:
 342                return "ICM fetch PCI error";
 343        case MLX5_HEALTH_SYNDR_HW_FTL_ERR:
 344                return "HW fatal error\n";
 345        case MLX5_HEALTH_SYNDR_ASYNC_EQ_OVERRUN_ERR:
 346                return "async EQ buffer overrun";
 347        case MLX5_HEALTH_SYNDR_EQ_ERR:
 348                return "EQ error";
 349        case MLX5_HEALTH_SYNDR_EQ_INV:
 350                return "Invalid EQ referenced";
 351        case MLX5_HEALTH_SYNDR_FFSER_ERR:
 352                return "FFSER error";
 353        case MLX5_HEALTH_SYNDR_HIGH_TEMP:
 354                return "High temperature";
 355        default:
 356                return "unrecognized error";
 357        }
 358}
 359
 360static void print_health_info(struct mlx5_core_dev *dev)
 361{
 362        struct mlx5_core_health *health = &dev->priv.health;
 363        struct health_buffer __iomem *h = health->health;
 364        char fw_str[18];
 365        u32 fw;
 366        int i;
 367
 368        /* If the syndrome is 0, the device is OK and no need to print buffer */
 369        if (!ioread8(&h->synd))
 370                return;
 371
 372        for (i = 0; i < ARRAY_SIZE(h->assert_var); i++)
 373                mlx5_core_err(dev, "assert_var[%d] 0x%08x\n", i,
 374                              ioread32be(h->assert_var + i));
 375
 376        mlx5_core_err(dev, "assert_exit_ptr 0x%08x\n",
 377                      ioread32be(&h->assert_exit_ptr));
 378        mlx5_core_err(dev, "assert_callra 0x%08x\n",
 379                      ioread32be(&h->assert_callra));
 380        sprintf(fw_str, "%d.%d.%d", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
 381        mlx5_core_err(dev, "fw_ver %s\n", fw_str);
 382        mlx5_core_err(dev, "hw_id 0x%08x\n", ioread32be(&h->hw_id));
 383        mlx5_core_err(dev, "irisc_index %d\n", ioread8(&h->irisc_index));
 384        mlx5_core_err(dev, "synd 0x%x: %s\n", ioread8(&h->synd),
 385                      hsynd_str(ioread8(&h->synd)));
 386        mlx5_core_err(dev, "ext_synd 0x%04x\n", ioread16be(&h->ext_synd));
 387        fw = ioread32be(&h->fw_ver);
 388        mlx5_core_err(dev, "raw fw_ver 0x%08x\n", fw);
 389}
 390
 391static int
 392mlx5_fw_reporter_diagnose(struct devlink_health_reporter *reporter,
 393                          struct devlink_fmsg *fmsg)
 394{
 395        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 396        struct mlx5_core_health *health = &dev->priv.health;
 397        struct health_buffer __iomem *h = health->health;
 398        u8 synd;
 399        int err;
 400
 401        synd = ioread8(&h->synd);
 402        err = devlink_fmsg_u8_pair_put(fmsg, "Syndrome", synd);
 403        if (err || !synd)
 404                return err;
 405        return devlink_fmsg_string_pair_put(fmsg, "Description", hsynd_str(synd));
 406}
 407
 408struct mlx5_fw_reporter_ctx {
 409        u8 err_synd;
 410        int miss_counter;
 411};
 412
 413static int
 414mlx5_fw_reporter_ctx_pairs_put(struct devlink_fmsg *fmsg,
 415                               struct mlx5_fw_reporter_ctx *fw_reporter_ctx)
 416{
 417        int err;
 418
 419        err = devlink_fmsg_u8_pair_put(fmsg, "syndrome",
 420                                       fw_reporter_ctx->err_synd);
 421        if (err)
 422                return err;
 423        err = devlink_fmsg_u32_pair_put(fmsg, "fw_miss_counter",
 424                                        fw_reporter_ctx->miss_counter);
 425        if (err)
 426                return err;
 427        return 0;
 428}
 429
 430static int
 431mlx5_fw_reporter_heath_buffer_data_put(struct mlx5_core_dev *dev,
 432                                       struct devlink_fmsg *fmsg)
 433{
 434        struct mlx5_core_health *health = &dev->priv.health;
 435        struct health_buffer __iomem *h = health->health;
 436        int err;
 437        int i;
 438
 439        if (!ioread8(&h->synd))
 440                return 0;
 441
 442        err = devlink_fmsg_pair_nest_start(fmsg, "health buffer");
 443        if (err)
 444                return err;
 445        err = devlink_fmsg_obj_nest_start(fmsg);
 446        if (err)
 447                return err;
 448        err = devlink_fmsg_arr_pair_nest_start(fmsg, "assert_var");
 449        if (err)
 450                return err;
 451
 452        for (i = 0; i < ARRAY_SIZE(h->assert_var); i++) {
 453                err = devlink_fmsg_u32_put(fmsg, ioread32be(h->assert_var + i));
 454                if (err)
 455                        return err;
 456        }
 457        err = devlink_fmsg_arr_pair_nest_end(fmsg);
 458        if (err)
 459                return err;
 460        err = devlink_fmsg_u32_pair_put(fmsg, "assert_exit_ptr",
 461                                        ioread32be(&h->assert_exit_ptr));
 462        if (err)
 463                return err;
 464        err = devlink_fmsg_u32_pair_put(fmsg, "assert_callra",
 465                                        ioread32be(&h->assert_callra));
 466        if (err)
 467                return err;
 468        err = devlink_fmsg_u32_pair_put(fmsg, "hw_id", ioread32be(&h->hw_id));
 469        if (err)
 470                return err;
 471        err = devlink_fmsg_u8_pair_put(fmsg, "irisc_index",
 472                                       ioread8(&h->irisc_index));
 473        if (err)
 474                return err;
 475        err = devlink_fmsg_u8_pair_put(fmsg, "synd", ioread8(&h->synd));
 476        if (err)
 477                return err;
 478        err = devlink_fmsg_u32_pair_put(fmsg, "ext_synd",
 479                                        ioread16be(&h->ext_synd));
 480        if (err)
 481                return err;
 482        err = devlink_fmsg_u32_pair_put(fmsg, "raw_fw_ver",
 483                                        ioread32be(&h->fw_ver));
 484        if (err)
 485                return err;
 486        err = devlink_fmsg_obj_nest_end(fmsg);
 487        if (err)
 488                return err;
 489        return devlink_fmsg_pair_nest_end(fmsg);
 490}
 491
 492static int
 493mlx5_fw_reporter_dump(struct devlink_health_reporter *reporter,
 494                      struct devlink_fmsg *fmsg, void *priv_ctx)
 495{
 496        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 497        int err;
 498
 499        err = mlx5_fw_tracer_trigger_core_dump_general(dev);
 500        if (err)
 501                return err;
 502
 503        if (priv_ctx) {
 504                struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
 505
 506                err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
 507                if (err)
 508                        return err;
 509        }
 510
 511        err = mlx5_fw_reporter_heath_buffer_data_put(dev, fmsg);
 512        if (err)
 513                return err;
 514        return mlx5_fw_tracer_get_saved_traces_objects(dev->tracer, fmsg);
 515}
 516
 517static void mlx5_fw_reporter_err_work(struct work_struct *work)
 518{
 519        struct mlx5_fw_reporter_ctx fw_reporter_ctx;
 520        struct mlx5_core_health *health;
 521
 522        health = container_of(work, struct mlx5_core_health, report_work);
 523
 524        if (IS_ERR_OR_NULL(health->fw_reporter))
 525                return;
 526
 527        fw_reporter_ctx.err_synd = health->synd;
 528        fw_reporter_ctx.miss_counter = health->miss_counter;
 529        if (fw_reporter_ctx.err_synd) {
 530                devlink_health_report(health->fw_reporter,
 531                                      "FW syndrom reported", &fw_reporter_ctx);
 532                return;
 533        }
 534        if (fw_reporter_ctx.miss_counter)
 535                devlink_health_report(health->fw_reporter,
 536                                      "FW miss counter reported",
 537                                      &fw_reporter_ctx);
 538}
 539
 540static const struct devlink_health_reporter_ops mlx5_fw_reporter_ops = {
 541                .name = "fw",
 542                .diagnose = mlx5_fw_reporter_diagnose,
 543                .dump = mlx5_fw_reporter_dump,
 544};
 545
 546static int
 547mlx5_fw_fatal_reporter_recover(struct devlink_health_reporter *reporter,
 548                               void *priv_ctx)
 549{
 550        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 551
 552        return mlx5_health_try_recover(dev);
 553}
 554
 555#define MLX5_CR_DUMP_CHUNK_SIZE 256
 556static int
 557mlx5_fw_fatal_reporter_dump(struct devlink_health_reporter *reporter,
 558                            struct devlink_fmsg *fmsg, void *priv_ctx)
 559{
 560        struct mlx5_core_dev *dev = devlink_health_reporter_priv(reporter);
 561        u32 crdump_size = dev->priv.health.crdump_size;
 562        u32 *cr_data;
 563        u32 data_size;
 564        u32 offset;
 565        int err;
 566
 567        if (!mlx5_core_is_pf(dev))
 568                return -EPERM;
 569
 570        cr_data = kvmalloc(crdump_size, GFP_KERNEL);
 571        if (!cr_data)
 572                return -ENOMEM;
 573        err = mlx5_crdump_collect(dev, cr_data);
 574        if (err)
 575                return err;
 576
 577        if (priv_ctx) {
 578                struct mlx5_fw_reporter_ctx *fw_reporter_ctx = priv_ctx;
 579
 580                err = mlx5_fw_reporter_ctx_pairs_put(fmsg, fw_reporter_ctx);
 581                if (err)
 582                        goto free_data;
 583        }
 584
 585        err = devlink_fmsg_arr_pair_nest_start(fmsg, "crdump_data");
 586        if (err)
 587                goto free_data;
 588        for (offset = 0; offset < crdump_size; offset += data_size) {
 589                if (crdump_size - offset < MLX5_CR_DUMP_CHUNK_SIZE)
 590                        data_size = crdump_size - offset;
 591                else
 592                        data_size = MLX5_CR_DUMP_CHUNK_SIZE;
 593                err = devlink_fmsg_binary_put(fmsg, (char *)cr_data + offset,
 594                                              data_size);
 595                if (err)
 596                        goto free_data;
 597        }
 598        err = devlink_fmsg_arr_pair_nest_end(fmsg);
 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, false, 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                                               true, 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 = 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