linux/drivers/scsi/cxlflash/main.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * CXL Flash Device Driver
   4 *
   5 * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation
   6 *             Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation
   7 *
   8 * Copyright (C) 2015 IBM Corporation
   9 */
  10
  11#include <linux/delay.h>
  12#include <linux/list.h>
  13#include <linux/module.h>
  14#include <linux/pci.h>
  15
  16#include <asm/unaligned.h>
  17
  18#include <scsi/scsi_cmnd.h>
  19#include <scsi/scsi_host.h>
  20#include <uapi/scsi/cxlflash_ioctl.h>
  21
  22#include "main.h"
  23#include "sislite.h"
  24#include "common.h"
  25
  26MODULE_DESCRIPTION(CXLFLASH_ADAPTER_NAME);
  27MODULE_AUTHOR("Manoj N. Kumar <manoj@linux.vnet.ibm.com>");
  28MODULE_AUTHOR("Matthew R. Ochs <mrochs@linux.vnet.ibm.com>");
  29MODULE_LICENSE("GPL");
  30
  31static struct class *cxlflash_class;
  32static u32 cxlflash_major;
  33static DECLARE_BITMAP(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
  34
  35/**
  36 * process_cmd_err() - command error handler
  37 * @cmd:        AFU command that experienced the error.
  38 * @scp:        SCSI command associated with the AFU command in error.
  39 *
  40 * Translates error bits from AFU command to SCSI command results.
  41 */
  42static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
  43{
  44        struct afu *afu = cmd->parent;
  45        struct cxlflash_cfg *cfg = afu->parent;
  46        struct device *dev = &cfg->dev->dev;
  47        struct sisl_ioasa *ioasa;
  48        u32 resid;
  49
  50        ioasa = &(cmd->sa);
  51
  52        if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
  53                resid = ioasa->resid;
  54                scsi_set_resid(scp, resid);
  55                dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
  56                        __func__, cmd, scp, resid);
  57        }
  58
  59        if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
  60                dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p\n",
  61                        __func__, cmd, scp);
  62                scp->result = (DID_ERROR << 16);
  63        }
  64
  65        dev_dbg(dev, "%s: cmd failed afu_rc=%02x scsi_rc=%02x fc_rc=%02x "
  66                "afu_extra=%02x scsi_extra=%02x fc_extra=%02x\n", __func__,
  67                ioasa->rc.afu_rc, ioasa->rc.scsi_rc, ioasa->rc.fc_rc,
  68                ioasa->afu_extra, ioasa->scsi_extra, ioasa->fc_extra);
  69
  70        if (ioasa->rc.scsi_rc) {
  71                /* We have a SCSI status */
  72                if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
  73                        memcpy(scp->sense_buffer, ioasa->sense_data,
  74                               SISL_SENSE_DATA_LEN);
  75                        scp->result = ioasa->rc.scsi_rc;
  76                } else
  77                        scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
  78        }
  79
  80        /*
  81         * We encountered an error. Set scp->result based on nature
  82         * of error.
  83         */
  84        if (ioasa->rc.fc_rc) {
  85                /* We have an FC status */
  86                switch (ioasa->rc.fc_rc) {
  87                case SISL_FC_RC_LINKDOWN:
  88                        scp->result = (DID_REQUEUE << 16);
  89                        break;
  90                case SISL_FC_RC_RESID:
  91                        /* This indicates an FCP resid underrun */
  92                        if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
  93                                /* If the SISL_RC_FLAGS_OVERRUN flag was set,
  94                                 * then we will handle this error else where.
  95                                 * If not then we must handle it here.
  96                                 * This is probably an AFU bug.
  97                                 */
  98                                scp->result = (DID_ERROR << 16);
  99                        }
 100                        break;
 101                case SISL_FC_RC_RESIDERR:
 102                        /* Resid mismatch between adapter and device */
 103                case SISL_FC_RC_TGTABORT:
 104                case SISL_FC_RC_ABORTOK:
 105                case SISL_FC_RC_ABORTFAIL:
 106                case SISL_FC_RC_NOLOGI:
 107                case SISL_FC_RC_ABORTPEND:
 108                case SISL_FC_RC_WRABORTPEND:
 109                case SISL_FC_RC_NOEXP:
 110                case SISL_FC_RC_INUSE:
 111                        scp->result = (DID_ERROR << 16);
 112                        break;
 113                }
 114        }
 115
 116        if (ioasa->rc.afu_rc) {
 117                /* We have an AFU error */
 118                switch (ioasa->rc.afu_rc) {
 119                case SISL_AFU_RC_NO_CHANNELS:
 120                        scp->result = (DID_NO_CONNECT << 16);
 121                        break;
 122                case SISL_AFU_RC_DATA_DMA_ERR:
 123                        switch (ioasa->afu_extra) {
 124                        case SISL_AFU_DMA_ERR_PAGE_IN:
 125                                /* Retry */
 126                                scp->result = (DID_IMM_RETRY << 16);
 127                                break;
 128                        case SISL_AFU_DMA_ERR_INVALID_EA:
 129                        default:
 130                                scp->result = (DID_ERROR << 16);
 131                        }
 132                        break;
 133                case SISL_AFU_RC_OUT_OF_DATA_BUFS:
 134                        /* Retry */
 135                        scp->result = (DID_ALLOC_FAILURE << 16);
 136                        break;
 137                default:
 138                        scp->result = (DID_ERROR << 16);
 139                }
 140        }
 141}
 142
 143/**
 144 * cmd_complete() - command completion handler
 145 * @cmd:        AFU command that has completed.
 146 *
 147 * For SCSI commands this routine prepares and submits commands that have
 148 * either completed or timed out to the SCSI stack. For internal commands
 149 * (TMF or AFU), this routine simply notifies the originator that the
 150 * command has completed.
 151 */
 152static void cmd_complete(struct afu_cmd *cmd)
 153{
 154        struct scsi_cmnd *scp;
 155        ulong lock_flags;
 156        struct afu *afu = cmd->parent;
 157        struct cxlflash_cfg *cfg = afu->parent;
 158        struct device *dev = &cfg->dev->dev;
 159        struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
 160
 161        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 162        list_del(&cmd->list);
 163        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 164
 165        if (cmd->scp) {
 166                scp = cmd->scp;
 167                if (unlikely(cmd->sa.ioasc))
 168                        process_cmd_err(cmd, scp);
 169                else
 170                        scp->result = (DID_OK << 16);
 171
 172                dev_dbg_ratelimited(dev, "%s:scp=%p result=%08x ioasc=%08x\n",
 173                                    __func__, scp, scp->result, cmd->sa.ioasc);
 174                scp->scsi_done(scp);
 175        } else if (cmd->cmd_tmf) {
 176                spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 177                cfg->tmf_active = false;
 178                wake_up_all_locked(&cfg->tmf_waitq);
 179                spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 180        } else
 181                complete(&cmd->cevent);
 182}
 183
 184/**
 185 * flush_pending_cmds() - flush all pending commands on this hardware queue
 186 * @hwq:        Hardware queue to flush.
 187 *
 188 * The hardware send queue lock associated with this hardware queue must be
 189 * held when calling this routine.
 190 */
 191static void flush_pending_cmds(struct hwq *hwq)
 192{
 193        struct cxlflash_cfg *cfg = hwq->afu->parent;
 194        struct afu_cmd *cmd, *tmp;
 195        struct scsi_cmnd *scp;
 196        ulong lock_flags;
 197
 198        list_for_each_entry_safe(cmd, tmp, &hwq->pending_cmds, list) {
 199                /* Bypass command when on a doneq, cmd_complete() will handle */
 200                if (!list_empty(&cmd->queue))
 201                        continue;
 202
 203                list_del(&cmd->list);
 204
 205                if (cmd->scp) {
 206                        scp = cmd->scp;
 207                        scp->result = (DID_IMM_RETRY << 16);
 208                        scp->scsi_done(scp);
 209                } else {
 210                        cmd->cmd_aborted = true;
 211
 212                        if (cmd->cmd_tmf) {
 213                                spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 214                                cfg->tmf_active = false;
 215                                wake_up_all_locked(&cfg->tmf_waitq);
 216                                spin_unlock_irqrestore(&cfg->tmf_slock,
 217                                                       lock_flags);
 218                        } else
 219                                complete(&cmd->cevent);
 220                }
 221        }
 222}
 223
 224/**
 225 * context_reset() - reset context via specified register
 226 * @hwq:        Hardware queue owning the context to be reset.
 227 * @reset_reg:  MMIO register to perform reset.
 228 *
 229 * When the reset is successful, the SISLite specification guarantees that
 230 * the AFU has aborted all currently pending I/O. Accordingly, these commands
 231 * must be flushed.
 232 *
 233 * Return: 0 on success, -errno on failure
 234 */
 235static int context_reset(struct hwq *hwq, __be64 __iomem *reset_reg)
 236{
 237        struct cxlflash_cfg *cfg = hwq->afu->parent;
 238        struct device *dev = &cfg->dev->dev;
 239        int rc = -ETIMEDOUT;
 240        int nretry = 0;
 241        u64 val = 0x1;
 242        ulong lock_flags;
 243
 244        dev_dbg(dev, "%s: hwq=%p\n", __func__, hwq);
 245
 246        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 247
 248        writeq_be(val, reset_reg);
 249        do {
 250                val = readq_be(reset_reg);
 251                if ((val & 0x1) == 0x0) {
 252                        rc = 0;
 253                        break;
 254                }
 255
 256                /* Double delay each time */
 257                udelay(1 << nretry);
 258        } while (nretry++ < MC_ROOM_RETRY_CNT);
 259
 260        if (!rc)
 261                flush_pending_cmds(hwq);
 262
 263        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 264
 265        dev_dbg(dev, "%s: returning rc=%d, val=%016llx nretry=%d\n",
 266                __func__, rc, val, nretry);
 267        return rc;
 268}
 269
 270/**
 271 * context_reset_ioarrin() - reset context via IOARRIN register
 272 * @hwq:        Hardware queue owning the context to be reset.
 273 *
 274 * Return: 0 on success, -errno on failure
 275 */
 276static int context_reset_ioarrin(struct hwq *hwq)
 277{
 278        return context_reset(hwq, &hwq->host_map->ioarrin);
 279}
 280
 281/**
 282 * context_reset_sq() - reset context via SQ_CONTEXT_RESET register
 283 * @hwq:        Hardware queue owning the context to be reset.
 284 *
 285 * Return: 0 on success, -errno on failure
 286 */
 287static int context_reset_sq(struct hwq *hwq)
 288{
 289        return context_reset(hwq, &hwq->host_map->sq_ctx_reset);
 290}
 291
 292/**
 293 * send_cmd_ioarrin() - sends an AFU command via IOARRIN register
 294 * @afu:        AFU associated with the host.
 295 * @cmd:        AFU command to send.
 296 *
 297 * Return:
 298 *      0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
 299 */
 300static int send_cmd_ioarrin(struct afu *afu, struct afu_cmd *cmd)
 301{
 302        struct cxlflash_cfg *cfg = afu->parent;
 303        struct device *dev = &cfg->dev->dev;
 304        struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
 305        int rc = 0;
 306        s64 room;
 307        ulong lock_flags;
 308
 309        /*
 310         * To avoid the performance penalty of MMIO, spread the update of
 311         * 'room' over multiple commands.
 312         */
 313        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 314        if (--hwq->room < 0) {
 315                room = readq_be(&hwq->host_map->cmd_room);
 316                if (room <= 0) {
 317                        dev_dbg_ratelimited(dev, "%s: no cmd_room to send "
 318                                            "0x%02X, room=0x%016llX\n",
 319                                            __func__, cmd->rcb.cdb[0], room);
 320                        hwq->room = 0;
 321                        rc = SCSI_MLQUEUE_HOST_BUSY;
 322                        goto out;
 323                }
 324                hwq->room = room - 1;
 325        }
 326
 327        list_add(&cmd->list, &hwq->pending_cmds);
 328        writeq_be((u64)&cmd->rcb, &hwq->host_map->ioarrin);
 329out:
 330        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 331        dev_dbg_ratelimited(dev, "%s: cmd=%p len=%u ea=%016llx rc=%d\n",
 332                __func__, cmd, cmd->rcb.data_len, cmd->rcb.data_ea, rc);
 333        return rc;
 334}
 335
 336/**
 337 * send_cmd_sq() - sends an AFU command via SQ ring
 338 * @afu:        AFU associated with the host.
 339 * @cmd:        AFU command to send.
 340 *
 341 * Return:
 342 *      0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
 343 */
 344static int send_cmd_sq(struct afu *afu, struct afu_cmd *cmd)
 345{
 346        struct cxlflash_cfg *cfg = afu->parent;
 347        struct device *dev = &cfg->dev->dev;
 348        struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
 349        int rc = 0;
 350        int newval;
 351        ulong lock_flags;
 352
 353        newval = atomic_dec_if_positive(&hwq->hsq_credits);
 354        if (newval <= 0) {
 355                rc = SCSI_MLQUEUE_HOST_BUSY;
 356                goto out;
 357        }
 358
 359        cmd->rcb.ioasa = &cmd->sa;
 360
 361        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 362
 363        *hwq->hsq_curr = cmd->rcb;
 364        if (hwq->hsq_curr < hwq->hsq_end)
 365                hwq->hsq_curr++;
 366        else
 367                hwq->hsq_curr = hwq->hsq_start;
 368
 369        list_add(&cmd->list, &hwq->pending_cmds);
 370        writeq_be((u64)hwq->hsq_curr, &hwq->host_map->sq_tail);
 371
 372        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 373out:
 374        dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx ioasa=%p rc=%d curr=%p "
 375               "head=%016llx tail=%016llx\n", __func__, cmd, cmd->rcb.data_len,
 376               cmd->rcb.data_ea, cmd->rcb.ioasa, rc, hwq->hsq_curr,
 377               readq_be(&hwq->host_map->sq_head),
 378               readq_be(&hwq->host_map->sq_tail));
 379        return rc;
 380}
 381
 382/**
 383 * wait_resp() - polls for a response or timeout to a sent AFU command
 384 * @afu:        AFU associated with the host.
 385 * @cmd:        AFU command that was sent.
 386 *
 387 * Return: 0 on success, -errno on failure
 388 */
 389static int wait_resp(struct afu *afu, struct afu_cmd *cmd)
 390{
 391        struct cxlflash_cfg *cfg = afu->parent;
 392        struct device *dev = &cfg->dev->dev;
 393        int rc = 0;
 394        ulong timeout = msecs_to_jiffies(cmd->rcb.timeout * 2 * 1000);
 395
 396        timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
 397        if (!timeout)
 398                rc = -ETIMEDOUT;
 399
 400        if (cmd->cmd_aborted)
 401                rc = -EAGAIN;
 402
 403        if (unlikely(cmd->sa.ioasc != 0)) {
 404                dev_err(dev, "%s: cmd %02x failed, ioasc=%08x\n",
 405                        __func__, cmd->rcb.cdb[0], cmd->sa.ioasc);
 406                rc = -EIO;
 407        }
 408
 409        return rc;
 410}
 411
 412/**
 413 * cmd_to_target_hwq() - selects a target hardware queue for a SCSI command
 414 * @host:       SCSI host associated with device.
 415 * @scp:        SCSI command to send.
 416 * @afu:        SCSI command to send.
 417 *
 418 * Hashes a command based upon the hardware queue mode.
 419 *
 420 * Return: Trusted index of target hardware queue
 421 */
 422static u32 cmd_to_target_hwq(struct Scsi_Host *host, struct scsi_cmnd *scp,
 423                             struct afu *afu)
 424{
 425        u32 tag;
 426        u32 hwq = 0;
 427
 428        if (afu->num_hwqs == 1)
 429                return 0;
 430
 431        switch (afu->hwq_mode) {
 432        case HWQ_MODE_RR:
 433                hwq = afu->hwq_rr_count++ % afu->num_hwqs;
 434                break;
 435        case HWQ_MODE_TAG:
 436                tag = blk_mq_unique_tag(scp->request);
 437                hwq = blk_mq_unique_tag_to_hwq(tag);
 438                break;
 439        case HWQ_MODE_CPU:
 440                hwq = smp_processor_id() % afu->num_hwqs;
 441                break;
 442        default:
 443                WARN_ON_ONCE(1);
 444        }
 445
 446        return hwq;
 447}
 448
 449/**
 450 * send_tmf() - sends a Task Management Function (TMF)
 451 * @cfg:        Internal structure associated with the host.
 452 * @sdev:       SCSI device destined for TMF.
 453 * @tmfcmd:     TMF command to send.
 454 *
 455 * Return:
 456 *      0 on success, SCSI_MLQUEUE_HOST_BUSY or -errno on failure
 457 */
 458static int send_tmf(struct cxlflash_cfg *cfg, struct scsi_device *sdev,
 459                    u64 tmfcmd)
 460{
 461        struct afu *afu = cfg->afu;
 462        struct afu_cmd *cmd = NULL;
 463        struct device *dev = &cfg->dev->dev;
 464        struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
 465        bool needs_deletion = false;
 466        char *buf = NULL;
 467        ulong lock_flags;
 468        int rc = 0;
 469        ulong to;
 470
 471        buf = kzalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
 472        if (unlikely(!buf)) {
 473                dev_err(dev, "%s: no memory for command\n", __func__);
 474                rc = -ENOMEM;
 475                goto out;
 476        }
 477
 478        cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
 479        INIT_LIST_HEAD(&cmd->queue);
 480
 481        /* When Task Management Function is active do not send another */
 482        spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 483        if (cfg->tmf_active)
 484                wait_event_interruptible_lock_irq(cfg->tmf_waitq,
 485                                                  !cfg->tmf_active,
 486                                                  cfg->tmf_slock);
 487        cfg->tmf_active = true;
 488        spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 489
 490        cmd->parent = afu;
 491        cmd->cmd_tmf = true;
 492        cmd->hwq_index = hwq->index;
 493
 494        cmd->rcb.ctx_id = hwq->ctx_hndl;
 495        cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
 496        cmd->rcb.port_sel = CHAN2PORTMASK(sdev->channel);
 497        cmd->rcb.lun_id = lun_to_lunid(sdev->lun);
 498        cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
 499                              SISL_REQ_FLAGS_SUP_UNDERRUN |
 500                              SISL_REQ_FLAGS_TMF_CMD);
 501        memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
 502
 503        rc = afu->send_cmd(afu, cmd);
 504        if (unlikely(rc)) {
 505                spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 506                cfg->tmf_active = false;
 507                spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 508                goto out;
 509        }
 510
 511        spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 512        to = msecs_to_jiffies(5000);
 513        to = wait_event_interruptible_lock_irq_timeout(cfg->tmf_waitq,
 514                                                       !cfg->tmf_active,
 515                                                       cfg->tmf_slock,
 516                                                       to);
 517        if (!to) {
 518                dev_err(dev, "%s: TMF timed out\n", __func__);
 519                rc = -ETIMEDOUT;
 520                needs_deletion = true;
 521        } else if (cmd->cmd_aborted) {
 522                dev_err(dev, "%s: TMF aborted\n", __func__);
 523                rc = -EAGAIN;
 524        } else if (cmd->sa.ioasc) {
 525                dev_err(dev, "%s: TMF failed ioasc=%08x\n",
 526                        __func__, cmd->sa.ioasc);
 527                rc = -EIO;
 528        }
 529        cfg->tmf_active = false;
 530        spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 531
 532        if (needs_deletion) {
 533                spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 534                list_del(&cmd->list);
 535                spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 536        }
 537out:
 538        kfree(buf);
 539        return rc;
 540}
 541
 542/**
 543 * cxlflash_driver_info() - information handler for this host driver
 544 * @host:       SCSI host associated with device.
 545 *
 546 * Return: A string describing the device.
 547 */
 548static const char *cxlflash_driver_info(struct Scsi_Host *host)
 549{
 550        return CXLFLASH_ADAPTER_NAME;
 551}
 552
 553/**
 554 * cxlflash_queuecommand() - sends a mid-layer request
 555 * @host:       SCSI host associated with device.
 556 * @scp:        SCSI command to send.
 557 *
 558 * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
 559 */
 560static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
 561{
 562        struct cxlflash_cfg *cfg = shost_priv(host);
 563        struct afu *afu = cfg->afu;
 564        struct device *dev = &cfg->dev->dev;
 565        struct afu_cmd *cmd = sc_to_afuci(scp);
 566        struct scatterlist *sg = scsi_sglist(scp);
 567        int hwq_index = cmd_to_target_hwq(host, scp, afu);
 568        struct hwq *hwq = get_hwq(afu, hwq_index);
 569        u16 req_flags = SISL_REQ_FLAGS_SUP_UNDERRUN;
 570        ulong lock_flags;
 571        int rc = 0;
 572
 573        dev_dbg_ratelimited(dev, "%s: (scp=%p) %d/%d/%d/%llu "
 574                            "cdb=(%08x-%08x-%08x-%08x)\n",
 575                            __func__, scp, host->host_no, scp->device->channel,
 576                            scp->device->id, scp->device->lun,
 577                            get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
 578                            get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
 579                            get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
 580                            get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
 581
 582        /*
 583         * If a Task Management Function is active, wait for it to complete
 584         * before continuing with regular commands.
 585         */
 586        spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 587        if (cfg->tmf_active) {
 588                spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 589                rc = SCSI_MLQUEUE_HOST_BUSY;
 590                goto out;
 591        }
 592        spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 593
 594        switch (cfg->state) {
 595        case STATE_PROBING:
 596        case STATE_PROBED:
 597        case STATE_RESET:
 598                dev_dbg_ratelimited(dev, "%s: device is in reset\n", __func__);
 599                rc = SCSI_MLQUEUE_HOST_BUSY;
 600                goto out;
 601        case STATE_FAILTERM:
 602                dev_dbg_ratelimited(dev, "%s: device has failed\n", __func__);
 603                scp->result = (DID_NO_CONNECT << 16);
 604                scp->scsi_done(scp);
 605                rc = 0;
 606                goto out;
 607        default:
 608                atomic_inc(&afu->cmds_active);
 609                break;
 610        }
 611
 612        if (likely(sg)) {
 613                cmd->rcb.data_len = sg->length;
 614                cmd->rcb.data_ea = (uintptr_t)sg_virt(sg);
 615        }
 616
 617        cmd->scp = scp;
 618        cmd->parent = afu;
 619        cmd->hwq_index = hwq_index;
 620
 621        cmd->sa.ioasc = 0;
 622        cmd->rcb.ctx_id = hwq->ctx_hndl;
 623        cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
 624        cmd->rcb.port_sel = CHAN2PORTMASK(scp->device->channel);
 625        cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
 626
 627        if (scp->sc_data_direction == DMA_TO_DEVICE)
 628                req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
 629
 630        cmd->rcb.req_flags = req_flags;
 631        memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
 632
 633        rc = afu->send_cmd(afu, cmd);
 634        atomic_dec(&afu->cmds_active);
 635out:
 636        return rc;
 637}
 638
 639/**
 640 * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
 641 * @cfg:        Internal structure associated with the host.
 642 */
 643static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
 644{
 645        struct pci_dev *pdev = cfg->dev;
 646
 647        if (pci_channel_offline(pdev))
 648                wait_event_timeout(cfg->reset_waitq,
 649                                   !pci_channel_offline(pdev),
 650                                   CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
 651}
 652
 653/**
 654 * free_mem() - free memory associated with the AFU
 655 * @cfg:        Internal structure associated with the host.
 656 */
 657static void free_mem(struct cxlflash_cfg *cfg)
 658{
 659        struct afu *afu = cfg->afu;
 660
 661        if (cfg->afu) {
 662                free_pages((ulong)afu, get_order(sizeof(struct afu)));
 663                cfg->afu = NULL;
 664        }
 665}
 666
 667/**
 668 * cxlflash_reset_sync() - synchronizing point for asynchronous resets
 669 * @cfg:        Internal structure associated with the host.
 670 */
 671static void cxlflash_reset_sync(struct cxlflash_cfg *cfg)
 672{
 673        if (cfg->async_reset_cookie == 0)
 674                return;
 675
 676        /* Wait until all async calls prior to this cookie have completed */
 677        async_synchronize_cookie(cfg->async_reset_cookie + 1);
 678        cfg->async_reset_cookie = 0;
 679}
 680
 681/**
 682 * stop_afu() - stops the AFU command timers and unmaps the MMIO space
 683 * @cfg:        Internal structure associated with the host.
 684 *
 685 * Safe to call with AFU in a partially allocated/initialized state.
 686 *
 687 * Cancels scheduled worker threads, waits for any active internal AFU
 688 * commands to timeout, disables IRQ polling and then unmaps the MMIO space.
 689 */
 690static void stop_afu(struct cxlflash_cfg *cfg)
 691{
 692        struct afu *afu = cfg->afu;
 693        struct hwq *hwq;
 694        int i;
 695
 696        cancel_work_sync(&cfg->work_q);
 697        if (!current_is_async())
 698                cxlflash_reset_sync(cfg);
 699
 700        if (likely(afu)) {
 701                while (atomic_read(&afu->cmds_active))
 702                        ssleep(1);
 703
 704                if (afu_is_irqpoll_enabled(afu)) {
 705                        for (i = 0; i < afu->num_hwqs; i++) {
 706                                hwq = get_hwq(afu, i);
 707
 708                                irq_poll_disable(&hwq->irqpoll);
 709                        }
 710                }
 711
 712                if (likely(afu->afu_map)) {
 713                        cfg->ops->psa_unmap(afu->afu_map);
 714                        afu->afu_map = NULL;
 715                }
 716        }
 717}
 718
 719/**
 720 * term_intr() - disables all AFU interrupts
 721 * @cfg:        Internal structure associated with the host.
 722 * @level:      Depth of allocation, where to begin waterfall tear down.
 723 * @index:      Index of the hardware queue.
 724 *
 725 * Safe to call with AFU/MC in partially allocated/initialized state.
 726 */
 727static void term_intr(struct cxlflash_cfg *cfg, enum undo_level level,
 728                      u32 index)
 729{
 730        struct afu *afu = cfg->afu;
 731        struct device *dev = &cfg->dev->dev;
 732        struct hwq *hwq;
 733
 734        if (!afu) {
 735                dev_err(dev, "%s: returning with NULL afu\n", __func__);
 736                return;
 737        }
 738
 739        hwq = get_hwq(afu, index);
 740
 741        if (!hwq->ctx_cookie) {
 742                dev_err(dev, "%s: returning with NULL MC\n", __func__);
 743                return;
 744        }
 745
 746        switch (level) {
 747        case UNMAP_THREE:
 748                /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
 749                if (index == PRIMARY_HWQ)
 750                        cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 3, hwq);
 751                fallthrough;
 752        case UNMAP_TWO:
 753                cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 2, hwq);
 754                fallthrough;
 755        case UNMAP_ONE:
 756                cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 1, hwq);
 757                fallthrough;
 758        case FREE_IRQ:
 759                cfg->ops->free_afu_irqs(hwq->ctx_cookie);
 760                fallthrough;
 761        case UNDO_NOOP:
 762                /* No action required */
 763                break;
 764        }
 765}
 766
 767/**
 768 * term_mc() - terminates the master context
 769 * @cfg:        Internal structure associated with the host.
 770 * @index:      Index of the hardware queue.
 771 *
 772 * Safe to call with AFU/MC in partially allocated/initialized state.
 773 */
 774static void term_mc(struct cxlflash_cfg *cfg, u32 index)
 775{
 776        struct afu *afu = cfg->afu;
 777        struct device *dev = &cfg->dev->dev;
 778        struct hwq *hwq;
 779        ulong lock_flags;
 780
 781        if (!afu) {
 782                dev_err(dev, "%s: returning with NULL afu\n", __func__);
 783                return;
 784        }
 785
 786        hwq = get_hwq(afu, index);
 787
 788        if (!hwq->ctx_cookie) {
 789                dev_err(dev, "%s: returning with NULL MC\n", __func__);
 790                return;
 791        }
 792
 793        WARN_ON(cfg->ops->stop_context(hwq->ctx_cookie));
 794        if (index != PRIMARY_HWQ)
 795                WARN_ON(cfg->ops->release_context(hwq->ctx_cookie));
 796        hwq->ctx_cookie = NULL;
 797
 798        spin_lock_irqsave(&hwq->hrrq_slock, lock_flags);
 799        hwq->hrrq_online = false;
 800        spin_unlock_irqrestore(&hwq->hrrq_slock, lock_flags);
 801
 802        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
 803        flush_pending_cmds(hwq);
 804        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
 805}
 806
 807/**
 808 * term_afu() - terminates the AFU
 809 * @cfg:        Internal structure associated with the host.
 810 *
 811 * Safe to call with AFU/MC in partially allocated/initialized state.
 812 */
 813static void term_afu(struct cxlflash_cfg *cfg)
 814{
 815        struct device *dev = &cfg->dev->dev;
 816        int k;
 817
 818        /*
 819         * Tear down is carefully orchestrated to ensure
 820         * no interrupts can come in when the problem state
 821         * area is unmapped.
 822         *
 823         * 1) Disable all AFU interrupts for each master
 824         * 2) Unmap the problem state area
 825         * 3) Stop each master context
 826         */
 827        for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
 828                term_intr(cfg, UNMAP_THREE, k);
 829
 830        stop_afu(cfg);
 831
 832        for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
 833                term_mc(cfg, k);
 834
 835        dev_dbg(dev, "%s: returning\n", __func__);
 836}
 837
 838/**
 839 * notify_shutdown() - notifies device of pending shutdown
 840 * @cfg:        Internal structure associated with the host.
 841 * @wait:       Whether to wait for shutdown processing to complete.
 842 *
 843 * This function will notify the AFU that the adapter is being shutdown
 844 * and will wait for shutdown processing to complete if wait is true.
 845 * This notification should flush pending I/Os to the device and halt
 846 * further I/Os until the next AFU reset is issued and device restarted.
 847 */
 848static void notify_shutdown(struct cxlflash_cfg *cfg, bool wait)
 849{
 850        struct afu *afu = cfg->afu;
 851        struct device *dev = &cfg->dev->dev;
 852        struct dev_dependent_vals *ddv;
 853        __be64 __iomem *fc_port_regs;
 854        u64 reg, status;
 855        int i, retry_cnt = 0;
 856
 857        ddv = (struct dev_dependent_vals *)cfg->dev_id->driver_data;
 858        if (!(ddv->flags & CXLFLASH_NOTIFY_SHUTDOWN))
 859                return;
 860
 861        if (!afu || !afu->afu_map) {
 862                dev_dbg(dev, "%s: Problem state area not mapped\n", __func__);
 863                return;
 864        }
 865
 866        /* Notify AFU */
 867        for (i = 0; i < cfg->num_fc_ports; i++) {
 868                fc_port_regs = get_fc_port_regs(cfg, i);
 869
 870                reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
 871                reg |= SISL_FC_SHUTDOWN_NORMAL;
 872                writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
 873        }
 874
 875        if (!wait)
 876                return;
 877
 878        /* Wait up to 1.5 seconds for shutdown processing to complete */
 879        for (i = 0; i < cfg->num_fc_ports; i++) {
 880                fc_port_regs = get_fc_port_regs(cfg, i);
 881                retry_cnt = 0;
 882
 883                while (true) {
 884                        status = readq_be(&fc_port_regs[FC_STATUS / 8]);
 885                        if (status & SISL_STATUS_SHUTDOWN_COMPLETE)
 886                                break;
 887                        if (++retry_cnt >= MC_RETRY_CNT) {
 888                                dev_dbg(dev, "%s: port %d shutdown processing "
 889                                        "not yet completed\n", __func__, i);
 890                                break;
 891                        }
 892                        msleep(100 * retry_cnt);
 893                }
 894        }
 895}
 896
 897/**
 898 * cxlflash_get_minor() - gets the first available minor number
 899 *
 900 * Return: Unique minor number that can be used to create the character device.
 901 */
 902static int cxlflash_get_minor(void)
 903{
 904        int minor;
 905        long bit;
 906
 907        bit = find_first_zero_bit(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
 908        if (bit >= CXLFLASH_MAX_ADAPTERS)
 909                return -1;
 910
 911        minor = bit & MINORMASK;
 912        set_bit(minor, cxlflash_minor);
 913        return minor;
 914}
 915
 916/**
 917 * cxlflash_put_minor() - releases the minor number
 918 * @minor:      Minor number that is no longer needed.
 919 */
 920static void cxlflash_put_minor(int minor)
 921{
 922        clear_bit(minor, cxlflash_minor);
 923}
 924
 925/**
 926 * cxlflash_release_chrdev() - release the character device for the host
 927 * @cfg:        Internal structure associated with the host.
 928 */
 929static void cxlflash_release_chrdev(struct cxlflash_cfg *cfg)
 930{
 931        device_unregister(cfg->chardev);
 932        cfg->chardev = NULL;
 933        cdev_del(&cfg->cdev);
 934        cxlflash_put_minor(MINOR(cfg->cdev.dev));
 935}
 936
 937/**
 938 * cxlflash_remove() - PCI entry point to tear down host
 939 * @pdev:       PCI device associated with the host.
 940 *
 941 * Safe to use as a cleanup in partially allocated/initialized state. Note that
 942 * the reset_waitq is flushed as part of the stop/termination of user contexts.
 943 */
 944static void cxlflash_remove(struct pci_dev *pdev)
 945{
 946        struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
 947        struct device *dev = &pdev->dev;
 948        ulong lock_flags;
 949
 950        if (!pci_is_enabled(pdev)) {
 951                dev_dbg(dev, "%s: Device is disabled\n", __func__);
 952                return;
 953        }
 954
 955        /* Yield to running recovery threads before continuing with remove */
 956        wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
 957                                     cfg->state != STATE_PROBING);
 958        spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
 959        if (cfg->tmf_active)
 960                wait_event_interruptible_lock_irq(cfg->tmf_waitq,
 961                                                  !cfg->tmf_active,
 962                                                  cfg->tmf_slock);
 963        spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
 964
 965        /* Notify AFU and wait for shutdown processing to complete */
 966        notify_shutdown(cfg, true);
 967
 968        cfg->state = STATE_FAILTERM;
 969        cxlflash_stop_term_user_contexts(cfg);
 970
 971        switch (cfg->init_state) {
 972        case INIT_STATE_CDEV:
 973                cxlflash_release_chrdev(cfg);
 974                fallthrough;
 975        case INIT_STATE_SCSI:
 976                cxlflash_term_local_luns(cfg);
 977                scsi_remove_host(cfg->host);
 978                fallthrough;
 979        case INIT_STATE_AFU:
 980                term_afu(cfg);
 981                fallthrough;
 982        case INIT_STATE_PCI:
 983                cfg->ops->destroy_afu(cfg->afu_cookie);
 984                pci_disable_device(pdev);
 985                fallthrough;
 986        case INIT_STATE_NONE:
 987                free_mem(cfg);
 988                scsi_host_put(cfg->host);
 989                break;
 990        }
 991
 992        dev_dbg(dev, "%s: returning\n", __func__);
 993}
 994
 995/**
 996 * alloc_mem() - allocates the AFU and its command pool
 997 * @cfg:        Internal structure associated with the host.
 998 *
 999 * A partially allocated state remains on failure.
1000 *
1001 * Return:
1002 *      0 on success
1003 *      -ENOMEM on failure to allocate memory
1004 */
1005static int alloc_mem(struct cxlflash_cfg *cfg)
1006{
1007        int rc = 0;
1008        struct device *dev = &cfg->dev->dev;
1009
1010        /* AFU is ~28k, i.e. only one 64k page or up to seven 4k pages */
1011        cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1012                                            get_order(sizeof(struct afu)));
1013        if (unlikely(!cfg->afu)) {
1014                dev_err(dev, "%s: cannot get %d free pages\n",
1015                        __func__, get_order(sizeof(struct afu)));
1016                rc = -ENOMEM;
1017                goto out;
1018        }
1019        cfg->afu->parent = cfg;
1020        cfg->afu->desired_hwqs = CXLFLASH_DEF_HWQS;
1021        cfg->afu->afu_map = NULL;
1022out:
1023        return rc;
1024}
1025
1026/**
1027 * init_pci() - initializes the host as a PCI device
1028 * @cfg:        Internal structure associated with the host.
1029 *
1030 * Return: 0 on success, -errno on failure
1031 */
1032static int init_pci(struct cxlflash_cfg *cfg)
1033{
1034        struct pci_dev *pdev = cfg->dev;
1035        struct device *dev = &cfg->dev->dev;
1036        int rc = 0;
1037
1038        rc = pci_enable_device(pdev);
1039        if (rc || pci_channel_offline(pdev)) {
1040                if (pci_channel_offline(pdev)) {
1041                        cxlflash_wait_for_pci_err_recovery(cfg);
1042                        rc = pci_enable_device(pdev);
1043                }
1044
1045                if (rc) {
1046                        dev_err(dev, "%s: Cannot enable adapter\n", __func__);
1047                        cxlflash_wait_for_pci_err_recovery(cfg);
1048                        goto out;
1049                }
1050        }
1051
1052out:
1053        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1054        return rc;
1055}
1056
1057/**
1058 * init_scsi() - adds the host to the SCSI stack and kicks off host scan
1059 * @cfg:        Internal structure associated with the host.
1060 *
1061 * Return: 0 on success, -errno on failure
1062 */
1063static int init_scsi(struct cxlflash_cfg *cfg)
1064{
1065        struct pci_dev *pdev = cfg->dev;
1066        struct device *dev = &cfg->dev->dev;
1067        int rc = 0;
1068
1069        rc = scsi_add_host(cfg->host, &pdev->dev);
1070        if (rc) {
1071                dev_err(dev, "%s: scsi_add_host failed rc=%d\n", __func__, rc);
1072                goto out;
1073        }
1074
1075        scsi_scan_host(cfg->host);
1076
1077out:
1078        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1079        return rc;
1080}
1081
1082/**
1083 * set_port_online() - transitions the specified host FC port to online state
1084 * @fc_regs:    Top of MMIO region defined for specified port.
1085 *
1086 * The provided MMIO region must be mapped prior to call. Online state means
1087 * that the FC link layer has synced, completed the handshaking process, and
1088 * is ready for login to start.
1089 */
1090static void set_port_online(__be64 __iomem *fc_regs)
1091{
1092        u64 cmdcfg;
1093
1094        cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1095        cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE); /* clear OFF_LINE */
1096        cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE);   /* set ON_LINE */
1097        writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1098}
1099
1100/**
1101 * set_port_offline() - transitions the specified host FC port to offline state
1102 * @fc_regs:    Top of MMIO region defined for specified port.
1103 *
1104 * The provided MMIO region must be mapped prior to call.
1105 */
1106static void set_port_offline(__be64 __iomem *fc_regs)
1107{
1108        u64 cmdcfg;
1109
1110        cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1111        cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE);  /* clear ON_LINE */
1112        cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE);  /* set OFF_LINE */
1113        writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1114}
1115
1116/**
1117 * wait_port_online() - waits for the specified host FC port come online
1118 * @fc_regs:    Top of MMIO region defined for specified port.
1119 * @delay_us:   Number of microseconds to delay between reading port status.
1120 * @nretry:     Number of cycles to retry reading port status.
1121 *
1122 * The provided MMIO region must be mapped prior to call. This will timeout
1123 * when the cable is not plugged in.
1124 *
1125 * Return:
1126 *      TRUE (1) when the specified port is online
1127 *      FALSE (0) when the specified port fails to come online after timeout
1128 */
1129static bool wait_port_online(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1130{
1131        u64 status;
1132
1133        WARN_ON(delay_us < 1000);
1134
1135        do {
1136                msleep(delay_us / 1000);
1137                status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1138                if (status == U64_MAX)
1139                        nretry /= 2;
1140        } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
1141                 nretry--);
1142
1143        return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
1144}
1145
1146/**
1147 * wait_port_offline() - waits for the specified host FC port go offline
1148 * @fc_regs:    Top of MMIO region defined for specified port.
1149 * @delay_us:   Number of microseconds to delay between reading port status.
1150 * @nretry:     Number of cycles to retry reading port status.
1151 *
1152 * The provided MMIO region must be mapped prior to call.
1153 *
1154 * Return:
1155 *      TRUE (1) when the specified port is offline
1156 *      FALSE (0) when the specified port fails to go offline after timeout
1157 */
1158static bool wait_port_offline(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1159{
1160        u64 status;
1161
1162        WARN_ON(delay_us < 1000);
1163
1164        do {
1165                msleep(delay_us / 1000);
1166                status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1167                if (status == U64_MAX)
1168                        nretry /= 2;
1169        } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1170                 nretry--);
1171
1172        return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
1173}
1174
1175/**
1176 * afu_set_wwpn() - configures the WWPN for the specified host FC port
1177 * @afu:        AFU associated with the host that owns the specified FC port.
1178 * @port:       Port number being configured.
1179 * @fc_regs:    Top of MMIO region defined for specified port.
1180 * @wwpn:       The world-wide-port-number previously discovered for port.
1181 *
1182 * The provided MMIO region must be mapped prior to call. As part of the
1183 * sequence to configure the WWPN, the port is toggled offline and then back
1184 * online. This toggling action can cause this routine to delay up to a few
1185 * seconds. When configured to use the internal LUN feature of the AFU, a
1186 * failure to come online is overridden.
1187 */
1188static void afu_set_wwpn(struct afu *afu, int port, __be64 __iomem *fc_regs,
1189                         u64 wwpn)
1190{
1191        struct cxlflash_cfg *cfg = afu->parent;
1192        struct device *dev = &cfg->dev->dev;
1193
1194        set_port_offline(fc_regs);
1195        if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1196                               FC_PORT_STATUS_RETRY_CNT)) {
1197                dev_dbg(dev, "%s: wait on port %d to go offline timed out\n",
1198                        __func__, port);
1199        }
1200
1201        writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1202
1203        set_port_online(fc_regs);
1204        if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1205                              FC_PORT_STATUS_RETRY_CNT)) {
1206                dev_dbg(dev, "%s: wait on port %d to go online timed out\n",
1207                        __func__, port);
1208        }
1209}
1210
1211/**
1212 * afu_link_reset() - resets the specified host FC port
1213 * @afu:        AFU associated with the host that owns the specified FC port.
1214 * @port:       Port number being configured.
1215 * @fc_regs:    Top of MMIO region defined for specified port.
1216 *
1217 * The provided MMIO region must be mapped prior to call. The sequence to
1218 * reset the port involves toggling it offline and then back online. This
1219 * action can cause this routine to delay up to a few seconds. An effort
1220 * is made to maintain link with the device by switching to host to use
1221 * the alternate port exclusively while the reset takes place.
1222 * failure to come online is overridden.
1223 */
1224static void afu_link_reset(struct afu *afu, int port, __be64 __iomem *fc_regs)
1225{
1226        struct cxlflash_cfg *cfg = afu->parent;
1227        struct device *dev = &cfg->dev->dev;
1228        u64 port_sel;
1229
1230        /* first switch the AFU to the other links, if any */
1231        port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1232        port_sel &= ~(1ULL << port);
1233        writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1234        cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1235
1236        set_port_offline(fc_regs);
1237        if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1238                               FC_PORT_STATUS_RETRY_CNT))
1239                dev_err(dev, "%s: wait on port %d to go offline timed out\n",
1240                        __func__, port);
1241
1242        set_port_online(fc_regs);
1243        if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1244                              FC_PORT_STATUS_RETRY_CNT))
1245                dev_err(dev, "%s: wait on port %d to go online timed out\n",
1246                        __func__, port);
1247
1248        /* switch back to include this port */
1249        port_sel |= (1ULL << port);
1250        writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1251        cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1252
1253        dev_dbg(dev, "%s: returning port_sel=%016llx\n", __func__, port_sel);
1254}
1255
1256/**
1257 * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1258 * @afu:        AFU associated with the host.
1259 */
1260static void afu_err_intr_init(struct afu *afu)
1261{
1262        struct cxlflash_cfg *cfg = afu->parent;
1263        __be64 __iomem *fc_port_regs;
1264        int i;
1265        struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
1266        u64 reg;
1267
1268        /* global async interrupts: AFU clears afu_ctrl on context exit
1269         * if async interrupts were sent to that context. This prevents
1270         * the AFU form sending further async interrupts when
1271         * there is
1272         * nobody to receive them.
1273         */
1274
1275        /* mask all */
1276        writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1277        /* set LISN# to send and point to primary master context */
1278        reg = ((u64) (((hwq->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1279
1280        if (afu->internal_lun)
1281                reg |= 1;       /* Bit 63 indicates local lun */
1282        writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1283        /* clear all */
1284        writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1285        /* unmask bits that are of interest */
1286        /* note: afu can send an interrupt after this step */
1287        writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1288        /* clear again in case a bit came on after previous clear but before */
1289        /* unmask */
1290        writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1291
1292        /* Clear/Set internal lun bits */
1293        fc_port_regs = get_fc_port_regs(cfg, 0);
1294        reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
1295        reg &= SISL_FC_INTERNAL_MASK;
1296        if (afu->internal_lun)
1297                reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1298        writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
1299
1300        /* now clear FC errors */
1301        for (i = 0; i < cfg->num_fc_ports; i++) {
1302                fc_port_regs = get_fc_port_regs(cfg, i);
1303
1304                writeq_be(0xFFFFFFFFU, &fc_port_regs[FC_ERROR / 8]);
1305                writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1306        }
1307
1308        /* sync interrupts for master's IOARRIN write */
1309        /* note that unlike asyncs, there can be no pending sync interrupts */
1310        /* at this time (this is a fresh context and master has not written */
1311        /* IOARRIN yet), so there is nothing to clear. */
1312
1313        /* set LISN#, it is always sent to the context that wrote IOARRIN */
1314        for (i = 0; i < afu->num_hwqs; i++) {
1315                hwq = get_hwq(afu, i);
1316
1317                reg = readq_be(&hwq->host_map->ctx_ctrl);
1318                WARN_ON((reg & SISL_CTX_CTRL_LISN_MASK) != 0);
1319                reg |= SISL_MSI_SYNC_ERROR;
1320                writeq_be(reg, &hwq->host_map->ctx_ctrl);
1321                writeq_be(SISL_ISTATUS_MASK, &hwq->host_map->intr_mask);
1322        }
1323}
1324
1325/**
1326 * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1327 * @irq:        Interrupt number.
1328 * @data:       Private data provided at interrupt registration, the AFU.
1329 *
1330 * Return: Always return IRQ_HANDLED.
1331 */
1332static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1333{
1334        struct hwq *hwq = (struct hwq *)data;
1335        struct cxlflash_cfg *cfg = hwq->afu->parent;
1336        struct device *dev = &cfg->dev->dev;
1337        u64 reg;
1338        u64 reg_unmasked;
1339
1340        reg = readq_be(&hwq->host_map->intr_status);
1341        reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1342
1343        if (reg_unmasked == 0UL) {
1344                dev_err(dev, "%s: spurious interrupt, intr_status=%016llx\n",
1345                        __func__, reg);
1346                goto cxlflash_sync_err_irq_exit;
1347        }
1348
1349        dev_err(dev, "%s: unexpected interrupt, intr_status=%016llx\n",
1350                __func__, reg);
1351
1352        writeq_be(reg_unmasked, &hwq->host_map->intr_clear);
1353
1354cxlflash_sync_err_irq_exit:
1355        return IRQ_HANDLED;
1356}
1357
1358/**
1359 * process_hrrq() - process the read-response queue
1360 * @hwq:        HWQ associated with the host.
1361 * @doneq:      Queue of commands harvested from the RRQ.
1362 * @budget:     Threshold of RRQ entries to process.
1363 *
1364 * This routine must be called holding the disabled RRQ spin lock.
1365 *
1366 * Return: The number of entries processed.
1367 */
1368static int process_hrrq(struct hwq *hwq, struct list_head *doneq, int budget)
1369{
1370        struct afu *afu = hwq->afu;
1371        struct afu_cmd *cmd;
1372        struct sisl_ioasa *ioasa;
1373        struct sisl_ioarcb *ioarcb;
1374        bool toggle = hwq->toggle;
1375        int num_hrrq = 0;
1376        u64 entry,
1377            *hrrq_start = hwq->hrrq_start,
1378            *hrrq_end = hwq->hrrq_end,
1379            *hrrq_curr = hwq->hrrq_curr;
1380
1381        /* Process ready RRQ entries up to the specified budget (if any) */
1382        while (true) {
1383                entry = *hrrq_curr;
1384
1385                if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1386                        break;
1387
1388                entry &= ~SISL_RESP_HANDLE_T_BIT;
1389
1390                if (afu_is_sq_cmd_mode(afu)) {
1391                        ioasa = (struct sisl_ioasa *)entry;
1392                        cmd = container_of(ioasa, struct afu_cmd, sa);
1393                } else {
1394                        ioarcb = (struct sisl_ioarcb *)entry;
1395                        cmd = container_of(ioarcb, struct afu_cmd, rcb);
1396                }
1397
1398                list_add_tail(&cmd->queue, doneq);
1399
1400                /* Advance to next entry or wrap and flip the toggle bit */
1401                if (hrrq_curr < hrrq_end)
1402                        hrrq_curr++;
1403                else {
1404                        hrrq_curr = hrrq_start;
1405                        toggle ^= SISL_RESP_HANDLE_T_BIT;
1406                }
1407
1408                atomic_inc(&hwq->hsq_credits);
1409                num_hrrq++;
1410
1411                if (budget > 0 && num_hrrq >= budget)
1412                        break;
1413        }
1414
1415        hwq->hrrq_curr = hrrq_curr;
1416        hwq->toggle = toggle;
1417
1418        return num_hrrq;
1419}
1420
1421/**
1422 * process_cmd_doneq() - process a queue of harvested RRQ commands
1423 * @doneq:      Queue of completed commands.
1424 *
1425 * Note that upon return the queue can no longer be trusted.
1426 */
1427static void process_cmd_doneq(struct list_head *doneq)
1428{
1429        struct afu_cmd *cmd, *tmp;
1430
1431        WARN_ON(list_empty(doneq));
1432
1433        list_for_each_entry_safe(cmd, tmp, doneq, queue)
1434                cmd_complete(cmd);
1435}
1436
1437/**
1438 * cxlflash_irqpoll() - process a queue of harvested RRQ commands
1439 * @irqpoll:    IRQ poll structure associated with queue to poll.
1440 * @budget:     Threshold of RRQ entries to process per poll.
1441 *
1442 * Return: The number of entries processed.
1443 */
1444static int cxlflash_irqpoll(struct irq_poll *irqpoll, int budget)
1445{
1446        struct hwq *hwq = container_of(irqpoll, struct hwq, irqpoll);
1447        unsigned long hrrq_flags;
1448        LIST_HEAD(doneq);
1449        int num_entries = 0;
1450
1451        spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1452
1453        num_entries = process_hrrq(hwq, &doneq, budget);
1454        if (num_entries < budget)
1455                irq_poll_complete(irqpoll);
1456
1457        spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1458
1459        process_cmd_doneq(&doneq);
1460        return num_entries;
1461}
1462
1463/**
1464 * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1465 * @irq:        Interrupt number.
1466 * @data:       Private data provided at interrupt registration, the AFU.
1467 *
1468 * Return: IRQ_HANDLED or IRQ_NONE when no ready entries found.
1469 */
1470static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1471{
1472        struct hwq *hwq = (struct hwq *)data;
1473        struct afu *afu = hwq->afu;
1474        unsigned long hrrq_flags;
1475        LIST_HEAD(doneq);
1476        int num_entries = 0;
1477
1478        spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1479
1480        /* Silently drop spurious interrupts when queue is not online */
1481        if (!hwq->hrrq_online) {
1482                spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1483                return IRQ_HANDLED;
1484        }
1485
1486        if (afu_is_irqpoll_enabled(afu)) {
1487                irq_poll_sched(&hwq->irqpoll);
1488                spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1489                return IRQ_HANDLED;
1490        }
1491
1492        num_entries = process_hrrq(hwq, &doneq, -1);
1493        spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1494
1495        if (num_entries == 0)
1496                return IRQ_NONE;
1497
1498        process_cmd_doneq(&doneq);
1499        return IRQ_HANDLED;
1500}
1501
1502/*
1503 * Asynchronous interrupt information table
1504 *
1505 * NOTE:
1506 *      - Order matters here as this array is indexed by bit position.
1507 *
1508 *      - The checkpatch script considers the BUILD_SISL_ASTATUS_FC_PORT macro
1509 *        as complex and complains due to a lack of parentheses/braces.
1510 */
1511#define ASTATUS_FC(_a, _b, _c, _d)                                       \
1512        { SISL_ASTATUS_FC##_a##_##_b, _c, _a, (_d) }
1513
1514#define BUILD_SISL_ASTATUS_FC_PORT(_a)                                   \
1515        ASTATUS_FC(_a, LINK_UP, "link up", 0),                           \
1516        ASTATUS_FC(_a, LINK_DN, "link down", 0),                         \
1517        ASTATUS_FC(_a, LOGI_S, "login succeeded", SCAN_HOST),            \
1518        ASTATUS_FC(_a, LOGI_F, "login failed", CLR_FC_ERROR),            \
1519        ASTATUS_FC(_a, LOGI_R, "login timed out, retrying", LINK_RESET), \
1520        ASTATUS_FC(_a, CRC_T, "CRC threshold exceeded", LINK_RESET),     \
1521        ASTATUS_FC(_a, LOGO, "target initiated LOGO", 0),                \
1522        ASTATUS_FC(_a, OTHER, "other error", CLR_FC_ERROR | LINK_RESET)
1523
1524static const struct asyc_intr_info ainfo[] = {
1525        BUILD_SISL_ASTATUS_FC_PORT(1),
1526        BUILD_SISL_ASTATUS_FC_PORT(0),
1527        BUILD_SISL_ASTATUS_FC_PORT(3),
1528        BUILD_SISL_ASTATUS_FC_PORT(2)
1529};
1530
1531/**
1532 * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1533 * @irq:        Interrupt number.
1534 * @data:       Private data provided at interrupt registration, the AFU.
1535 *
1536 * Return: Always return IRQ_HANDLED.
1537 */
1538static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1539{
1540        struct hwq *hwq = (struct hwq *)data;
1541        struct afu *afu = hwq->afu;
1542        struct cxlflash_cfg *cfg = afu->parent;
1543        struct device *dev = &cfg->dev->dev;
1544        const struct asyc_intr_info *info;
1545        struct sisl_global_map __iomem *global = &afu->afu_map->global;
1546        __be64 __iomem *fc_port_regs;
1547        u64 reg_unmasked;
1548        u64 reg;
1549        u64 bit;
1550        u8 port;
1551
1552        reg = readq_be(&global->regs.aintr_status);
1553        reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1554
1555        if (unlikely(reg_unmasked == 0)) {
1556                dev_err(dev, "%s: spurious interrupt, aintr_status=%016llx\n",
1557                        __func__, reg);
1558                goto out;
1559        }
1560
1561        /* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1562        writeq_be(reg_unmasked, &global->regs.aintr_clear);
1563
1564        /* Check each bit that is on */
1565        for_each_set_bit(bit, (ulong *)&reg_unmasked, BITS_PER_LONG) {
1566                if (unlikely(bit >= ARRAY_SIZE(ainfo))) {
1567                        WARN_ON_ONCE(1);
1568                        continue;
1569                }
1570
1571                info = &ainfo[bit];
1572                if (unlikely(info->status != 1ULL << bit)) {
1573                        WARN_ON_ONCE(1);
1574                        continue;
1575                }
1576
1577                port = info->port;
1578                fc_port_regs = get_fc_port_regs(cfg, port);
1579
1580                dev_err(dev, "%s: FC Port %d -> %s, fc_status=%016llx\n",
1581                        __func__, port, info->desc,
1582                       readq_be(&fc_port_regs[FC_STATUS / 8]));
1583
1584                /*
1585                 * Do link reset first, some OTHER errors will set FC_ERROR
1586                 * again if cleared before or w/o a reset
1587                 */
1588                if (info->action & LINK_RESET) {
1589                        dev_err(dev, "%s: FC Port %d: resetting link\n",
1590                                __func__, port);
1591                        cfg->lr_state = LINK_RESET_REQUIRED;
1592                        cfg->lr_port = port;
1593                        schedule_work(&cfg->work_q);
1594                }
1595
1596                if (info->action & CLR_FC_ERROR) {
1597                        reg = readq_be(&fc_port_regs[FC_ERROR / 8]);
1598
1599                        /*
1600                         * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1601                         * should be the same and tracing one is sufficient.
1602                         */
1603
1604                        dev_err(dev, "%s: fc %d: clearing fc_error=%016llx\n",
1605                                __func__, port, reg);
1606
1607                        writeq_be(reg, &fc_port_regs[FC_ERROR / 8]);
1608                        writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1609                }
1610
1611                if (info->action & SCAN_HOST) {
1612                        atomic_inc(&cfg->scan_host_needed);
1613                        schedule_work(&cfg->work_q);
1614                }
1615        }
1616
1617out:
1618        return IRQ_HANDLED;
1619}
1620
1621/**
1622 * read_vpd() - obtains the WWPNs from VPD
1623 * @cfg:        Internal structure associated with the host.
1624 * @wwpn:       Array of size MAX_FC_PORTS to pass back WWPNs
1625 *
1626 * Return: 0 on success, -errno on failure
1627 */
1628static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1629{
1630        struct device *dev = &cfg->dev->dev;
1631        struct pci_dev *pdev = cfg->dev;
1632        int rc = 0;
1633        int ro_start, ro_size, i, j, k;
1634        ssize_t vpd_size;
1635        char vpd_data[CXLFLASH_VPD_LEN];
1636        char tmp_buf[WWPN_BUF_LEN] = { 0 };
1637        const struct dev_dependent_vals *ddv = (struct dev_dependent_vals *)
1638                                                cfg->dev_id->driver_data;
1639        const bool wwpn_vpd_required = ddv->flags & CXLFLASH_WWPN_VPD_REQUIRED;
1640        const char *wwpn_vpd_tags[MAX_FC_PORTS] = { "V5", "V6", "V7", "V8" };
1641
1642        /* Get the VPD data from the device */
1643        vpd_size = cfg->ops->read_adapter_vpd(pdev, vpd_data, sizeof(vpd_data));
1644        if (unlikely(vpd_size <= 0)) {
1645                dev_err(dev, "%s: Unable to read VPD (size = %ld)\n",
1646                        __func__, vpd_size);
1647                rc = -ENODEV;
1648                goto out;
1649        }
1650
1651        /* Get the read only section offset */
1652        ro_start = pci_vpd_find_tag(vpd_data, vpd_size, PCI_VPD_LRDT_RO_DATA);
1653        if (unlikely(ro_start < 0)) {
1654                dev_err(dev, "%s: VPD Read-only data not found\n", __func__);
1655                rc = -ENODEV;
1656                goto out;
1657        }
1658
1659        /* Get the read only section size, cap when extends beyond read VPD */
1660        ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1661        j = ro_size;
1662        i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1663        if (unlikely((i + j) > vpd_size)) {
1664                dev_dbg(dev, "%s: Might need to read more VPD (%d > %ld)\n",
1665                        __func__, (i + j), vpd_size);
1666                ro_size = vpd_size - i;
1667        }
1668
1669        /*
1670         * Find the offset of the WWPN tag within the read only
1671         * VPD data and validate the found field (partials are
1672         * no good to us). Convert the ASCII data to an integer
1673         * value. Note that we must copy to a temporary buffer
1674         * because the conversion service requires that the ASCII
1675         * string be terminated.
1676         *
1677         * Allow for WWPN not being found for all devices, setting
1678         * the returned WWPN to zero when not found. Notify with a
1679         * log error for cards that should have had WWPN keywords
1680         * in the VPD - cards requiring WWPN will not have their
1681         * ports programmed and operate in an undefined state.
1682         */
1683        for (k = 0; k < cfg->num_fc_ports; k++) {
1684                j = ro_size;
1685                i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1686
1687                i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1688                if (i < 0) {
1689                        if (wwpn_vpd_required)
1690                                dev_err(dev, "%s: Port %d WWPN not found\n",
1691                                        __func__, k);
1692                        wwpn[k] = 0ULL;
1693                        continue;
1694                }
1695
1696                j = pci_vpd_info_field_size(&vpd_data[i]);
1697                i += PCI_VPD_INFO_FLD_HDR_SIZE;
1698                if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1699                        dev_err(dev, "%s: Port %d WWPN incomplete or bad VPD\n",
1700                                __func__, k);
1701                        rc = -ENODEV;
1702                        goto out;
1703                }
1704
1705                memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1706                rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1707                if (unlikely(rc)) {
1708                        dev_err(dev, "%s: WWPN conversion failed for port %d\n",
1709                                __func__, k);
1710                        rc = -ENODEV;
1711                        goto out;
1712                }
1713
1714                dev_dbg(dev, "%s: wwpn%d=%016llx\n", __func__, k, wwpn[k]);
1715        }
1716
1717out:
1718        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1719        return rc;
1720}
1721
1722/**
1723 * init_pcr() - initialize the provisioning and control registers
1724 * @cfg:        Internal structure associated with the host.
1725 *
1726 * Also sets up fast access to the mapped registers and initializes AFU
1727 * command fields that never change.
1728 */
1729static void init_pcr(struct cxlflash_cfg *cfg)
1730{
1731        struct afu *afu = cfg->afu;
1732        struct sisl_ctrl_map __iomem *ctrl_map;
1733        struct hwq *hwq;
1734        void *cookie;
1735        int i;
1736
1737        for (i = 0; i < MAX_CONTEXT; i++) {
1738                ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1739                /* Disrupt any clients that could be running */
1740                /* e.g. clients that survived a master restart */
1741                writeq_be(0, &ctrl_map->rht_start);
1742                writeq_be(0, &ctrl_map->rht_cnt_id);
1743                writeq_be(0, &ctrl_map->ctx_cap);
1744        }
1745
1746        /* Copy frequently used fields into hwq */
1747        for (i = 0; i < afu->num_hwqs; i++) {
1748                hwq = get_hwq(afu, i);
1749                cookie = hwq->ctx_cookie;
1750
1751                hwq->ctx_hndl = (u16) cfg->ops->process_element(cookie);
1752                hwq->host_map = &afu->afu_map->hosts[hwq->ctx_hndl].host;
1753                hwq->ctrl_map = &afu->afu_map->ctrls[hwq->ctx_hndl].ctrl;
1754
1755                /* Program the Endian Control for the master context */
1756                writeq_be(SISL_ENDIAN_CTRL, &hwq->host_map->endian_ctrl);
1757        }
1758}
1759
1760/**
1761 * init_global() - initialize AFU global registers
1762 * @cfg:        Internal structure associated with the host.
1763 */
1764static int init_global(struct cxlflash_cfg *cfg)
1765{
1766        struct afu *afu = cfg->afu;
1767        struct device *dev = &cfg->dev->dev;
1768        struct hwq *hwq;
1769        struct sisl_host_map __iomem *hmap;
1770        __be64 __iomem *fc_port_regs;
1771        u64 wwpn[MAX_FC_PORTS]; /* wwpn of AFU ports */
1772        int i = 0, num_ports = 0;
1773        int rc = 0;
1774        int j;
1775        void *ctx;
1776        u64 reg;
1777
1778        rc = read_vpd(cfg, &wwpn[0]);
1779        if (rc) {
1780                dev_err(dev, "%s: could not read vpd rc=%d\n", __func__, rc);
1781                goto out;
1782        }
1783
1784        /* Set up RRQ and SQ in HWQ for master issued cmds */
1785        for (i = 0; i < afu->num_hwqs; i++) {
1786                hwq = get_hwq(afu, i);
1787                hmap = hwq->host_map;
1788
1789                writeq_be((u64) hwq->hrrq_start, &hmap->rrq_start);
1790                writeq_be((u64) hwq->hrrq_end, &hmap->rrq_end);
1791                hwq->hrrq_online = true;
1792
1793                if (afu_is_sq_cmd_mode(afu)) {
1794                        writeq_be((u64)hwq->hsq_start, &hmap->sq_start);
1795                        writeq_be((u64)hwq->hsq_end, &hmap->sq_end);
1796                }
1797        }
1798
1799        /* AFU configuration */
1800        reg = readq_be(&afu->afu_map->global.regs.afu_config);
1801        reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1802        /* enable all auto retry options and control endianness */
1803        /* leave others at default: */
1804        /* CTX_CAP write protected, mbox_r does not clear on read and */
1805        /* checker on if dual afu */
1806        writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1807
1808        /* Global port select: select either port */
1809        if (afu->internal_lun) {
1810                /* Only use port 0 */
1811                writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1812                num_ports = 0;
1813        } else {
1814                writeq_be(PORT_MASK(cfg->num_fc_ports),
1815                          &afu->afu_map->global.regs.afu_port_sel);
1816                num_ports = cfg->num_fc_ports;
1817        }
1818
1819        for (i = 0; i < num_ports; i++) {
1820                fc_port_regs = get_fc_port_regs(cfg, i);
1821
1822                /* Unmask all errors (but they are still masked at AFU) */
1823                writeq_be(0, &fc_port_regs[FC_ERRMSK / 8]);
1824                /* Clear CRC error cnt & set a threshold */
1825                (void)readq_be(&fc_port_regs[FC_CNT_CRCERR / 8]);
1826                writeq_be(MC_CRC_THRESH, &fc_port_regs[FC_CRC_THRESH / 8]);
1827
1828                /* Set WWPNs. If already programmed, wwpn[i] is 0 */
1829                if (wwpn[i] != 0)
1830                        afu_set_wwpn(afu, i, &fc_port_regs[0], wwpn[i]);
1831                /* Programming WWPN back to back causes additional
1832                 * offline/online transitions and a PLOGI
1833                 */
1834                msleep(100);
1835        }
1836
1837        if (afu_is_ocxl_lisn(afu)) {
1838                /* Set up the LISN effective address for each master */
1839                for (i = 0; i < afu->num_hwqs; i++) {
1840                        hwq = get_hwq(afu, i);
1841                        ctx = hwq->ctx_cookie;
1842
1843                        for (j = 0; j < hwq->num_irqs; j++) {
1844                                reg = cfg->ops->get_irq_objhndl(ctx, j);
1845                                writeq_be(reg, &hwq->ctrl_map->lisn_ea[j]);
1846                        }
1847
1848                        reg = hwq->ctx_hndl;
1849                        writeq_be(SISL_LISN_PASID(reg, reg),
1850                                  &hwq->ctrl_map->lisn_pasid[0]);
1851                        writeq_be(SISL_LISN_PASID(0UL, reg),
1852                                  &hwq->ctrl_map->lisn_pasid[1]);
1853                }
1854        }
1855
1856        /* Set up master's own CTX_CAP to allow real mode, host translation */
1857        /* tables, afu cmds and read/write GSCSI cmds. */
1858        /* First, unlock ctx_cap write by reading mbox */
1859        for (i = 0; i < afu->num_hwqs; i++) {
1860                hwq = get_hwq(afu, i);
1861
1862                (void)readq_be(&hwq->ctrl_map->mbox_r); /* unlock ctx_cap */
1863                writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1864                        SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1865                        SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1866                        &hwq->ctrl_map->ctx_cap);
1867        }
1868
1869        /*
1870         * Determine write-same unmap support for host by evaluating the unmap
1871         * sector support bit of the context control register associated with
1872         * the primary hardware queue. Note that while this status is reflected
1873         * in a context register, the outcome can be assumed to be host-wide.
1874         */
1875        hwq = get_hwq(afu, PRIMARY_HWQ);
1876        reg = readq_be(&hwq->host_map->ctx_ctrl);
1877        if (reg & SISL_CTX_CTRL_UNMAP_SECTOR)
1878                cfg->ws_unmap = true;
1879
1880        /* Initialize heartbeat */
1881        afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1882out:
1883        return rc;
1884}
1885
1886/**
1887 * start_afu() - initializes and starts the AFU
1888 * @cfg:        Internal structure associated with the host.
1889 */
1890static int start_afu(struct cxlflash_cfg *cfg)
1891{
1892        struct afu *afu = cfg->afu;
1893        struct device *dev = &cfg->dev->dev;
1894        struct hwq *hwq;
1895        int rc = 0;
1896        int i;
1897
1898        init_pcr(cfg);
1899
1900        /* Initialize each HWQ */
1901        for (i = 0; i < afu->num_hwqs; i++) {
1902                hwq = get_hwq(afu, i);
1903
1904                /* After an AFU reset, RRQ entries are stale, clear them */
1905                memset(&hwq->rrq_entry, 0, sizeof(hwq->rrq_entry));
1906
1907                /* Initialize RRQ pointers */
1908                hwq->hrrq_start = &hwq->rrq_entry[0];
1909                hwq->hrrq_end = &hwq->rrq_entry[NUM_RRQ_ENTRY - 1];
1910                hwq->hrrq_curr = hwq->hrrq_start;
1911                hwq->toggle = 1;
1912
1913                /* Initialize spin locks */
1914                spin_lock_init(&hwq->hrrq_slock);
1915                spin_lock_init(&hwq->hsq_slock);
1916
1917                /* Initialize SQ */
1918                if (afu_is_sq_cmd_mode(afu)) {
1919                        memset(&hwq->sq, 0, sizeof(hwq->sq));
1920                        hwq->hsq_start = &hwq->sq[0];
1921                        hwq->hsq_end = &hwq->sq[NUM_SQ_ENTRY - 1];
1922                        hwq->hsq_curr = hwq->hsq_start;
1923
1924                        atomic_set(&hwq->hsq_credits, NUM_SQ_ENTRY - 1);
1925                }
1926
1927                /* Initialize IRQ poll */
1928                if (afu_is_irqpoll_enabled(afu))
1929                        irq_poll_init(&hwq->irqpoll, afu->irqpoll_weight,
1930                                      cxlflash_irqpoll);
1931
1932        }
1933
1934        rc = init_global(cfg);
1935
1936        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1937        return rc;
1938}
1939
1940/**
1941 * init_intr() - setup interrupt handlers for the master context
1942 * @cfg:        Internal structure associated with the host.
1943 * @hwq:        Hardware queue to initialize.
1944 *
1945 * Return: 0 on success, -errno on failure
1946 */
1947static enum undo_level init_intr(struct cxlflash_cfg *cfg,
1948                                 struct hwq *hwq)
1949{
1950        struct device *dev = &cfg->dev->dev;
1951        void *ctx = hwq->ctx_cookie;
1952        int rc = 0;
1953        enum undo_level level = UNDO_NOOP;
1954        bool is_primary_hwq = (hwq->index == PRIMARY_HWQ);
1955        int num_irqs = hwq->num_irqs;
1956
1957        rc = cfg->ops->allocate_afu_irqs(ctx, num_irqs);
1958        if (unlikely(rc)) {
1959                dev_err(dev, "%s: allocate_afu_irqs failed rc=%d\n",
1960                        __func__, rc);
1961                level = UNDO_NOOP;
1962                goto out;
1963        }
1964
1965        rc = cfg->ops->map_afu_irq(ctx, 1, cxlflash_sync_err_irq, hwq,
1966                                   "SISL_MSI_SYNC_ERROR");
1967        if (unlikely(rc <= 0)) {
1968                dev_err(dev, "%s: SISL_MSI_SYNC_ERROR map failed\n", __func__);
1969                level = FREE_IRQ;
1970                goto out;
1971        }
1972
1973        rc = cfg->ops->map_afu_irq(ctx, 2, cxlflash_rrq_irq, hwq,
1974                                   "SISL_MSI_RRQ_UPDATED");
1975        if (unlikely(rc <= 0)) {
1976                dev_err(dev, "%s: SISL_MSI_RRQ_UPDATED map failed\n", __func__);
1977                level = UNMAP_ONE;
1978                goto out;
1979        }
1980
1981        /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
1982        if (!is_primary_hwq)
1983                goto out;
1984
1985        rc = cfg->ops->map_afu_irq(ctx, 3, cxlflash_async_err_irq, hwq,
1986                                   "SISL_MSI_ASYNC_ERROR");
1987        if (unlikely(rc <= 0)) {
1988                dev_err(dev, "%s: SISL_MSI_ASYNC_ERROR map failed\n", __func__);
1989                level = UNMAP_TWO;
1990                goto out;
1991        }
1992out:
1993        return level;
1994}
1995
1996/**
1997 * init_mc() - create and register as the master context
1998 * @cfg:        Internal structure associated with the host.
1999 * @index:      HWQ Index of the master context.
2000 *
2001 * Return: 0 on success, -errno on failure
2002 */
2003static int init_mc(struct cxlflash_cfg *cfg, u32 index)
2004{
2005        void *ctx;
2006        struct device *dev = &cfg->dev->dev;
2007        struct hwq *hwq = get_hwq(cfg->afu, index);
2008        int rc = 0;
2009        int num_irqs;
2010        enum undo_level level;
2011
2012        hwq->afu = cfg->afu;
2013        hwq->index = index;
2014        INIT_LIST_HEAD(&hwq->pending_cmds);
2015
2016        if (index == PRIMARY_HWQ) {
2017                ctx = cfg->ops->get_context(cfg->dev, cfg->afu_cookie);
2018                num_irqs = 3;
2019        } else {
2020                ctx = cfg->ops->dev_context_init(cfg->dev, cfg->afu_cookie);
2021                num_irqs = 2;
2022        }
2023        if (IS_ERR_OR_NULL(ctx)) {
2024                rc = -ENOMEM;
2025                goto err1;
2026        }
2027
2028        WARN_ON(hwq->ctx_cookie);
2029        hwq->ctx_cookie = ctx;
2030        hwq->num_irqs = num_irqs;
2031
2032        /* Set it up as a master with the CXL */
2033        cfg->ops->set_master(ctx);
2034
2035        /* Reset AFU when initializing primary context */
2036        if (index == PRIMARY_HWQ) {
2037                rc = cfg->ops->afu_reset(ctx);
2038                if (unlikely(rc)) {
2039                        dev_err(dev, "%s: AFU reset failed rc=%d\n",
2040                                      __func__, rc);
2041                        goto err1;
2042                }
2043        }
2044
2045        level = init_intr(cfg, hwq);
2046        if (unlikely(level)) {
2047                dev_err(dev, "%s: interrupt init failed rc=%d\n", __func__, rc);
2048                goto err2;
2049        }
2050
2051        /* Finally, activate the context by starting it */
2052        rc = cfg->ops->start_context(hwq->ctx_cookie);
2053        if (unlikely(rc)) {
2054                dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
2055                level = UNMAP_THREE;
2056                goto err2;
2057        }
2058
2059out:
2060        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2061        return rc;
2062err2:
2063        term_intr(cfg, level, index);
2064        if (index != PRIMARY_HWQ)
2065                cfg->ops->release_context(ctx);
2066err1:
2067        hwq->ctx_cookie = NULL;
2068        goto out;
2069}
2070
2071/**
2072 * get_num_afu_ports() - determines and configures the number of AFU ports
2073 * @cfg:        Internal structure associated with the host.
2074 *
2075 * This routine determines the number of AFU ports by converting the global
2076 * port selection mask. The converted value is only valid following an AFU
2077 * reset (explicit or power-on). This routine must be invoked shortly after
2078 * mapping as other routines are dependent on the number of ports during the
2079 * initialization sequence.
2080 *
2081 * To support legacy AFUs that might not have reflected an initial global
2082 * port mask (value read is 0), default to the number of ports originally
2083 * supported by the cxlflash driver (2) before hardware with other port
2084 * offerings was introduced.
2085 */
2086static void get_num_afu_ports(struct cxlflash_cfg *cfg)
2087{
2088        struct afu *afu = cfg->afu;
2089        struct device *dev = &cfg->dev->dev;
2090        u64 port_mask;
2091        int num_fc_ports = LEGACY_FC_PORTS;
2092
2093        port_mask = readq_be(&afu->afu_map->global.regs.afu_port_sel);
2094        if (port_mask != 0ULL)
2095                num_fc_ports = min(ilog2(port_mask) + 1, MAX_FC_PORTS);
2096
2097        dev_dbg(dev, "%s: port_mask=%016llx num_fc_ports=%d\n",
2098                __func__, port_mask, num_fc_ports);
2099
2100        cfg->num_fc_ports = num_fc_ports;
2101        cfg->host->max_channel = PORTNUM2CHAN(num_fc_ports);
2102}
2103
2104/**
2105 * init_afu() - setup as master context and start AFU
2106 * @cfg:        Internal structure associated with the host.
2107 *
2108 * This routine is a higher level of control for configuring the
2109 * AFU on probe and reset paths.
2110 *
2111 * Return: 0 on success, -errno on failure
2112 */
2113static int init_afu(struct cxlflash_cfg *cfg)
2114{
2115        u64 reg;
2116        int rc = 0;
2117        struct afu *afu = cfg->afu;
2118        struct device *dev = &cfg->dev->dev;
2119        struct hwq *hwq;
2120        int i;
2121
2122        cfg->ops->perst_reloads_same_image(cfg->afu_cookie, true);
2123
2124        mutex_init(&afu->sync_active);
2125        afu->num_hwqs = afu->desired_hwqs;
2126        for (i = 0; i < afu->num_hwqs; i++) {
2127                rc = init_mc(cfg, i);
2128                if (rc) {
2129                        dev_err(dev, "%s: init_mc failed rc=%d index=%d\n",
2130                                __func__, rc, i);
2131                        goto err1;
2132                }
2133        }
2134
2135        /* Map the entire MMIO space of the AFU using the first context */
2136        hwq = get_hwq(afu, PRIMARY_HWQ);
2137        afu->afu_map = cfg->ops->psa_map(hwq->ctx_cookie);
2138        if (!afu->afu_map) {
2139                dev_err(dev, "%s: psa_map failed\n", __func__);
2140                rc = -ENOMEM;
2141                goto err1;
2142        }
2143
2144        /* No byte reverse on reading afu_version or string will be backwards */
2145        reg = readq(&afu->afu_map->global.regs.afu_version);
2146        memcpy(afu->version, &reg, sizeof(reg));
2147        afu->interface_version =
2148            readq_be(&afu->afu_map->global.regs.interface_version);
2149        if ((afu->interface_version + 1) == 0) {
2150                dev_err(dev, "Back level AFU, please upgrade. AFU version %s "
2151                        "interface version %016llx\n", afu->version,
2152                       afu->interface_version);
2153                rc = -EINVAL;
2154                goto err1;
2155        }
2156
2157        if (afu_is_sq_cmd_mode(afu)) {
2158                afu->send_cmd = send_cmd_sq;
2159                afu->context_reset = context_reset_sq;
2160        } else {
2161                afu->send_cmd = send_cmd_ioarrin;
2162                afu->context_reset = context_reset_ioarrin;
2163        }
2164
2165        dev_dbg(dev, "%s: afu_ver=%s interface_ver=%016llx\n", __func__,
2166                afu->version, afu->interface_version);
2167
2168        get_num_afu_ports(cfg);
2169
2170        rc = start_afu(cfg);
2171        if (rc) {
2172                dev_err(dev, "%s: start_afu failed, rc=%d\n", __func__, rc);
2173                goto err1;
2174        }
2175
2176        afu_err_intr_init(cfg->afu);
2177        for (i = 0; i < afu->num_hwqs; i++) {
2178                hwq = get_hwq(afu, i);
2179
2180                hwq->room = readq_be(&hwq->host_map->cmd_room);
2181        }
2182
2183        /* Restore the LUN mappings */
2184        cxlflash_restore_luntable(cfg);
2185out:
2186        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2187        return rc;
2188
2189err1:
2190        for (i = afu->num_hwqs - 1; i >= 0; i--) {
2191                term_intr(cfg, UNMAP_THREE, i);
2192                term_mc(cfg, i);
2193        }
2194        goto out;
2195}
2196
2197/**
2198 * afu_reset() - resets the AFU
2199 * @cfg:        Internal structure associated with the host.
2200 *
2201 * Return: 0 on success, -errno on failure
2202 */
2203static int afu_reset(struct cxlflash_cfg *cfg)
2204{
2205        struct device *dev = &cfg->dev->dev;
2206        int rc = 0;
2207
2208        /* Stop the context before the reset. Since the context is
2209         * no longer available restart it after the reset is complete
2210         */
2211        term_afu(cfg);
2212
2213        rc = init_afu(cfg);
2214
2215        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2216        return rc;
2217}
2218
2219/**
2220 * drain_ioctls() - wait until all currently executing ioctls have completed
2221 * @cfg:        Internal structure associated with the host.
2222 *
2223 * Obtain write access to read/write semaphore that wraps ioctl
2224 * handling to 'drain' ioctls currently executing.
2225 */
2226static void drain_ioctls(struct cxlflash_cfg *cfg)
2227{
2228        down_write(&cfg->ioctl_rwsem);
2229        up_write(&cfg->ioctl_rwsem);
2230}
2231
2232/**
2233 * cxlflash_async_reset_host() - asynchronous host reset handler
2234 * @data:       Private data provided while scheduling reset.
2235 * @cookie:     Cookie that can be used for checkpointing.
2236 */
2237static void cxlflash_async_reset_host(void *data, async_cookie_t cookie)
2238{
2239        struct cxlflash_cfg *cfg = data;
2240        struct device *dev = &cfg->dev->dev;
2241        int rc = 0;
2242
2243        if (cfg->state != STATE_RESET) {
2244                dev_dbg(dev, "%s: Not performing a reset, state=%d\n",
2245                        __func__, cfg->state);
2246                goto out;
2247        }
2248
2249        drain_ioctls(cfg);
2250        cxlflash_mark_contexts_error(cfg);
2251        rc = afu_reset(cfg);
2252        if (rc)
2253                cfg->state = STATE_FAILTERM;
2254        else
2255                cfg->state = STATE_NORMAL;
2256        wake_up_all(&cfg->reset_waitq);
2257
2258out:
2259        scsi_unblock_requests(cfg->host);
2260}
2261
2262/**
2263 * cxlflash_schedule_async_reset() - schedule an asynchronous host reset
2264 * @cfg:        Internal structure associated with the host.
2265 */
2266static void cxlflash_schedule_async_reset(struct cxlflash_cfg *cfg)
2267{
2268        struct device *dev = &cfg->dev->dev;
2269
2270        if (cfg->state != STATE_NORMAL) {
2271                dev_dbg(dev, "%s: Not performing reset state=%d\n",
2272                        __func__, cfg->state);
2273                return;
2274        }
2275
2276        cfg->state = STATE_RESET;
2277        scsi_block_requests(cfg->host);
2278        cfg->async_reset_cookie = async_schedule(cxlflash_async_reset_host,
2279                                                 cfg);
2280}
2281
2282/**
2283 * send_afu_cmd() - builds and sends an internal AFU command
2284 * @afu:        AFU associated with the host.
2285 * @rcb:        Pre-populated IOARCB describing command to send.
2286 *
2287 * The AFU can only take one internal AFU command at a time. This limitation is
2288 * enforced by using a mutex to provide exclusive access to the AFU during the
2289 * operation. This design point requires calling threads to not be on interrupt
2290 * context due to the possibility of sleeping during concurrent AFU operations.
2291 *
2292 * The command status is optionally passed back to the caller when the caller
2293 * populates the IOASA field of the IOARCB with a pointer to an IOASA structure.
2294 *
2295 * Return:
2296 *      0 on success, -errno on failure
2297 */
2298static int send_afu_cmd(struct afu *afu, struct sisl_ioarcb *rcb)
2299{
2300        struct cxlflash_cfg *cfg = afu->parent;
2301        struct device *dev = &cfg->dev->dev;
2302        struct afu_cmd *cmd = NULL;
2303        struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
2304        ulong lock_flags;
2305        char *buf = NULL;
2306        int rc = 0;
2307        int nretry = 0;
2308
2309        if (cfg->state != STATE_NORMAL) {
2310                dev_dbg(dev, "%s: Sync not required state=%u\n",
2311                        __func__, cfg->state);
2312                return 0;
2313        }
2314
2315        mutex_lock(&afu->sync_active);
2316        atomic_inc(&afu->cmds_active);
2317        buf = kmalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
2318        if (unlikely(!buf)) {
2319                dev_err(dev, "%s: no memory for command\n", __func__);
2320                rc = -ENOMEM;
2321                goto out;
2322        }
2323
2324        cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
2325
2326retry:
2327        memset(cmd, 0, sizeof(*cmd));
2328        memcpy(&cmd->rcb, rcb, sizeof(*rcb));
2329        INIT_LIST_HEAD(&cmd->queue);
2330        init_completion(&cmd->cevent);
2331        cmd->parent = afu;
2332        cmd->hwq_index = hwq->index;
2333        cmd->rcb.ctx_id = hwq->ctx_hndl;
2334
2335        dev_dbg(dev, "%s: afu=%p cmd=%p type=%02x nretry=%d\n",
2336                __func__, afu, cmd, cmd->rcb.cdb[0], nretry);
2337
2338        rc = afu->send_cmd(afu, cmd);
2339        if (unlikely(rc)) {
2340                rc = -ENOBUFS;
2341                goto out;
2342        }
2343
2344        rc = wait_resp(afu, cmd);
2345        switch (rc) {
2346        case -ETIMEDOUT:
2347                rc = afu->context_reset(hwq);
2348                if (rc) {
2349                        /* Delete the command from pending_cmds list */
2350                        spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
2351                        list_del(&cmd->list);
2352                        spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
2353
2354                        cxlflash_schedule_async_reset(cfg);
2355                        break;
2356                }
2357                fallthrough;    /* to retry */
2358        case -EAGAIN:
2359                if (++nretry < 2)
2360                        goto retry;
2361                fallthrough;    /* to exit */
2362        default:
2363                break;
2364        }
2365
2366        if (rcb->ioasa)
2367                *rcb->ioasa = cmd->sa;
2368out:
2369        atomic_dec(&afu->cmds_active);
2370        mutex_unlock(&afu->sync_active);
2371        kfree(buf);
2372        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2373        return rc;
2374}
2375
2376/**
2377 * cxlflash_afu_sync() - builds and sends an AFU sync command
2378 * @afu:        AFU associated with the host.
2379 * @ctx:        Identifies context requesting sync.
2380 * @res:        Identifies resource requesting sync.
2381 * @mode:       Type of sync to issue (lightweight, heavyweight, global).
2382 *
2383 * AFU sync operations are only necessary and allowed when the device is
2384 * operating normally. When not operating normally, sync requests can occur as
2385 * part of cleaning up resources associated with an adapter prior to removal.
2386 * In this scenario, these requests are simply ignored (safe due to the AFU
2387 * going away).
2388 *
2389 * Return:
2390 *      0 on success, -errno on failure
2391 */
2392int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx, res_hndl_t res, u8 mode)
2393{
2394        struct cxlflash_cfg *cfg = afu->parent;
2395        struct device *dev = &cfg->dev->dev;
2396        struct sisl_ioarcb rcb = { 0 };
2397
2398        dev_dbg(dev, "%s: afu=%p ctx=%u res=%u mode=%u\n",
2399                __func__, afu, ctx, res, mode);
2400
2401        rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
2402        rcb.msi = SISL_MSI_RRQ_UPDATED;
2403        rcb.timeout = MC_AFU_SYNC_TIMEOUT;
2404
2405        rcb.cdb[0] = SISL_AFU_CMD_SYNC;
2406        rcb.cdb[1] = mode;
2407        put_unaligned_be16(ctx, &rcb.cdb[2]);
2408        put_unaligned_be32(res, &rcb.cdb[4]);
2409
2410        return send_afu_cmd(afu, &rcb);
2411}
2412
2413/**
2414 * cxlflash_eh_abort_handler() - abort a SCSI command
2415 * @scp:        SCSI command to abort.
2416 *
2417 * CXL Flash devices do not support a single command abort. Reset the context
2418 * as per SISLite specification. Flush any pending commands in the hardware
2419 * queue before the reset.
2420 *
2421 * Return: SUCCESS/FAILED as defined in scsi/scsi.h
2422 */
2423static int cxlflash_eh_abort_handler(struct scsi_cmnd *scp)
2424{
2425        int rc = FAILED;
2426        struct Scsi_Host *host = scp->device->host;
2427        struct cxlflash_cfg *cfg = shost_priv(host);
2428        struct afu_cmd *cmd = sc_to_afuc(scp);
2429        struct device *dev = &cfg->dev->dev;
2430        struct afu *afu = cfg->afu;
2431        struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
2432
2433        dev_dbg(dev, "%s: (scp=%p) %d/%d/%d/%llu "
2434                "cdb=(%08x-%08x-%08x-%08x)\n", __func__, scp, host->host_no,
2435                scp->device->channel, scp->device->id, scp->device->lun,
2436                get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
2437                get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
2438                get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
2439                get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
2440
2441        /* When the state is not normal, another reset/reload is in progress.
2442         * Return failed and the mid-layer will invoke host reset handler.
2443         */
2444        if (cfg->state != STATE_NORMAL) {
2445                dev_dbg(dev, "%s: Invalid state for abort, state=%d\n",
2446                        __func__, cfg->state);
2447                goto out;
2448        }
2449
2450        rc = afu->context_reset(hwq);
2451        if (unlikely(rc))
2452                goto out;
2453
2454        rc = SUCCESS;
2455
2456out:
2457        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2458        return rc;
2459}
2460
2461/**
2462 * cxlflash_eh_device_reset_handler() - reset a single LUN
2463 * @scp:        SCSI command to send.
2464 *
2465 * Return:
2466 *      SUCCESS as defined in scsi/scsi.h
2467 *      FAILED as defined in scsi/scsi.h
2468 */
2469static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
2470{
2471        int rc = SUCCESS;
2472        struct scsi_device *sdev = scp->device;
2473        struct Scsi_Host *host = sdev->host;
2474        struct cxlflash_cfg *cfg = shost_priv(host);
2475        struct device *dev = &cfg->dev->dev;
2476        int rcr = 0;
2477
2478        dev_dbg(dev, "%s: %d/%d/%d/%llu\n", __func__,
2479                host->host_no, sdev->channel, sdev->id, sdev->lun);
2480retry:
2481        switch (cfg->state) {
2482        case STATE_NORMAL:
2483                rcr = send_tmf(cfg, sdev, TMF_LUN_RESET);
2484                if (unlikely(rcr))
2485                        rc = FAILED;
2486                break;
2487        case STATE_RESET:
2488                wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2489                goto retry;
2490        default:
2491                rc = FAILED;
2492                break;
2493        }
2494
2495        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2496        return rc;
2497}
2498
2499/**
2500 * cxlflash_eh_host_reset_handler() - reset the host adapter
2501 * @scp:        SCSI command from stack identifying host.
2502 *
2503 * Following a reset, the state is evaluated again in case an EEH occurred
2504 * during the reset. In such a scenario, the host reset will either yield
2505 * until the EEH recovery is complete or return success or failure based
2506 * upon the current device state.
2507 *
2508 * Return:
2509 *      SUCCESS as defined in scsi/scsi.h
2510 *      FAILED as defined in scsi/scsi.h
2511 */
2512static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
2513{
2514        int rc = SUCCESS;
2515        int rcr = 0;
2516        struct Scsi_Host *host = scp->device->host;
2517        struct cxlflash_cfg *cfg = shost_priv(host);
2518        struct device *dev = &cfg->dev->dev;
2519
2520        dev_dbg(dev, "%s: %d\n", __func__, host->host_no);
2521
2522        switch (cfg->state) {
2523        case STATE_NORMAL:
2524                cfg->state = STATE_RESET;
2525                drain_ioctls(cfg);
2526                cxlflash_mark_contexts_error(cfg);
2527                rcr = afu_reset(cfg);
2528                if (rcr) {
2529                        rc = FAILED;
2530                        cfg->state = STATE_FAILTERM;
2531                } else
2532                        cfg->state = STATE_NORMAL;
2533                wake_up_all(&cfg->reset_waitq);
2534                ssleep(1);
2535                fallthrough;
2536        case STATE_RESET:
2537                wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2538                if (cfg->state == STATE_NORMAL)
2539                        break;
2540                fallthrough;
2541        default:
2542                rc = FAILED;
2543                break;
2544        }
2545
2546        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2547        return rc;
2548}
2549
2550/**
2551 * cxlflash_change_queue_depth() - change the queue depth for the device
2552 * @sdev:       SCSI device destined for queue depth change.
2553 * @qdepth:     Requested queue depth value to set.
2554 *
2555 * The requested queue depth is capped to the maximum supported value.
2556 *
2557 * Return: The actual queue depth set.
2558 */
2559static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
2560{
2561
2562        if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
2563                qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
2564
2565        scsi_change_queue_depth(sdev, qdepth);
2566        return sdev->queue_depth;
2567}
2568
2569/**
2570 * cxlflash_show_port_status() - queries and presents the current port status
2571 * @port:       Desired port for status reporting.
2572 * @cfg:        Internal structure associated with the host.
2573 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2574 *
2575 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2576 */
2577static ssize_t cxlflash_show_port_status(u32 port,
2578                                         struct cxlflash_cfg *cfg,
2579                                         char *buf)
2580{
2581        struct device *dev = &cfg->dev->dev;
2582        char *disp_status;
2583        u64 status;
2584        __be64 __iomem *fc_port_regs;
2585
2586        WARN_ON(port >= MAX_FC_PORTS);
2587
2588        if (port >= cfg->num_fc_ports) {
2589                dev_info(dev, "%s: Port %d not supported on this card.\n",
2590                        __func__, port);
2591                return -EINVAL;
2592        }
2593
2594        fc_port_regs = get_fc_port_regs(cfg, port);
2595        status = readq_be(&fc_port_regs[FC_MTIP_STATUS / 8]);
2596        status &= FC_MTIP_STATUS_MASK;
2597
2598        if (status == FC_MTIP_STATUS_ONLINE)
2599                disp_status = "online";
2600        else if (status == FC_MTIP_STATUS_OFFLINE)
2601                disp_status = "offline";
2602        else
2603                disp_status = "unknown";
2604
2605        return scnprintf(buf, PAGE_SIZE, "%s\n", disp_status);
2606}
2607
2608/**
2609 * port0_show() - queries and presents the current status of port 0
2610 * @dev:        Generic device associated with the host owning the port.
2611 * @attr:       Device attribute representing the port.
2612 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2613 *
2614 * Return: The size of the ASCII string returned in @buf.
2615 */
2616static ssize_t port0_show(struct device *dev,
2617                          struct device_attribute *attr,
2618                          char *buf)
2619{
2620        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2621
2622        return cxlflash_show_port_status(0, cfg, buf);
2623}
2624
2625/**
2626 * port1_show() - queries and presents the current status of port 1
2627 * @dev:        Generic device associated with the host owning the port.
2628 * @attr:       Device attribute representing the port.
2629 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2630 *
2631 * Return: The size of the ASCII string returned in @buf.
2632 */
2633static ssize_t port1_show(struct device *dev,
2634                          struct device_attribute *attr,
2635                          char *buf)
2636{
2637        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2638
2639        return cxlflash_show_port_status(1, cfg, buf);
2640}
2641
2642/**
2643 * port2_show() - queries and presents the current status of port 2
2644 * @dev:        Generic device associated with the host owning the port.
2645 * @attr:       Device attribute representing the port.
2646 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2647 *
2648 * Return: The size of the ASCII string returned in @buf.
2649 */
2650static ssize_t port2_show(struct device *dev,
2651                          struct device_attribute *attr,
2652                          char *buf)
2653{
2654        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2655
2656        return cxlflash_show_port_status(2, cfg, buf);
2657}
2658
2659/**
2660 * port3_show() - queries and presents the current status of port 3
2661 * @dev:        Generic device associated with the host owning the port.
2662 * @attr:       Device attribute representing the port.
2663 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2664 *
2665 * Return: The size of the ASCII string returned in @buf.
2666 */
2667static ssize_t port3_show(struct device *dev,
2668                          struct device_attribute *attr,
2669                          char *buf)
2670{
2671        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2672
2673        return cxlflash_show_port_status(3, cfg, buf);
2674}
2675
2676/**
2677 * lun_mode_show() - presents the current LUN mode of the host
2678 * @dev:        Generic device associated with the host.
2679 * @attr:       Device attribute representing the LUN mode.
2680 * @buf:        Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2681 *
2682 * Return: The size of the ASCII string returned in @buf.
2683 */
2684static ssize_t lun_mode_show(struct device *dev,
2685                             struct device_attribute *attr, char *buf)
2686{
2687        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2688        struct afu *afu = cfg->afu;
2689
2690        return scnprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
2691}
2692
2693/**
2694 * lun_mode_store() - sets the LUN mode of the host
2695 * @dev:        Generic device associated with the host.
2696 * @attr:       Device attribute representing the LUN mode.
2697 * @buf:        Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2698 * @count:      Length of data resizing in @buf.
2699 *
2700 * The CXL Flash AFU supports a dummy LUN mode where the external
2701 * links and storage are not required. Space on the FPGA is used
2702 * to create 1 or 2 small LUNs which are presented to the system
2703 * as if they were a normal storage device. This feature is useful
2704 * during development and also provides manufacturing with a way
2705 * to test the AFU without an actual device.
2706 *
2707 * 0 = external LUN[s] (default)
2708 * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2709 * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2710 * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2711 * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2712 *
2713 * Return: The size of the ASCII string returned in @buf.
2714 */
2715static ssize_t lun_mode_store(struct device *dev,
2716                              struct device_attribute *attr,
2717                              const char *buf, size_t count)
2718{
2719        struct Scsi_Host *shost = class_to_shost(dev);
2720        struct cxlflash_cfg *cfg = shost_priv(shost);
2721        struct afu *afu = cfg->afu;
2722        int rc;
2723        u32 lun_mode;
2724
2725        rc = kstrtouint(buf, 10, &lun_mode);
2726        if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
2727                afu->internal_lun = lun_mode;
2728
2729                /*
2730                 * When configured for internal LUN, there is only one channel,
2731                 * channel number 0, else there will be one less than the number
2732                 * of fc ports for this card.
2733                 */
2734                if (afu->internal_lun)
2735                        shost->max_channel = 0;
2736                else
2737                        shost->max_channel = PORTNUM2CHAN(cfg->num_fc_ports);
2738
2739                afu_reset(cfg);
2740                scsi_scan_host(cfg->host);
2741        }
2742
2743        return count;
2744}
2745
2746/**
2747 * ioctl_version_show() - presents the current ioctl version of the host
2748 * @dev:        Generic device associated with the host.
2749 * @attr:       Device attribute representing the ioctl version.
2750 * @buf:        Buffer of length PAGE_SIZE to report back the ioctl version.
2751 *
2752 * Return: The size of the ASCII string returned in @buf.
2753 */
2754static ssize_t ioctl_version_show(struct device *dev,
2755                                  struct device_attribute *attr, char *buf)
2756{
2757        ssize_t bytes = 0;
2758
2759        bytes = scnprintf(buf, PAGE_SIZE,
2760                          "disk: %u\n", DK_CXLFLASH_VERSION_0);
2761        bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2762                           "host: %u\n", HT_CXLFLASH_VERSION_0);
2763
2764        return bytes;
2765}
2766
2767/**
2768 * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2769 * @port:       Desired port for status reporting.
2770 * @cfg:        Internal structure associated with the host.
2771 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2772 *
2773 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2774 */
2775static ssize_t cxlflash_show_port_lun_table(u32 port,
2776                                            struct cxlflash_cfg *cfg,
2777                                            char *buf)
2778{
2779        struct device *dev = &cfg->dev->dev;
2780        __be64 __iomem *fc_port_luns;
2781        int i;
2782        ssize_t bytes = 0;
2783
2784        WARN_ON(port >= MAX_FC_PORTS);
2785
2786        if (port >= cfg->num_fc_ports) {
2787                dev_info(dev, "%s: Port %d not supported on this card.\n",
2788                        __func__, port);
2789                return -EINVAL;
2790        }
2791
2792        fc_port_luns = get_fc_port_luns(cfg, port);
2793
2794        for (i = 0; i < CXLFLASH_NUM_VLUNS; i++)
2795                bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2796                                   "%03d: %016llx\n",
2797                                   i, readq_be(&fc_port_luns[i]));
2798        return bytes;
2799}
2800
2801/**
2802 * port0_lun_table_show() - presents the current LUN table of port 0
2803 * @dev:        Generic device associated with the host owning the port.
2804 * @attr:       Device attribute representing the port.
2805 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2806 *
2807 * Return: The size of the ASCII string returned in @buf.
2808 */
2809static ssize_t port0_lun_table_show(struct device *dev,
2810                                    struct device_attribute *attr,
2811                                    char *buf)
2812{
2813        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2814
2815        return cxlflash_show_port_lun_table(0, cfg, buf);
2816}
2817
2818/**
2819 * port1_lun_table_show() - presents the current LUN table of port 1
2820 * @dev:        Generic device associated with the host owning the port.
2821 * @attr:       Device attribute representing the port.
2822 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2823 *
2824 * Return: The size of the ASCII string returned in @buf.
2825 */
2826static ssize_t port1_lun_table_show(struct device *dev,
2827                                    struct device_attribute *attr,
2828                                    char *buf)
2829{
2830        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2831
2832        return cxlflash_show_port_lun_table(1, cfg, buf);
2833}
2834
2835/**
2836 * port2_lun_table_show() - presents the current LUN table of port 2
2837 * @dev:        Generic device associated with the host owning the port.
2838 * @attr:       Device attribute representing the port.
2839 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2840 *
2841 * Return: The size of the ASCII string returned in @buf.
2842 */
2843static ssize_t port2_lun_table_show(struct device *dev,
2844                                    struct device_attribute *attr,
2845                                    char *buf)
2846{
2847        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2848
2849        return cxlflash_show_port_lun_table(2, cfg, buf);
2850}
2851
2852/**
2853 * port3_lun_table_show() - presents the current LUN table of port 3
2854 * @dev:        Generic device associated with the host owning the port.
2855 * @attr:       Device attribute representing the port.
2856 * @buf:        Buffer of length PAGE_SIZE to report back port status in ASCII.
2857 *
2858 * Return: The size of the ASCII string returned in @buf.
2859 */
2860static ssize_t port3_lun_table_show(struct device *dev,
2861                                    struct device_attribute *attr,
2862                                    char *buf)
2863{
2864        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2865
2866        return cxlflash_show_port_lun_table(3, cfg, buf);
2867}
2868
2869/**
2870 * irqpoll_weight_show() - presents the current IRQ poll weight for the host
2871 * @dev:        Generic device associated with the host.
2872 * @attr:       Device attribute representing the IRQ poll weight.
2873 * @buf:        Buffer of length PAGE_SIZE to report back the current IRQ poll
2874 *              weight in ASCII.
2875 *
2876 * An IRQ poll weight of 0 indicates polling is disabled.
2877 *
2878 * Return: The size of the ASCII string returned in @buf.
2879 */
2880static ssize_t irqpoll_weight_show(struct device *dev,
2881                                   struct device_attribute *attr, char *buf)
2882{
2883        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2884        struct afu *afu = cfg->afu;
2885
2886        return scnprintf(buf, PAGE_SIZE, "%u\n", afu->irqpoll_weight);
2887}
2888
2889/**
2890 * irqpoll_weight_store() - sets the current IRQ poll weight for the host
2891 * @dev:        Generic device associated with the host.
2892 * @attr:       Device attribute representing the IRQ poll weight.
2893 * @buf:        Buffer of length PAGE_SIZE containing the desired IRQ poll
2894 *              weight in ASCII.
2895 * @count:      Length of data resizing in @buf.
2896 *
2897 * An IRQ poll weight of 0 indicates polling is disabled.
2898 *
2899 * Return: The size of the ASCII string returned in @buf.
2900 */
2901static ssize_t irqpoll_weight_store(struct device *dev,
2902                                    struct device_attribute *attr,
2903                                    const char *buf, size_t count)
2904{
2905        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2906        struct device *cfgdev = &cfg->dev->dev;
2907        struct afu *afu = cfg->afu;
2908        struct hwq *hwq;
2909        u32 weight;
2910        int rc, i;
2911
2912        rc = kstrtouint(buf, 10, &weight);
2913        if (rc)
2914                return -EINVAL;
2915
2916        if (weight > 256) {
2917                dev_info(cfgdev,
2918                         "Invalid IRQ poll weight. It must be 256 or less.\n");
2919                return -EINVAL;
2920        }
2921
2922        if (weight == afu->irqpoll_weight) {
2923                dev_info(cfgdev,
2924                         "Current IRQ poll weight has the same weight.\n");
2925                return -EINVAL;
2926        }
2927
2928        if (afu_is_irqpoll_enabled(afu)) {
2929                for (i = 0; i < afu->num_hwqs; i++) {
2930                        hwq = get_hwq(afu, i);
2931
2932                        irq_poll_disable(&hwq->irqpoll);
2933                }
2934        }
2935
2936        afu->irqpoll_weight = weight;
2937
2938        if (weight > 0) {
2939                for (i = 0; i < afu->num_hwqs; i++) {
2940                        hwq = get_hwq(afu, i);
2941
2942                        irq_poll_init(&hwq->irqpoll, weight, cxlflash_irqpoll);
2943                }
2944        }
2945
2946        return count;
2947}
2948
2949/**
2950 * num_hwqs_show() - presents the number of hardware queues for the host
2951 * @dev:        Generic device associated with the host.
2952 * @attr:       Device attribute representing the number of hardware queues.
2953 * @buf:        Buffer of length PAGE_SIZE to report back the number of hardware
2954 *              queues in ASCII.
2955 *
2956 * Return: The size of the ASCII string returned in @buf.
2957 */
2958static ssize_t num_hwqs_show(struct device *dev,
2959                             struct device_attribute *attr, char *buf)
2960{
2961        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2962        struct afu *afu = cfg->afu;
2963
2964        return scnprintf(buf, PAGE_SIZE, "%u\n", afu->num_hwqs);
2965}
2966
2967/**
2968 * num_hwqs_store() - sets the number of hardware queues for the host
2969 * @dev:        Generic device associated with the host.
2970 * @attr:       Device attribute representing the number of hardware queues.
2971 * @buf:        Buffer of length PAGE_SIZE containing the number of hardware
2972 *              queues in ASCII.
2973 * @count:      Length of data resizing in @buf.
2974 *
2975 * n > 0: num_hwqs = n
2976 * n = 0: num_hwqs = num_online_cpus()
2977 * n < 0: num_online_cpus() / abs(n)
2978 *
2979 * Return: The size of the ASCII string returned in @buf.
2980 */
2981static ssize_t num_hwqs_store(struct device *dev,
2982                              struct device_attribute *attr,
2983                              const char *buf, size_t count)
2984{
2985        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2986        struct afu *afu = cfg->afu;
2987        int rc;
2988        int nhwqs, num_hwqs;
2989
2990        rc = kstrtoint(buf, 10, &nhwqs);
2991        if (rc)
2992                return -EINVAL;
2993
2994        if (nhwqs >= 1)
2995                num_hwqs = nhwqs;
2996        else if (nhwqs == 0)
2997                num_hwqs = num_online_cpus();
2998        else
2999                num_hwqs = num_online_cpus() / abs(nhwqs);
3000
3001        afu->desired_hwqs = min(num_hwqs, CXLFLASH_MAX_HWQS);
3002        WARN_ON_ONCE(afu->desired_hwqs == 0);
3003
3004retry:
3005        switch (cfg->state) {
3006        case STATE_NORMAL:
3007                cfg->state = STATE_RESET;
3008                drain_ioctls(cfg);
3009                cxlflash_mark_contexts_error(cfg);
3010                rc = afu_reset(cfg);
3011                if (rc)
3012                        cfg->state = STATE_FAILTERM;
3013                else
3014                        cfg->state = STATE_NORMAL;
3015                wake_up_all(&cfg->reset_waitq);
3016                break;
3017        case STATE_RESET:
3018                wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
3019                if (cfg->state == STATE_NORMAL)
3020                        goto retry;
3021                fallthrough;
3022        default:
3023                /* Ideally should not happen */
3024                dev_err(dev, "%s: Device is not ready, state=%d\n",
3025                        __func__, cfg->state);
3026                break;
3027        }
3028
3029        return count;
3030}
3031
3032static const char *hwq_mode_name[MAX_HWQ_MODE] = { "rr", "tag", "cpu" };
3033
3034/**
3035 * hwq_mode_show() - presents the HWQ steering mode for the host
3036 * @dev:        Generic device associated with the host.
3037 * @attr:       Device attribute representing the HWQ steering mode.
3038 * @buf:        Buffer of length PAGE_SIZE to report back the HWQ steering mode
3039 *              as a character string.
3040 *
3041 * Return: The size of the ASCII string returned in @buf.
3042 */
3043static ssize_t hwq_mode_show(struct device *dev,
3044                             struct device_attribute *attr, char *buf)
3045{
3046        struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
3047        struct afu *afu = cfg->afu;
3048
3049        return scnprintf(buf, PAGE_SIZE, "%s\n", hwq_mode_name[afu->hwq_mode]);
3050}
3051
3052/**
3053 * hwq_mode_store() - sets the HWQ steering mode for the host
3054 * @dev:        Generic device associated with the host.
3055 * @attr:       Device attribute representing the HWQ steering mode.
3056 * @buf:        Buffer of length PAGE_SIZE containing the HWQ steering mode
3057 *              as a character string.
3058 * @count:      Length of data resizing in @buf.
3059 *
3060 * rr = Round-Robin
3061 * tag = Block MQ Tagging
3062 * cpu = CPU Affinity
3063 *
3064 * Return: The size of the ASCII string returned in @buf.
3065 */
3066static ssize_t hwq_mode_store(struct device *dev,
3067                              struct device_attribute *attr,
3068                              const char *buf, size_t count)
3069{
3070        struct Scsi_Host *shost = class_to_shost(dev);
3071        struct cxlflash_cfg *cfg = shost_priv(shost);
3072        struct device *cfgdev = &cfg->dev->dev;
3073        struct afu *afu = cfg->afu;
3074        int i;
3075        u32 mode = MAX_HWQ_MODE;
3076
3077        for (i = 0; i < MAX_HWQ_MODE; i++) {
3078                if (!strncmp(hwq_mode_name[i], buf, strlen(hwq_mode_name[i]))) {
3079                        mode = i;
3080                        break;
3081                }
3082        }
3083
3084        if (mode >= MAX_HWQ_MODE) {
3085                dev_info(cfgdev, "Invalid HWQ steering mode.\n");
3086                return -EINVAL;
3087        }
3088
3089        afu->hwq_mode = mode;
3090
3091        return count;
3092}
3093
3094/**
3095 * mode_show() - presents the current mode of the device
3096 * @dev:        Generic device associated with the device.
3097 * @attr:       Device attribute representing the device mode.
3098 * @buf:        Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
3099 *
3100 * Return: The size of the ASCII string returned in @buf.
3101 */
3102static ssize_t mode_show(struct device *dev,
3103                         struct device_attribute *attr, char *buf)
3104{
3105        struct scsi_device *sdev = to_scsi_device(dev);
3106
3107        return scnprintf(buf, PAGE_SIZE, "%s\n",
3108                         sdev->hostdata ? "superpipe" : "legacy");
3109}
3110
3111/*
3112 * Host attributes
3113 */
3114static DEVICE_ATTR_RO(port0);
3115static DEVICE_ATTR_RO(port1);
3116static DEVICE_ATTR_RO(port2);
3117static DEVICE_ATTR_RO(port3);
3118static DEVICE_ATTR_RW(lun_mode);
3119static DEVICE_ATTR_RO(ioctl_version);
3120static DEVICE_ATTR_RO(port0_lun_table);
3121static DEVICE_ATTR_RO(port1_lun_table);
3122static DEVICE_ATTR_RO(port2_lun_table);
3123static DEVICE_ATTR_RO(port3_lun_table);
3124static DEVICE_ATTR_RW(irqpoll_weight);
3125static DEVICE_ATTR_RW(num_hwqs);
3126static DEVICE_ATTR_RW(hwq_mode);
3127
3128static struct device_attribute *cxlflash_host_attrs[] = {
3129        &dev_attr_port0,
3130        &dev_attr_port1,
3131        &dev_attr_port2,
3132        &dev_attr_port3,
3133        &dev_attr_lun_mode,
3134        &dev_attr_ioctl_version,
3135        &dev_attr_port0_lun_table,
3136        &dev_attr_port1_lun_table,
3137        &dev_attr_port2_lun_table,
3138        &dev_attr_port3_lun_table,
3139        &dev_attr_irqpoll_weight,
3140        &dev_attr_num_hwqs,
3141        &dev_attr_hwq_mode,
3142        NULL
3143};
3144
3145/*
3146 * Device attributes
3147 */
3148static DEVICE_ATTR_RO(mode);
3149
3150static struct device_attribute *cxlflash_dev_attrs[] = {
3151        &dev_attr_mode,
3152        NULL
3153};
3154
3155/*
3156 * Host template
3157 */
3158static struct scsi_host_template driver_template = {
3159        .module = THIS_MODULE,
3160        .name = CXLFLASH_ADAPTER_NAME,
3161        .info = cxlflash_driver_info,
3162        .ioctl = cxlflash_ioctl,
3163        .proc_name = CXLFLASH_NAME,
3164        .queuecommand = cxlflash_queuecommand,
3165        .eh_abort_handler = cxlflash_eh_abort_handler,
3166        .eh_device_reset_handler = cxlflash_eh_device_reset_handler,
3167        .eh_host_reset_handler = cxlflash_eh_host_reset_handler,
3168        .change_queue_depth = cxlflash_change_queue_depth,
3169        .cmd_per_lun = CXLFLASH_MAX_CMDS_PER_LUN,
3170        .can_queue = CXLFLASH_MAX_CMDS,
3171        .cmd_size = sizeof(struct afu_cmd) + __alignof__(struct afu_cmd) - 1,
3172        .this_id = -1,
3173        .sg_tablesize = 1,      /* No scatter gather support */
3174        .max_sectors = CXLFLASH_MAX_SECTORS,
3175        .shost_attrs = cxlflash_host_attrs,
3176        .sdev_attrs = cxlflash_dev_attrs,
3177};
3178
3179/*
3180 * Device dependent values
3181 */
3182static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS,
3183                                        CXLFLASH_WWPN_VPD_REQUIRED };
3184static struct dev_dependent_vals dev_flash_gt_vals = { CXLFLASH_MAX_SECTORS,
3185                                        CXLFLASH_NOTIFY_SHUTDOWN };
3186static struct dev_dependent_vals dev_briard_vals = { CXLFLASH_MAX_SECTORS,
3187                                        (CXLFLASH_NOTIFY_SHUTDOWN |
3188                                        CXLFLASH_OCXL_DEV) };
3189
3190/*
3191 * PCI device binding table
3192 */
3193static struct pci_device_id cxlflash_pci_table[] = {
3194        {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
3195         PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
3196        {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_FLASH_GT,
3197         PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_flash_gt_vals},
3198        {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_BRIARD,
3199         PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_briard_vals},
3200        {}
3201};
3202
3203MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
3204
3205/**
3206 * cxlflash_worker_thread() - work thread handler for the AFU
3207 * @work:       Work structure contained within cxlflash associated with host.
3208 *
3209 * Handles the following events:
3210 * - Link reset which cannot be performed on interrupt context due to
3211 * blocking up to a few seconds
3212 * - Rescan the host
3213 */
3214static void cxlflash_worker_thread(struct work_struct *work)
3215{
3216        struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
3217                                                work_q);
3218        struct afu *afu = cfg->afu;
3219        struct device *dev = &cfg->dev->dev;
3220        __be64 __iomem *fc_port_regs;
3221        int port;
3222        ulong lock_flags;
3223
3224        /* Avoid MMIO if the device has failed */
3225
3226        if (cfg->state != STATE_NORMAL)
3227                return;
3228
3229        spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3230
3231        if (cfg->lr_state == LINK_RESET_REQUIRED) {
3232                port = cfg->lr_port;
3233                if (port < 0)
3234                        dev_err(dev, "%s: invalid port index %d\n",
3235                                __func__, port);
3236                else {
3237                        spin_unlock_irqrestore(cfg->host->host_lock,
3238                                               lock_flags);
3239
3240                        /* The reset can block... */
3241                        fc_port_regs = get_fc_port_regs(cfg, port);
3242                        afu_link_reset(afu, port, fc_port_regs);
3243                        spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3244                }
3245
3246                cfg->lr_state = LINK_RESET_COMPLETE;
3247        }
3248
3249        spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
3250
3251        if (atomic_dec_if_positive(&cfg->scan_host_needed) >= 0)
3252                scsi_scan_host(cfg->host);
3253}
3254
3255/**
3256 * cxlflash_chr_open() - character device open handler
3257 * @inode:      Device inode associated with this character device.
3258 * @file:       File pointer for this device.
3259 *
3260 * Only users with admin privileges are allowed to open the character device.
3261 *
3262 * Return: 0 on success, -errno on failure
3263 */
3264static int cxlflash_chr_open(struct inode *inode, struct file *file)
3265{
3266        struct cxlflash_cfg *cfg;
3267
3268        if (!capable(CAP_SYS_ADMIN))
3269                return -EACCES;
3270
3271        cfg = container_of(inode->i_cdev, struct cxlflash_cfg, cdev);
3272        file->private_data = cfg;
3273
3274        return 0;
3275}
3276
3277/**
3278 * decode_hioctl() - translates encoded host ioctl to easily identifiable string
3279 * @cmd:        The host ioctl command to decode.
3280 *
3281 * Return: A string identifying the decoded host ioctl.
3282 */
3283static char *decode_hioctl(unsigned int cmd)
3284{
3285        switch (cmd) {
3286        case HT_CXLFLASH_LUN_PROVISION:
3287                return __stringify_1(HT_CXLFLASH_LUN_PROVISION);
3288        }
3289
3290        return "UNKNOWN";
3291}
3292
3293/**
3294 * cxlflash_lun_provision() - host LUN provisioning handler
3295 * @cfg:        Internal structure associated with the host.
3296 * @lunprov:    Kernel copy of userspace ioctl data structure.
3297 *
3298 * Return: 0 on success, -errno on failure
3299 */
3300static int cxlflash_lun_provision(struct cxlflash_cfg *cfg,
3301                                  struct ht_cxlflash_lun_provision *lunprov)
3302{
3303        struct afu *afu = cfg->afu;
3304        struct device *dev = &cfg->dev->dev;
3305        struct sisl_ioarcb rcb;
3306        struct sisl_ioasa asa;
3307        __be64 __iomem *fc_port_regs;
3308        u16 port = lunprov->port;
3309        u16 scmd = lunprov->hdr.subcmd;
3310        u16 type;
3311        u64 reg;
3312        u64 size;
3313        u64 lun_id;
3314        int rc = 0;
3315
3316        if (!afu_is_lun_provision(afu)) {
3317                rc = -ENOTSUPP;
3318                goto out;
3319        }
3320
3321        if (port >= cfg->num_fc_ports) {
3322                rc = -EINVAL;
3323                goto out;
3324        }
3325
3326        switch (scmd) {
3327        case HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN:
3328                type = SISL_AFU_LUN_PROVISION_CREATE;
3329                size = lunprov->size;
3330                lun_id = 0;
3331                break;
3332        case HT_CXLFLASH_LUN_PROVISION_SUBCMD_DELETE_LUN:
3333                type = SISL_AFU_LUN_PROVISION_DELETE;
3334                size = 0;
3335                lun_id = lunprov->lun_id;
3336                break;
3337        case HT_CXLFLASH_LUN_PROVISION_SUBCMD_QUERY_PORT:
3338                fc_port_regs = get_fc_port_regs(cfg, port);
3339
3340                reg = readq_be(&fc_port_regs[FC_MAX_NUM_LUNS / 8]);
3341                lunprov->max_num_luns = reg;
3342                reg = readq_be(&fc_port_regs[FC_CUR_NUM_LUNS / 8]);
3343                lunprov->cur_num_luns = reg;
3344                reg = readq_be(&fc_port_regs[FC_MAX_CAP_PORT / 8]);
3345                lunprov->max_cap_port = reg;
3346                reg = readq_be(&fc_port_regs[FC_CUR_CAP_PORT / 8]);
3347                lunprov->cur_cap_port = reg;
3348
3349                goto out;
3350        default:
3351                rc = -EINVAL;
3352                goto out;
3353        }
3354
3355        memset(&rcb, 0, sizeof(rcb));
3356        memset(&asa, 0, sizeof(asa));
3357        rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
3358        rcb.lun_id = lun_id;
3359        rcb.msi = SISL_MSI_RRQ_UPDATED;
3360        rcb.timeout = MC_LUN_PROV_TIMEOUT;
3361        rcb.ioasa = &asa;
3362
3363        rcb.cdb[0] = SISL_AFU_CMD_LUN_PROVISION;
3364        rcb.cdb[1] = type;
3365        rcb.cdb[2] = port;
3366        put_unaligned_be64(size, &rcb.cdb[8]);
3367
3368        rc = send_afu_cmd(afu, &rcb);
3369        if (rc) {
3370                dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3371                        __func__, rc, asa.ioasc, asa.afu_extra);
3372                goto out;
3373        }
3374
3375        if (scmd == HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN) {
3376                lunprov->lun_id = (u64)asa.lunid_hi << 32 | asa.lunid_lo;
3377                memcpy(lunprov->wwid, asa.wwid, sizeof(lunprov->wwid));
3378        }
3379out:
3380        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3381        return rc;
3382}
3383
3384/**
3385 * cxlflash_afu_debug() - host AFU debug handler
3386 * @cfg:        Internal structure associated with the host.
3387 * @afu_dbg:    Kernel copy of userspace ioctl data structure.
3388 *
3389 * For debug requests requiring a data buffer, always provide an aligned
3390 * (cache line) buffer to the AFU to appease any alignment requirements.
3391 *
3392 * Return: 0 on success, -errno on failure
3393 */
3394static int cxlflash_afu_debug(struct cxlflash_cfg *cfg,
3395                              struct ht_cxlflash_afu_debug *afu_dbg)
3396{
3397        struct afu *afu = cfg->afu;
3398        struct device *dev = &cfg->dev->dev;
3399        struct sisl_ioarcb rcb;
3400        struct sisl_ioasa asa;
3401        char *buf = NULL;
3402        char *kbuf = NULL;
3403        void __user *ubuf = (__force void __user *)afu_dbg->data_ea;
3404        u16 req_flags = SISL_REQ_FLAGS_AFU_CMD;
3405        u32 ulen = afu_dbg->data_len;
3406        bool is_write = afu_dbg->hdr.flags & HT_CXLFLASH_HOST_WRITE;
3407        int rc = 0;
3408
3409        if (!afu_is_afu_debug(afu)) {
3410                rc = -ENOTSUPP;
3411                goto out;
3412        }
3413
3414        if (ulen) {
3415                req_flags |= SISL_REQ_FLAGS_SUP_UNDERRUN;
3416
3417                if (ulen > HT_CXLFLASH_AFU_DEBUG_MAX_DATA_LEN) {
3418                        rc = -EINVAL;
3419                        goto out;
3420                }
3421
3422                buf = kmalloc(ulen + cache_line_size() - 1, GFP_KERNEL);
3423                if (unlikely(!buf)) {
3424                        rc = -ENOMEM;
3425                        goto out;
3426                }
3427
3428                kbuf = PTR_ALIGN(buf, cache_line_size());
3429
3430                if (is_write) {
3431                        req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
3432
3433                        if (copy_from_user(kbuf, ubuf, ulen)) {
3434                                rc = -EFAULT;
3435                                goto out;
3436                        }
3437                }
3438        }
3439
3440        memset(&rcb, 0, sizeof(rcb));
3441        memset(&asa, 0, sizeof(asa));
3442
3443        rcb.req_flags = req_flags;
3444        rcb.msi = SISL_MSI_RRQ_UPDATED;
3445        rcb.timeout = MC_AFU_DEBUG_TIMEOUT;
3446        rcb.ioasa = &asa;
3447
3448        if (ulen) {
3449                rcb.data_len = ulen;
3450                rcb.data_ea = (uintptr_t)kbuf;
3451        }
3452
3453        rcb.cdb[0] = SISL_AFU_CMD_DEBUG;
3454        memcpy(&rcb.cdb[4], afu_dbg->afu_subcmd,
3455               HT_CXLFLASH_AFU_DEBUG_SUBCMD_LEN);
3456
3457        rc = send_afu_cmd(afu, &rcb);
3458        if (rc) {
3459                dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3460                        __func__, rc, asa.ioasc, asa.afu_extra);
3461                goto out;
3462        }
3463
3464        if (ulen && !is_write) {
3465                if (copy_to_user(ubuf, kbuf, ulen))
3466                        rc = -EFAULT;
3467        }
3468out:
3469        kfree(buf);
3470        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3471        return rc;
3472}
3473
3474/**
3475 * cxlflash_chr_ioctl() - character device IOCTL handler
3476 * @file:       File pointer for this device.
3477 * @cmd:        IOCTL command.
3478 * @arg:        Userspace ioctl data structure.
3479 *
3480 * A read/write semaphore is used to implement a 'drain' of currently
3481 * running ioctls. The read semaphore is taken at the beginning of each
3482 * ioctl thread and released upon concluding execution. Additionally the
3483 * semaphore should be released and then reacquired in any ioctl execution
3484 * path which will wait for an event to occur that is outside the scope of
3485 * the ioctl (i.e. an adapter reset). To drain the ioctls currently running,
3486 * a thread simply needs to acquire the write semaphore.
3487 *
3488 * Return: 0 on success, -errno on failure
3489 */
3490static long cxlflash_chr_ioctl(struct file *file, unsigned int cmd,
3491                               unsigned long arg)
3492{
3493        typedef int (*hioctl) (struct cxlflash_cfg *, void *);
3494
3495        struct cxlflash_cfg *cfg = file->private_data;
3496        struct device *dev = &cfg->dev->dev;
3497        char buf[sizeof(union cxlflash_ht_ioctls)];
3498        void __user *uarg = (void __user *)arg;
3499        struct ht_cxlflash_hdr *hdr;
3500        size_t size = 0;
3501        bool known_ioctl = false;
3502        int idx = 0;
3503        int rc = 0;
3504        hioctl do_ioctl = NULL;
3505
3506        static const struct {
3507                size_t size;
3508                hioctl ioctl;
3509        } ioctl_tbl[] = {       /* NOTE: order matters here */
3510        { sizeof(struct ht_cxlflash_lun_provision),
3511                (hioctl)cxlflash_lun_provision },
3512        { sizeof(struct ht_cxlflash_afu_debug),
3513                (hioctl)cxlflash_afu_debug },
3514        };
3515
3516        /* Hold read semaphore so we can drain if needed */
3517        down_read(&cfg->ioctl_rwsem);
3518
3519        dev_dbg(dev, "%s: cmd=%u idx=%d tbl_size=%lu\n",
3520                __func__, cmd, idx, sizeof(ioctl_tbl));
3521
3522        switch (cmd) {
3523        case HT_CXLFLASH_LUN_PROVISION:
3524        case HT_CXLFLASH_AFU_DEBUG:
3525                known_ioctl = true;
3526                idx = _IOC_NR(HT_CXLFLASH_LUN_PROVISION) - _IOC_NR(cmd);
3527                size = ioctl_tbl[idx].size;
3528                do_ioctl = ioctl_tbl[idx].ioctl;
3529
3530                if (likely(do_ioctl))
3531                        break;
3532
3533                fallthrough;
3534        default:
3535                rc = -EINVAL;
3536                goto out;
3537        }
3538
3539        if (unlikely(copy_from_user(&buf, uarg, size))) {
3540                dev_err(dev, "%s: copy_from_user() fail "
3541                        "size=%lu cmd=%d (%s) uarg=%p\n",
3542                        __func__, size, cmd, decode_hioctl(cmd), uarg);
3543                rc = -EFAULT;
3544                goto out;
3545        }
3546
3547        hdr = (struct ht_cxlflash_hdr *)&buf;
3548        if (hdr->version != HT_CXLFLASH_VERSION_0) {
3549                dev_dbg(dev, "%s: Version %u not supported for %s\n",
3550                        __func__, hdr->version, decode_hioctl(cmd));
3551                rc = -EINVAL;
3552                goto out;
3553        }
3554
3555        if (hdr->rsvd[0] || hdr->rsvd[1] || hdr->return_flags) {
3556                dev_dbg(dev, "%s: Reserved/rflags populated\n", __func__);
3557                rc = -EINVAL;
3558                goto out;
3559        }
3560
3561        rc = do_ioctl(cfg, (void *)&buf);
3562        if (likely(!rc))
3563                if (unlikely(copy_to_user(uarg, &buf, size))) {
3564                        dev_err(dev, "%s: copy_to_user() fail "
3565                                "size=%lu cmd=%d (%s) uarg=%p\n",
3566                                __func__, size, cmd, decode_hioctl(cmd), uarg);
3567                        rc = -EFAULT;
3568                }
3569
3570        /* fall through to exit */
3571
3572out:
3573        up_read(&cfg->ioctl_rwsem);
3574        if (unlikely(rc && known_ioctl))
3575                dev_err(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3576                        __func__, decode_hioctl(cmd), cmd, rc);
3577        else
3578                dev_dbg(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3579                        __func__, decode_hioctl(cmd), cmd, rc);
3580        return rc;
3581}
3582
3583/*
3584 * Character device file operations
3585 */
3586static const struct file_operations cxlflash_chr_fops = {
3587        .owner          = THIS_MODULE,
3588        .open           = cxlflash_chr_open,
3589        .unlocked_ioctl = cxlflash_chr_ioctl,
3590        .compat_ioctl   = compat_ptr_ioctl,
3591};
3592
3593/**
3594 * init_chrdev() - initialize the character device for the host
3595 * @cfg:        Internal structure associated with the host.
3596 *
3597 * Return: 0 on success, -errno on failure
3598 */
3599static int init_chrdev(struct cxlflash_cfg *cfg)
3600{
3601        struct device *dev = &cfg->dev->dev;
3602        struct device *char_dev;
3603        dev_t devno;
3604        int minor;
3605        int rc = 0;
3606
3607        minor = cxlflash_get_minor();
3608        if (unlikely(minor < 0)) {
3609                dev_err(dev, "%s: Exhausted allowed adapters\n", __func__);
3610                rc = -ENOSPC;
3611                goto out;
3612        }
3613
3614        devno = MKDEV(cxlflash_major, minor);
3615        cdev_init(&cfg->cdev, &cxlflash_chr_fops);
3616
3617        rc = cdev_add(&cfg->cdev, devno, 1);
3618        if (rc) {
3619                dev_err(dev, "%s: cdev_add failed rc=%d\n", __func__, rc);
3620                goto err1;
3621        }
3622
3623        char_dev = device_create(cxlflash_class, NULL, devno,
3624                                 NULL, "cxlflash%d", minor);
3625        if (IS_ERR(char_dev)) {
3626                rc = PTR_ERR(char_dev);
3627                dev_err(dev, "%s: device_create failed rc=%d\n",
3628                        __func__, rc);
3629                goto err2;
3630        }
3631
3632        cfg->chardev = char_dev;
3633out:
3634        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3635        return rc;
3636err2:
3637        cdev_del(&cfg->cdev);
3638err1:
3639        cxlflash_put_minor(minor);
3640        goto out;
3641}
3642
3643/**
3644 * cxlflash_probe() - PCI entry point to add host
3645 * @pdev:       PCI device associated with the host.
3646 * @dev_id:     PCI device id associated with device.
3647 *
3648 * The device will initially start out in a 'probing' state and
3649 * transition to the 'normal' state at the end of a successful
3650 * probe. Should an EEH event occur during probe, the notification
3651 * thread (error_detected()) will wait until the probe handler
3652 * is nearly complete. At that time, the device will be moved to
3653 * a 'probed' state and the EEH thread woken up to drive the slot
3654 * reset and recovery (device moves to 'normal' state). Meanwhile,
3655 * the probe will be allowed to exit successfully.
3656 *
3657 * Return: 0 on success, -errno on failure
3658 */
3659static int cxlflash_probe(struct pci_dev *pdev,
3660                          const struct pci_device_id *dev_id)
3661{
3662        struct Scsi_Host *host;
3663        struct cxlflash_cfg *cfg = NULL;
3664        struct device *dev = &pdev->dev;
3665        struct dev_dependent_vals *ddv;
3666        int rc = 0;
3667        int k;
3668
3669        dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
3670                __func__, pdev->irq);
3671
3672        ddv = (struct dev_dependent_vals *)dev_id->driver_data;
3673        driver_template.max_sectors = ddv->max_sectors;
3674
3675        host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
3676        if (!host) {
3677                dev_err(dev, "%s: scsi_host_alloc failed\n", __func__);
3678                rc = -ENOMEM;
3679                goto out;
3680        }
3681
3682        host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
3683        host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
3684        host->unique_id = host->host_no;
3685        host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
3686
3687        cfg = shost_priv(host);
3688        cfg->state = STATE_PROBING;
3689        cfg->host = host;
3690        rc = alloc_mem(cfg);
3691        if (rc) {
3692                dev_err(dev, "%s: alloc_mem failed\n", __func__);
3693                rc = -ENOMEM;
3694                scsi_host_put(cfg->host);
3695                goto out;
3696        }
3697
3698        cfg->init_state = INIT_STATE_NONE;
3699        cfg->dev = pdev;
3700        cfg->cxl_fops = cxlflash_cxl_fops;
3701        cfg->ops = cxlflash_assign_ops(ddv);
3702        WARN_ON_ONCE(!cfg->ops);
3703
3704        /*
3705         * Promoted LUNs move to the top of the LUN table. The rest stay on
3706         * the bottom half. The bottom half grows from the end (index = 255),
3707         * whereas the top half grows from the beginning (index = 0).
3708         *
3709         * Initialize the last LUN index for all possible ports.
3710         */
3711        cfg->promote_lun_index = 0;
3712
3713        for (k = 0; k < MAX_FC_PORTS; k++)
3714                cfg->last_lun_index[k] = CXLFLASH_NUM_VLUNS/2 - 1;
3715
3716        cfg->dev_id = (struct pci_device_id *)dev_id;
3717
3718        init_waitqueue_head(&cfg->tmf_waitq);
3719        init_waitqueue_head(&cfg->reset_waitq);
3720
3721        INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
3722        cfg->lr_state = LINK_RESET_INVALID;
3723        cfg->lr_port = -1;
3724        spin_lock_init(&cfg->tmf_slock);
3725        mutex_init(&cfg->ctx_tbl_list_mutex);
3726        mutex_init(&cfg->ctx_recovery_mutex);
3727        init_rwsem(&cfg->ioctl_rwsem);
3728        INIT_LIST_HEAD(&cfg->ctx_err_recovery);
3729        INIT_LIST_HEAD(&cfg->lluns);
3730
3731        pci_set_drvdata(pdev, cfg);
3732
3733        rc = init_pci(cfg);
3734        if (rc) {
3735                dev_err(dev, "%s: init_pci failed rc=%d\n", __func__, rc);
3736                goto out_remove;
3737        }
3738        cfg->init_state = INIT_STATE_PCI;
3739
3740        cfg->afu_cookie = cfg->ops->create_afu(pdev);
3741        if (unlikely(!cfg->afu_cookie)) {
3742                dev_err(dev, "%s: create_afu failed\n", __func__);
3743                rc = -ENOMEM;
3744                goto out_remove;
3745        }
3746
3747        rc = init_afu(cfg);
3748        if (rc && !wq_has_sleeper(&cfg->reset_waitq)) {
3749                dev_err(dev, "%s: init_afu failed rc=%d\n", __func__, rc);
3750                goto out_remove;
3751        }
3752        cfg->init_state = INIT_STATE_AFU;
3753
3754        rc = init_scsi(cfg);
3755        if (rc) {
3756                dev_err(dev, "%s: init_scsi failed rc=%d\n", __func__, rc);
3757                goto out_remove;
3758        }
3759        cfg->init_state = INIT_STATE_SCSI;
3760
3761        rc = init_chrdev(cfg);
3762        if (rc) {
3763                dev_err(dev, "%s: init_chrdev failed rc=%d\n", __func__, rc);
3764                goto out_remove;
3765        }
3766        cfg->init_state = INIT_STATE_CDEV;
3767
3768        if (wq_has_sleeper(&cfg->reset_waitq)) {
3769                cfg->state = STATE_PROBED;
3770                wake_up_all(&cfg->reset_waitq);
3771        } else
3772                cfg->state = STATE_NORMAL;
3773out:
3774        dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3775        return rc;
3776
3777out_remove:
3778        cfg->state = STATE_PROBED;
3779        cxlflash_remove(pdev);
3780        goto out;
3781}
3782
3783/**
3784 * cxlflash_pci_error_detected() - called when a PCI error is detected
3785 * @pdev:       PCI device struct.
3786 * @state:      PCI channel state.
3787 *
3788 * When an EEH occurs during an active reset, wait until the reset is
3789 * complete and then take action based upon the device state.
3790 *
3791 * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
3792 */
3793static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
3794                                                    pci_channel_state_t state)
3795{
3796        int rc = 0;
3797        struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3798        struct device *dev = &cfg->dev->dev;
3799
3800        dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
3801
3802        switch (state) {
3803        case pci_channel_io_frozen:
3804                wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
3805                                             cfg->state != STATE_PROBING);
3806                if (cfg->state == STATE_FAILTERM)
3807                        return PCI_ERS_RESULT_DISCONNECT;
3808
3809                cfg->state = STATE_RESET;
3810                scsi_block_requests(cfg->host);
3811                drain_ioctls(cfg);
3812                rc = cxlflash_mark_contexts_error(cfg);
3813                if (unlikely(rc))
3814                        dev_err(dev, "%s: Failed to mark user contexts rc=%d\n",
3815                                __func__, rc);
3816                term_afu(cfg);
3817                return PCI_ERS_RESULT_NEED_RESET;
3818        case pci_channel_io_perm_failure:
3819                cfg->state = STATE_FAILTERM;
3820                wake_up_all(&cfg->reset_waitq);
3821                scsi_unblock_requests(cfg->host);
3822                return PCI_ERS_RESULT_DISCONNECT;
3823        default:
3824                break;
3825        }
3826        return PCI_ERS_RESULT_NEED_RESET;
3827}
3828
3829/**
3830 * cxlflash_pci_slot_reset() - called when PCI slot has been reset
3831 * @pdev:       PCI device struct.
3832 *
3833 * This routine is called by the pci error recovery code after the PCI
3834 * slot has been reset, just before we should resume normal operations.
3835 *
3836 * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
3837 */
3838static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
3839{
3840        int rc = 0;
3841        struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3842        struct device *dev = &cfg->dev->dev;
3843
3844        dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3845
3846        rc = init_afu(cfg);
3847        if (unlikely(rc)) {
3848                dev_err(dev, "%s: EEH recovery failed rc=%d\n", __func__, rc);
3849                return PCI_ERS_RESULT_DISCONNECT;
3850        }
3851
3852        return PCI_ERS_RESULT_RECOVERED;
3853}
3854
3855/**
3856 * cxlflash_pci_resume() - called when normal operation can resume
3857 * @pdev:       PCI device struct
3858 */
3859static void cxlflash_pci_resume(struct pci_dev *pdev)
3860{
3861        struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3862        struct device *dev = &cfg->dev->dev;
3863
3864        dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3865
3866        cfg->state = STATE_NORMAL;
3867        wake_up_all(&cfg->reset_waitq);
3868        scsi_unblock_requests(cfg->host);
3869}
3870
3871/**
3872 * cxlflash_devnode() - provides devtmpfs for devices in the cxlflash class
3873 * @dev:        Character device.
3874 * @mode:       Mode that can be used to verify access.
3875 *
3876 * Return: Allocated string describing the devtmpfs structure.
3877 */
3878static char *cxlflash_devnode(struct device *dev, umode_t *mode)
3879{
3880        return kasprintf(GFP_KERNEL, "cxlflash/%s", dev_name(dev));
3881}
3882
3883/**
3884 * cxlflash_class_init() - create character device class
3885 *
3886 * Return: 0 on success, -errno on failure
3887 */
3888static int cxlflash_class_init(void)
3889{
3890        dev_t devno;
3891        int rc = 0;
3892
3893        rc = alloc_chrdev_region(&devno, 0, CXLFLASH_MAX_ADAPTERS, "cxlflash");
3894        if (unlikely(rc)) {
3895                pr_err("%s: alloc_chrdev_region failed rc=%d\n", __func__, rc);
3896                goto out;
3897        }
3898
3899        cxlflash_major = MAJOR(devno);
3900
3901        cxlflash_class = class_create(THIS_MODULE, "cxlflash");
3902        if (IS_ERR(cxlflash_class)) {
3903                rc = PTR_ERR(cxlflash_class);
3904                pr_err("%s: class_create failed rc=%d\n", __func__, rc);
3905                goto err;
3906        }
3907
3908        cxlflash_class->devnode = cxlflash_devnode;
3909out:
3910        pr_debug("%s: returning rc=%d\n", __func__, rc);
3911        return rc;
3912err:
3913        unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3914        goto out;
3915}
3916
3917/**
3918 * cxlflash_class_exit() - destroy character device class
3919 */
3920static void cxlflash_class_exit(void)
3921{
3922        dev_t devno = MKDEV(cxlflash_major, 0);
3923
3924        class_destroy(cxlflash_class);
3925        unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3926}
3927
3928static const struct pci_error_handlers cxlflash_err_handler = {
3929        .error_detected = cxlflash_pci_error_detected,
3930        .slot_reset = cxlflash_pci_slot_reset,
3931        .resume = cxlflash_pci_resume,
3932};
3933
3934/*
3935 * PCI device structure
3936 */
3937static struct pci_driver cxlflash_driver = {
3938        .name = CXLFLASH_NAME,
3939        .id_table = cxlflash_pci_table,
3940        .probe = cxlflash_probe,
3941        .remove = cxlflash_remove,
3942        .shutdown = cxlflash_remove,
3943        .err_handler = &cxlflash_err_handler,
3944};
3945
3946/**
3947 * init_cxlflash() - module entry point
3948 *
3949 * Return: 0 on success, -errno on failure
3950 */
3951static int __init init_cxlflash(void)
3952{
3953        int rc;
3954
3955        check_sizes();
3956        cxlflash_list_init();
3957        rc = cxlflash_class_init();
3958        if (unlikely(rc))
3959                goto out;
3960
3961        rc = pci_register_driver(&cxlflash_driver);
3962        if (unlikely(rc))
3963                goto err;
3964out:
3965        pr_debug("%s: returning rc=%d\n", __func__, rc);
3966        return rc;
3967err:
3968        cxlflash_class_exit();
3969        goto out;
3970}
3971
3972/**
3973 * exit_cxlflash() - module exit point
3974 */
3975static void __exit exit_cxlflash(void)
3976{
3977        cxlflash_term_global_luns();
3978        cxlflash_free_errpage();
3979
3980        pci_unregister_driver(&cxlflash_driver);
3981        cxlflash_class_exit();
3982}
3983
3984module_init(init_cxlflash);
3985module_exit(exit_cxlflash);
3986