drivers/scsi/lpfc/lpfc_nvme.c

   1 /*******************************************************************
   2  * This file is part of the Emulex Linux Device Driver for         *
   3  * Fibre Channel Host Bus Adapters.                                *
   4  * Copyright (C) 2017-2018 Broadcom. All Rights Reserved. The term *
   5  * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.  *
   6  * Copyright (C) 2004-2016 Emulex.  All rights reserved.           *
   7  * EMULEX and SLI are trademarks of Emulex.                        *
   8  * www.broadcom.com                                                *
   9  * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
  10  *                                                                 *
  11  * This program is free software; you can redistribute it and/or   *
  12  * modify it under the terms of version 2 of the GNU General       *
  13  * Public License as published by the Free Software Foundation.    *
  14  * This program is distributed in the hope that it will be useful. *
  15  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
  16  * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
  17  * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
  18  * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
  19  * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
  20  * more details, a copy of which can be found in the file COPYING  *
  21  * included with this package.                                     *
  22  ********************************************************************/
  23 #include <linux/pci.h>
  24 #include <linux/slab.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/delay.h>
  27 #include <asm/unaligned.h>
  28 #include <linux/crc-t10dif.h>
  29 #include <net/checksum.h>
  30
  31 #include <scsi/scsi.h>
  32 #include <scsi/scsi_device.h>
  33 #include <scsi/scsi_eh.h>
  34 #include <scsi/scsi_host.h>
  35 #include <scsi/scsi_tcq.h>
  36 #include <scsi/scsi_transport_fc.h>
  37 #include <scsi/fc/fc_fs.h>
  38
  39 #include <linux/nvme.h>
  40 #include <linux/nvme-fc-driver.h>
  41 #include <linux/nvme-fc.h>
  42 #include "lpfc_version.h"
  43 #include "lpfc_hw4.h"
  44 #include "lpfc_hw.h"
  45 #include "lpfc_sli.h"
  46 #include "lpfc_sli4.h"
  47 #include "lpfc_nl.h"
  48 #include "lpfc_disc.h"
  49 #include "lpfc.h"
  50 #include "lpfc_nvme.h"
  51 #include "lpfc_scsi.h"
  52 #include "lpfc_logmsg.h"
  53 #include "lpfc_crtn.h"
  54 #include "lpfc_vport.h"
  55 #include "lpfc_debugfs.h"
  56
  57 /* NVME initiator-based functions */
  58
  59 static struct lpfc_nvme_buf *
  60 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
  61                   int idx, int expedite);
  62
  63 static void
  64 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_buf *);
  65
  66 static struct nvme_fc_port_template lpfc_nvme_template;
  67
  68 static union lpfc_wqe128 lpfc_iread_cmd_template;
  69 static union lpfc_wqe128 lpfc_iwrite_cmd_template;
  70 static union lpfc_wqe128 lpfc_icmnd_cmd_template;
  71
  72 /* Setup WQE templates for NVME IOs */
  73 void
  74 lpfc_nvme_cmd_template(void)
  75 {
  76         union lpfc_wqe128 *wqe;
  77
  78         /* IREAD template */
  79         wqe = &lpfc_iread_cmd_template;
  80         memset(wqe, 0, sizeof(union lpfc_wqe128));
  81
  82         /* Word 0, 1, 2 - BDE is variable */
  83
  84         /* Word 3 - cmd_buff_len, payload_offset_len is zero */
  85
  86         /* Word 4 - total_xfer_len is variable */
  87
  88         /* Word 5 - is zero */
  89
  90         /* Word 6 - ctxt_tag, xri_tag is variable */
  91
  92         /* Word 7 */
  93         bf_set(wqe_cmnd, &wqe->fcp_iread.wqe_com, CMD_FCP_IREAD64_WQE);
  94         bf_set(wqe_pu, &wqe->fcp_iread.wqe_com, PARM_READ_CHECK);
  95         bf_set(wqe_class, &wqe->fcp_iread.wqe_com, CLASS3);
  96         bf_set(wqe_ct, &wqe->fcp_iread.wqe_com, SLI4_CT_RPI);
  97
  98         /* Word 8 - abort_tag is variable */
  99
 100         /* Word 9  - reqtag is variable */
 101
 102         /* Word 10 - dbde, wqes is variable */
 103         bf_set(wqe_qosd, &wqe->fcp_iread.wqe_com, 0);
 104         bf_set(wqe_nvme, &wqe->fcp_iread.wqe_com, 1);
 105         bf_set(wqe_iod, &wqe->fcp_iread.wqe_com, LPFC_WQE_IOD_READ);
 106         bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com, LPFC_WQE_LENLOC_WORD4);
 107         bf_set(wqe_dbde, &wqe->fcp_iread.wqe_com, 0);
 108         bf_set(wqe_wqes, &wqe->fcp_iread.wqe_com, 1);
 109
 110         /* Word 11 - pbde is variable */
 111         bf_set(wqe_cmd_type, &wqe->fcp_iread.wqe_com, NVME_READ_CMD);
 112         bf_set(wqe_cqid, &wqe->fcp_iread.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
 113         bf_set(wqe_pbde, &wqe->fcp_iread.wqe_com, 1);
 114
 115         /* Word 12 - is zero */
 116
 117         /* Word 13, 14, 15 - PBDE is variable */
 118
 119         /* IWRITE template */
 120         wqe = &lpfc_iwrite_cmd_template;
 121         memset(wqe, 0, sizeof(union lpfc_wqe128));
 122
 123         /* Word 0, 1, 2 - BDE is variable */
 124
 125         /* Word 3 - cmd_buff_len, payload_offset_len is zero */
 126
 127         /* Word 4 - total_xfer_len is variable */
 128
 129         /* Word 5 - initial_xfer_len is variable */
 130
 131         /* Word 6 - ctxt_tag, xri_tag is variable */
 132
 133         /* Word 7 */
 134         bf_set(wqe_cmnd, &wqe->fcp_iwrite.wqe_com, CMD_FCP_IWRITE64_WQE);
 135         bf_set(wqe_pu, &wqe->fcp_iwrite.wqe_com, PARM_READ_CHECK);
 136         bf_set(wqe_class, &wqe->fcp_iwrite.wqe_com, CLASS3);
 137         bf_set(wqe_ct, &wqe->fcp_iwrite.wqe_com, SLI4_CT_RPI);
 138
 139         /* Word 8 - abort_tag is variable */
 140
 141         /* Word 9  - reqtag is variable */
 142
 143         /* Word 10 - dbde, wqes is variable */
 144         bf_set(wqe_qosd, &wqe->fcp_iwrite.wqe_com, 0);
 145         bf_set(wqe_nvme, &wqe->fcp_iwrite.wqe_com, 1);
 146         bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_IOD_WRITE);
 147         bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_LENLOC_WORD4);
 148         bf_set(wqe_dbde, &wqe->fcp_iwrite.wqe_com, 0);
 149         bf_set(wqe_wqes, &wqe->fcp_iwrite.wqe_com, 1);
 150
 151         /* Word 11 - pbde is variable */
 152         bf_set(wqe_cmd_type, &wqe->fcp_iwrite.wqe_com, NVME_WRITE_CMD);
 153         bf_set(wqe_cqid, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
 154         bf_set(wqe_pbde, &wqe->fcp_iwrite.wqe_com, 1);
 155
 156         /* Word 12 - is zero */
 157
 158         /* Word 13, 14, 15 - PBDE is variable */
 159
 160         /* ICMND template */
 161         wqe = &lpfc_icmnd_cmd_template;
 162         memset(wqe, 0, sizeof(union lpfc_wqe128));
 163
 164         /* Word 0, 1, 2 - BDE is variable */
 165
 166         /* Word 3 - payload_offset_len is variable */
 167
 168         /* Word 4, 5 - is zero */
 169
 170         /* Word 6 - ctxt_tag, xri_tag is variable */
 171
 172         /* Word 7 */
 173         bf_set(wqe_cmnd, &wqe->fcp_icmd.wqe_com, CMD_FCP_ICMND64_WQE);
 174         bf_set(wqe_pu, &wqe->fcp_icmd.wqe_com, 0);
 175         bf_set(wqe_class, &wqe->fcp_icmd.wqe_com, CLASS3);
 176         bf_set(wqe_ct, &wqe->fcp_icmd.wqe_com, SLI4_CT_RPI);
 177
 178         /* Word 8 - abort_tag is variable */
 179
 180         /* Word 9  - reqtag is variable */
 181
 182         /* Word 10 - dbde, wqes is variable */
 183         bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1);
 184         bf_set(wqe_nvme, &wqe->fcp_icmd.wqe_com, 1);
 185         bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_NONE);
 186         bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com, LPFC_WQE_LENLOC_NONE);
 187         bf_set(wqe_dbde, &wqe->fcp_icmd.wqe_com, 0);
 188         bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1);
 189
 190         /* Word 11 */
 191         bf_set(wqe_cmd_type, &wqe->fcp_icmd.wqe_com, FCP_COMMAND);
 192         bf_set(wqe_cqid, &wqe->fcp_icmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
 193         bf_set(wqe_pbde, &wqe->fcp_icmd.wqe_com, 0);
 194
 195         /* Word 12, 13, 14, 15 - is zero */
 196 }
 197
 198 /**
 199  * lpfc_nvme_create_queue -
 200  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 201  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
 202  * @handle: An opaque driver handle used in follow-up calls.
 203  *
 204  * Driver registers this routine to preallocate and initialize any
 205  * internal data structures to bind the @qidx to its internal IO queues.
 206  * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ.
 207  *
 208  * Return value :
 209  *   0 - Success
 210  *   -EINVAL - Unsupported input value.
 211  *   -ENOMEM - Could not alloc necessary memory
 212  **/
 213 static int
 214 lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport,
 215                        unsigned int qidx, u16 qsize,
 216                        void **handle)
 217 {
 218         struct lpfc_nvme_lport *lport;
 219         struct lpfc_vport *vport;
 220         struct lpfc_nvme_qhandle *qhandle;
 221         char *str;
 222
 223         if (!pnvme_lport->private)
 224                 return -ENOMEM;
 225
 226         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 227         vport = lport->vport;
 228         qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
 229         if (qhandle == NULL)
 230                 return -ENOMEM;
 231
 232         qhandle->cpu_id = smp_processor_id();
 233         qhandle->qidx = qidx;
 234         /*
 235          * NVME qidx == 0 is the admin queue, so both admin queue
 236          * and first IO queue will use MSI-X vector and associated
 237          * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
 238          */
 239         if (qidx) {
 240                 str = "IO ";  /* IO queue */
 241                 qhandle->index = ((qidx - 1) %
 242                         vport->phba->cfg_hdw_queue);
 243         } else {
 244                 str = "ADM";  /* Admin queue */
 245                 qhandle->index = qidx;
 246         }
 247
 248         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
 249                          "6073 Binding %s HdwQueue %d  (cpu %d) to "
 250                          "hdw_queue %d qhandle %p\n", str,
 251                          qidx, qhandle->cpu_id, qhandle->index, qhandle);
 252         *handle = (void *)qhandle;
 253         return 0;
 254 }
 255
 256 /**
 257  * lpfc_nvme_delete_queue -
 258  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 259  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
 260  * @handle: An opaque driver handle from lpfc_nvme_create_queue
 261  *
 262  * Driver registers this routine to free
 263  * any internal data structures to bind the @qidx to its internal
 264  * IO queues.
 265  *
 266  * Return value :
 267  *   0 - Success
 268  *   TODO:  What are the failure codes.
 269  **/
 270 static void
 271 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
 272                        unsigned int qidx,
 273                        void *handle)
 274 {
 275         struct lpfc_nvme_lport *lport;
 276         struct lpfc_vport *vport;
 277
 278         if (!pnvme_lport->private)
 279                 return;
 280
 281         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 282         vport = lport->vport;
 283
 284         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
 285                         "6001 ENTER.  lpfc_pnvme %p, qidx x%x qhandle %p\n",
 286                         lport, qidx, handle);
 287         kfree(handle);
 288 }
 289
 290 static void
 291 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
 292 {
 293         struct lpfc_nvme_lport *lport = localport->private;
 294
 295         lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
 296                          "6173 localport %p delete complete\n",
 297                          lport);
 298
 299         /* release any threads waiting for the unreg to complete */
 300         complete(&lport->lport_unreg_done);
 301 }
 302
 303 /* lpfc_nvme_remoteport_delete
 304  *
 305  * @remoteport: Pointer to an nvme transport remoteport instance.
 306  *
 307  * This is a template downcall.  NVME transport calls this function
 308  * when it has completed the unregistration of a previously
 309  * registered remoteport.
 310  *
 311  * Return value :
 312  * None
 313  */
 314 void
 315 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
 316 {
 317         struct lpfc_nvme_rport *rport = remoteport->private;
 318         struct lpfc_vport *vport;
 319         struct lpfc_nodelist *ndlp;
 320
 321         ndlp = rport->ndlp;
 322         if (!ndlp)
 323                 goto rport_err;
 324
 325         vport = ndlp->vport;
 326         if (!vport)
 327                 goto rport_err;
 328
 329         /* Remove this rport from the lport's list - memory is owned by the
 330          * transport. Remove the ndlp reference for the NVME transport before
 331          * calling state machine to remove the node.
 332          */
 333         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 334                         "6146 remoteport delete of remoteport %p\n",
 335                         remoteport);
 336         spin_lock_irq(&vport->phba->hbalock);
 337
 338         /* The register rebind might have occurred before the delete
 339          * downcall.  Guard against this race.
 340          */
 341         if (ndlp->upcall_flags & NLP_WAIT_FOR_UNREG) {
 342                 ndlp->nrport = NULL;
 343                 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
 344         }
 345         spin_unlock_irq(&vport->phba->hbalock);
 346
 347         /* Remove original register reference. The host transport
 348          * won't reference this rport/remoteport any further.
 349          */
 350         lpfc_nlp_put(ndlp);
 351
 352  rport_err:
 353         return;
 354 }
 355
 356 static void
 357 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
 358                        struct lpfc_wcqe_complete *wcqe)
 359 {
 360         struct lpfc_vport *vport = cmdwqe->vport;
 361         struct lpfc_nvme_lport *lport;
 362         uint32_t status;
 363         struct nvmefc_ls_req *pnvme_lsreq;
 364         struct lpfc_dmabuf *buf_ptr;
 365         struct lpfc_nodelist *ndlp;
 366
 367         pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
 368         status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
 369
 370         if (vport->localport) {
 371                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
 372                 if (lport) {
 373                         atomic_inc(&lport->fc4NvmeLsCmpls);
 374                         if (status) {
 375                                 if (bf_get(lpfc_wcqe_c_xb, wcqe))
 376                                         atomic_inc(&lport->cmpl_ls_xb);
 377                                 atomic_inc(&lport->cmpl_ls_err);
 378                         }
 379                 }
 380         }
 381
 382         ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
 383         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 384                          "6047 nvme cmpl Enter "
 385                          "Data %p DID %x Xri: %x status %x reason x%x cmd:%p "
 386                          "lsreg:%p bmp:%p ndlp:%p\n",
 387                          pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
 388                          cmdwqe->sli4_xritag, status,
 389                          (wcqe->parameter & 0xffff),
 390                          cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
 391
 392         lpfc_nvmeio_data(phba, "NVME LS  CMPL: xri x%x stat x%x parm x%x\n",
 393                          cmdwqe->sli4_xritag, status, wcqe->parameter);
 394
 395         if (cmdwqe->context3) {
 396                 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
 397                 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
 398                 kfree(buf_ptr);
 399                 cmdwqe->context3 = NULL;
 400         }
 401         if (pnvme_lsreq->done)
 402                 pnvme_lsreq->done(pnvme_lsreq, status);
 403         else
 404                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 405                                  "6046 nvme cmpl without done call back? "
 406                                  "Data %p DID %x Xri: %x status %x\n",
 407                                 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
 408                                 cmdwqe->sli4_xritag, status);
 409         if (ndlp) {
 410                 lpfc_nlp_put(ndlp);
 411                 cmdwqe->context1 = NULL;
 412         }
 413         lpfc_sli_release_iocbq(phba, cmdwqe);
 414 }
 415
 416 static int
 417 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
 418                   struct lpfc_dmabuf *inp,
 419                   struct nvmefc_ls_req *pnvme_lsreq,
 420                   void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
 421                                struct lpfc_wcqe_complete *),
 422                   struct lpfc_nodelist *ndlp, uint32_t num_entry,
 423                   uint32_t tmo, uint8_t retry)
 424 {
 425         struct lpfc_hba *phba = vport->phba;
 426         union lpfc_wqe128 *wqe;
 427         struct lpfc_iocbq *genwqe;
 428         struct ulp_bde64 *bpl;
 429         struct ulp_bde64 bde;
 430         int i, rc, xmit_len, first_len;
 431
 432         /* Allocate buffer for  command WQE */
 433         genwqe = lpfc_sli_get_iocbq(phba);
 434         if (genwqe == NULL)
 435                 return 1;
 436
 437         wqe = &genwqe->wqe;
 438         memset(wqe, 0, sizeof(union lpfc_wqe));
 439
 440         genwqe->context3 = (uint8_t *)bmp;
 441         genwqe->iocb_flag |= LPFC_IO_NVME_LS;
 442
 443         /* Save for completion so we can release these resources */
 444         genwqe->context1 = lpfc_nlp_get(ndlp);
 445         genwqe->context2 = (uint8_t *)pnvme_lsreq;
 446         /* Fill in payload, bp points to frame payload */
 447
 448         if (!tmo)
 449                 /* FC spec states we need 3 * ratov for CT requests */
 450                 tmo = (3 * phba->fc_ratov);
 451
 452         /* For this command calculate the xmit length of the request bde. */
 453         xmit_len = 0;
 454         first_len = 0;
 455         bpl = (struct ulp_bde64 *)bmp->virt;
 456         for (i = 0; i < num_entry; i++) {
 457                 bde.tus.w = bpl[i].tus.w;
 458                 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
 459                         break;
 460                 xmit_len += bde.tus.f.bdeSize;
 461                 if (i == 0)
 462                         first_len = xmit_len;
 463         }
 464
 465         genwqe->rsvd2 = num_entry;
 466         genwqe->hba_wqidx = 0;
 467
 468         /* Words 0 - 2 */
 469         wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
 470         wqe->generic.bde.tus.f.bdeSize = first_len;
 471         wqe->generic.bde.addrLow = bpl[0].addrLow;
 472         wqe->generic.bde.addrHigh = bpl[0].addrHigh;
 473
 474         /* Word 3 */
 475         wqe->gen_req.request_payload_len = first_len;
 476
 477         /* Word 4 */
 478
 479         /* Word 5 */
 480         bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
 481         bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
 482         bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
 483         bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
 484         bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
 485
 486         /* Word 6 */
 487         bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
 488                phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
 489         bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
 490
 491         /* Word 7 */
 492         bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1));
 493         bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
 494         bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
 495         bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
 496
 497         /* Word 8 */
 498         wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
 499
 500         /* Word 9 */
 501         bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
 502
 503         /* Word 10 */
 504         bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
 505         bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
 506         bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
 507         bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
 508         bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
 509
 510         /* Word 11 */
 511         bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
 512         bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
 513
 514
 515         /* Issue GEN REQ WQE for NPORT <did> */
 516         lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
 517                          "6050 Issue GEN REQ WQE to NPORT x%x "
 518                          "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n",
 519                          ndlp->nlp_DID, genwqe->iotag,
 520                          vport->port_state,
 521                         genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
 522         genwqe->wqe_cmpl = cmpl;
 523         genwqe->iocb_cmpl = NULL;
 524         genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
 525         genwqe->vport = vport;
 526         genwqe->retry = retry;
 527
 528         lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
 529                          genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
 530
 531         rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe);
 532         if (rc) {
 533                 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
 534                                  "6045 Issue GEN REQ WQE to NPORT x%x "
 535                                  "Data: x%x x%x\n",
 536                                  ndlp->nlp_DID, genwqe->iotag,
 537                                  vport->port_state);
 538                 lpfc_sli_release_iocbq(phba, genwqe);
 539                 return 1;
 540         }
 541         return 0;
 542 }
 543
 544 /**
 545  * lpfc_nvme_ls_req - Issue an Link Service request
 546  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 547  * @lpfc_nvme_lport: Pointer to the driver's local port data
 548  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 549  *
 550  * Driver registers this routine to handle any link service request
 551  * from the nvme_fc transport to a remote nvme-aware port.
 552  *
 553  * Return value :
 554  *   0 - Success
 555  *   TODO: What are the failure codes.
 556  **/
 557 static int
 558 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
 559                  struct nvme_fc_remote_port *pnvme_rport,
 560                  struct nvmefc_ls_req *pnvme_lsreq)
 561 {
 562         int ret = 0;
 563         struct lpfc_nvme_lport *lport;
 564         struct lpfc_nvme_rport *rport;
 565         struct lpfc_vport *vport;
 566         struct lpfc_nodelist *ndlp;
 567         struct ulp_bde64 *bpl;
 568         struct lpfc_dmabuf *bmp;
 569         uint16_t ntype, nstate;
 570
 571         /* there are two dma buf in the request, actually there is one and
 572          * the second one is just the start address + cmd size.
 573          * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
 574          * in a lpfc_dmabuf struct. When freeing we just free the wrapper
 575          * because the nvem layer owns the data bufs.
 576          * We do not have to break these packets open, we don't care what is in
 577          * them. And we do not have to look at the resonse data, we only care
 578          * that we got a response. All of the caring is going to happen in the
 579          * nvme-fc layer.
 580          */
 581
 582         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 583         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
 584         if (unlikely(!lport) || unlikely(!rport))
 585                 return -EINVAL;
 586
 587         vport = lport->vport;
 588
 589         if (vport->load_flag & FC_UNLOADING)
 590                 return -ENODEV;
 591
 592         /* Need the ndlp.  It is stored in the driver's rport. */
 593         ndlp = rport->ndlp;
 594         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
 595                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 596                                  "6051 Remoteport %p, rport has invalid ndlp. "
 597                                  "Failing LS Req\n", pnvme_rport);
 598                 return -ENODEV;
 599         }
 600
 601         /* The remote node has to be a mapped nvme target or an
 602          * unmapped nvme initiator or it's an error.
 603          */
 604         ntype = ndlp->nlp_type;
 605         nstate = ndlp->nlp_state;
 606         if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
 607             (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
 608                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 609                                  "6088 DID x%06x not ready for "
 610                                  "IO. State x%x, Type x%x\n",
 611                                  pnvme_rport->port_id,
 612                                  ndlp->nlp_state, ndlp->nlp_type);
 613                 return -ENODEV;
 614         }
 615         bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
 616         if (!bmp) {
 617
 618                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 619                                  "6044 Could not find node for DID %x\n",
 620                                  pnvme_rport->port_id);
 621                 return 2;
 622         }
 623         INIT_LIST_HEAD(&bmp->list);
 624         bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
 625         if (!bmp->virt) {
 626                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 627                                  "6042 Could not find node for DID %x\n",
 628                                  pnvme_rport->port_id);
 629                 kfree(bmp);
 630                 return 3;
 631         }
 632         bpl = (struct ulp_bde64 *)bmp->virt;
 633         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
 634         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
 635         bpl->tus.f.bdeFlags = 0;
 636         bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
 637         bpl->tus.w = le32_to_cpu(bpl->tus.w);
 638         bpl++;
 639
 640         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
 641         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
 642         bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
 643         bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
 644         bpl->tus.w = le32_to_cpu(bpl->tus.w);
 645
 646         /* Expand print to include key fields. */
 647         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 648                          "6149 Issue LS Req to DID 0x%06x lport %p, rport %p "
 649                          "lsreq%p rqstlen:%d rsplen:%d %pad %pad\n",
 650                          ndlp->nlp_DID,
 651                          pnvme_lport, pnvme_rport,
 652                          pnvme_lsreq, pnvme_lsreq->rqstlen,
 653                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
 654                          &pnvme_lsreq->rspdma);
 655
 656         atomic_inc(&lport->fc4NvmeLsRequests);
 657
 658         /* Hardcode the wait to 30 seconds.  Connections are failing otherwise.
 659          * This code allows it all to work.
 660          */
 661         ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
 662                                 pnvme_lsreq, lpfc_nvme_cmpl_gen_req,
 663                                 ndlp, 2, 30, 0);
 664         if (ret != WQE_SUCCESS) {
 665                 atomic_inc(&lport->xmt_ls_err);
 666                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 667                                  "6052 EXIT. issue ls wqe failed lport %p, "
 668                                  "rport %p lsreq%p Status %x DID %x\n",
 669                                  pnvme_lport, pnvme_rport, pnvme_lsreq,
 670                                  ret, ndlp->nlp_DID);
 671                 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
 672                 kfree(bmp);
 673                 return ret;
 674         }
 675
 676         /* Stub in routine and return 0 for now. */
 677         return ret;
 678 }
 679
 680 /**
 681  * lpfc_nvme_ls_abort - Issue an Link Service request
 682  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 683  * @lpfc_nvme_lport: Pointer to the driver's local port data
 684  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 685  *
 686  * Driver registers this routine to handle any link service request
 687  * from the nvme_fc transport to a remote nvme-aware port.
 688  *
 689  * Return value :
 690  *   0 - Success
 691  *   TODO: What are the failure codes.
 692  **/
 693 static void
 694 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
 695                    struct nvme_fc_remote_port *pnvme_rport,
 696                    struct nvmefc_ls_req *pnvme_lsreq)
 697 {
 698         struct lpfc_nvme_lport *lport;
 699         struct lpfc_vport *vport;
 700         struct lpfc_hba *phba;
 701         struct lpfc_nodelist *ndlp;
 702         LIST_HEAD(abort_list);
 703         struct lpfc_sli_ring *pring;
 704         struct lpfc_iocbq *wqe, *next_wqe;
 705
 706         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 707         if (unlikely(!lport))
 708                 return;
 709         vport = lport->vport;
 710         phba = vport->phba;
 711
 712         if (vport->load_flag & FC_UNLOADING)
 713                 return;
 714
 715         ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
 716         if (!ndlp) {
 717                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
 718                                  "6049 Could not find node for DID %x\n",
 719                                  pnvme_rport->port_id);
 720                 return;
 721         }
 722
 723         /* Expand print to include key fields. */
 724         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
 725                          "6040 ENTER.  lport %p, rport %p lsreq %p rqstlen:%d "
 726                          "rsplen:%d %pad %pad\n",
 727                          pnvme_lport, pnvme_rport,
 728                          pnvme_lsreq, pnvme_lsreq->rqstlen,
 729                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
 730                          &pnvme_lsreq->rspdma);
 731
 732         /*
 733          * Lock the ELS ring txcmplq and build a local list of all ELS IOs
 734          * that need an ABTS.  The IOs need to stay on the txcmplq so that
 735          * the abort operation completes them successfully.
 736          */
 737         pring = phba->sli4_hba.nvmels_wq->pring;
 738         spin_lock_irq(&phba->hbalock);
 739         spin_lock(&pring->ring_lock);
 740         list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
 741                 /* Add to abort_list on on NDLP match. */
 742                 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) {
 743                         wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
 744                         list_add_tail(&wqe->dlist, &abort_list);
 745                 }
 746         }
 747         spin_unlock(&pring->ring_lock);
 748         spin_unlock_irq(&phba->hbalock);
 749
 750         /* Abort the targeted IOs and remove them from the abort list. */
 751         list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) {
 752                 atomic_inc(&lport->xmt_ls_abort);
 753                 spin_lock_irq(&phba->hbalock);
 754                 list_del_init(&wqe->dlist);
 755                 lpfc_sli_issue_abort_iotag(phba, pring, wqe);
 756                 spin_unlock_irq(&phba->hbalock);
 757         }
 758 }
 759
 760 /* Fix up the existing sgls for NVME IO. */
 761 static inline void
 762 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
 763                        struct lpfc_nvme_buf *lpfc_ncmd,
 764                        struct nvmefc_fcp_req *nCmd)
 765 {
 766         struct lpfc_hba  *phba = vport->phba;
 767         struct sli4_sge *sgl;
 768         union lpfc_wqe128 *wqe;
 769         uint32_t *wptr, *dptr;
 770
 771         /*
 772          * Get a local pointer to the built-in wqe and correct
 773          * the cmd size to match NVME's 96 bytes and fix
 774          * the dma address.
 775          */
 776
 777         wqe = &lpfc_ncmd->cur_iocbq.wqe;
 778
 779         /*
 780          * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
 781          * match NVME.  NVME sends 96 bytes. Also, use the
 782          * nvme commands command and response dma addresses
 783          * rather than the virtual memory to ease the restore
 784          * operation.
 785          */
 786         sgl = lpfc_ncmd->dma_sgl;
 787         sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
 788         if (phba->cfg_nvme_embed_cmd) {
 789                 sgl->addr_hi = 0;
 790                 sgl->addr_lo = 0;
 791
 792                 /* Word 0-2 - NVME CMND IU (embedded payload) */
 793                 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
 794                 wqe->generic.bde.tus.f.bdeSize = 56;
 795                 wqe->generic.bde.addrHigh = 0;
 796                 wqe->generic.bde.addrLow =  64;  /* Word 16 */
 797
 798                 /* Word 10  - dbde is 0, wqes is 1 in template */
 799
 800                 /*
 801                  * Embed the payload in the last half of the WQE
 802                  * WQE words 16-30 get the NVME CMD IU payload
 803                  *
 804                  * WQE words 16-19 get payload Words 1-4
 805                  * WQE words 20-21 get payload Words 6-7
 806                  * WQE words 22-29 get payload Words 16-23
 807                  */
 808                 wptr = &wqe->words[16];  /* WQE ptr */
 809                 dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
 810                 dptr++;                 /* Skip Word 0 in payload */
 811
 812                 *wptr++ = *dptr++;      /* Word 1 */
 813                 *wptr++ = *dptr++;      /* Word 2 */
 814                 *wptr++ = *dptr++;      /* Word 3 */
 815                 *wptr++ = *dptr++;      /* Word 4 */
 816                 dptr++;                 /* Skip Word 5 in payload */
 817                 *wptr++ = *dptr++;      /* Word 6 */
 818                 *wptr++ = *dptr++;      /* Word 7 */
 819                 dptr += 8;              /* Skip Words 8-15 in payload */
 820                 *wptr++ = *dptr++;      /* Word 16 */
 821                 *wptr++ = *dptr++;      /* Word 17 */
 822                 *wptr++ = *dptr++;      /* Word 18 */
 823                 *wptr++ = *dptr++;      /* Word 19 */
 824                 *wptr++ = *dptr++;      /* Word 20 */
 825                 *wptr++ = *dptr++;      /* Word 21 */
 826                 *wptr++ = *dptr++;      /* Word 22 */
 827                 *wptr   = *dptr;        /* Word 23 */
 828         } else {
 829                 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
 830                 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
 831
 832                 /* Word 0-2 - NVME CMND IU Inline BDE */
 833                 wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
 834                 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
 835                 wqe->generic.bde.addrHigh = sgl->addr_hi;
 836                 wqe->generic.bde.addrLow =  sgl->addr_lo;
 837
 838                 /* Word 10 */
 839                 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
 840                 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
 841         }
 842
 843         sgl++;
 844
 845         /* Setup the physical region for the FCP RSP */
 846         sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
 847         sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
 848         sgl->word2 = le32_to_cpu(sgl->word2);
 849         if (nCmd->sg_cnt)
 850                 bf_set(lpfc_sli4_sge_last, sgl, 0);
 851         else
 852                 bf_set(lpfc_sli4_sge_last, sgl, 1);
 853         sgl->word2 = cpu_to_le32(sgl->word2);
 854         sgl->sge_len = cpu_to_le32(nCmd->rsplen);
 855 }
 856
 857 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
 858 static void
 859 lpfc_nvme_ktime(struct lpfc_hba *phba,
 860                 struct lpfc_nvme_buf *lpfc_ncmd)
 861 {
 862         uint64_t seg1, seg2, seg3, seg4;
 863         uint64_t segsum;
 864
 865         if (!lpfc_ncmd->ts_last_cmd ||
 866             !lpfc_ncmd->ts_cmd_start ||
 867             !lpfc_ncmd->ts_cmd_wqput ||
 868             !lpfc_ncmd->ts_isr_cmpl ||
 869             !lpfc_ncmd->ts_data_nvme)
 870                 return;
 871
 872         if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_cmd_start)
 873                 return;
 874         if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd)
 875                 return;
 876         if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start)
 877                 return;
 878         if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput)
 879                 return;
 880         if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl)
 881                 return;
 882         /*
 883          * Segment 1 - Time from Last FCP command cmpl is handed
 884          * off to NVME Layer to start of next command.
 885          * Segment 2 - Time from Driver receives a IO cmd start
 886          * from NVME Layer to WQ put is done on IO cmd.
 887          * Segment 3 - Time from Driver WQ put is done on IO cmd
 888          * to MSI-X ISR for IO cmpl.
 889          * Segment 4 - Time from MSI-X ISR for IO cmpl to when
 890          * cmpl is handled off to the NVME Layer.
 891          */
 892         seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd;
 893         if (seg1 > 5000000)  /* 5 ms - for sequential IOs only */
 894                 seg1 = 0;
 895
 896         /* Calculate times relative to start of IO */
 897         seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start);
 898         segsum = seg2;
 899         seg3 = lpfc_ncmd->ts_isr_cmpl - lpfc_ncmd->ts_cmd_start;
 900         if (segsum > seg3)
 901                 return;
 902         seg3 -= segsum;
 903         segsum += seg3;
 904
 905         seg4 = lpfc_ncmd->ts_data_nvme - lpfc_ncmd->ts_cmd_start;
 906         if (segsum > seg4)
 907                 return;
 908         seg4 -= segsum;
 909
 910         phba->ktime_data_samples++;
 911         phba->ktime_seg1_total += seg1;
 912         if (seg1 < phba->ktime_seg1_min)
 913                 phba->ktime_seg1_min = seg1;
 914         else if (seg1 > phba->ktime_seg1_max)
 915                 phba->ktime_seg1_max = seg1;
 916         phba->ktime_seg2_total += seg2;
 917         if (seg2 < phba->ktime_seg2_min)
 918                 phba->ktime_seg2_min = seg2;
 919         else if (seg2 > phba->ktime_seg2_max)
 920                 phba->ktime_seg2_max = seg2;
 921         phba->ktime_seg3_total += seg3;
 922         if (seg3 < phba->ktime_seg3_min)
 923                 phba->ktime_seg3_min = seg3;
 924         else if (seg3 > phba->ktime_seg3_max)
 925                 phba->ktime_seg3_max = seg3;
 926         phba->ktime_seg4_total += seg4;
 927         if (seg4 < phba->ktime_seg4_min)
 928                 phba->ktime_seg4_min = seg4;
 929         else if (seg4 > phba->ktime_seg4_max)
 930                 phba->ktime_seg4_max = seg4;
 931
 932         lpfc_ncmd->ts_last_cmd = 0;
 933         lpfc_ncmd->ts_cmd_start = 0;
 934         lpfc_ncmd->ts_cmd_wqput  = 0;
 935         lpfc_ncmd->ts_isr_cmpl = 0;
 936         lpfc_ncmd->ts_data_nvme = 0;
 937 }
 938 #endif
 939
 940 /**
 941  * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
 942  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 943  * @lpfc_nvme_lport: Pointer to the driver's local port data
 944  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 945  *
 946  * Driver registers this routine as it io request handler.  This
 947  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
 948  * data structure to the rport indicated in @lpfc_nvme_rport.
 949  *
 950  * Return value :
 951  *   0 - Success
 952  *   TODO: What are the failure codes.
 953  **/
 954 static void
 955 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
 956                           struct lpfc_wcqe_complete *wcqe)
 957 {
 958         struct lpfc_nvme_buf *lpfc_ncmd =
 959                 (struct lpfc_nvme_buf *)pwqeIn->context1;
 960         struct lpfc_vport *vport = pwqeIn->vport;
 961         struct nvmefc_fcp_req *nCmd;
 962         struct nvme_fc_ersp_iu *ep;
 963         struct nvme_fc_cmd_iu *cp;
 964         struct lpfc_nodelist *ndlp;
 965         struct lpfc_nvme_fcpreq_priv *freqpriv;
 966         struct lpfc_nvme_lport *lport;
 967         struct lpfc_nvme_ctrl_stat *cstat;
 968         uint32_t code, status, idx;
 969         uint16_t cid, sqhd, data;
 970         uint32_t *ptr;
 971
 972         /* Sanity check on return of outstanding command */
 973         if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd) {
 974                 if (!lpfc_ncmd) {
 975                         lpfc_printf_vlog(vport, KERN_ERR,
 976                                          LOG_NODE | LOG_NVME_IOERR,
 977                                          "6071 Null lpfc_ncmd pointer. No "
 978                                          "release, skip completion\n");
 979                         return;
 980                 }
 981
 982                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 983                                  "6066 Missing cmpl ptrs: lpfc_ncmd %p, "
 984                                  "nvmeCmd %p\n",
 985                                  lpfc_ncmd, lpfc_ncmd->nvmeCmd);
 986
 987                 /* Release the lpfc_ncmd regardless of the missing elements. */
 988                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
 989                 return;
 990         }
 991         nCmd = lpfc_ncmd->nvmeCmd;
 992         status = bf_get(lpfc_wcqe_c_status, wcqe);
 993
 994         if (vport->localport) {
 995                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
 996                 if (lport) {
 997                         idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
 998                         cstat = &lport->cstat[idx];
 999                         atomic_inc(&cstat->fc4NvmeIoCmpls);
1000                         if (status) {
1001                                 if (bf_get(lpfc_wcqe_c_xb, wcqe))
1002                                         atomic_inc(&lport->cmpl_fcp_xb);
1003                                 atomic_inc(&lport->cmpl_fcp_err);
1004                         }
1005                 }
1006         }
1007
1008         lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
1009                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1010                          status, wcqe->parameter);
1011         /*
1012          * Catch race where our node has transitioned, but the
1013          * transport is still transitioning.
1014          */
1015         ndlp = lpfc_ncmd->ndlp;
1016         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1017                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1018                                  "6062 Ignoring NVME cmpl.  No ndlp\n");
1019                 goto out_err;
1020         }
1021
1022         code = bf_get(lpfc_wcqe_c_code, wcqe);
1023         if (code == CQE_CODE_NVME_ERSP) {
1024                 /* For this type of CQE, we need to rebuild the rsp */
1025                 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1026
1027                 /*
1028                  * Get Command Id from cmd to plug into response. This
1029                  * code is not needed in the next NVME Transport drop.
1030                  */
1031                 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1032                 cid = cp->sqe.common.command_id;
1033
1034                 /*
1035                  * RSN is in CQE word 2
1036                  * SQHD is in CQE Word 3 bits 15:0
1037                  * Cmd Specific info is in CQE Word 1
1038                  * and in CQE Word 0 bits 15:0
1039                  */
1040                 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1041
1042                 /* Now lets build the NVME ERSP IU */
1043                 ep->iu_len = cpu_to_be16(8);
1044                 ep->rsn = wcqe->parameter;
1045                 ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1046                 ep->rsvd12 = 0;
1047                 ptr = (uint32_t *)&ep->cqe.result.u64;
1048                 *ptr++ = wcqe->total_data_placed;
1049                 data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1050                 *ptr = (uint32_t)data;
1051                 ep->cqe.sq_head = sqhd;
1052                 ep->cqe.sq_id =  nCmd->sqid;
1053                 ep->cqe.command_id = cid;
1054                 ep->cqe.status = 0;
1055
1056                 lpfc_ncmd->status = IOSTAT_SUCCESS;
1057                 lpfc_ncmd->result = 0;
1058                 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1059                 nCmd->transferred_length = nCmd->payload_length;
1060         } else {
1061                 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1062                 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1063
1064                 /* For NVME, the only failure path that results in an
1065                  * IO error is when the adapter rejects it.  All other
1066                  * conditions are a success case and resolved by the
1067                  * transport.
1068                  * IOSTAT_FCP_RSP_ERROR means:
1069                  * 1. Length of data received doesn't match total
1070                  *    transfer length in WQE
1071                  * 2. If the RSP payload does NOT match these cases:
1072                  *    a. RSP length 12/24 bytes and all zeros
1073                  *    b. NVME ERSP
1074                  */
1075                 switch (lpfc_ncmd->status) {
1076                 case IOSTAT_SUCCESS:
1077                         nCmd->transferred_length = wcqe->total_data_placed;
1078                         nCmd->rcv_rsplen = 0;
1079                         nCmd->status = 0;
1080                         break;
1081                 case IOSTAT_FCP_RSP_ERROR:
1082                         nCmd->transferred_length = wcqe->total_data_placed;
1083                         nCmd->rcv_rsplen = wcqe->parameter;
1084                         nCmd->status = 0;
1085                         /* Sanity check */
1086                         if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN)
1087                                 break;
1088                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1089                                          "6081 NVME Completion Protocol Error: "
1090                                          "xri %x status x%x result x%x "
1091                                          "placed x%x\n",
1092                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1093                                          lpfc_ncmd->status, lpfc_ncmd->result,
1094                                          wcqe->total_data_placed);
1095                         break;
1096                 case IOSTAT_LOCAL_REJECT:
1097                         /* Let fall through to set command final state. */
1098                         if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1099                                 lpfc_printf_vlog(vport, KERN_INFO,
1100                                          LOG_NVME_IOERR,
1101                                          "6032 Delay Aborted cmd %p "
1102                                          "nvme cmd %p, xri x%x, "
1103                                          "xb %d\n",
1104                                          lpfc_ncmd, nCmd,
1105                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1106                                          bf_get(lpfc_wcqe_c_xb, wcqe));
1107                 default:
1108 out_err:
1109                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1110                                          "6072 NVME Completion Error: xri %x "
1111                                          "status x%x result x%x placed x%x\n",
1112                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1113                                          lpfc_ncmd->status, lpfc_ncmd->result,
1114                                          wcqe->total_data_placed);
1115                         nCmd->transferred_length = 0;
1116                         nCmd->rcv_rsplen = 0;
1117                         nCmd->status = NVME_SC_INTERNAL;
1118                 }
1119         }
1120
1121         /* pick up SLI4 exhange busy condition */
1122         if (bf_get(lpfc_wcqe_c_xb, wcqe))
1123                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1124         else
1125                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1126
1127         /* Update stats and complete the IO.  There is
1128          * no need for dma unprep because the nvme_transport
1129          * owns the dma address.
1130          */
1131 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1132         if (lpfc_ncmd->ts_cmd_start) {
1133                 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1134                 lpfc_ncmd->ts_data_nvme = ktime_get_ns();
1135                 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme;
1136                 lpfc_nvme_ktime(phba, lpfc_ncmd);
1137         }
1138         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1139                 if (lpfc_ncmd->cpu != smp_processor_id())
1140                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1141                                          "6701 CPU Check cmpl: "
1142                                          "cpu %d expect %d\n",
1143                                          smp_processor_id(), lpfc_ncmd->cpu);
1144                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
1145                         phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++;
1146         }
1147 #endif
1148
1149         /* NVME targets need completion held off until the abort exchange
1150          * completes unless the NVME Rport is getting unregistered.
1151          */
1152
1153         if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1154                 freqpriv = nCmd->private;
1155                 freqpriv->nvme_buf = NULL;
1156                 nCmd->done(nCmd);
1157                 lpfc_ncmd->nvmeCmd = NULL;
1158         }
1159
1160         /* Call release with XB=1 to queue the IO into the abort list. */
1161         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1162 }
1163
1164
1165 /**
1166  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1167  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1168  * @lpfc_nvme_lport: Pointer to the driver's local port data
1169  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1170  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1171  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1172  *
1173  * Driver registers this routine as it io request handler.  This
1174  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1175  * data structure to the rport indicated in @lpfc_nvme_rport.
1176  *
1177  * Return value :
1178  *   0 - Success
1179  *   TODO: What are the failure codes.
1180  **/
1181 static int
1182 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1183                       struct lpfc_nvme_buf *lpfc_ncmd,
1184                       struct lpfc_nodelist *pnode,
1185                       struct lpfc_nvme_ctrl_stat *cstat)
1186 {
1187         struct lpfc_hba *phba = vport->phba;
1188         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1189         struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1190         union lpfc_wqe128 *wqe = &pwqeq->wqe;
1191         uint32_t req_len;
1192
1193         if (!NLP_CHK_NODE_ACT(pnode))
1194                 return -EINVAL;
1195
1196         /*
1197          * There are three possibilities here - use scatter-gather segment, use
1198          * the single mapping, or neither.
1199          */
1200         if (nCmd->sg_cnt) {
1201                 if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1202                         /* From the iwrite template, initialize words 7 - 11 */
1203                         memcpy(&wqe->words[7],
1204                                &lpfc_iwrite_cmd_template.words[7],
1205                                sizeof(uint32_t) * 5);
1206
1207                         /* Word 4 */
1208                         wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1209
1210                         /* Word 5 */
1211                         if ((phba->cfg_nvme_enable_fb) &&
1212                             (pnode->nlp_flag & NLP_FIRSTBURST)) {
1213                                 req_len = lpfc_ncmd->nvmeCmd->payload_length;
1214                                 if (req_len < pnode->nvme_fb_size)
1215                                         wqe->fcp_iwrite.initial_xfer_len =
1216                                                 req_len;
1217                                 else
1218                                         wqe->fcp_iwrite.initial_xfer_len =
1219                                                 pnode->nvme_fb_size;
1220                         } else {
1221                                 wqe->fcp_iwrite.initial_xfer_len = 0;
1222                         }
1223                         atomic_inc(&cstat->fc4NvmeOutputRequests);
1224                 } else {
1225                         /* From the iread template, initialize words 7 - 11 */
1226                         memcpy(&wqe->words[7],
1227                                &lpfc_iread_cmd_template.words[7],
1228                                sizeof(uint32_t) * 5);
1229
1230                         /* Word 4 */
1231                         wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1232
1233                         /* Word 5 */
1234                         wqe->fcp_iread.rsrvd5 = 0;
1235
1236                         atomic_inc(&cstat->fc4NvmeInputRequests);
1237                 }
1238         } else {
1239                 /* From the icmnd template, initialize words 4 - 11 */
1240                 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1241                        sizeof(uint32_t) * 8);
1242                 atomic_inc(&cstat->fc4NvmeControlRequests);
1243         }
1244         /*
1245          * Finish initializing those WQE fields that are independent
1246          * of the nvme_cmnd request_buffer
1247          */
1248
1249         /* Word 3 */
1250         bf_set(payload_offset_len, &wqe->fcp_icmd,
1251                (nCmd->rsplen + nCmd->cmdlen));
1252
1253         /* Word 6 */
1254         bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1255                phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1256         bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1257
1258         /* Word 8 */
1259         wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1260
1261         /* Word 9 */
1262         bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1263
1264         /* Words 13 14 15 are for PBDE support */
1265
1266         pwqeq->vport = vport;
1267         return 0;
1268 }
1269
1270
1271 /**
1272  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1273  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1274  * @lpfc_nvme_lport: Pointer to the driver's local port data
1275  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1276  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1277  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1278  *
1279  * Driver registers this routine as it io request handler.  This
1280  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1281  * data structure to the rport indicated in @lpfc_nvme_rport.
1282  *
1283  * Return value :
1284  *   0 - Success
1285  *   TODO: What are the failure codes.
1286  **/
1287 static int
1288 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1289                       struct lpfc_nvme_buf *lpfc_ncmd)
1290 {
1291         struct lpfc_hba *phba = vport->phba;
1292         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1293         union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1294         struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1295         struct scatterlist *data_sg;
1296         struct sli4_sge *first_data_sgl;
1297         struct ulp_bde64 *bde;
1298         dma_addr_t physaddr;
1299         uint32_t num_bde = 0;
1300         uint32_t dma_len;
1301         uint32_t dma_offset = 0;
1302         int nseg, i;
1303
1304         /* Fix up the command and response DMA stuff. */
1305         lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1306
1307         /*
1308          * There are three possibilities here - use scatter-gather segment, use
1309          * the single mapping, or neither.
1310          */
1311         if (nCmd->sg_cnt) {
1312                 /*
1313                  * Jump over the cmd and rsp SGEs.  The fix routine
1314                  * has already adjusted for this.
1315                  */
1316                 sgl += 2;
1317
1318                 first_data_sgl = sgl;
1319                 lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1320                 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1321                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1322                                         "6058 Too many sg segments from "
1323                                         "NVME Transport.  Max %d, "
1324                                         "nvmeIO sg_cnt %d\n",
1325                                         phba->cfg_nvme_seg_cnt + 1,
1326                                         lpfc_ncmd->seg_cnt);
1327                         lpfc_ncmd->seg_cnt = 0;
1328                         return 1;
1329                 }
1330
1331                 /*
1332                  * The driver established a maximum scatter-gather segment count
1333                  * during probe that limits the number of sg elements in any
1334                  * single nvme command.  Just run through the seg_cnt and format
1335                  * the sge's.
1336                  */
1337                 nseg = nCmd->sg_cnt;
1338                 data_sg = nCmd->first_sgl;
1339                 for (i = 0; i < nseg; i++) {
1340                         if (data_sg == NULL) {
1341                                 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1342                                                 "6059 dptr err %d, nseg %d\n",
1343                                                 i, nseg);
1344                                 lpfc_ncmd->seg_cnt = 0;
1345                                 return 1;
1346                         }
1347                         physaddr = data_sg->dma_address;
1348                         dma_len = data_sg->length;
1349                         sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
1350                         sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
1351                         sgl->word2 = le32_to_cpu(sgl->word2);
1352                         if ((num_bde + 1) == nseg)
1353                                 bf_set(lpfc_sli4_sge_last, sgl, 1);
1354                         else
1355                                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1356                         bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1357                         bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA);
1358                         sgl->word2 = cpu_to_le32(sgl->word2);
1359                         sgl->sge_len = cpu_to_le32(dma_len);
1360
1361                         dma_offset += dma_len;
1362                         data_sg = sg_next(data_sg);
1363                         sgl++;
1364                 }
1365                 if (phba->cfg_enable_pbde) {
1366                         /* Use PBDE support for first SGL only, offset == 0 */
1367                         /* Words 13-15 */
1368                         bde = (struct ulp_bde64 *)
1369                                 &wqe->words[13];
1370                         bde->addrLow = first_data_sgl->addr_lo;
1371                         bde->addrHigh = first_data_sgl->addr_hi;
1372                         bde->tus.f.bdeSize =
1373                                 le32_to_cpu(first_data_sgl->sge_len);
1374                         bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1375                         bde->tus.w = cpu_to_le32(bde->tus.w);
1376                         /* wqe_pbde is 1 in template */
1377                 } else {
1378                         memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1379                         bf_set(wqe_pbde, &wqe->generic.wqe_com, 0);
1380                 }
1381
1382         } else {
1383                 lpfc_ncmd->seg_cnt = 0;
1384
1385                 /* For this clause to be valid, the payload_length
1386                  * and sg_cnt must zero.
1387                  */
1388                 if (nCmd->payload_length != 0) {
1389                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1390                                         "6063 NVME DMA Prep Err: sg_cnt %d "
1391                                         "payload_length x%x\n",
1392                                         nCmd->sg_cnt, nCmd->payload_length);
1393                         return 1;
1394                 }
1395         }
1396         return 0;
1397 }
1398
1399 /**
1400  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1401  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1402  * @lpfc_nvme_lport: Pointer to the driver's local port data
1403  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1404  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1405  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1406  *
1407  * Driver registers this routine as it io request handler.  This
1408  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1409  * data structure to the rport
1410  indicated in @lpfc_nvme_rport.
1411  *
1412  * Return value :
1413  *   0 - Success
1414  *   TODO: What are the failure codes.
1415  **/
1416 static int
1417 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1418                         struct nvme_fc_remote_port *pnvme_rport,
1419                         void *hw_queue_handle,
1420                         struct nvmefc_fcp_req *pnvme_fcreq)
1421 {
1422         int ret = 0;
1423         int expedite = 0;
1424         int idx;
1425         struct lpfc_nvme_lport *lport;
1426         struct lpfc_nvme_ctrl_stat *cstat;
1427         struct lpfc_vport *vport;
1428         struct lpfc_hba *phba;
1429         struct lpfc_nodelist *ndlp;
1430         struct lpfc_nvme_buf *lpfc_ncmd;
1431         struct lpfc_nvme_rport *rport;
1432         struct lpfc_nvme_qhandle *lpfc_queue_info;
1433         struct lpfc_nvme_fcpreq_priv *freqpriv;
1434         struct nvme_common_command *sqe;
1435 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1436         uint64_t start = 0;
1437 #endif
1438
1439         /* Validate pointers. LLDD fault handling with transport does
1440          * have timing races.
1441          */
1442         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1443         if (unlikely(!lport)) {
1444                 ret = -EINVAL;
1445                 goto out_fail;
1446         }
1447
1448         vport = lport->vport;
1449
1450         if (unlikely(!hw_queue_handle)) {
1451                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1452                                  "6117 Fail IO, NULL hw_queue_handle\n");
1453                 atomic_inc(&lport->xmt_fcp_err);
1454                 ret = -EBUSY;
1455                 goto out_fail;
1456         }
1457
1458         phba = vport->phba;
1459
1460         if (vport->load_flag & FC_UNLOADING) {
1461                 ret = -ENODEV;
1462                 goto out_fail;
1463         }
1464
1465         if (vport->load_flag & FC_UNLOADING) {
1466                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1467                                  "6124 Fail IO, Driver unload\n");
1468                 atomic_inc(&lport->xmt_fcp_err);
1469                 ret = -ENODEV;
1470                 goto out_fail;
1471         }
1472
1473         freqpriv = pnvme_fcreq->private;
1474         if (unlikely(!freqpriv)) {
1475                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1476                                  "6158 Fail IO, NULL request data\n");
1477                 atomic_inc(&lport->xmt_fcp_err);
1478                 ret = -EINVAL;
1479                 goto out_fail;
1480         }
1481
1482 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1483         if (phba->ktime_on)
1484                 start = ktime_get_ns();
1485 #endif
1486         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1487         lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1488
1489         /*
1490          * Catch race where our node has transitioned, but the
1491          * transport is still transitioning.
1492          */
1493         ndlp = rport->ndlp;
1494         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1495                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1496                                  "6053 Fail IO, ndlp not ready: rport %p "
1497                                   "ndlp %p, DID x%06x\n",
1498                                  rport, ndlp, pnvme_rport->port_id);
1499                 atomic_inc(&lport->xmt_fcp_err);
1500                 ret = -EBUSY;
1501                 goto out_fail;
1502         }
1503
1504         /* The remote node has to be a mapped target or it's an error. */
1505         if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1506             (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1507                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1508                                  "6036 Fail IO, DID x%06x not ready for "
1509                                  "IO. State x%x, Type x%x Flg x%x\n",
1510                                  pnvme_rport->port_id,
1511                                  ndlp->nlp_state, ndlp->nlp_type,
1512                                  ndlp->upcall_flags);
1513                 atomic_inc(&lport->xmt_fcp_bad_ndlp);
1514                 ret = -EBUSY;
1515                 goto out_fail;
1516
1517         }
1518
1519         /* Currently only NVME Keep alive commands should be expedited
1520          * if the driver runs out of a resource. These should only be
1521          * issued on the admin queue, qidx 0
1522          */
1523         if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1524                 sqe = &((struct nvme_fc_cmd_iu *)
1525                         pnvme_fcreq->cmdaddr)->sqe.common;
1526                 if (sqe->opcode == nvme_admin_keep_alive)
1527                         expedite = 1;
1528         }
1529
1530         /* The node is shared with FCP IO, make sure the IO pending count does
1531          * not exceed the programmed depth.
1532          */
1533         if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1534                 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1535                     !expedite) {
1536                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1537                                          "6174 Fail IO, ndlp qdepth exceeded: "
1538                                          "idx %d DID %x pend %d qdepth %d\n",
1539                                          lpfc_queue_info->index, ndlp->nlp_DID,
1540                                          atomic_read(&ndlp->cmd_pending),
1541                                          ndlp->cmd_qdepth);
1542                         atomic_inc(&lport->xmt_fcp_qdepth);
1543                         ret = -EBUSY;
1544                         goto out_fail;
1545                 }
1546         }
1547
1548         lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp,
1549                                       lpfc_queue_info->index, expedite);
1550         if (lpfc_ncmd == NULL) {
1551                 atomic_inc(&lport->xmt_fcp_noxri);
1552                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1553                                  "6065 Fail IO, driver buffer pool is empty: "
1554                                  "idx %d DID %x\n",
1555                                  lpfc_queue_info->index, ndlp->nlp_DID);
1556                 ret = -EBUSY;
1557                 goto out_fail;
1558         }
1559 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1560         if (start) {
1561                 lpfc_ncmd->ts_cmd_start = start;
1562                 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1563         } else {
1564                 lpfc_ncmd->ts_cmd_start = 0;
1565         }
1566 #endif
1567
1568         /*
1569          * Store the data needed by the driver to issue, abort, and complete
1570          * an IO.
1571          * Do not let the IO hang out forever.  There is no midlayer issuing
1572          * an abort so inform the FW of the maximum IO pending time.
1573          */
1574         freqpriv->nvme_buf = lpfc_ncmd;
1575         lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1576         lpfc_ncmd->ndlp = ndlp;
1577         lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1578
1579         /*
1580          * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1581          * This identfier was create in our hardware queue create callback
1582          * routine. The driver now is dependent on the IO queue steering from
1583          * the transport.  We are trusting the upper NVME layers know which
1584          * index to use and that they have affinitized a CPU to this hardware
1585          * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1586          */
1587         idx = lpfc_queue_info->index;
1588         lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1589         cstat = &lport->cstat[idx];
1590
1591         lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1592         ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1593         if (ret) {
1594                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1595                                  "6175 Fail IO, Prep DMA: "
1596                                  "idx %d DID %x\n",
1597                                  lpfc_queue_info->index, ndlp->nlp_DID);
1598                 atomic_inc(&lport->xmt_fcp_err);
1599                 ret = -ENOMEM;
1600                 goto out_free_nvme_buf;
1601         }
1602
1603         lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1604                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1605                          lpfc_queue_info->index, ndlp->nlp_DID);
1606
1607         ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq);
1608         if (ret) {
1609                 atomic_inc(&lport->xmt_fcp_wqerr);
1610                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1611                                  "6113 Fail IO, Could not issue WQE err %x "
1612                                  "sid: x%x did: x%x oxid: x%x\n",
1613                                  ret, vport->fc_myDID, ndlp->nlp_DID,
1614                                  lpfc_ncmd->cur_iocbq.sli4_xritag);
1615                 goto out_free_nvme_buf;
1616         }
1617
1618 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1619         if (lpfc_ncmd->ts_cmd_start)
1620                 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1621
1622         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1623                 lpfc_ncmd->cpu = smp_processor_id();
1624                 if (lpfc_ncmd->cpu != lpfc_queue_info->index) {
1625                         /* Check for admin queue */
1626                         if (lpfc_queue_info->qidx) {
1627                                 lpfc_printf_vlog(vport,
1628                                                  KERN_ERR, LOG_NVME_IOERR,
1629                                                 "6702 CPU Check cmd: "
1630                                                 "cpu %d wq %d\n",
1631                                                 lpfc_ncmd->cpu,
1632                                                 lpfc_queue_info->index);
1633                         }
1634                         lpfc_ncmd->cpu = lpfc_queue_info->index;
1635                 }
1636                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
1637                         phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++;
1638         }
1639 #endif
1640         return 0;
1641
1642  out_free_nvme_buf:
1643         if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1644                 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1645                         atomic_dec(&cstat->fc4NvmeOutputRequests);
1646                 else
1647                         atomic_dec(&cstat->fc4NvmeInputRequests);
1648         } else
1649                 atomic_dec(&cstat->fc4NvmeControlRequests);
1650         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1651  out_fail:
1652         return ret;
1653 }
1654
1655 /**
1656  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1657  * @phba: Pointer to HBA context object
1658  * @cmdiocb: Pointer to command iocb object.
1659  * @rspiocb: Pointer to response iocb object.
1660  *
1661  * This is the callback function for any NVME FCP IO that was aborted.
1662  *
1663  * Return value:
1664  *   None
1665  **/
1666 void
1667 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1668                            struct lpfc_wcqe_complete *abts_cmpl)
1669 {
1670         lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1671                         "6145 ABORT_XRI_CN completing on rpi x%x "
1672                         "original iotag x%x, abort cmd iotag x%x "
1673                         "req_tag x%x, status x%x, hwstatus x%x\n",
1674                         cmdiocb->iocb.un.acxri.abortContextTag,
1675                         cmdiocb->iocb.un.acxri.abortIoTag,
1676                         cmdiocb->iotag,
1677                         bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1678                         bf_get(lpfc_wcqe_c_status, abts_cmpl),
1679                         bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1680         lpfc_sli_release_iocbq(phba, cmdiocb);
1681 }
1682
1683 /**
1684  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1685  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1686  * @lpfc_nvme_lport: Pointer to the driver's local port data
1687  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1688  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1689  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1690  *
1691  * Driver registers this routine as its nvme request io abort handler.  This
1692  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1693  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1694  * is executed asynchronously - one the target is validated as "MAPPED" and
1695  * ready for IO, the driver issues the abort request and returns.
1696  *
1697  * Return value:
1698  *   None
1699  **/
1700 static void
1701 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1702                     struct nvme_fc_remote_port *pnvme_rport,
1703                     void *hw_queue_handle,
1704                     struct nvmefc_fcp_req *pnvme_fcreq)
1705 {
1706         struct lpfc_nvme_lport *lport;
1707         struct lpfc_vport *vport;
1708         struct lpfc_hba *phba;
1709         struct lpfc_nvme_buf *lpfc_nbuf;
1710         struct lpfc_iocbq *abts_buf;
1711         struct lpfc_iocbq *nvmereq_wqe;
1712         struct lpfc_nvme_fcpreq_priv *freqpriv;
1713         union lpfc_wqe128 *abts_wqe;
1714         unsigned long flags;
1715         int ret_val;
1716
1717         /* Validate pointers. LLDD fault handling with transport does
1718          * have timing races.
1719          */
1720         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1721         if (unlikely(!lport))
1722                 return;
1723
1724         vport = lport->vport;
1725
1726         if (unlikely(!hw_queue_handle)) {
1727                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1728                                  "6129 Fail Abort, HW Queue Handle NULL.\n");
1729                 return;
1730         }
1731
1732         phba = vport->phba;
1733         freqpriv = pnvme_fcreq->private;
1734
1735         if (unlikely(!freqpriv))
1736                 return;
1737         if (vport->load_flag & FC_UNLOADING)
1738                 return;
1739
1740         /* Announce entry to new IO submit field. */
1741         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1742                          "6002 Abort Request to rport DID x%06x "
1743                          "for nvme_fc_req %p\n",
1744                          pnvme_rport->port_id,
1745                          pnvme_fcreq);
1746
1747         /* If the hba is getting reset, this flag is set.  It is
1748          * cleared when the reset is complete and rings reestablished.
1749          */
1750         spin_lock_irqsave(&phba->hbalock, flags);
1751         /* driver queued commands are in process of being flushed */
1752         if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) {
1753                 spin_unlock_irqrestore(&phba->hbalock, flags);
1754                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1755                                  "6139 Driver in reset cleanup - flushing "
1756                                  "NVME Req now.  hba_flag x%x\n",
1757                                  phba->hba_flag);
1758                 return;
1759         }
1760
1761         lpfc_nbuf = freqpriv->nvme_buf;
1762         if (!lpfc_nbuf) {
1763                 spin_unlock_irqrestore(&phba->hbalock, flags);
1764                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1765                                  "6140 NVME IO req has no matching lpfc nvme "
1766                                  "io buffer.  Skipping abort req.\n");
1767                 return;
1768         } else if (!lpfc_nbuf->nvmeCmd) {
1769                 spin_unlock_irqrestore(&phba->hbalock, flags);
1770                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1771                                  "6141 lpfc NVME IO req has no nvme_fcreq "
1772                                  "io buffer.  Skipping abort req.\n");
1773                 return;
1774         }
1775         nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1776
1777         /*
1778          * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1779          * state must match the nvme_fcreq passed by the nvme
1780          * transport.  If they don't match, it is likely the driver
1781          * has already completed the NVME IO and the nvme transport
1782          * has not seen it yet.
1783          */
1784         if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1785                 spin_unlock_irqrestore(&phba->hbalock, flags);
1786                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1787                                  "6143 NVME req mismatch: "
1788                                  "lpfc_nbuf %p nvmeCmd %p, "
1789                                  "pnvme_fcreq %p.  Skipping Abort xri x%x\n",
1790                                  lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1791                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1792                 return;
1793         }
1794
1795         /* Don't abort IOs no longer on the pending queue. */
1796         if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
1797                 spin_unlock_irqrestore(&phba->hbalock, flags);
1798                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1799                                  "6142 NVME IO req %p not queued - skipping "
1800                                  "abort req xri x%x\n",
1801                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1802                 return;
1803         }
1804
1805         atomic_inc(&lport->xmt_fcp_abort);
1806         lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1807                          nvmereq_wqe->sli4_xritag,
1808                          nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1809
1810         /* Outstanding abort is in progress */
1811         if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
1812                 spin_unlock_irqrestore(&phba->hbalock, flags);
1813                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1814                                  "6144 Outstanding NVME I/O Abort Request "
1815                                  "still pending on nvme_fcreq %p, "
1816                                  "lpfc_ncmd %p xri x%x\n",
1817                                  pnvme_fcreq, lpfc_nbuf,
1818                                  nvmereq_wqe->sli4_xritag);
1819                 return;
1820         }
1821
1822         abts_buf = __lpfc_sli_get_iocbq(phba);
1823         if (!abts_buf) {
1824                 spin_unlock_irqrestore(&phba->hbalock, flags);
1825                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1826                                  "6136 No available abort wqes. Skipping "
1827                                  "Abts req for nvme_fcreq %p xri x%x\n",
1828                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1829                 return;
1830         }
1831
1832         /* Ready - mark outstanding as aborted by driver. */
1833         nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
1834
1835         /* Complete prepping the abort wqe and issue to the FW. */
1836         abts_wqe = &abts_buf->wqe;
1837
1838         /* WQEs are reused.  Clear stale data and set key fields to
1839          * zero like ia, iaab, iaar, xri_tag, and ctxt_tag.
1840          */
1841         memset(abts_wqe, 0, sizeof(union lpfc_wqe));
1842         bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
1843
1844         /* word 7 */
1845         bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
1846         bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
1847                nvmereq_wqe->iocb.ulpClass);
1848
1849         /* word 8 - tell the FW to abort the IO associated with this
1850          * outstanding exchange ID.
1851          */
1852         abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag;
1853
1854         /* word 9 - this is the iotag for the abts_wqe completion. */
1855         bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
1856                abts_buf->iotag);
1857
1858         /* word 10 */
1859         bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
1860         bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
1861
1862         /* word 11 */
1863         bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
1864         bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
1865         bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
1866
1867         /* ABTS WQE must go to the same WQ as the WQE to be aborted */
1868         abts_buf->iocb_flag |= LPFC_IO_NVME;
1869         abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx;
1870         abts_buf->vport = vport;
1871         abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
1872         ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf);
1873         spin_unlock_irqrestore(&phba->hbalock, flags);
1874         if (ret_val) {
1875                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1876                                  "6137 Failed abts issue_wqe with status x%x "
1877                                  "for nvme_fcreq %p.\n",
1878                                  ret_val, pnvme_fcreq);
1879                 lpfc_sli_release_iocbq(phba, abts_buf);
1880                 return;
1881         }
1882
1883         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1884                          "6138 Transport Abort NVME Request Issued for "
1885                          "ox_id x%x on reqtag x%x\n",
1886                          nvmereq_wqe->sli4_xritag,
1887                          abts_buf->iotag);
1888 }
1889
1890 /* Declare and initialization an instance of the FC NVME template. */
1891 static struct nvme_fc_port_template lpfc_nvme_template = {
1892         /* initiator-based functions */
1893         .localport_delete  = lpfc_nvme_localport_delete,
1894         .remoteport_delete = lpfc_nvme_remoteport_delete,
1895         .create_queue = lpfc_nvme_create_queue,
1896         .delete_queue = lpfc_nvme_delete_queue,
1897         .ls_req       = lpfc_nvme_ls_req,
1898         .fcp_io       = lpfc_nvme_fcp_io_submit,
1899         .ls_abort     = lpfc_nvme_ls_abort,
1900         .fcp_abort    = lpfc_nvme_fcp_abort,
1901
1902         .max_hw_queues = 1,
1903         .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1904         .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1905         .dma_boundary = 0xFFFFFFFF,
1906
1907         /* Sizes of additional private data for data structures.
1908          * No use for the last two sizes at this time.
1909          */
1910         .local_priv_sz = sizeof(struct lpfc_nvme_lport),
1911         .remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1912         .lsrqst_priv_sz = 0,
1913         .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1914 };
1915
1916 static inline struct lpfc_nvme_buf *
1917 lpfc_nvme_buf(struct lpfc_hba *phba, int idx)
1918 {
1919         struct lpfc_sli4_hdw_queue *qp;
1920         struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
1921
1922         qp = &phba->sli4_hba.hdwq[idx];
1923         list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
1924                                  &qp->lpfc_io_buf_list_get, list) {
1925                 list_del_init(&lpfc_ncmd->list);
1926                 qp->get_io_bufs--;
1927                 return lpfc_ncmd;
1928         }
1929         return NULL;
1930 }
1931
1932 /**
1933  * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
1934  * @phba: The HBA for which this call is being executed.
1935  *
1936  * This routine removes a nvme buffer from head of @hdwq io_buf_list
1937  * and returns to caller.
1938  *
1939  * Return codes:
1940  *   NULL - Error
1941  *   Pointer to lpfc_nvme_buf - Success
1942  **/
1943 static struct lpfc_nvme_buf *
1944 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
1945                   int idx, int expedite)
1946 {
1947         struct lpfc_nvme_buf *lpfc_ncmd = NULL;
1948         struct lpfc_sli4_hdw_queue *qp;
1949         struct sli4_sge *sgl;
1950         struct lpfc_iocbq *pwqeq;
1951         union lpfc_wqe128 *wqe;
1952         unsigned long iflag = 0;
1953
1954         qp = &phba->sli4_hba.hdwq[idx];
1955         spin_lock_irqsave(&qp->io_buf_list_get_lock, iflag);
1956         if (qp->get_io_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite)
1957                 lpfc_ncmd = lpfc_nvme_buf(phba, idx);
1958         if (!lpfc_ncmd) {
1959                 spin_lock(&qp->io_buf_list_put_lock);
1960                 list_splice(&qp->lpfc_io_buf_list_put,
1961                             &qp->lpfc_io_buf_list_get);
1962                 qp->get_io_bufs += qp->put_io_bufs;
1963                 INIT_LIST_HEAD(&qp->lpfc_io_buf_list_put);
1964                 qp->put_io_bufs = 0;
1965                 spin_unlock(&qp->io_buf_list_put_lock);
1966                 if (qp->get_io_bufs > LPFC_NVME_EXPEDITE_XRICNT ||
1967                     expedite)
1968                         lpfc_ncmd = lpfc_nvme_buf(phba, idx);
1969         }
1970         spin_unlock_irqrestore(&qp->io_buf_list_get_lock, iflag);
1971
1972         if (lpfc_ncmd) {
1973                 pwqeq = &(lpfc_ncmd->cur_iocbq);
1974                 wqe = &pwqeq->wqe;
1975
1976                 /* Setup key fields in buffer that may have been changed
1977                  * if other protocols used this buffer.
1978                  */
1979                 pwqeq->iocb_flag = LPFC_IO_NVME;
1980                 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
1981                 lpfc_ncmd->start_time = jiffies;
1982                 lpfc_ncmd->flags = 0;
1983                 lpfc_ncmd->hdwq = idx;
1984
1985                 /* Rsp SGE will be filled in when we rcv an IO
1986                  * from the NVME Layer to be sent.
1987                  * The cmd is going to be embedded so we need a SKIP SGE.
1988                  */
1989                 sgl = lpfc_ncmd->dma_sgl;
1990                 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1991                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1992                 sgl->word2 = cpu_to_le32(sgl->word2);
1993                 /* Fill in word 3 / sgl_len during cmd submission */
1994
1995                 /* Initialize WQE */
1996                 memset(wqe, 0, sizeof(union lpfc_wqe));
1997
1998                 if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1999                         atomic_inc(&ndlp->cmd_pending);
2000                         lpfc_ncmd->flags |= LPFC_BUMP_QDEPTH;
2001                 }
2002
2003         } else
2004                 qp->empty_io_bufs++;
2005
2006         return  lpfc_ncmd;
2007 }
2008
2009 /**
2010  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2011  * @phba: The Hba for which this call is being executed.
2012  * @lpfc_ncmd: The nvme buffer which is being released.
2013  *
2014  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2015  * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2016  * and cannot be reused for at least RA_TOV amount of time if it was
2017  * aborted.
2018  **/
2019 static void
2020 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd)
2021 {
2022         struct lpfc_sli4_hdw_queue *qp;
2023         unsigned long iflag = 0;
2024
2025         if ((lpfc_ncmd->flags & LPFC_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2026                 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2027
2028         lpfc_ncmd->ndlp = NULL;
2029         lpfc_ncmd->flags &= ~LPFC_BUMP_QDEPTH;
2030
2031         qp = &phba->sli4_hba.hdwq[lpfc_ncmd->hdwq];
2032         if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) {
2033                 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2034                                 "6310 XB release deferred for "
2035                                 "ox_id x%x on reqtag x%x\n",
2036                                 lpfc_ncmd->cur_iocbq.sli4_xritag,
2037                                 lpfc_ncmd->cur_iocbq.iotag);
2038
2039                 spin_lock_irqsave(&qp->abts_nvme_buf_list_lock, iflag);
2040                 list_add_tail(&lpfc_ncmd->list,
2041                         &qp->lpfc_abts_nvme_buf_list);
2042                 qp->abts_nvme_io_bufs++;
2043                 spin_unlock_irqrestore(&qp->abts_nvme_buf_list_lock, iflag);
2044         } else {
2045                 /* MUST zero fields if buffer is reused by another protocol */
2046                 lpfc_ncmd->nvmeCmd = NULL;
2047                 lpfc_ncmd->cur_iocbq.wqe_cmpl = NULL;
2048
2049                 spin_lock_irqsave(&qp->io_buf_list_put_lock, iflag);
2050                 list_add_tail(&lpfc_ncmd->list,
2051                               &qp->lpfc_io_buf_list_put);
2052                 qp->put_io_bufs++;
2053                 spin_unlock_irqrestore(&qp->io_buf_list_put_lock, iflag);
2054         }
2055 }
2056
2057 /**
2058  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2059  * @pvport - the lpfc_vport instance requesting a localport.
2060  *
2061  * This routine is invoked to create an nvme localport instance to bind
2062  * to the nvme_fc_transport.  It is called once during driver load
2063  * like lpfc_create_shost after all other services are initialized.
2064  * It requires a vport, vpi, and wwns at call time.  Other localport
2065  * parameters are modified as the driver's FCID and the Fabric WWN
2066  * are established.
2067  *
2068  * Return codes
2069  *      0 - successful
2070  *      -ENOMEM - no heap memory available
2071  *      other values - from nvme registration upcall
2072  **/
2073 int
2074 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2075 {
2076         int ret = 0;
2077         struct lpfc_hba  *phba = vport->phba;
2078         struct nvme_fc_port_info nfcp_info;
2079         struct nvme_fc_local_port *localport;
2080         struct lpfc_nvme_lport *lport;
2081         struct lpfc_nvme_ctrl_stat *cstat;
2082         int i;
2083
2084         /* Initialize this localport instance.  The vport wwn usage ensures
2085          * that NPIV is accounted for.
2086          */
2087         memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2088         nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2089         nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2090         nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2091
2092         /* We need to tell the transport layer + 1 because it takes page
2093          * alignment into account. When space for the SGL is allocated we
2094          * allocate + 3, one for cmd, one for rsp and one for this alignment
2095          */
2096         lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2097         lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2098
2099         cstat = kmalloc((sizeof(struct lpfc_nvme_ctrl_stat) *
2100                         phba->cfg_hdw_queue), GFP_KERNEL);
2101         if (!cstat)
2102                 return -ENOMEM;
2103
2104         /* localport is allocated from the stack, but the registration
2105          * call allocates heap memory as well as the private area.
2106          */
2107 #if (IS_ENABLED(CONFIG_NVME_FC))
2108         ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2109                                          &vport->phba->pcidev->dev, &localport);
2110 #else
2111         ret = -ENOMEM;
2112 #endif
2113         if (!ret) {
2114                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2115                                  "6005 Successfully registered local "
2116                                  "NVME port num %d, localP %p, private %p, "
2117                                  "sg_seg %d\n",
2118                                  localport->port_num, localport,
2119                                  localport->private,
2120                                  lpfc_nvme_template.max_sgl_segments);
2121
2122                 /* Private is our lport size declared in the template. */
2123                 lport = (struct lpfc_nvme_lport *)localport->private;
2124                 vport->localport = localport;
2125                 lport->vport = vport;
2126                 lport->cstat = cstat;
2127                 vport->nvmei_support = 1;
2128
2129                 atomic_set(&lport->xmt_fcp_noxri, 0);
2130                 atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2131                 atomic_set(&lport->xmt_fcp_qdepth, 0);
2132                 atomic_set(&lport->xmt_fcp_err, 0);
2133                 atomic_set(&lport->xmt_fcp_wqerr, 0);
2134                 atomic_set(&lport->xmt_fcp_abort, 0);
2135                 atomic_set(&lport->xmt_ls_abort, 0);
2136                 atomic_set(&lport->xmt_ls_err, 0);
2137                 atomic_set(&lport->cmpl_fcp_xb, 0);
2138                 atomic_set(&lport->cmpl_fcp_err, 0);
2139                 atomic_set(&lport->cmpl_ls_xb, 0);
2140                 atomic_set(&lport->cmpl_ls_err, 0);
2141                 atomic_set(&lport->fc4NvmeLsRequests, 0);
2142                 atomic_set(&lport->fc4NvmeLsCmpls, 0);
2143
2144                 for (i = 0; i < phba->cfg_hdw_queue; i++) {
2145                         cstat = &lport->cstat[i];
2146                         atomic_set(&cstat->fc4NvmeInputRequests, 0);
2147                         atomic_set(&cstat->fc4NvmeOutputRequests, 0);
2148                         atomic_set(&cstat->fc4NvmeControlRequests, 0);
2149                         atomic_set(&cstat->fc4NvmeIoCmpls, 0);
2150                 }
2151         } else
2152                 kfree(cstat);
2153
2154         return ret;
2155 }
2156
2157 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2158  *
2159  * The driver has to wait for the host nvme transport to callback
2160  * indicating the localport has successfully unregistered all
2161  * resources.  Since this is an uninterruptible wait, loop every ten
2162  * seconds and print a message indicating no progress.
2163  *
2164  * An uninterruptible wait is used because of the risk of transport-to-
2165  * driver state mismatch.
2166  */
2167 void
2168 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2169                            struct lpfc_nvme_lport *lport)
2170 {
2171 #if (IS_ENABLED(CONFIG_NVME_FC))
2172         u32 wait_tmo;
2173         int ret;
2174
2175         /* Host transport has to clean up and confirm requiring an indefinite
2176          * wait. Print a message if a 10 second wait expires and renew the
2177          * wait. This is unexpected.
2178          */
2179         wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2180         while (true) {
2181                 ret = wait_for_completion_timeout(&lport->lport_unreg_done,
2182                                                   wait_tmo);
2183                 if (unlikely(!ret)) {
2184                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
2185                                          "6176 Lport %p Localport %p wait "
2186                                          "timed out. Renewing.\n",
2187                                          lport, vport->localport);
2188                         continue;
2189                 }
2190                 break;
2191         }
2192         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2193                          "6177 Lport %p Localport %p Complete Success\n",
2194                          lport, vport->localport);
2195 #endif
2196 }
2197
2198 /**
2199  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2200  * @pnvme: pointer to lpfc nvme data structure.
2201  *
2202  * This routine is invoked to destroy all lports bound to the phba.
2203  * The lport memory was allocated by the nvme fc transport and is
2204  * released there.  This routine ensures all rports bound to the
2205  * lport have been disconnected.
2206  *
2207  **/
2208 void
2209 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2210 {
2211 #if (IS_ENABLED(CONFIG_NVME_FC))
2212         struct nvme_fc_local_port *localport;
2213         struct lpfc_nvme_lport *lport;
2214         struct lpfc_nvme_ctrl_stat *cstat;
2215         int ret;
2216
2217         if (vport->nvmei_support == 0)
2218                 return;
2219
2220         localport = vport->localport;
2221         vport->localport = NULL;
2222         lport = (struct lpfc_nvme_lport *)localport->private;
2223         cstat = lport->cstat;
2224
2225         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2226                          "6011 Destroying NVME localport %p\n",
2227                          localport);
2228
2229         /* lport's rport list is clear.  Unregister
2230          * lport and release resources.
2231          */
2232         init_completion(&lport->lport_unreg_done);
2233         ret = nvme_fc_unregister_localport(localport);
2234
2235         /* Wait for completion.  This either blocks
2236          * indefinitely or succeeds
2237          */
2238         lpfc_nvme_lport_unreg_wait(vport, lport);
2239         kfree(cstat);
2240
2241         /* Regardless of the unregister upcall response, clear
2242          * nvmei_support.  All rports are unregistered and the
2243          * driver will clean up.
2244          */
2245         vport->nvmei_support = 0;
2246         if (ret == 0) {
2247                 lpfc_printf_vlog(vport,
2248                                  KERN_INFO, LOG_NVME_DISC,
2249                                  "6009 Unregistered lport Success\n");
2250         } else {
2251                 lpfc_printf_vlog(vport,
2252                                  KERN_INFO, LOG_NVME_DISC,
2253                                  "6010 Unregistered lport "
2254                                  "Failed, status x%x\n",
2255                                  ret);
2256         }
2257 #endif
2258 }
2259
2260 void
2261 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2262 {
2263 #if (IS_ENABLED(CONFIG_NVME_FC))
2264         struct nvme_fc_local_port *localport;
2265         struct lpfc_nvme_lport *lport;
2266
2267         localport = vport->localport;
2268         if (!localport) {
2269                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2270                                  "6710 Update NVME fail. No localport\n");
2271                 return;
2272         }
2273         lport = (struct lpfc_nvme_lport *)localport->private;
2274         if (!lport) {
2275                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2276                                  "6171 Update NVME fail. localP %p, No lport\n",
2277                                  localport);
2278                 return;
2279         }
2280         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2281                          "6012 Update NVME lport %p did x%x\n",
2282                          localport, vport->fc_myDID);
2283
2284         localport->port_id = vport->fc_myDID;
2285         if (localport->port_id == 0)
2286                 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2287         else
2288                 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2289
2290         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2291                          "6030 bound lport %p to DID x%06x\n",
2292                          lport, localport->port_id);
2293 #endif
2294 }
2295
2296 int
2297 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2298 {
2299 #if (IS_ENABLED(CONFIG_NVME_FC))
2300         int ret = 0;
2301         struct nvme_fc_local_port *localport;
2302         struct lpfc_nvme_lport *lport;
2303         struct lpfc_nvme_rport *rport;
2304         struct lpfc_nvme_rport *oldrport;
2305         struct nvme_fc_remote_port *remote_port;
2306         struct nvme_fc_port_info rpinfo;
2307         struct lpfc_nodelist *prev_ndlp = NULL;
2308
2309         lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2310                          "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2311                          ndlp->nlp_DID, ndlp->nlp_type);
2312
2313         localport = vport->localport;
2314         if (!localport)
2315                 return 0;
2316
2317         lport = (struct lpfc_nvme_lport *)localport->private;
2318
2319         /* NVME rports are not preserved across devloss.
2320          * Just register this instance.  Note, rpinfo->dev_loss_tmo
2321          * is left 0 to indicate accept transport defaults.  The
2322          * driver communicates port role capabilities consistent
2323          * with the PRLI response data.
2324          */
2325         memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2326         rpinfo.port_id = ndlp->nlp_DID;
2327         if (ndlp->nlp_type & NLP_NVME_TARGET)
2328                 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2329         if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2330                 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2331
2332         if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2333                 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2334
2335         rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2336         rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2337
2338         spin_lock_irq(&vport->phba->hbalock);
2339         oldrport = lpfc_ndlp_get_nrport(ndlp);
2340         spin_unlock_irq(&vport->phba->hbalock);
2341         if (!oldrport)
2342                 lpfc_nlp_get(ndlp);
2343
2344         ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2345         if (!ret) {
2346                 /* If the ndlp already has an nrport, this is just
2347                  * a resume of the existing rport.  Else this is a
2348                  * new rport.
2349                  */
2350                 /* Guard against an unregister/reregister
2351                  * race that leaves the WAIT flag set.
2352                  */
2353                 spin_lock_irq(&vport->phba->hbalock);
2354                 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2355                 spin_unlock_irq(&vport->phba->hbalock);
2356                 rport = remote_port->private;
2357                 if (oldrport) {
2358                         /* New remoteport record does not guarantee valid
2359                          * host private memory area.
2360                          */
2361                         prev_ndlp = oldrport->ndlp;
2362                         if (oldrport == remote_port->private) {
2363                                 /* Same remoteport - ndlp should match.
2364                                  * Just reuse.
2365                                  */
2366                                 lpfc_printf_vlog(ndlp->vport, KERN_INFO,
2367                                                  LOG_NVME_DISC,
2368                                                  "6014 Rebinding lport to "
2369                                                  "remoteport %p wwpn 0x%llx, "
2370                                                  "Data: x%x x%x %p %p x%x x%06x\n",
2371                                                  remote_port,
2372                                                  remote_port->port_name,
2373                                                  remote_port->port_id,
2374                                                  remote_port->port_role,
2375                                                  prev_ndlp,
2376                                                  ndlp,
2377                                                  ndlp->nlp_type,
2378                                                  ndlp->nlp_DID);
2379                                 return 0;
2380                         }
2381
2382                         /* Sever the ndlp<->rport association
2383                          * before dropping the ndlp ref from
2384                          * register.
2385                          */
2386                         spin_lock_irq(&vport->phba->hbalock);
2387                         ndlp->nrport = NULL;
2388                         ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2389                         spin_unlock_irq(&vport->phba->hbalock);
2390                         rport->ndlp = NULL;
2391                         rport->remoteport = NULL;
2392
2393                         /* Reference only removed if previous NDLP is no longer
2394                          * active. It might be just a swap and removing the
2395                          * reference would cause a premature cleanup.
2396                          */
2397                         if (prev_ndlp && prev_ndlp != ndlp) {
2398                                 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) ||
2399                                     (!prev_ndlp->nrport))
2400                                         lpfc_nlp_put(prev_ndlp);
2401                         }
2402                 }
2403
2404                 /* Clean bind the rport to the ndlp. */
2405                 rport->remoteport = remote_port;
2406                 rport->lport = lport;
2407                 rport->ndlp = ndlp;
2408                 spin_lock_irq(&vport->phba->hbalock);
2409                 ndlp->nrport = rport;
2410                 spin_unlock_irq(&vport->phba->hbalock);
2411                 lpfc_printf_vlog(vport, KERN_INFO,
2412                                  LOG_NVME_DISC | LOG_NODE,
2413                                  "6022 Binding new rport to "
2414                                  "lport %p Remoteport %p rport %p WWNN 0x%llx, "
2415                                  "Rport WWPN 0x%llx DID "
2416                                  "x%06x Role x%x, ndlp %p prev_ndlp %p\n",
2417                                  lport, remote_port, rport,
2418                                  rpinfo.node_name, rpinfo.port_name,
2419                                  rpinfo.port_id, rpinfo.port_role,
2420                                  ndlp, prev_ndlp);
2421         } else {
2422                 lpfc_printf_vlog(vport, KERN_ERR,
2423                                  LOG_NVME_DISC | LOG_NODE,
2424                                  "6031 RemotePort Registration failed "
2425                                  "err: %d, DID x%06x\n",
2426                                  ret, ndlp->nlp_DID);
2427         }
2428
2429         return ret;
2430 #else
2431         return 0;
2432 #endif
2433 }
2434
2435 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2436  *
2437  * There is no notion of Devloss or rport recovery from the current
2438  * nvme_transport perspective.  Loss of an rport just means IO cannot
2439  * be sent and recovery is completely up to the initator.
2440  * For now, the driver just unbinds the DID and port_role so that
2441  * no further IO can be issued.  Changes are planned for later.
2442  *
2443  * Notes - the ndlp reference count is not decremented here since
2444  * since there is no nvme_transport api for devloss.  Node ref count
2445  * is only adjusted in driver unload.
2446  */
2447 void
2448 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2449 {
2450 #if (IS_ENABLED(CONFIG_NVME_FC))
2451         int ret;
2452         struct nvme_fc_local_port *localport;
2453         struct lpfc_nvme_lport *lport;
2454         struct lpfc_nvme_rport *rport;
2455         struct nvme_fc_remote_port *remoteport = NULL;
2456
2457         localport = vport->localport;
2458
2459         /* This is fundamental error.  The localport is always
2460          * available until driver unload.  Just exit.
2461          */
2462         if (!localport)
2463                 return;
2464
2465         lport = (struct lpfc_nvme_lport *)localport->private;
2466         if (!lport)
2467                 goto input_err;
2468
2469         spin_lock_irq(&vport->phba->hbalock);
2470         rport = lpfc_ndlp_get_nrport(ndlp);
2471         if (rport)
2472                 remoteport = rport->remoteport;
2473         spin_unlock_irq(&vport->phba->hbalock);
2474         if (!remoteport)
2475                 goto input_err;
2476
2477         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2478                          "6033 Unreg nvme remoteport %p, portname x%llx, "
2479                          "port_id x%06x, portstate x%x port type x%x\n",
2480                          remoteport, remoteport->port_name,
2481                          remoteport->port_id, remoteport->port_state,
2482                          ndlp->nlp_type);
2483
2484         /* Sanity check ndlp type.  Only call for NVME ports. Don't
2485          * clear any rport state until the transport calls back.
2486          */
2487
2488         if (ndlp->nlp_type & NLP_NVME_TARGET) {
2489                 /* No concern about the role change on the nvme remoteport.
2490                  * The transport will update it.
2491                  */
2492                 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG;
2493
2494                 /* Don't let the host nvme transport keep sending keep-alives
2495                  * on this remoteport. Vport is unloading, no recovery. The
2496                  * return values is ignored.  The upcall is a courtesy to the
2497                  * transport.
2498                  */
2499                 if (vport->load_flag & FC_UNLOADING)
2500                         (void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2501
2502                 ret = nvme_fc_unregister_remoteport(remoteport);
2503                 if (ret != 0) {
2504                         lpfc_nlp_put(ndlp);
2505                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2506                                          "6167 NVME unregister failed %d "
2507                                          "port_state x%x\n",
2508                                          ret, remoteport->port_state);
2509                 }
2510         }
2511         return;
2512
2513  input_err:
2514 #endif
2515         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2516                          "6168 State error: lport %p, rport%p FCID x%06x\n",
2517                          vport->localport, ndlp->rport, ndlp->nlp_DID);
2518 }
2519
2520 /**
2521  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2522  * @phba: pointer to lpfc hba data structure.
2523  * @axri: pointer to the fcp xri abort wcqe structure.
2524  *
2525  * This routine is invoked by the worker thread to process a SLI4 fast-path
2526  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2527  * here.
2528  **/
2529 void
2530 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2531                            struct sli4_wcqe_xri_aborted *axri, int idx)
2532 {
2533         uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2534         struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd;
2535         struct nvmefc_fcp_req *nvme_cmd = NULL;
2536         struct lpfc_nodelist *ndlp;
2537         struct lpfc_sli4_hdw_queue *qp;
2538         unsigned long iflag = 0;
2539
2540         if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
2541                 return;
2542         qp = &phba->sli4_hba.hdwq[idx];
2543         spin_lock_irqsave(&phba->hbalock, iflag);
2544         spin_lock(&qp->abts_nvme_buf_list_lock);
2545         list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd,
2546                                  &qp->lpfc_abts_nvme_buf_list, list) {
2547                 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) {
2548                         list_del_init(&lpfc_ncmd->list);
2549                         qp->abts_nvme_io_bufs--;
2550                         lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2551                         lpfc_ncmd->status = IOSTAT_SUCCESS;
2552                         spin_unlock(&qp->abts_nvme_buf_list_lock);
2553
2554                         spin_unlock_irqrestore(&phba->hbalock, iflag);
2555                         ndlp = lpfc_ncmd->ndlp;
2556                         if (ndlp)
2557                                 lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2558
2559                         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2560                                         "6311 nvme_cmd %p xri x%x tag x%x "
2561                                         "abort complete and xri released\n",
2562                                         lpfc_ncmd->nvmeCmd, xri,
2563                                         lpfc_ncmd->cur_iocbq.iotag);
2564
2565                         /* Aborted NVME commands are required to not complete
2566                          * before the abort exchange command fully completes.
2567                          * Once completed, it is available via the put list.
2568                          */
2569                         if (lpfc_ncmd->nvmeCmd) {
2570                                 nvme_cmd = lpfc_ncmd->nvmeCmd;
2571                                 nvme_cmd->done(nvme_cmd);
2572                                 lpfc_ncmd->nvmeCmd = NULL;
2573                         }
2574                         lpfc_release_nvme_buf(phba, lpfc_ncmd);
2575                         return;
2576                 }
2577         }
2578         spin_unlock(&qp->abts_nvme_buf_list_lock);
2579         spin_unlock_irqrestore(&phba->hbalock, iflag);
2580
2581         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2582                         "6312 XRI Aborted xri x%x not found\n", xri);
2583
2584 }
2585
2586 /**
2587  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2588  * @phba: Pointer to HBA context object.
2589  *
2590  * This function flushes all wqes in the nvme rings and frees all resources
2591  * in the txcmplq. This function does not issue abort wqes for the IO
2592  * commands in txcmplq, they will just be returned with
2593  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2594  * slot has been permanently disabled.
2595  **/
2596 void
2597 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2598 {
2599         struct lpfc_sli_ring  *pring;
2600         u32 i, wait_cnt = 0;
2601
2602         if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2603                 return;
2604
2605         /* Cycle through all NVME rings and make sure all outstanding
2606          * WQEs have been removed from the txcmplqs.
2607          */
2608         for (i = 0; i < phba->cfg_hdw_queue; i++) {
2609                 pring = phba->sli4_hba.hdwq[i].nvme_wq->pring;
2610
2611                 if (!pring)
2612                         continue;
2613
2614                 /* Retrieve everything on the txcmplq */
2615                 while (!list_empty(&pring->txcmplq)) {
2616                         msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2617                         wait_cnt++;
2618
2619                         /* The sleep is 10mS.  Every ten seconds,
2620                          * dump a message.  Something is wrong.
2621                          */
2622                         if ((wait_cnt % 1000) == 0) {
2623                                 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
2624                                                 "6178 NVME IO not empty, "
2625                                                 "cnt %d\n", wait_cnt);
2626                         }
2627                 }
2628         }
2629 }