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 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_nvme_io_channel);
 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                          "io_channel %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         if (lport->vport->localport)
 301                 complete(lport->lport_unreg_cmp);
 302 }
 303
 304 /* lpfc_nvme_remoteport_delete
 305  *
 306  * @remoteport: Pointer to an nvme transport remoteport instance.
 307  *
 308  * This is a template downcall.  NVME transport calls this function
 309  * when it has completed the unregistration of a previously
 310  * registered remoteport.
 311  *
 312  * Return value :
 313  * None
 314  */
 315 void
 316 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
 317 {
 318         struct lpfc_nvme_rport *rport = remoteport->private;
 319         struct lpfc_vport *vport;
 320         struct lpfc_nodelist *ndlp;
 321
 322         ndlp = rport->ndlp;
 323         if (!ndlp)
 324                 goto rport_err;
 325
 326         vport = ndlp->vport;
 327         if (!vport)
 328                 goto rport_err;
 329
 330         /* Remove this rport from the lport's list - memory is owned by the
 331          * transport. Remove the ndlp reference for the NVME transport before
 332          * calling state machine to remove the node.
 333          */
 334         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 335                         "6146 remoteport delete of remoteport %p\n",
 336                         remoteport);
 337         spin_lock_irq(&vport->phba->hbalock);
 338
 339         /* The register rebind might have occurred before the delete
 340          * downcall.  Guard against this race.
 341          */
 342         if (ndlp->upcall_flags & NLP_WAIT_FOR_UNREG) {
 343                 ndlp->nrport = NULL;
 344                 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
 345         }
 346         spin_unlock_irq(&vport->phba->hbalock);
 347
 348         /* Remove original register reference. The host transport
 349          * won't reference this rport/remoteport any further.
 350          */
 351         lpfc_nlp_put(ndlp);
 352
 353  rport_err:
 354         return;
 355 }
 356
 357 static void
 358 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
 359                        struct lpfc_wcqe_complete *wcqe)
 360 {
 361         struct lpfc_vport *vport = cmdwqe->vport;
 362         struct lpfc_nvme_lport *lport;
 363         uint32_t status;
 364         struct nvmefc_ls_req *pnvme_lsreq;
 365         struct lpfc_dmabuf *buf_ptr;
 366         struct lpfc_nodelist *ndlp;
 367
 368         pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
 369         status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
 370
 371         if (vport->localport) {
 372                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
 373                 if (lport) {
 374                         atomic_inc(&lport->fc4NvmeLsCmpls);
 375                         if (status) {
 376                                 if (bf_get(lpfc_wcqe_c_xb, wcqe))
 377                                         atomic_inc(&lport->cmpl_ls_xb);
 378                                 atomic_inc(&lport->cmpl_ls_err);
 379                         }
 380                 }
 381         }
 382
 383         ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
 384         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 385                          "6047 nvme cmpl Enter "
 386                          "Data %p DID %x Xri: %x status %x reason x%x cmd:%p "
 387                          "lsreg:%p bmp:%p ndlp:%p\n",
 388                          pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
 389                          cmdwqe->sli4_xritag, status,
 390                          (wcqe->parameter & 0xffff),
 391                          cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
 392
 393         lpfc_nvmeio_data(phba, "NVME LS  CMPL: xri x%x stat x%x parm x%x\n",
 394                          cmdwqe->sli4_xritag, status, wcqe->parameter);
 395
 396         if (cmdwqe->context3) {
 397                 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
 398                 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
 399                 kfree(buf_ptr);
 400                 cmdwqe->context3 = NULL;
 401         }
 402         if (pnvme_lsreq->done)
 403                 pnvme_lsreq->done(pnvme_lsreq, status);
 404         else
 405                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 406                                  "6046 nvme cmpl without done call back? "
 407                                  "Data %p DID %x Xri: %x status %x\n",
 408                                 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
 409                                 cmdwqe->sli4_xritag, status);
 410         if (ndlp) {
 411                 lpfc_nlp_put(ndlp);
 412                 cmdwqe->context1 = NULL;
 413         }
 414         lpfc_sli_release_iocbq(phba, cmdwqe);
 415 }
 416
 417 static int
 418 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
 419                   struct lpfc_dmabuf *inp,
 420                   struct nvmefc_ls_req *pnvme_lsreq,
 421                   void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
 422                                struct lpfc_wcqe_complete *),
 423                   struct lpfc_nodelist *ndlp, uint32_t num_entry,
 424                   uint32_t tmo, uint8_t retry)
 425 {
 426         struct lpfc_hba *phba = vport->phba;
 427         union lpfc_wqe128 *wqe;
 428         struct lpfc_iocbq *genwqe;
 429         struct ulp_bde64 *bpl;
 430         struct ulp_bde64 bde;
 431         int i, rc, xmit_len, first_len;
 432
 433         /* Allocate buffer for  command WQE */
 434         genwqe = lpfc_sli_get_iocbq(phba);
 435         if (genwqe == NULL)
 436                 return 1;
 437
 438         wqe = &genwqe->wqe;
 439         memset(wqe, 0, sizeof(union lpfc_wqe));
 440
 441         genwqe->context3 = (uint8_t *)bmp;
 442         genwqe->iocb_flag |= LPFC_IO_NVME_LS;
 443
 444         /* Save for completion so we can release these resources */
 445         genwqe->context1 = lpfc_nlp_get(ndlp);
 446         genwqe->context2 = (uint8_t *)pnvme_lsreq;
 447         /* Fill in payload, bp points to frame payload */
 448
 449         if (!tmo)
 450                 /* FC spec states we need 3 * ratov for CT requests */
 451                 tmo = (3 * phba->fc_ratov);
 452
 453         /* For this command calculate the xmit length of the request bde. */
 454         xmit_len = 0;
 455         first_len = 0;
 456         bpl = (struct ulp_bde64 *)bmp->virt;
 457         for (i = 0; i < num_entry; i++) {
 458                 bde.tus.w = bpl[i].tus.w;
 459                 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
 460                         break;
 461                 xmit_len += bde.tus.f.bdeSize;
 462                 if (i == 0)
 463                         first_len = xmit_len;
 464         }
 465
 466         genwqe->rsvd2 = num_entry;
 467         genwqe->hba_wqidx = 0;
 468
 469         /* Words 0 - 2 */
 470         wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
 471         wqe->generic.bde.tus.f.bdeSize = first_len;
 472         wqe->generic.bde.addrLow = bpl[0].addrLow;
 473         wqe->generic.bde.addrHigh = bpl[0].addrHigh;
 474
 475         /* Word 3 */
 476         wqe->gen_req.request_payload_len = first_len;
 477
 478         /* Word 4 */
 479
 480         /* Word 5 */
 481         bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
 482         bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
 483         bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
 484         bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
 485         bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
 486
 487         /* Word 6 */
 488         bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
 489                phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
 490         bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
 491
 492         /* Word 7 */
 493         bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1));
 494         bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
 495         bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
 496         bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
 497
 498         /* Word 8 */
 499         wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
 500
 501         /* Word 9 */
 502         bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
 503
 504         /* Word 10 */
 505         bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
 506         bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
 507         bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
 508         bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
 509         bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
 510
 511         /* Word 11 */
 512         bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
 513         bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
 514
 515
 516         /* Issue GEN REQ WQE for NPORT <did> */
 517         lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
 518                          "6050 Issue GEN REQ WQE to NPORT x%x "
 519                          "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n",
 520                          ndlp->nlp_DID, genwqe->iotag,
 521                          vport->port_state,
 522                         genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
 523         genwqe->wqe_cmpl = cmpl;
 524         genwqe->iocb_cmpl = NULL;
 525         genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
 526         genwqe->vport = vport;
 527         genwqe->retry = retry;
 528
 529         lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
 530                          genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
 531
 532         rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe);
 533         if (rc) {
 534                 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
 535                                  "6045 Issue GEN REQ WQE to NPORT x%x "
 536                                  "Data: x%x x%x\n",
 537                                  ndlp->nlp_DID, genwqe->iotag,
 538                                  vport->port_state);
 539                 lpfc_sli_release_iocbq(phba, genwqe);
 540                 return 1;
 541         }
 542         return 0;
 543 }
 544
 545 /**
 546  * lpfc_nvme_ls_req - Issue an Link Service request
 547  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 548  * @lpfc_nvme_lport: Pointer to the driver's local port data
 549  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 550  *
 551  * Driver registers this routine to handle any link service request
 552  * from the nvme_fc transport to a remote nvme-aware port.
 553  *
 554  * Return value :
 555  *   0 - Success
 556  *   TODO: What are the failure codes.
 557  **/
 558 static int
 559 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
 560                  struct nvme_fc_remote_port *pnvme_rport,
 561                  struct nvmefc_ls_req *pnvme_lsreq)
 562 {
 563         int ret = 0;
 564         struct lpfc_nvme_lport *lport;
 565         struct lpfc_nvme_rport *rport;
 566         struct lpfc_vport *vport;
 567         struct lpfc_nodelist *ndlp;
 568         struct ulp_bde64 *bpl;
 569         struct lpfc_dmabuf *bmp;
 570         uint16_t ntype, nstate;
 571
 572         /* there are two dma buf in the request, actually there is one and
 573          * the second one is just the start address + cmd size.
 574          * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
 575          * in a lpfc_dmabuf struct. When freeing we just free the wrapper
 576          * because the nvem layer owns the data bufs.
 577          * We do not have to break these packets open, we don't care what is in
 578          * them. And we do not have to look at the resonse data, we only care
 579          * that we got a response. All of the caring is going to happen in the
 580          * nvme-fc layer.
 581          */
 582
 583         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 584         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
 585         if (unlikely(!lport) || unlikely(!rport))
 586                 return -EINVAL;
 587
 588         vport = lport->vport;
 589
 590         if (vport->load_flag & FC_UNLOADING)
 591                 return -ENODEV;
 592
 593         /* Need the ndlp.  It is stored in the driver's rport. */
 594         ndlp = rport->ndlp;
 595         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
 596                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 597                                  "6051 Remoteport %p, rport has invalid ndlp. "
 598                                  "Failing LS Req\n", pnvme_rport);
 599                 return -ENODEV;
 600         }
 601
 602         /* The remote node has to be a mapped nvme target or an
 603          * unmapped nvme initiator or it's an error.
 604          */
 605         ntype = ndlp->nlp_type;
 606         nstate = ndlp->nlp_state;
 607         if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
 608             (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
 609                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 610                                  "6088 DID x%06x not ready for "
 611                                  "IO. State x%x, Type x%x\n",
 612                                  pnvme_rport->port_id,
 613                                  ndlp->nlp_state, ndlp->nlp_type);
 614                 return -ENODEV;
 615         }
 616         bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
 617         if (!bmp) {
 618
 619                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 620                                  "6044 Could not find node for DID %x\n",
 621                                  pnvme_rport->port_id);
 622                 return 2;
 623         }
 624         INIT_LIST_HEAD(&bmp->list);
 625         bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
 626         if (!bmp->virt) {
 627                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 628                                  "6042 Could not find node for DID %x\n",
 629                                  pnvme_rport->port_id);
 630                 kfree(bmp);
 631                 return 3;
 632         }
 633         bpl = (struct ulp_bde64 *)bmp->virt;
 634         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
 635         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
 636         bpl->tus.f.bdeFlags = 0;
 637         bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
 638         bpl->tus.w = le32_to_cpu(bpl->tus.w);
 639         bpl++;
 640
 641         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
 642         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
 643         bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
 644         bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
 645         bpl->tus.w = le32_to_cpu(bpl->tus.w);
 646
 647         /* Expand print to include key fields. */
 648         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
 649                          "6149 Issue LS Req to DID 0x%06x lport %p, rport %p "
 650                          "lsreq%p rqstlen:%d rsplen:%d %pad %pad\n",
 651                          ndlp->nlp_DID,
 652                          pnvme_lport, pnvme_rport,
 653                          pnvme_lsreq, pnvme_lsreq->rqstlen,
 654                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
 655                          &pnvme_lsreq->rspdma);
 656
 657         atomic_inc(&lport->fc4NvmeLsRequests);
 658
 659         /* Hardcode the wait to 30 seconds.  Connections are failing otherwise.
 660          * This code allows it all to work.
 661          */
 662         ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
 663                                 pnvme_lsreq, lpfc_nvme_cmpl_gen_req,
 664                                 ndlp, 2, 30, 0);
 665         if (ret != WQE_SUCCESS) {
 666                 atomic_inc(&lport->xmt_ls_err);
 667                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
 668                                  "6052 EXIT. issue ls wqe failed lport %p, "
 669                                  "rport %p lsreq%p Status %x DID %x\n",
 670                                  pnvme_lport, pnvme_rport, pnvme_lsreq,
 671                                  ret, ndlp->nlp_DID);
 672                 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
 673                 kfree(bmp);
 674                 return ret;
 675         }
 676
 677         /* Stub in routine and return 0 for now. */
 678         return ret;
 679 }
 680
 681 /**
 682  * lpfc_nvme_ls_abort - Issue an Link Service request
 683  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 684  * @lpfc_nvme_lport: Pointer to the driver's local port data
 685  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 686  *
 687  * Driver registers this routine to handle any link service request
 688  * from the nvme_fc transport to a remote nvme-aware port.
 689  *
 690  * Return value :
 691  *   0 - Success
 692  *   TODO: What are the failure codes.
 693  **/
 694 static void
 695 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
 696                    struct nvme_fc_remote_port *pnvme_rport,
 697                    struct nvmefc_ls_req *pnvme_lsreq)
 698 {
 699         struct lpfc_nvme_lport *lport;
 700         struct lpfc_vport *vport;
 701         struct lpfc_hba *phba;
 702         struct lpfc_nodelist *ndlp;
 703         LIST_HEAD(abort_list);
 704         struct lpfc_sli_ring *pring;
 705         struct lpfc_iocbq *wqe, *next_wqe;
 706
 707         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
 708         if (unlikely(!lport))
 709                 return;
 710         vport = lport->vport;
 711         phba = vport->phba;
 712
 713         if (vport->load_flag & FC_UNLOADING)
 714                 return;
 715
 716         ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
 717         if (!ndlp) {
 718                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
 719                                  "6049 Could not find node for DID %x\n",
 720                                  pnvme_rport->port_id);
 721                 return;
 722         }
 723
 724         /* Expand print to include key fields. */
 725         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
 726                          "6040 ENTER.  lport %p, rport %p lsreq %p rqstlen:%d "
 727                          "rsplen:%d %pad %pad\n",
 728                          pnvme_lport, pnvme_rport,
 729                          pnvme_lsreq, pnvme_lsreq->rqstlen,
 730                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
 731                          &pnvme_lsreq->rspdma);
 732
 733         /*
 734          * Lock the ELS ring txcmplq and build a local list of all ELS IOs
 735          * that need an ABTS.  The IOs need to stay on the txcmplq so that
 736          * the abort operation completes them successfully.
 737          */
 738         pring = phba->sli4_hba.nvmels_wq->pring;
 739         spin_lock_irq(&phba->hbalock);
 740         spin_lock(&pring->ring_lock);
 741         list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
 742                 /* Add to abort_list on on NDLP match. */
 743                 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) {
 744                         wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
 745                         list_add_tail(&wqe->dlist, &abort_list);
 746                 }
 747         }
 748         spin_unlock(&pring->ring_lock);
 749         spin_unlock_irq(&phba->hbalock);
 750
 751         /* Abort the targeted IOs and remove them from the abort list. */
 752         list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) {
 753                 atomic_inc(&lport->xmt_ls_abort);
 754                 spin_lock_irq(&phba->hbalock);
 755                 list_del_init(&wqe->dlist);
 756                 lpfc_sli_issue_abort_iotag(phba, pring, wqe);
 757                 spin_unlock_irq(&phba->hbalock);
 758         }
 759 }
 760
 761 /* Fix up the existing sgls for NVME IO. */
 762 static inline void
 763 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
 764                        struct lpfc_nvme_buf *lpfc_ncmd,
 765                        struct nvmefc_fcp_req *nCmd)
 766 {
 767         struct lpfc_hba  *phba = vport->phba;
 768         struct sli4_sge *sgl;
 769         union lpfc_wqe128 *wqe;
 770         uint32_t *wptr, *dptr;
 771
 772         /*
 773          * Get a local pointer to the built-in wqe and correct
 774          * the cmd size to match NVME's 96 bytes and fix
 775          * the dma address.
 776          */
 777
 778         wqe = &lpfc_ncmd->cur_iocbq.wqe;
 779
 780         /*
 781          * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
 782          * match NVME.  NVME sends 96 bytes. Also, use the
 783          * nvme commands command and response dma addresses
 784          * rather than the virtual memory to ease the restore
 785          * operation.
 786          */
 787         sgl = lpfc_ncmd->nvme_sgl;
 788         sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
 789         if (phba->cfg_nvme_embed_cmd) {
 790                 sgl->addr_hi = 0;
 791                 sgl->addr_lo = 0;
 792
 793                 /* Word 0-2 - NVME CMND IU (embedded payload) */
 794                 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
 795                 wqe->generic.bde.tus.f.bdeSize = 56;
 796                 wqe->generic.bde.addrHigh = 0;
 797                 wqe->generic.bde.addrLow =  64;  /* Word 16 */
 798
 799                 /* Word 10  - dbde is 0, wqes is 1 in template */
 800
 801                 /*
 802                  * Embed the payload in the last half of the WQE
 803                  * WQE words 16-30 get the NVME CMD IU payload
 804                  *
 805                  * WQE words 16-19 get payload Words 1-4
 806                  * WQE words 20-21 get payload Words 6-7
 807                  * WQE words 22-29 get payload Words 16-23
 808                  */
 809                 wptr = &wqe->words[16];  /* WQE ptr */
 810                 dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
 811                 dptr++;                 /* Skip Word 0 in payload */
 812
 813                 *wptr++ = *dptr++;      /* Word 1 */
 814                 *wptr++ = *dptr++;      /* Word 2 */
 815                 *wptr++ = *dptr++;      /* Word 3 */
 816                 *wptr++ = *dptr++;      /* Word 4 */
 817                 dptr++;                 /* Skip Word 5 in payload */
 818                 *wptr++ = *dptr++;      /* Word 6 */
 819                 *wptr++ = *dptr++;      /* Word 7 */
 820                 dptr += 8;              /* Skip Words 8-15 in payload */
 821                 *wptr++ = *dptr++;      /* Word 16 */
 822                 *wptr++ = *dptr++;      /* Word 17 */
 823                 *wptr++ = *dptr++;      /* Word 18 */
 824                 *wptr++ = *dptr++;      /* Word 19 */
 825                 *wptr++ = *dptr++;      /* Word 20 */
 826                 *wptr++ = *dptr++;      /* Word 21 */
 827                 *wptr++ = *dptr++;      /* Word 22 */
 828                 *wptr   = *dptr;        /* Word 23 */
 829         } else {
 830                 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
 831                 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
 832
 833                 /* Word 0-2 - NVME CMND IU Inline BDE */
 834                 wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
 835                 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
 836                 wqe->generic.bde.addrHigh = sgl->addr_hi;
 837                 wqe->generic.bde.addrLow =  sgl->addr_lo;
 838
 839                 /* Word 10 */
 840                 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
 841                 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
 842         }
 843
 844         sgl++;
 845
 846         /* Setup the physical region for the FCP RSP */
 847         sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
 848         sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
 849         sgl->word2 = le32_to_cpu(sgl->word2);
 850         if (nCmd->sg_cnt)
 851                 bf_set(lpfc_sli4_sge_last, sgl, 0);
 852         else
 853                 bf_set(lpfc_sli4_sge_last, sgl, 1);
 854         sgl->word2 = cpu_to_le32(sgl->word2);
 855         sgl->sge_len = cpu_to_le32(nCmd->rsplen);
 856 }
 857
 858 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
 859 static void
 860 lpfc_nvme_ktime(struct lpfc_hba *phba,
 861                 struct lpfc_nvme_buf *lpfc_ncmd)
 862 {
 863         uint64_t seg1, seg2, seg3, seg4;
 864         uint64_t segsum;
 865
 866         if (!lpfc_ncmd->ts_last_cmd ||
 867             !lpfc_ncmd->ts_cmd_start ||
 868             !lpfc_ncmd->ts_cmd_wqput ||
 869             !lpfc_ncmd->ts_isr_cmpl ||
 870             !lpfc_ncmd->ts_data_nvme)
 871                 return;
 872
 873         if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_cmd_start)
 874                 return;
 875         if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd)
 876                 return;
 877         if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start)
 878                 return;
 879         if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput)
 880                 return;
 881         if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl)
 882                 return;
 883         /*
 884          * Segment 1 - Time from Last FCP command cmpl is handed
 885          * off to NVME Layer to start of next command.
 886          * Segment 2 - Time from Driver receives a IO cmd start
 887          * from NVME Layer to WQ put is done on IO cmd.
 888          * Segment 3 - Time from Driver WQ put is done on IO cmd
 889          * to MSI-X ISR for IO cmpl.
 890          * Segment 4 - Time from MSI-X ISR for IO cmpl to when
 891          * cmpl is handled off to the NVME Layer.
 892          */
 893         seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd;
 894         if (seg1 > 5000000)  /* 5 ms - for sequential IOs only */
 895                 seg1 = 0;
 896
 897         /* Calculate times relative to start of IO */
 898         seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start);
 899         segsum = seg2;
 900         seg3 = lpfc_ncmd->ts_isr_cmpl - lpfc_ncmd->ts_cmd_start;
 901         if (segsum > seg3)
 902                 return;
 903         seg3 -= segsum;
 904         segsum += seg3;
 905
 906         seg4 = lpfc_ncmd->ts_data_nvme - lpfc_ncmd->ts_cmd_start;
 907         if (segsum > seg4)
 908                 return;
 909         seg4 -= segsum;
 910
 911         phba->ktime_data_samples++;
 912         phba->ktime_seg1_total += seg1;
 913         if (seg1 < phba->ktime_seg1_min)
 914                 phba->ktime_seg1_min = seg1;
 915         else if (seg1 > phba->ktime_seg1_max)
 916                 phba->ktime_seg1_max = seg1;
 917         phba->ktime_seg2_total += seg2;
 918         if (seg2 < phba->ktime_seg2_min)
 919                 phba->ktime_seg2_min = seg2;
 920         else if (seg2 > phba->ktime_seg2_max)
 921                 phba->ktime_seg2_max = seg2;
 922         phba->ktime_seg3_total += seg3;
 923         if (seg3 < phba->ktime_seg3_min)
 924                 phba->ktime_seg3_min = seg3;
 925         else if (seg3 > phba->ktime_seg3_max)
 926                 phba->ktime_seg3_max = seg3;
 927         phba->ktime_seg4_total += seg4;
 928         if (seg4 < phba->ktime_seg4_min)
 929                 phba->ktime_seg4_min = seg4;
 930         else if (seg4 > phba->ktime_seg4_max)
 931                 phba->ktime_seg4_max = seg4;
 932
 933         lpfc_ncmd->ts_last_cmd = 0;
 934         lpfc_ncmd->ts_cmd_start = 0;
 935         lpfc_ncmd->ts_cmd_wqput  = 0;
 936         lpfc_ncmd->ts_isr_cmpl = 0;
 937         lpfc_ncmd->ts_data_nvme = 0;
 938 }
 939 #endif
 940
 941 /**
 942  * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
 943  * @lpfc_pnvme: Pointer to the driver's nvme instance data
 944  * @lpfc_nvme_lport: Pointer to the driver's local port data
 945  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
 946  *
 947  * Driver registers this routine as it io request handler.  This
 948  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
 949  * data structure to the rport indicated in @lpfc_nvme_rport.
 950  *
 951  * Return value :
 952  *   0 - Success
 953  *   TODO: What are the failure codes.
 954  **/
 955 static void
 956 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
 957                           struct lpfc_wcqe_complete *wcqe)
 958 {
 959         struct lpfc_nvme_buf *lpfc_ncmd =
 960                 (struct lpfc_nvme_buf *)pwqeIn->context1;
 961         struct lpfc_vport *vport = pwqeIn->vport;
 962         struct nvmefc_fcp_req *nCmd;
 963         struct nvme_fc_ersp_iu *ep;
 964         struct nvme_fc_cmd_iu *cp;
 965         struct lpfc_nvme_rport *rport;
 966         struct lpfc_nodelist *ndlp;
 967         struct lpfc_nvme_fcpreq_priv *freqpriv;
 968         struct lpfc_nvme_lport *lport;
 969         struct lpfc_nvme_ctrl_stat *cstat;
 970         unsigned long flags;
 971         uint32_t code, status, idx;
 972         uint16_t cid, sqhd, data;
 973         uint32_t *ptr;
 974
 975         /* Sanity check on return of outstanding command */
 976         if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd || !lpfc_ncmd->nrport) {
 977                 if (!lpfc_ncmd) {
 978                         lpfc_printf_vlog(vport, KERN_ERR,
 979                                          LOG_NODE | LOG_NVME_IOERR,
 980                                          "6071 Null lpfc_ncmd pointer. No "
 981                                          "release, skip completion\n");
 982                         return;
 983                 }
 984
 985                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
 986                                  "6066 Missing cmpl ptrs: lpfc_ncmd %p, "
 987                                  "nvmeCmd %p nrport %p\n",
 988                                  lpfc_ncmd, lpfc_ncmd->nvmeCmd,
 989                                  lpfc_ncmd->nrport);
 990
 991                 /* Release the lpfc_ncmd regardless of the missing elements. */
 992                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
 993                 return;
 994         }
 995         nCmd = lpfc_ncmd->nvmeCmd;
 996         rport = lpfc_ncmd->nrport;
 997         status = bf_get(lpfc_wcqe_c_status, wcqe);
 998
 999         if (vport->localport) {
1000                 lport = (struct lpfc_nvme_lport *)vport->localport->private;
1001                 if (lport) {
1002                         idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
1003                         cstat = &lport->cstat[idx];
1004                         atomic_inc(&cstat->fc4NvmeIoCmpls);
1005                         if (status) {
1006                                 if (bf_get(lpfc_wcqe_c_xb, wcqe))
1007                                         atomic_inc(&lport->cmpl_fcp_xb);
1008                                 atomic_inc(&lport->cmpl_fcp_err);
1009                         }
1010                 }
1011         }
1012
1013         lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
1014                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1015                          status, wcqe->parameter);
1016         /*
1017          * Catch race where our node has transitioned, but the
1018          * transport is still transitioning.
1019          */
1020         ndlp = rport->ndlp;
1021         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1022                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
1023                                  "6061 rport %p,  DID x%06x node not ready.\n",
1024                                  rport, rport->remoteport->port_id);
1025
1026                 ndlp = lpfc_findnode_did(vport, rport->remoteport->port_id);
1027                 if (!ndlp) {
1028                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1029                                          "6062 Ignoring NVME cmpl.  No ndlp\n");
1030                         goto out_err;
1031                 }
1032         }
1033
1034         code = bf_get(lpfc_wcqe_c_code, wcqe);
1035         if (code == CQE_CODE_NVME_ERSP) {
1036                 /* For this type of CQE, we need to rebuild the rsp */
1037                 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1038
1039                 /*
1040                  * Get Command Id from cmd to plug into response. This
1041                  * code is not needed in the next NVME Transport drop.
1042                  */
1043                 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1044                 cid = cp->sqe.common.command_id;
1045
1046                 /*
1047                  * RSN is in CQE word 2
1048                  * SQHD is in CQE Word 3 bits 15:0
1049                  * Cmd Specific info is in CQE Word 1
1050                  * and in CQE Word 0 bits 15:0
1051                  */
1052                 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1053
1054                 /* Now lets build the NVME ERSP IU */
1055                 ep->iu_len = cpu_to_be16(8);
1056                 ep->rsn = wcqe->parameter;
1057                 ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1058                 ep->rsvd12 = 0;
1059                 ptr = (uint32_t *)&ep->cqe.result.u64;
1060                 *ptr++ = wcqe->total_data_placed;
1061                 data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1062                 *ptr = (uint32_t)data;
1063                 ep->cqe.sq_head = sqhd;
1064                 ep->cqe.sq_id =  nCmd->sqid;
1065                 ep->cqe.command_id = cid;
1066                 ep->cqe.status = 0;
1067
1068                 lpfc_ncmd->status = IOSTAT_SUCCESS;
1069                 lpfc_ncmd->result = 0;
1070                 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1071                 nCmd->transferred_length = nCmd->payload_length;
1072         } else {
1073                 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1074                 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1075
1076                 /* For NVME, the only failure path that results in an
1077                  * IO error is when the adapter rejects it.  All other
1078                  * conditions are a success case and resolved by the
1079                  * transport.
1080                  * IOSTAT_FCP_RSP_ERROR means:
1081                  * 1. Length of data received doesn't match total
1082                  *    transfer length in WQE
1083                  * 2. If the RSP payload does NOT match these cases:
1084                  *    a. RSP length 12/24 bytes and all zeros
1085                  *    b. NVME ERSP
1086                  */
1087                 switch (lpfc_ncmd->status) {
1088                 case IOSTAT_SUCCESS:
1089                         nCmd->transferred_length = wcqe->total_data_placed;
1090                         nCmd->rcv_rsplen = 0;
1091                         nCmd->status = 0;
1092                         break;
1093                 case IOSTAT_FCP_RSP_ERROR:
1094                         nCmd->transferred_length = wcqe->total_data_placed;
1095                         nCmd->rcv_rsplen = wcqe->parameter;
1096                         nCmd->status = 0;
1097                         /* Sanity check */
1098                         if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN)
1099                                 break;
1100                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1101                                          "6081 NVME Completion Protocol Error: "
1102                                          "xri %x status x%x result x%x "
1103                                          "placed x%x\n",
1104                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1105                                          lpfc_ncmd->status, lpfc_ncmd->result,
1106                                          wcqe->total_data_placed);
1107                         break;
1108                 case IOSTAT_LOCAL_REJECT:
1109                         /* Let fall through to set command final state. */
1110                         if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1111                                 lpfc_printf_vlog(vport, KERN_INFO,
1112                                          LOG_NVME_IOERR,
1113                                          "6032 Delay Aborted cmd %p "
1114                                          "nvme cmd %p, xri x%x, "
1115                                          "xb %d\n",
1116                                          lpfc_ncmd, nCmd,
1117                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1118                                          bf_get(lpfc_wcqe_c_xb, wcqe));
1119                 default:
1120 out_err:
1121                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1122                                          "6072 NVME Completion Error: xri %x "
1123                                          "status x%x result x%x placed x%x\n",
1124                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1125                                          lpfc_ncmd->status, lpfc_ncmd->result,
1126                                          wcqe->total_data_placed);
1127                         nCmd->transferred_length = 0;
1128                         nCmd->rcv_rsplen = 0;
1129                         nCmd->status = NVME_SC_INTERNAL;
1130                 }
1131         }
1132
1133         /* pick up SLI4 exhange busy condition */
1134         if (bf_get(lpfc_wcqe_c_xb, wcqe))
1135                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1136         else
1137                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1138
1139         /* Update stats and complete the IO.  There is
1140          * no need for dma unprep because the nvme_transport
1141          * owns the dma address.
1142          */
1143 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1144         if (lpfc_ncmd->ts_cmd_start) {
1145                 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1146                 lpfc_ncmd->ts_data_nvme = ktime_get_ns();
1147                 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme;
1148                 lpfc_nvme_ktime(phba, lpfc_ncmd);
1149         }
1150         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1151                 if (lpfc_ncmd->cpu != smp_processor_id())
1152                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1153                                          "6701 CPU Check cmpl: "
1154                                          "cpu %d expect %d\n",
1155                                          smp_processor_id(), lpfc_ncmd->cpu);
1156                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
1157                         phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++;
1158         }
1159 #endif
1160
1161         /* NVME targets need completion held off until the abort exchange
1162          * completes unless the NVME Rport is getting unregistered.
1163          */
1164
1165         if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1166                 freqpriv = nCmd->private;
1167                 freqpriv->nvme_buf = NULL;
1168                 nCmd->done(nCmd);
1169                 lpfc_ncmd->nvmeCmd = NULL;
1170         }
1171
1172         spin_lock_irqsave(&phba->hbalock, flags);
1173         lpfc_ncmd->nrport = NULL;
1174         spin_unlock_irqrestore(&phba->hbalock, flags);
1175
1176         /* Call release with XB=1 to queue the IO into the abort list. */
1177         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1178 }
1179
1180
1181 /**
1182  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1183  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1184  * @lpfc_nvme_lport: Pointer to the driver's local port data
1185  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1186  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1187  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1188  *
1189  * Driver registers this routine as it io request handler.  This
1190  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1191  * data structure to the rport indicated in @lpfc_nvme_rport.
1192  *
1193  * Return value :
1194  *   0 - Success
1195  *   TODO: What are the failure codes.
1196  **/
1197 static int
1198 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1199                       struct lpfc_nvme_buf *lpfc_ncmd,
1200                       struct lpfc_nodelist *pnode,
1201                       struct lpfc_nvme_ctrl_stat *cstat)
1202 {
1203         struct lpfc_hba *phba = vport->phba;
1204         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1205         struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1206         union lpfc_wqe128 *wqe = &pwqeq->wqe;
1207         uint32_t req_len;
1208
1209         if (!pnode || !NLP_CHK_NODE_ACT(pnode))
1210                 return -EINVAL;
1211
1212         /*
1213          * There are three possibilities here - use scatter-gather segment, use
1214          * the single mapping, or neither.
1215          */
1216         if (nCmd->sg_cnt) {
1217                 if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1218                         /* From the iwrite template, initialize words 7 - 11 */
1219                         memcpy(&wqe->words[7],
1220                                &lpfc_iwrite_cmd_template.words[7],
1221                                sizeof(uint32_t) * 5);
1222
1223                         /* Word 4 */
1224                         wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1225
1226                         /* Word 5 */
1227                         if ((phba->cfg_nvme_enable_fb) &&
1228                             (pnode->nlp_flag & NLP_FIRSTBURST)) {
1229                                 req_len = lpfc_ncmd->nvmeCmd->payload_length;
1230                                 if (req_len < pnode->nvme_fb_size)
1231                                         wqe->fcp_iwrite.initial_xfer_len =
1232                                                 req_len;
1233                                 else
1234                                         wqe->fcp_iwrite.initial_xfer_len =
1235                                                 pnode->nvme_fb_size;
1236                         } else {
1237                                 wqe->fcp_iwrite.initial_xfer_len = 0;
1238                         }
1239                         atomic_inc(&cstat->fc4NvmeOutputRequests);
1240                 } else {
1241                         /* From the iread template, initialize words 7 - 11 */
1242                         memcpy(&wqe->words[7],
1243                                &lpfc_iread_cmd_template.words[7],
1244                                sizeof(uint32_t) * 5);
1245
1246                         /* Word 4 */
1247                         wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1248
1249                         /* Word 5 */
1250                         wqe->fcp_iread.rsrvd5 = 0;
1251
1252                         atomic_inc(&cstat->fc4NvmeInputRequests);
1253                 }
1254         } else {
1255                 /* From the icmnd template, initialize words 4 - 11 */
1256                 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1257                        sizeof(uint32_t) * 8);
1258                 atomic_inc(&cstat->fc4NvmeControlRequests);
1259         }
1260         /*
1261          * Finish initializing those WQE fields that are independent
1262          * of the nvme_cmnd request_buffer
1263          */
1264
1265         /* Word 3 */
1266         bf_set(payload_offset_len, &wqe->fcp_icmd,
1267                (nCmd->rsplen + nCmd->cmdlen));
1268
1269         /* Word 6 */
1270         bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1271                phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1272         bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1273
1274         /* Word 8 */
1275         wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1276
1277         /* Word 9 */
1278         bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1279
1280         /* Words 13 14 15 are for PBDE support */
1281
1282         pwqeq->vport = vport;
1283         return 0;
1284 }
1285
1286
1287 /**
1288  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1289  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1290  * @lpfc_nvme_lport: Pointer to the driver's local port data
1291  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1292  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1293  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1294  *
1295  * Driver registers this routine as it io request handler.  This
1296  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1297  * data structure to the rport indicated in @lpfc_nvme_rport.
1298  *
1299  * Return value :
1300  *   0 - Success
1301  *   TODO: What are the failure codes.
1302  **/
1303 static int
1304 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1305                       struct lpfc_nvme_buf *lpfc_ncmd)
1306 {
1307         struct lpfc_hba *phba = vport->phba;
1308         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1309         union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1310         struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl;
1311         struct scatterlist *data_sg;
1312         struct sli4_sge *first_data_sgl;
1313         struct ulp_bde64 *bde;
1314         dma_addr_t physaddr;
1315         uint32_t num_bde = 0;
1316         uint32_t dma_len;
1317         uint32_t dma_offset = 0;
1318         int nseg, i;
1319
1320         /* Fix up the command and response DMA stuff. */
1321         lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1322
1323         /*
1324          * There are three possibilities here - use scatter-gather segment, use
1325          * the single mapping, or neither.
1326          */
1327         if (nCmd->sg_cnt) {
1328                 /*
1329                  * Jump over the cmd and rsp SGEs.  The fix routine
1330                  * has already adjusted for this.
1331                  */
1332                 sgl += 2;
1333
1334                 first_data_sgl = sgl;
1335                 lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1336                 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1337                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1338                                         "6058 Too many sg segments from "
1339                                         "NVME Transport.  Max %d, "
1340                                         "nvmeIO sg_cnt %d\n",
1341                                         phba->cfg_nvme_seg_cnt + 1,
1342                                         lpfc_ncmd->seg_cnt);
1343                         lpfc_ncmd->seg_cnt = 0;
1344                         return 1;
1345                 }
1346
1347                 /*
1348                  * The driver established a maximum scatter-gather segment count
1349                  * during probe that limits the number of sg elements in any
1350                  * single nvme command.  Just run through the seg_cnt and format
1351                  * the sge's.
1352                  */
1353                 nseg = nCmd->sg_cnt;
1354                 data_sg = nCmd->first_sgl;
1355                 for (i = 0; i < nseg; i++) {
1356                         if (data_sg == NULL) {
1357                                 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1358                                                 "6059 dptr err %d, nseg %d\n",
1359                                                 i, nseg);
1360                                 lpfc_ncmd->seg_cnt = 0;
1361                                 return 1;
1362                         }
1363                         physaddr = data_sg->dma_address;
1364                         dma_len = data_sg->length;
1365                         sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
1366                         sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
1367                         sgl->word2 = le32_to_cpu(sgl->word2);
1368                         if ((num_bde + 1) == nseg)
1369                                 bf_set(lpfc_sli4_sge_last, sgl, 1);
1370                         else
1371                                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1372                         bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1373                         bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA);
1374                         sgl->word2 = cpu_to_le32(sgl->word2);
1375                         sgl->sge_len = cpu_to_le32(dma_len);
1376
1377                         dma_offset += dma_len;
1378                         data_sg = sg_next(data_sg);
1379                         sgl++;
1380                 }
1381                 if (phba->cfg_enable_pbde) {
1382                         /* Use PBDE support for first SGL only, offset == 0 */
1383                         /* Words 13-15 */
1384                         bde = (struct ulp_bde64 *)
1385                                 &wqe->words[13];
1386                         bde->addrLow = first_data_sgl->addr_lo;
1387                         bde->addrHigh = first_data_sgl->addr_hi;
1388                         bde->tus.f.bdeSize =
1389                                 le32_to_cpu(first_data_sgl->sge_len);
1390                         bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1391                         bde->tus.w = cpu_to_le32(bde->tus.w);
1392                         /* wqe_pbde is 1 in template */
1393                 } else {
1394                         memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1395                         bf_set(wqe_pbde, &wqe->generic.wqe_com, 0);
1396                 }
1397
1398         } else {
1399                 /* For this clause to be valid, the payload_length
1400                  * and sg_cnt must zero.
1401                  */
1402                 if (nCmd->payload_length != 0) {
1403                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1404                                         "6063 NVME DMA Prep Err: sg_cnt %d "
1405                                         "payload_length x%x\n",
1406                                         nCmd->sg_cnt, nCmd->payload_length);
1407                         return 1;
1408                 }
1409         }
1410         return 0;
1411 }
1412
1413 /**
1414  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1415  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1416  * @lpfc_nvme_lport: Pointer to the driver's local port data
1417  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1418  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1419  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1420  *
1421  * Driver registers this routine as it io request handler.  This
1422  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1423  * data structure to the rport
1424  indicated in @lpfc_nvme_rport.
1425  *
1426  * Return value :
1427  *   0 - Success
1428  *   TODO: What are the failure codes.
1429  **/
1430 static int
1431 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1432                         struct nvme_fc_remote_port *pnvme_rport,
1433                         void *hw_queue_handle,
1434                         struct nvmefc_fcp_req *pnvme_fcreq)
1435 {
1436         int ret = 0;
1437         int expedite = 0;
1438         int idx;
1439         struct lpfc_nvme_lport *lport;
1440         struct lpfc_nvme_ctrl_stat *cstat;
1441         struct lpfc_vport *vport;
1442         struct lpfc_hba *phba;
1443         struct lpfc_nodelist *ndlp;
1444         struct lpfc_nvme_buf *lpfc_ncmd;
1445         struct lpfc_nvme_rport *rport;
1446         struct lpfc_nvme_qhandle *lpfc_queue_info;
1447         struct lpfc_nvme_fcpreq_priv *freqpriv;
1448         struct nvme_common_command *sqe;
1449 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1450         uint64_t start = 0;
1451 #endif
1452
1453         /* Validate pointers. LLDD fault handling with transport does
1454          * have timing races.
1455          */
1456         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1457         if (unlikely(!lport)) {
1458                 ret = -EINVAL;
1459                 goto out_fail;
1460         }
1461
1462         vport = lport->vport;
1463
1464         if (unlikely(!hw_queue_handle)) {
1465                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1466                                  "6117 Fail IO, NULL hw_queue_handle\n");
1467                 atomic_inc(&lport->xmt_fcp_err);
1468                 ret = -EBUSY;
1469                 goto out_fail;
1470         }
1471
1472         phba = vport->phba;
1473
1474         if (vport->load_flag & FC_UNLOADING) {
1475                 ret = -ENODEV;
1476                 goto out_fail;
1477         }
1478
1479         if (vport->load_flag & FC_UNLOADING) {
1480                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1481                                  "6124 Fail IO, Driver unload\n");
1482                 atomic_inc(&lport->xmt_fcp_err);
1483                 ret = -ENODEV;
1484                 goto out_fail;
1485         }
1486
1487         freqpriv = pnvme_fcreq->private;
1488         if (unlikely(!freqpriv)) {
1489                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1490                                  "6158 Fail IO, NULL request data\n");
1491                 atomic_inc(&lport->xmt_fcp_err);
1492                 ret = -EINVAL;
1493                 goto out_fail;
1494         }
1495
1496 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1497         if (phba->ktime_on)
1498                 start = ktime_get_ns();
1499 #endif
1500         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1501         lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1502
1503         /*
1504          * Catch race where our node has transitioned, but the
1505          * transport is still transitioning.
1506          */
1507         ndlp = rport->ndlp;
1508         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1509                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1510                                  "6053 Fail IO, ndlp not ready: rport %p "
1511                                   "ndlp %p, DID x%06x\n",
1512                                  rport, ndlp, pnvme_rport->port_id);
1513                 atomic_inc(&lport->xmt_fcp_err);
1514                 ret = -EBUSY;
1515                 goto out_fail;
1516         }
1517
1518         /* The remote node has to be a mapped target or it's an error. */
1519         if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1520             (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1521                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1522                                  "6036 Fail IO, DID x%06x not ready for "
1523                                  "IO. State x%x, Type x%x Flg x%x\n",
1524                                  pnvme_rport->port_id,
1525                                  ndlp->nlp_state, ndlp->nlp_type,
1526                                  ndlp->upcall_flags);
1527                 atomic_inc(&lport->xmt_fcp_bad_ndlp);
1528                 ret = -EBUSY;
1529                 goto out_fail;
1530
1531         }
1532
1533         /* Currently only NVME Keep alive commands should be expedited
1534          * if the driver runs out of a resource. These should only be
1535          * issued on the admin queue, qidx 0
1536          */
1537         if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1538                 sqe = &((struct nvme_fc_cmd_iu *)
1539                         pnvme_fcreq->cmdaddr)->sqe.common;
1540                 if (sqe->opcode == nvme_admin_keep_alive)
1541                         expedite = 1;
1542         }
1543
1544         /* The node is shared with FCP IO, make sure the IO pending count does
1545          * not exceed the programmed depth.
1546          */
1547         if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1548                 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1549                     !expedite) {
1550                         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1551                                          "6174 Fail IO, ndlp qdepth exceeded: "
1552                                          "idx %d DID %x pend %d qdepth %d\n",
1553                                          lpfc_queue_info->index, ndlp->nlp_DID,
1554                                          atomic_read(&ndlp->cmd_pending),
1555                                          ndlp->cmd_qdepth);
1556                         atomic_inc(&lport->xmt_fcp_qdepth);
1557                         ret = -EBUSY;
1558                         goto out_fail;
1559                 }
1560         }
1561
1562         lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, expedite);
1563         if (lpfc_ncmd == NULL) {
1564                 atomic_inc(&lport->xmt_fcp_noxri);
1565                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1566                                  "6065 Fail IO, driver buffer pool is empty: "
1567                                  "idx %d DID %x\n",
1568                                  lpfc_queue_info->index, ndlp->nlp_DID);
1569                 ret = -EBUSY;
1570                 goto out_fail;
1571         }
1572 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1573         if (start) {
1574                 lpfc_ncmd->ts_cmd_start = start;
1575                 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1576         } else {
1577                 lpfc_ncmd->ts_cmd_start = 0;
1578         }
1579 #endif
1580
1581         /*
1582          * Store the data needed by the driver to issue, abort, and complete
1583          * an IO.
1584          * Do not let the IO hang out forever.  There is no midlayer issuing
1585          * an abort so inform the FW of the maximum IO pending time.
1586          */
1587         freqpriv->nvme_buf = lpfc_ncmd;
1588         lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1589         lpfc_ncmd->nrport = rport;
1590         lpfc_ncmd->ndlp = ndlp;
1591         lpfc_ncmd->start_time = jiffies;
1592
1593         /*
1594          * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1595          * This identfier was create in our hardware queue create callback
1596          * routine. The driver now is dependent on the IO queue steering from
1597          * the transport.  We are trusting the upper NVME layers know which
1598          * index to use and that they have affinitized a CPU to this hardware
1599          * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1600          */
1601         idx = lpfc_queue_info->index;
1602         lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1603         cstat = &lport->cstat[idx];
1604
1605         lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1606         ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1607         if (ret) {
1608                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1609                                  "6175 Fail IO, Prep DMA: "
1610                                  "idx %d DID %x\n",
1611                                  lpfc_queue_info->index, ndlp->nlp_DID);
1612                 atomic_inc(&lport->xmt_fcp_err);
1613                 ret = -ENOMEM;
1614                 goto out_free_nvme_buf;
1615         }
1616
1617         lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1618                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1619                          lpfc_queue_info->index, ndlp->nlp_DID);
1620
1621         ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq);
1622         if (ret) {
1623                 atomic_inc(&lport->xmt_fcp_wqerr);
1624                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1625                                  "6113 Fail IO, Could not issue WQE err %x "
1626                                  "sid: x%x did: x%x oxid: x%x\n",
1627                                  ret, vport->fc_myDID, ndlp->nlp_DID,
1628                                  lpfc_ncmd->cur_iocbq.sli4_xritag);
1629                 goto out_free_nvme_buf;
1630         }
1631
1632 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1633         if (lpfc_ncmd->ts_cmd_start)
1634                 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1635
1636         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1637                 lpfc_ncmd->cpu = smp_processor_id();
1638                 if (lpfc_ncmd->cpu != lpfc_queue_info->index) {
1639                         /* Check for admin queue */
1640                         if (lpfc_queue_info->qidx) {
1641                                 lpfc_printf_vlog(vport,
1642                                                  KERN_ERR, LOG_NVME_IOERR,
1643                                                 "6702 CPU Check cmd: "
1644                                                 "cpu %d wq %d\n",
1645                                                 lpfc_ncmd->cpu,
1646                                                 lpfc_queue_info->index);
1647                         }
1648                         lpfc_ncmd->cpu = lpfc_queue_info->index;
1649                 }
1650                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
1651                         phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++;
1652         }
1653 #endif
1654         return 0;
1655
1656  out_free_nvme_buf:
1657         if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1658                 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1659                         atomic_dec(&cstat->fc4NvmeOutputRequests);
1660                 else
1661                         atomic_dec(&cstat->fc4NvmeInputRequests);
1662         } else
1663                 atomic_dec(&cstat->fc4NvmeControlRequests);
1664         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1665  out_fail:
1666         return ret;
1667 }
1668
1669 /**
1670  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1671  * @phba: Pointer to HBA context object
1672  * @cmdiocb: Pointer to command iocb object.
1673  * @rspiocb: Pointer to response iocb object.
1674  *
1675  * This is the callback function for any NVME FCP IO that was aborted.
1676  *
1677  * Return value:
1678  *   None
1679  **/
1680 void
1681 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1682                            struct lpfc_wcqe_complete *abts_cmpl)
1683 {
1684         lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1685                         "6145 ABORT_XRI_CN completing on rpi x%x "
1686                         "original iotag x%x, abort cmd iotag x%x "
1687                         "req_tag x%x, status x%x, hwstatus x%x\n",
1688                         cmdiocb->iocb.un.acxri.abortContextTag,
1689                         cmdiocb->iocb.un.acxri.abortIoTag,
1690                         cmdiocb->iotag,
1691                         bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1692                         bf_get(lpfc_wcqe_c_status, abts_cmpl),
1693                         bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1694         lpfc_sli_release_iocbq(phba, cmdiocb);
1695 }
1696
1697 /**
1698  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1699  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1700  * @lpfc_nvme_lport: Pointer to the driver's local port data
1701  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1702  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1703  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1704  *
1705  * Driver registers this routine as its nvme request io abort handler.  This
1706  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1707  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1708  * is executed asynchronously - one the target is validated as "MAPPED" and
1709  * ready for IO, the driver issues the abort request and returns.
1710  *
1711  * Return value:
1712  *   None
1713  **/
1714 static void
1715 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1716                     struct nvme_fc_remote_port *pnvme_rport,
1717                     void *hw_queue_handle,
1718                     struct nvmefc_fcp_req *pnvme_fcreq)
1719 {
1720         struct lpfc_nvme_lport *lport;
1721         struct lpfc_vport *vport;
1722         struct lpfc_hba *phba;
1723         struct lpfc_nvme_buf *lpfc_nbuf;
1724         struct lpfc_iocbq *abts_buf;
1725         struct lpfc_iocbq *nvmereq_wqe;
1726         struct lpfc_nvme_fcpreq_priv *freqpriv;
1727         union lpfc_wqe128 *abts_wqe;
1728         unsigned long flags;
1729         int ret_val;
1730
1731         /* Validate pointers. LLDD fault handling with transport does
1732          * have timing races.
1733          */
1734         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1735         if (unlikely(!lport))
1736                 return;
1737
1738         vport = lport->vport;
1739
1740         if (unlikely(!hw_queue_handle)) {
1741                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1742                                  "6129 Fail Abort, HW Queue Handle NULL.\n");
1743                 return;
1744         }
1745
1746         phba = vport->phba;
1747         freqpriv = pnvme_fcreq->private;
1748
1749         if (unlikely(!freqpriv))
1750                 return;
1751         if (vport->load_flag & FC_UNLOADING)
1752                 return;
1753
1754         /* Announce entry to new IO submit field. */
1755         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1756                          "6002 Abort Request to rport DID x%06x "
1757                          "for nvme_fc_req %p\n",
1758                          pnvme_rport->port_id,
1759                          pnvme_fcreq);
1760
1761         /* If the hba is getting reset, this flag is set.  It is
1762          * cleared when the reset is complete and rings reestablished.
1763          */
1764         spin_lock_irqsave(&phba->hbalock, flags);
1765         /* driver queued commands are in process of being flushed */
1766         if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) {
1767                 spin_unlock_irqrestore(&phba->hbalock, flags);
1768                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1769                                  "6139 Driver in reset cleanup - flushing "
1770                                  "NVME Req now.  hba_flag x%x\n",
1771                                  phba->hba_flag);
1772                 return;
1773         }
1774
1775         lpfc_nbuf = freqpriv->nvme_buf;
1776         if (!lpfc_nbuf) {
1777                 spin_unlock_irqrestore(&phba->hbalock, flags);
1778                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1779                                  "6140 NVME IO req has no matching lpfc nvme "
1780                                  "io buffer.  Skipping abort req.\n");
1781                 return;
1782         } else if (!lpfc_nbuf->nvmeCmd) {
1783                 spin_unlock_irqrestore(&phba->hbalock, flags);
1784                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1785                                  "6141 lpfc NVME IO req has no nvme_fcreq "
1786                                  "io buffer.  Skipping abort req.\n");
1787                 return;
1788         }
1789         nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1790
1791         /*
1792          * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1793          * state must match the nvme_fcreq passed by the nvme
1794          * transport.  If they don't match, it is likely the driver
1795          * has already completed the NVME IO and the nvme transport
1796          * has not seen it yet.
1797          */
1798         if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1799                 spin_unlock_irqrestore(&phba->hbalock, flags);
1800                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1801                                  "6143 NVME req mismatch: "
1802                                  "lpfc_nbuf %p nvmeCmd %p, "
1803                                  "pnvme_fcreq %p.  Skipping Abort xri x%x\n",
1804                                  lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1805                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1806                 return;
1807         }
1808
1809         /* Don't abort IOs no longer on the pending queue. */
1810         if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
1811                 spin_unlock_irqrestore(&phba->hbalock, flags);
1812                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1813                                  "6142 NVME IO req %p not queued - skipping "
1814                                  "abort req xri x%x\n",
1815                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1816                 return;
1817         }
1818
1819         atomic_inc(&lport->xmt_fcp_abort);
1820         lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1821                          nvmereq_wqe->sli4_xritag,
1822                          nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1823
1824         /* Outstanding abort is in progress */
1825         if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
1826                 spin_unlock_irqrestore(&phba->hbalock, flags);
1827                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1828                                  "6144 Outstanding NVME I/O Abort Request "
1829                                  "still pending on nvme_fcreq %p, "
1830                                  "lpfc_ncmd %p xri x%x\n",
1831                                  pnvme_fcreq, lpfc_nbuf,
1832                                  nvmereq_wqe->sli4_xritag);
1833                 return;
1834         }
1835
1836         abts_buf = __lpfc_sli_get_iocbq(phba);
1837         if (!abts_buf) {
1838                 spin_unlock_irqrestore(&phba->hbalock, flags);
1839                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1840                                  "6136 No available abort wqes. Skipping "
1841                                  "Abts req for nvme_fcreq %p xri x%x\n",
1842                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1843                 return;
1844         }
1845
1846         /* Ready - mark outstanding as aborted by driver. */
1847         nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
1848
1849         /* Complete prepping the abort wqe and issue to the FW. */
1850         abts_wqe = &abts_buf->wqe;
1851
1852         /* WQEs are reused.  Clear stale data and set key fields to
1853          * zero like ia, iaab, iaar, xri_tag, and ctxt_tag.
1854          */
1855         memset(abts_wqe, 0, sizeof(union lpfc_wqe));
1856         bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
1857
1858         /* word 7 */
1859         bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
1860         bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
1861                nvmereq_wqe->iocb.ulpClass);
1862
1863         /* word 8 - tell the FW to abort the IO associated with this
1864          * outstanding exchange ID.
1865          */
1866         abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag;
1867
1868         /* word 9 - this is the iotag for the abts_wqe completion. */
1869         bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
1870                abts_buf->iotag);
1871
1872         /* word 10 */
1873         bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
1874         bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
1875
1876         /* word 11 */
1877         bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
1878         bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
1879         bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
1880
1881         /* ABTS WQE must go to the same WQ as the WQE to be aborted */
1882         abts_buf->iocb_flag |= LPFC_IO_NVME;
1883         abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx;
1884         abts_buf->vport = vport;
1885         abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
1886         ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf);
1887         spin_unlock_irqrestore(&phba->hbalock, flags);
1888         if (ret_val) {
1889                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1890                                  "6137 Failed abts issue_wqe with status x%x "
1891                                  "for nvme_fcreq %p.\n",
1892                                  ret_val, pnvme_fcreq);
1893                 lpfc_sli_release_iocbq(phba, abts_buf);
1894                 return;
1895         }
1896
1897         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1898                          "6138 Transport Abort NVME Request Issued for "
1899                          "ox_id x%x on reqtag x%x\n",
1900                          nvmereq_wqe->sli4_xritag,
1901                          abts_buf->iotag);
1902 }
1903
1904 /* Declare and initialization an instance of the FC NVME template. */
1905 static struct nvme_fc_port_template lpfc_nvme_template = {
1906         /* initiator-based functions */
1907         .localport_delete  = lpfc_nvme_localport_delete,
1908         .remoteport_delete = lpfc_nvme_remoteport_delete,
1909         .create_queue = lpfc_nvme_create_queue,
1910         .delete_queue = lpfc_nvme_delete_queue,
1911         .ls_req       = lpfc_nvme_ls_req,
1912         .fcp_io       = lpfc_nvme_fcp_io_submit,
1913         .ls_abort     = lpfc_nvme_ls_abort,
1914         .fcp_abort    = lpfc_nvme_fcp_abort,
1915
1916         .max_hw_queues = 1,
1917         .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1918         .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1919         .dma_boundary = 0xFFFFFFFF,
1920
1921         /* Sizes of additional private data for data structures.
1922          * No use for the last two sizes at this time.
1923          */
1924         .local_priv_sz = sizeof(struct lpfc_nvme_lport),
1925         .remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1926         .lsrqst_priv_sz = 0,
1927         .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1928 };
1929
1930 /**
1931  * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware
1932  * @phba: pointer to lpfc hba data structure.
1933  * @nblist: pointer to nvme buffer list.
1934  * @count: number of scsi buffers on the list.
1935  *
1936  * This routine is invoked to post a block of @count scsi sgl pages from a
1937  * SCSI buffer list @nblist to the HBA using non-embedded mailbox command.
1938  * No Lock is held.
1939  *
1940  **/
1941 static int
1942 lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba,
1943                               struct list_head *nblist,
1944                               int count)
1945 {
1946         struct lpfc_nvme_buf *lpfc_ncmd;
1947         struct lpfc_mbx_post_uembed_sgl_page1 *sgl;
1948         struct sgl_page_pairs *sgl_pg_pairs;
1949         void *viraddr;
1950         LPFC_MBOXQ_t *mbox;
1951         uint32_t reqlen, alloclen, pg_pairs;
1952         uint32_t mbox_tmo;
1953         uint16_t xritag_start = 0;
1954         int rc = 0;
1955         uint32_t shdr_status, shdr_add_status;
1956         dma_addr_t pdma_phys_bpl1;
1957         union lpfc_sli4_cfg_shdr *shdr;
1958
1959         /* Calculate the requested length of the dma memory */
1960         reqlen = count * sizeof(struct sgl_page_pairs) +
1961                  sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t);
1962         if (reqlen > SLI4_PAGE_SIZE) {
1963                 lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
1964                                 "6118 Block sgl registration required DMA "
1965                                 "size (%d) great than a page\n", reqlen);
1966                 return -ENOMEM;
1967         }
1968         mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
1969         if (!mbox) {
1970                 lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
1971                                 "6119 Failed to allocate mbox cmd memory\n");
1972                 return -ENOMEM;
1973         }
1974
1975         /* Allocate DMA memory and set up the non-embedded mailbox command */
1976         alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
1977                                 LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen,
1978                                 LPFC_SLI4_MBX_NEMBED);
1979
1980         if (alloclen < reqlen) {
1981                 lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
1982                                 "6120 Allocated DMA memory size (%d) is "
1983                                 "less than the requested DMA memory "
1984                                 "size (%d)\n", alloclen, reqlen);
1985                 lpfc_sli4_mbox_cmd_free(phba, mbox);
1986                 return -ENOMEM;
1987         }
1988
1989         /* Get the first SGE entry from the non-embedded DMA memory */
1990         viraddr = mbox->sge_array->addr[0];
1991
1992         /* Set up the SGL pages in the non-embedded DMA pages */
1993         sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr;
1994         sgl_pg_pairs = &sgl->sgl_pg_pairs;
1995
1996         pg_pairs = 0;
1997         list_for_each_entry(lpfc_ncmd, nblist, list) {
1998                 /* Set up the sge entry */
1999                 sgl_pg_pairs->sgl_pg0_addr_lo =
2000                         cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl));
2001                 sgl_pg_pairs->sgl_pg0_addr_hi =
2002                         cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl));
2003                 if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
2004                         pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl +
2005                                                 SGL_PAGE_SIZE;
2006                 else
2007                         pdma_phys_bpl1 = 0;
2008                 sgl_pg_pairs->sgl_pg1_addr_lo =
2009                         cpu_to_le32(putPaddrLow(pdma_phys_bpl1));
2010                 sgl_pg_pairs->sgl_pg1_addr_hi =
2011                         cpu_to_le32(putPaddrHigh(pdma_phys_bpl1));
2012                 /* Keep the first xritag on the list */
2013                 if (pg_pairs == 0)
2014                         xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag;
2015                 sgl_pg_pairs++;
2016                 pg_pairs++;
2017         }
2018         bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start);
2019         bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs);
2020         /* Perform endian conversion if necessary */
2021         sgl->word0 = cpu_to_le32(sgl->word0);
2022
2023         if (!phba->sli4_hba.intr_enable)
2024                 rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
2025         else {
2026                 mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
2027                 rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
2028         }
2029         shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr;
2030         shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
2031         shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
2032         if (rc != MBX_TIMEOUT)
2033                 lpfc_sli4_mbox_cmd_free(phba, mbox);
2034         if (shdr_status || shdr_add_status || rc) {
2035                 lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
2036                                 "6125 POST_SGL_BLOCK mailbox command failed "
2037                                 "status x%x add_status x%x mbx status x%x\n",
2038                                 shdr_status, shdr_add_status, rc);
2039                 rc = -ENXIO;
2040         }
2041         return rc;
2042 }
2043
2044 /**
2045  * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list
2046  * @phba: pointer to lpfc hba data structure.
2047  * @post_nblist: pointer to the nvme buffer list.
2048  *
2049  * This routine walks a list of nvme buffers that was passed in. It attempts
2050  * to construct blocks of nvme buffer sgls which contains contiguous xris and
2051  * uses the non-embedded SGL block post mailbox commands to post to the port.
2052  * For single NVME buffer sgl with non-contiguous xri, if any, it shall use
2053  * embedded SGL post mailbox command for posting. The @post_nblist passed in
2054  * must be local list, thus no lock is needed when manipulate the list.
2055  *
2056  * Returns: 0 = failure, non-zero number of successfully posted buffers.
2057  **/
2058 static int
2059 lpfc_post_nvme_sgl_list(struct lpfc_hba *phba,
2060                              struct list_head *post_nblist, int sb_count)
2061 {
2062         struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
2063         int status, sgl_size;
2064         int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0;
2065         dma_addr_t pdma_phys_sgl1;
2066         int last_xritag = NO_XRI;
2067         int cur_xritag;
2068         LIST_HEAD(prep_nblist);
2069         LIST_HEAD(blck_nblist);
2070         LIST_HEAD(nvme_nblist);
2071
2072         /* sanity check */
2073         if (sb_count <= 0)
2074                 return -EINVAL;
2075
2076         sgl_size = phba->cfg_sg_dma_buf_size;
2077
2078         list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) {
2079                 list_del_init(&lpfc_ncmd->list);
2080                 block_cnt++;
2081                 if ((last_xritag != NO_XRI) &&
2082                     (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) {
2083                         /* a hole in xri block, form a sgl posting block */
2084                         list_splice_init(&prep_nblist, &blck_nblist);
2085                         post_cnt = block_cnt - 1;
2086                         /* prepare list for next posting block */
2087                         list_add_tail(&lpfc_ncmd->list, &prep_nblist);
2088                         block_cnt = 1;
2089                 } else {
2090                         /* prepare list for next posting block */
2091                         list_add_tail(&lpfc_ncmd->list, &prep_nblist);
2092                         /* enough sgls for non-embed sgl mbox command */
2093                         if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) {
2094                                 list_splice_init(&prep_nblist, &blck_nblist);
2095                                 post_cnt = block_cnt;
2096                                 block_cnt = 0;
2097                         }
2098                 }
2099                 num_posting++;
2100                 last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
2101
2102                 /* end of repost sgl list condition for NVME buffers */
2103                 if (num_posting == sb_count) {
2104                         if (post_cnt == 0) {
2105                                 /* last sgl posting block */
2106                                 list_splice_init(&prep_nblist, &blck_nblist);
2107                                 post_cnt = block_cnt;
2108                         } else if (block_cnt == 1) {
2109                                 /* last single sgl with non-contiguous xri */
2110                                 if (sgl_size > SGL_PAGE_SIZE)
2111                                         pdma_phys_sgl1 =
2112                                                 lpfc_ncmd->dma_phys_sgl +
2113                                                 SGL_PAGE_SIZE;
2114                                 else
2115                                         pdma_phys_sgl1 = 0;
2116                                 cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
2117                                 status = lpfc_sli4_post_sgl(phba,
2118                                                 lpfc_ncmd->dma_phys_sgl,
2119                                                 pdma_phys_sgl1, cur_xritag);
2120                                 if (status) {
2121                                         /* failure, put on abort nvme list */
2122                                         lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
2123                                 } else {
2124                                         /* success, put on NVME buffer list */
2125                                         lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2126                                         lpfc_ncmd->status = IOSTAT_SUCCESS;
2127                                         num_posted++;
2128                                 }
2129                                 /* success, put on NVME buffer sgl list */
2130                                 list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
2131                         }
2132                 }
2133
2134                 /* continue until a nembed page worth of sgls */
2135                 if (post_cnt == 0)
2136                         continue;
2137
2138                 /* post block of NVME buffer list sgls */
2139                 status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist,
2140                                                        post_cnt);
2141
2142                 /* don't reset xirtag due to hole in xri block */
2143                 if (block_cnt == 0)
2144                         last_xritag = NO_XRI;
2145
2146                 /* reset NVME buffer post count for next round of posting */
2147                 post_cnt = 0;
2148
2149                 /* put posted NVME buffer-sgl posted on NVME buffer sgl list */
2150                 while (!list_empty(&blck_nblist)) {
2151                         list_remove_head(&blck_nblist, lpfc_ncmd,
2152                                          struct lpfc_nvme_buf, list);
2153                         if (status) {
2154                                 /* failure, put on abort nvme list */
2155                                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
2156                         } else {
2157                                 /* success, put on NVME buffer list */
2158                                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2159                                 lpfc_ncmd->status = IOSTAT_SUCCESS;
2160                                 num_posted++;
2161                         }
2162                         list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
2163                 }
2164         }
2165         /* Push NVME buffers with sgl posted to the available list */
2166         while (!list_empty(&nvme_nblist)) {
2167                 list_remove_head(&nvme_nblist, lpfc_ncmd,
2168                                  struct lpfc_nvme_buf, list);
2169                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
2170         }
2171         return num_posted;
2172 }
2173
2174 /**
2175  * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls
2176  * @phba: pointer to lpfc hba data structure.
2177  *
2178  * This routine walks the list of nvme buffers that have been allocated and
2179  * repost them to the port by using SGL block post. This is needed after a
2180  * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine
2181  * is responsible for moving all nvme buffers on the lpfc_abts_nvme_sgl_list
2182  * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers.
2183  *
2184  * Returns: 0 = success, non-zero failure.
2185  **/
2186 int
2187 lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba)
2188 {
2189         LIST_HEAD(post_nblist);
2190         int num_posted, rc = 0;
2191
2192         /* get all NVME buffers need to repost to a local list */
2193         spin_lock_irq(&phba->nvme_buf_list_get_lock);
2194         spin_lock(&phba->nvme_buf_list_put_lock);
2195         list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist);
2196         list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist);
2197         phba->get_nvme_bufs = 0;
2198         phba->put_nvme_bufs = 0;
2199         spin_unlock(&phba->nvme_buf_list_put_lock);
2200         spin_unlock_irq(&phba->nvme_buf_list_get_lock);
2201
2202         /* post the list of nvme buffer sgls to port if available */
2203         if (!list_empty(&post_nblist)) {
2204                 num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist,
2205                                                 phba->sli4_hba.nvme_xri_cnt);
2206                 /* failed to post any nvme buffer, return error */
2207                 if (num_posted == 0)
2208                         rc = -EIO;
2209         }
2210         return rc;
2211 }
2212
2213 /**
2214  * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec
2215  * @vport: The virtual port for which this call being executed.
2216  * @num_to_allocate: The requested number of buffers to allocate.
2217  *
2218  * This routine allocates nvme buffers for device with SLI-4 interface spec,
2219  * the nvme buffer contains all the necessary information needed to initiate
2220  * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put
2221  * them on a list, it post them to the port by using SGL block post.
2222  *
2223  * Return codes:
2224  *   int - number of nvme buffers that were allocated and posted.
2225  *   0 = failure, less than num_to_alloc is a partial failure.
2226  **/
2227 static int
2228 lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc)
2229 {
2230         struct lpfc_hba *phba = vport->phba;
2231         struct lpfc_nvme_buf *lpfc_ncmd;
2232         struct lpfc_iocbq *pwqeq;
2233         union lpfc_wqe128 *wqe;
2234         struct sli4_sge *sgl;
2235         dma_addr_t pdma_phys_sgl;
2236         uint16_t iotag, lxri = 0;
2237         int bcnt, num_posted;
2238         LIST_HEAD(prep_nblist);
2239         LIST_HEAD(post_nblist);
2240         LIST_HEAD(nvme_nblist);
2241
2242         for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
2243                 lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL);
2244                 if (!lpfc_ncmd)
2245                         break;
2246                 /*
2247                  * Get memory from the pci pool to map the virt space to
2248                  * pci bus space for an I/O. The DMA buffer includes the
2249                  * number of SGE's necessary to support the sg_tablesize.
2250                  */
2251                 lpfc_ncmd->data = dma_pool_zalloc(phba->lpfc_sg_dma_buf_pool,
2252                                                   GFP_KERNEL,
2253                                                   &lpfc_ncmd->dma_handle);
2254                 if (!lpfc_ncmd->data) {
2255                         kfree(lpfc_ncmd);
2256                         break;
2257                 }
2258
2259                 lxri = lpfc_sli4_next_xritag(phba);
2260                 if (lxri == NO_XRI) {
2261                         dma_pool_free(phba->lpfc_sg_dma_buf_pool,
2262                                       lpfc_ncmd->data, lpfc_ncmd->dma_handle);
2263                         kfree(lpfc_ncmd);
2264                         break;
2265                 }
2266                 pwqeq = &(lpfc_ncmd->cur_iocbq);
2267                 wqe = &pwqeq->wqe;
2268
2269                 /* Allocate iotag for lpfc_ncmd->cur_iocbq. */
2270                 iotag = lpfc_sli_next_iotag(phba, pwqeq);
2271                 if (iotag == 0) {
2272                         dma_pool_free(phba->lpfc_sg_dma_buf_pool,
2273                                       lpfc_ncmd->data, lpfc_ncmd->dma_handle);
2274                         kfree(lpfc_ncmd);
2275                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
2276                                         "6121 Failed to allocated IOTAG for"
2277                                         " XRI:0x%x\n", lxri);
2278                         lpfc_sli4_free_xri(phba, lxri);
2279                         break;
2280                 }
2281                 pwqeq->sli4_lxritag = lxri;
2282                 pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri];
2283                 pwqeq->iocb_flag |= LPFC_IO_NVME;
2284                 pwqeq->context1 = lpfc_ncmd;
2285                 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
2286
2287                 /* Initialize local short-hand pointers. */
2288                 lpfc_ncmd->nvme_sgl = lpfc_ncmd->data;
2289                 sgl = lpfc_ncmd->nvme_sgl;
2290                 pdma_phys_sgl = lpfc_ncmd->dma_handle;
2291                 lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl;
2292
2293                 /* Rsp SGE will be filled in when we rcv an IO
2294                  * from the NVME Layer to be sent.
2295                  * The cmd is going to be embedded so we need a SKIP SGE.
2296                  */
2297                 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
2298                 bf_set(lpfc_sli4_sge_last, sgl, 0);
2299                 sgl->word2 = cpu_to_le32(sgl->word2);
2300                 /* Fill in word 3 / sgl_len during cmd submission */
2301
2302                 lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd;
2303
2304                 /* Initialize WQE */
2305                 memset(wqe, 0, sizeof(union lpfc_wqe));
2306
2307                 /* add the nvme buffer to a post list */
2308                 list_add_tail(&lpfc_ncmd->list, &post_nblist);
2309                 spin_lock_irq(&phba->nvme_buf_list_get_lock);
2310                 phba->sli4_hba.nvme_xri_cnt++;
2311                 spin_unlock_irq(&phba->nvme_buf_list_get_lock);
2312         }
2313         lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
2314                         "6114 Allocate %d out of %d requested new NVME "
2315                         "buffers\n", bcnt, num_to_alloc);
2316
2317         /* post the list of nvme buffer sgls to port if available */
2318         if (!list_empty(&post_nblist))
2319                 num_posted = lpfc_post_nvme_sgl_list(phba,
2320                                                      &post_nblist, bcnt);
2321         else
2322                 num_posted = 0;
2323
2324         return num_posted;
2325 }
2326
2327 static inline struct lpfc_nvme_buf *
2328 lpfc_nvme_buf(struct lpfc_hba *phba)
2329 {
2330         struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
2331
2332         list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
2333                                  &phba->lpfc_nvme_buf_list_get, list) {
2334                 list_del_init(&lpfc_ncmd->list);
2335                 phba->get_nvme_bufs--;
2336                 return lpfc_ncmd;
2337         }
2338         return NULL;
2339 }
2340
2341 /**
2342  * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA
2343  * @phba: The HBA for which this call is being executed.
2344  *
2345  * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list
2346  * and returns to caller.
2347  *
2348  * Return codes:
2349  *   NULL - Error
2350  *   Pointer to lpfc_nvme_buf - Success
2351  **/
2352 static struct lpfc_nvme_buf *
2353 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
2354                   int expedite)
2355 {
2356         struct lpfc_nvme_buf *lpfc_ncmd = NULL;
2357         unsigned long iflag = 0;
2358
2359         spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag);
2360         if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite)
2361                 lpfc_ncmd = lpfc_nvme_buf(phba);
2362         if (!lpfc_ncmd) {
2363                 spin_lock(&phba->nvme_buf_list_put_lock);
2364                 list_splice(&phba->lpfc_nvme_buf_list_put,
2365                             &phba->lpfc_nvme_buf_list_get);
2366                 phba->get_nvme_bufs += phba->put_nvme_bufs;
2367                 INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
2368                 phba->put_nvme_bufs = 0;
2369                 spin_unlock(&phba->nvme_buf_list_put_lock);
2370                 if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite)
2371                         lpfc_ncmd = lpfc_nvme_buf(phba);
2372         }
2373         spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag);
2374
2375         if (lpfc_ndlp_check_qdepth(phba, ndlp) && lpfc_ncmd) {
2376                 atomic_inc(&ndlp->cmd_pending);
2377                 lpfc_ncmd->flags |= LPFC_BUMP_QDEPTH;
2378         }
2379         return  lpfc_ncmd;
2380 }
2381
2382 /**
2383  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2384  * @phba: The Hba for which this call is being executed.
2385  * @lpfc_ncmd: The nvme buffer which is being released.
2386  *
2387  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2388  * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2389  * and cannot be reused for at least RA_TOV amount of time if it was
2390  * aborted.
2391  **/
2392 static void
2393 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd)
2394 {
2395         unsigned long iflag = 0;
2396
2397         if ((lpfc_ncmd->flags & LPFC_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2398                 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2399
2400         lpfc_ncmd->nonsg_phys = 0;
2401         lpfc_ncmd->ndlp = NULL;
2402         lpfc_ncmd->flags &= ~LPFC_BUMP_QDEPTH;
2403
2404         if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) {
2405                 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2406                                 "6310 XB release deferred for "
2407                                 "ox_id x%x on reqtag x%x\n",
2408                                 lpfc_ncmd->cur_iocbq.sli4_xritag,
2409                                 lpfc_ncmd->cur_iocbq.iotag);
2410
2411                 spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock,
2412                                         iflag);
2413                 list_add_tail(&lpfc_ncmd->list,
2414                         &phba->sli4_hba.lpfc_abts_nvme_buf_list);
2415                 spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock,
2416                                         iflag);
2417         } else {
2418                 lpfc_ncmd->nvmeCmd = NULL;
2419                 lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME;
2420                 spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag);
2421                 list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put);
2422                 phba->put_nvme_bufs++;
2423                 spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag);
2424         }
2425 }
2426
2427 /**
2428  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2429  * @pvport - the lpfc_vport instance requesting a localport.
2430  *
2431  * This routine is invoked to create an nvme localport instance to bind
2432  * to the nvme_fc_transport.  It is called once during driver load
2433  * like lpfc_create_shost after all other services are initialized.
2434  * It requires a vport, vpi, and wwns at call time.  Other localport
2435  * parameters are modified as the driver's FCID and the Fabric WWN
2436  * are established.
2437  *
2438  * Return codes
2439  *      0 - successful
2440  *      -ENOMEM - no heap memory available
2441  *      other values - from nvme registration upcall
2442  **/
2443 int
2444 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2445 {
2446         int ret = 0;
2447         struct lpfc_hba  *phba = vport->phba;
2448         struct nvme_fc_port_info nfcp_info;
2449         struct nvme_fc_local_port *localport;
2450         struct lpfc_nvme_lport *lport;
2451         struct lpfc_nvme_ctrl_stat *cstat;
2452         int len, i;
2453
2454         /* Initialize this localport instance.  The vport wwn usage ensures
2455          * that NPIV is accounted for.
2456          */
2457         memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2458         nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2459         nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2460         nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2461
2462         /* We need to tell the transport layer + 1 because it takes page
2463          * alignment into account. When space for the SGL is allocated we
2464          * allocate + 3, one for cmd, one for rsp and one for this alignment
2465          */
2466         lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2467         lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_io_channel;
2468
2469         cstat = kmalloc((sizeof(struct lpfc_nvme_ctrl_stat) *
2470                         phba->cfg_nvme_io_channel), GFP_KERNEL);
2471         if (!cstat)
2472                 return -ENOMEM;
2473
2474         /* localport is allocated from the stack, but the registration
2475          * call allocates heap memory as well as the private area.
2476          */
2477 #if (IS_ENABLED(CONFIG_NVME_FC))
2478         ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2479                                          &vport->phba->pcidev->dev, &localport);
2480 #else
2481         ret = -ENOMEM;
2482 #endif
2483         if (!ret) {
2484                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2485                                  "6005 Successfully registered local "
2486                                  "NVME port num %d, localP %p, private %p, "
2487                                  "sg_seg %d\n",
2488                                  localport->port_num, localport,
2489                                  localport->private,
2490                                  lpfc_nvme_template.max_sgl_segments);
2491
2492                 /* Private is our lport size declared in the template. */
2493                 lport = (struct lpfc_nvme_lport *)localport->private;
2494                 vport->localport = localport;
2495                 lport->vport = vport;
2496                 lport->cstat = cstat;
2497                 vport->nvmei_support = 1;
2498
2499                 atomic_set(&lport->xmt_fcp_noxri, 0);
2500                 atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2501                 atomic_set(&lport->xmt_fcp_qdepth, 0);
2502                 atomic_set(&lport->xmt_fcp_err, 0);
2503                 atomic_set(&lport->xmt_fcp_wqerr, 0);
2504                 atomic_set(&lport->xmt_fcp_abort, 0);
2505                 atomic_set(&lport->xmt_ls_abort, 0);
2506                 atomic_set(&lport->xmt_ls_err, 0);
2507                 atomic_set(&lport->cmpl_fcp_xb, 0);
2508                 atomic_set(&lport->cmpl_fcp_err, 0);
2509                 atomic_set(&lport->cmpl_ls_xb, 0);
2510                 atomic_set(&lport->cmpl_ls_err, 0);
2511                 atomic_set(&lport->fc4NvmeLsRequests, 0);
2512                 atomic_set(&lport->fc4NvmeLsCmpls, 0);
2513
2514                 for (i = 0; i < phba->cfg_nvme_io_channel; i++) {
2515                         cstat = &lport->cstat[i];
2516                         atomic_set(&cstat->fc4NvmeInputRequests, 0);
2517                         atomic_set(&cstat->fc4NvmeOutputRequests, 0);
2518                         atomic_set(&cstat->fc4NvmeControlRequests, 0);
2519                         atomic_set(&cstat->fc4NvmeIoCmpls, 0);
2520                 }
2521
2522                 /* Don't post more new bufs if repost already recovered
2523                  * the nvme sgls.
2524                  */
2525                 if (phba->sli4_hba.nvme_xri_cnt == 0) {
2526                         len  = lpfc_new_nvme_buf(vport,
2527                                                  phba->sli4_hba.nvme_xri_max);
2528                         vport->phba->total_nvme_bufs += len;
2529                 }
2530         } else {
2531                 kfree(cstat);
2532         }
2533
2534         return ret;
2535 }
2536
2537 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2538  *
2539  * The driver has to wait for the host nvme transport to callback
2540  * indicating the localport has successfully unregistered all
2541  * resources.  Since this is an uninterruptible wait, loop every ten
2542  * seconds and print a message indicating no progress.
2543  *
2544  * An uninterruptible wait is used because of the risk of transport-to-
2545  * driver state mismatch.
2546  */
2547 void
2548 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2549                            struct lpfc_nvme_lport *lport,
2550                            struct completion *lport_unreg_cmp)
2551 {
2552 #if (IS_ENABLED(CONFIG_NVME_FC))
2553         u32 wait_tmo;
2554         int ret;
2555
2556         /* Host transport has to clean up and confirm requiring an indefinite
2557          * wait. Print a message if a 10 second wait expires and renew the
2558          * wait. This is unexpected.
2559          */
2560         wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2561         while (true) {
2562                 ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2563                 if (unlikely(!ret)) {
2564                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
2565                                          "6176 Lport %p Localport %p wait "
2566                                          "timed out. Renewing.\n",
2567                                          lport, vport->localport);
2568                         continue;
2569                 }
2570                 break;
2571         }
2572         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2573                          "6177 Lport %p Localport %p Complete Success\n",
2574                          lport, vport->localport);
2575 #endif
2576 }
2577
2578 /**
2579  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2580  * @pnvme: pointer to lpfc nvme data structure.
2581  *
2582  * This routine is invoked to destroy all lports bound to the phba.
2583  * The lport memory was allocated by the nvme fc transport and is
2584  * released there.  This routine ensures all rports bound to the
2585  * lport have been disconnected.
2586  *
2587  **/
2588 void
2589 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2590 {
2591 #if (IS_ENABLED(CONFIG_NVME_FC))
2592         struct nvme_fc_local_port *localport;
2593         struct lpfc_nvme_lport *lport;
2594         struct lpfc_nvme_ctrl_stat *cstat;
2595         int ret;
2596         DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2597
2598         if (vport->nvmei_support == 0)
2599                 return;
2600
2601         localport = vport->localport;
2602         lport = (struct lpfc_nvme_lport *)localport->private;
2603         cstat = lport->cstat;
2604
2605         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2606                          "6011 Destroying NVME localport %p\n",
2607                          localport);
2608
2609         /* lport's rport list is clear.  Unregister
2610          * lport and release resources.
2611          */
2612         lport->lport_unreg_cmp = &lport_unreg_cmp;
2613         ret = nvme_fc_unregister_localport(localport);
2614
2615         /* Wait for completion.  This either blocks
2616          * indefinitely or succeeds
2617          */
2618         lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2619         vport->localport = NULL;
2620         kfree(cstat);
2621
2622         /* Regardless of the unregister upcall response, clear
2623          * nvmei_support.  All rports are unregistered and the
2624          * driver will clean up.
2625          */
2626         vport->nvmei_support = 0;
2627         if (ret == 0) {
2628                 lpfc_printf_vlog(vport,
2629                                  KERN_INFO, LOG_NVME_DISC,
2630                                  "6009 Unregistered lport Success\n");
2631         } else {
2632                 lpfc_printf_vlog(vport,
2633                                  KERN_INFO, LOG_NVME_DISC,
2634                                  "6010 Unregistered lport "
2635                                  "Failed, status x%x\n",
2636                                  ret);
2637         }
2638 #endif
2639 }
2640
2641 void
2642 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2643 {
2644 #if (IS_ENABLED(CONFIG_NVME_FC))
2645         struct nvme_fc_local_port *localport;
2646         struct lpfc_nvme_lport *lport;
2647
2648         localport = vport->localport;
2649         if (!localport) {
2650                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2651                                  "6710 Update NVME fail. No localport\n");
2652                 return;
2653         }
2654         lport = (struct lpfc_nvme_lport *)localport->private;
2655         if (!lport) {
2656                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2657                                  "6171 Update NVME fail. localP %p, No lport\n",
2658                                  localport);
2659                 return;
2660         }
2661         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2662                          "6012 Update NVME lport %p did x%x\n",
2663                          localport, vport->fc_myDID);
2664
2665         localport->port_id = vport->fc_myDID;
2666         if (localport->port_id == 0)
2667                 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2668         else
2669                 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2670
2671         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2672                          "6030 bound lport %p to DID x%06x\n",
2673                          lport, localport->port_id);
2674 #endif
2675 }
2676
2677 int
2678 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2679 {
2680 #if (IS_ENABLED(CONFIG_NVME_FC))
2681         int ret = 0;
2682         struct nvme_fc_local_port *localport;
2683         struct lpfc_nvme_lport *lport;
2684         struct lpfc_nvme_rport *rport;
2685         struct lpfc_nvme_rport *oldrport;
2686         struct nvme_fc_remote_port *remote_port;
2687         struct nvme_fc_port_info rpinfo;
2688         struct lpfc_nodelist *prev_ndlp = NULL;
2689
2690         lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2691                          "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2692                          ndlp->nlp_DID, ndlp->nlp_type);
2693
2694         localport = vport->localport;
2695         if (!localport)
2696                 return 0;
2697
2698         lport = (struct lpfc_nvme_lport *)localport->private;
2699
2700         /* NVME rports are not preserved across devloss.
2701          * Just register this instance.  Note, rpinfo->dev_loss_tmo
2702          * is left 0 to indicate accept transport defaults.  The
2703          * driver communicates port role capabilities consistent
2704          * with the PRLI response data.
2705          */
2706         memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2707         rpinfo.port_id = ndlp->nlp_DID;
2708         if (ndlp->nlp_type & NLP_NVME_TARGET)
2709                 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2710         if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2711                 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2712
2713         if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2714                 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2715
2716         rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2717         rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2718
2719         spin_lock_irq(&vport->phba->hbalock);
2720         oldrport = lpfc_ndlp_get_nrport(ndlp);
2721         spin_unlock_irq(&vport->phba->hbalock);
2722         if (!oldrport)
2723                 lpfc_nlp_get(ndlp);
2724
2725         ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2726         if (!ret) {
2727                 /* If the ndlp already has an nrport, this is just
2728                  * a resume of the existing rport.  Else this is a
2729                  * new rport.
2730                  */
2731                 /* Guard against an unregister/reregister
2732                  * race that leaves the WAIT flag set.
2733                  */
2734                 spin_lock_irq(&vport->phba->hbalock);
2735                 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2736                 spin_unlock_irq(&vport->phba->hbalock);
2737                 rport = remote_port->private;
2738                 if (oldrport) {
2739                         /* New remoteport record does not guarantee valid
2740                          * host private memory area.
2741                          */
2742                         prev_ndlp = oldrport->ndlp;
2743                         if (oldrport == remote_port->private) {
2744                                 /* Same remoteport - ndlp should match.
2745                                  * Just reuse.
2746                                  */
2747                                 lpfc_printf_vlog(ndlp->vport, KERN_INFO,
2748                                                  LOG_NVME_DISC,
2749                                                  "6014 Rebinding lport to "
2750                                                  "remoteport %p wwpn 0x%llx, "
2751                                                  "Data: x%x x%x %p %p x%x x%06x\n",
2752                                                  remote_port,
2753                                                  remote_port->port_name,
2754                                                  remote_port->port_id,
2755                                                  remote_port->port_role,
2756                                                  prev_ndlp,
2757                                                  ndlp,
2758                                                  ndlp->nlp_type,
2759                                                  ndlp->nlp_DID);
2760                                 return 0;
2761                         }
2762
2763                         /* Sever the ndlp<->rport association
2764                          * before dropping the ndlp ref from
2765                          * register.
2766                          */
2767                         spin_lock_irq(&vport->phba->hbalock);
2768                         ndlp->nrport = NULL;
2769                         ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2770                         spin_unlock_irq(&vport->phba->hbalock);
2771                         rport->ndlp = NULL;
2772                         rport->remoteport = NULL;
2773
2774                         /* Reference only removed if previous NDLP is no longer
2775                          * active. It might be just a swap and removing the
2776                          * reference would cause a premature cleanup.
2777                          */
2778                         if (prev_ndlp && prev_ndlp != ndlp) {
2779                                 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) ||
2780                                     (!prev_ndlp->nrport))
2781                                         lpfc_nlp_put(prev_ndlp);
2782                         }
2783                 }
2784
2785                 /* Clean bind the rport to the ndlp. */
2786                 rport->remoteport = remote_port;
2787                 rport->lport = lport;
2788                 rport->ndlp = ndlp;
2789                 spin_lock_irq(&vport->phba->hbalock);
2790                 ndlp->nrport = rport;
2791                 spin_unlock_irq(&vport->phba->hbalock);
2792                 lpfc_printf_vlog(vport, KERN_INFO,
2793                                  LOG_NVME_DISC | LOG_NODE,
2794                                  "6022 Binding new rport to "
2795                                  "lport %p Remoteport %p rport %p WWNN 0x%llx, "
2796                                  "Rport WWPN 0x%llx DID "
2797                                  "x%06x Role x%x, ndlp %p prev_ndlp %p\n",
2798                                  lport, remote_port, rport,
2799                                  rpinfo.node_name, rpinfo.port_name,
2800                                  rpinfo.port_id, rpinfo.port_role,
2801                                  ndlp, prev_ndlp);
2802         } else {
2803                 lpfc_printf_vlog(vport, KERN_ERR,
2804                                  LOG_NVME_DISC | LOG_NODE,
2805                                  "6031 RemotePort Registration failed "
2806                                  "err: %d, DID x%06x\n",
2807                                  ret, ndlp->nlp_DID);
2808         }
2809
2810         return ret;
2811 #else
2812         return 0;
2813 #endif
2814 }
2815
2816 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2817  *
2818  * There is no notion of Devloss or rport recovery from the current
2819  * nvme_transport perspective.  Loss of an rport just means IO cannot
2820  * be sent and recovery is completely up to the initator.
2821  * For now, the driver just unbinds the DID and port_role so that
2822  * no further IO can be issued.  Changes are planned for later.
2823  *
2824  * Notes - the ndlp reference count is not decremented here since
2825  * since there is no nvme_transport api for devloss.  Node ref count
2826  * is only adjusted in driver unload.
2827  */
2828 void
2829 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2830 {
2831 #if (IS_ENABLED(CONFIG_NVME_FC))
2832         int ret;
2833         struct nvme_fc_local_port *localport;
2834         struct lpfc_nvme_lport *lport;
2835         struct lpfc_nvme_rport *rport;
2836         struct nvme_fc_remote_port *remoteport = NULL;
2837
2838         localport = vport->localport;
2839
2840         /* This is fundamental error.  The localport is always
2841          * available until driver unload.  Just exit.
2842          */
2843         if (!localport)
2844                 return;
2845
2846         lport = (struct lpfc_nvme_lport *)localport->private;
2847         if (!lport)
2848                 goto input_err;
2849
2850         spin_lock_irq(&vport->phba->hbalock);
2851         rport = lpfc_ndlp_get_nrport(ndlp);
2852         if (rport)
2853                 remoteport = rport->remoteport;
2854         spin_unlock_irq(&vport->phba->hbalock);
2855         if (!remoteport)
2856                 goto input_err;
2857
2858         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2859                          "6033 Unreg nvme remoteport %p, portname x%llx, "
2860                          "port_id x%06x, portstate x%x port type x%x\n",
2861                          remoteport, remoteport->port_name,
2862                          remoteport->port_id, remoteport->port_state,
2863                          ndlp->nlp_type);
2864
2865         /* Sanity check ndlp type.  Only call for NVME ports. Don't
2866          * clear any rport state until the transport calls back.
2867          */
2868
2869         if (ndlp->nlp_type & NLP_NVME_TARGET) {
2870                 /* No concern about the role change on the nvme remoteport.
2871                  * The transport will update it.
2872                  */
2873                 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG;
2874
2875                 /* Don't let the host nvme transport keep sending keep-alives
2876                  * on this remoteport. Vport is unloading, no recovery. The
2877                  * return values is ignored.  The upcall is a courtesy to the
2878                  * transport.
2879                  */
2880                 if (vport->load_flag & FC_UNLOADING)
2881                         (void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2882
2883                 ret = nvme_fc_unregister_remoteport(remoteport);
2884                 if (ret != 0) {
2885                         lpfc_nlp_put(ndlp);
2886                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2887                                          "6167 NVME unregister failed %d "
2888                                          "port_state x%x\n",
2889                                          ret, remoteport->port_state);
2890                 }
2891         }
2892         return;
2893
2894  input_err:
2895 #endif
2896         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2897                          "6168 State error: lport %p, rport%p FCID x%06x\n",
2898                          vport->localport, ndlp->rport, ndlp->nlp_DID);
2899 }
2900
2901 /**
2902  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2903  * @phba: pointer to lpfc hba data structure.
2904  * @axri: pointer to the fcp xri abort wcqe structure.
2905  *
2906  * This routine is invoked by the worker thread to process a SLI4 fast-path
2907  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2908  * here.
2909  **/
2910 void
2911 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2912                            struct sli4_wcqe_xri_aborted *axri)
2913 {
2914         uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2915         struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd;
2916         struct nvmefc_fcp_req *nvme_cmd = NULL;
2917         struct lpfc_nodelist *ndlp;
2918         unsigned long iflag = 0;
2919
2920         if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
2921                 return;
2922         spin_lock_irqsave(&phba->hbalock, iflag);
2923         spin_lock(&phba->sli4_hba.abts_nvme_buf_list_lock);
2924         list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd,
2925                                  &phba->sli4_hba.lpfc_abts_nvme_buf_list,
2926                                  list) {
2927                 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) {
2928                         list_del_init(&lpfc_ncmd->list);
2929                         lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2930                         lpfc_ncmd->status = IOSTAT_SUCCESS;
2931                         spin_unlock(
2932                                 &phba->sli4_hba.abts_nvme_buf_list_lock);
2933
2934                         spin_unlock_irqrestore(&phba->hbalock, iflag);
2935                         ndlp = lpfc_ncmd->ndlp;
2936                         if (ndlp)
2937                                 lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2938
2939                         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2940                                         "6311 nvme_cmd %p xri x%x tag x%x "
2941                                         "abort complete and xri released\n",
2942                                         lpfc_ncmd->nvmeCmd, xri,
2943                                         lpfc_ncmd->cur_iocbq.iotag);
2944
2945                         /* Aborted NVME commands are required to not complete
2946                          * before the abort exchange command fully completes.
2947                          * Once completed, it is available via the put list.
2948                          */
2949                         if (lpfc_ncmd->nvmeCmd) {
2950                                 nvme_cmd = lpfc_ncmd->nvmeCmd;
2951                                 nvme_cmd->done(nvme_cmd);
2952                                 lpfc_ncmd->nvmeCmd = NULL;
2953                         }
2954                         lpfc_release_nvme_buf(phba, lpfc_ncmd);
2955                         return;
2956                 }
2957         }
2958         spin_unlock(&phba->sli4_hba.abts_nvme_buf_list_lock);
2959         spin_unlock_irqrestore(&phba->hbalock, iflag);
2960
2961         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2962                         "6312 XRI Aborted xri x%x not found\n", xri);
2963
2964 }
2965
2966 /**
2967  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2968  * @phba: Pointer to HBA context object.
2969  *
2970  * This function flushes all wqes in the nvme rings and frees all resources
2971  * in the txcmplq. This function does not issue abort wqes for the IO
2972  * commands in txcmplq, they will just be returned with
2973  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2974  * slot has been permanently disabled.
2975  **/
2976 void
2977 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2978 {
2979         struct lpfc_sli_ring  *pring;
2980         u32 i, wait_cnt = 0;
2981
2982         if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.nvme_wq)
2983                 return;
2984
2985         /* Cycle through all NVME rings and make sure all outstanding
2986          * WQEs have been removed from the txcmplqs.
2987          */
2988         for (i = 0; i < phba->cfg_nvme_io_channel; i++) {
2989                 pring = phba->sli4_hba.nvme_wq[i]->pring;
2990
2991                 if (!pring)
2992                         continue;
2993
2994                 /* Retrieve everything on the txcmplq */
2995                 while (!list_empty(&pring->txcmplq)) {
2996                         msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2997                         wait_cnt++;
2998
2999                         /* The sleep is 10mS.  Every ten seconds,
3000                          * dump a message.  Something is wrong.
3001                          */
3002                         if ((wait_cnt % 1000) == 0) {
3003                                 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
3004                                                 "6178 NVME IO not empty, "
3005                                                 "cnt %d\n", wait_cnt);
3006                         }
3007                 }
3008         }
3009 }