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