1 /******************************************************************* 2 * This file is part of the Emulex Linux Device Driver for * 3 * Fibre Channel Host Bus Adapters. * 4 * Copyright (C) 2017-2021 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 "lpfc_version.h" 40 #include "lpfc_hw4.h" 41 #include "lpfc_hw.h" 42 #include "lpfc_sli.h" 43 #include "lpfc_sli4.h" 44 #include "lpfc_nl.h" 45 #include "lpfc_disc.h" 46 #include "lpfc.h" 47 #include "lpfc_nvme.h" 48 #include "lpfc_scsi.h" 49 #include "lpfc_logmsg.h" 50 #include "lpfc_crtn.h" 51 #include "lpfc_vport.h" 52 #include "lpfc_debugfs.h" 53 54 /* NVME initiator-based functions */ 55 56 static struct lpfc_io_buf * 57 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 58 int idx, int expedite); 59 60 static void 61 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_io_buf *); 62 63 static struct nvme_fc_port_template lpfc_nvme_template; 64 65 /** 66 * lpfc_nvme_create_queue - 67 * @pnvme_lport: Transport localport that LS is to be issued from 68 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors. 69 * @qsize: Size of the queue in bytes 70 * @handle: An opaque driver handle used in follow-up calls. 71 * 72 * Driver registers this routine to preallocate and initialize any 73 * internal data structures to bind the @qidx to its internal IO queues. 74 * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ. 75 * 76 * Return value : 77 * 0 - Success 78 * -EINVAL - Unsupported input value. 79 * -ENOMEM - Could not alloc necessary memory 80 **/ 81 static int 82 lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport, 83 unsigned int qidx, u16 qsize, 84 void **handle) 85 { 86 struct lpfc_nvme_lport *lport; 87 struct lpfc_vport *vport; 88 struct lpfc_nvme_qhandle *qhandle; 89 char *str; 90 91 if (!pnvme_lport->private) 92 return -ENOMEM; 93 94 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 95 vport = lport->vport; 96 qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL); 97 if (qhandle == NULL) 98 return -ENOMEM; 99 100 qhandle->cpu_id = raw_smp_processor_id(); 101 qhandle->qidx = qidx; 102 /* 103 * NVME qidx == 0 is the admin queue, so both admin queue 104 * and first IO queue will use MSI-X vector and associated 105 * EQ/CQ/WQ at index 0. After that they are sequentially assigned. 106 */ 107 if (qidx) { 108 str = "IO "; /* IO queue */ 109 qhandle->index = ((qidx - 1) % 110 lpfc_nvme_template.max_hw_queues); 111 } else { 112 str = "ADM"; /* Admin queue */ 113 qhandle->index = qidx; 114 } 115 116 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 117 "6073 Binding %s HdwQueue %d (cpu %d) to " 118 "hdw_queue %d qhandle x%px\n", str, 119 qidx, qhandle->cpu_id, qhandle->index, qhandle); 120 *handle = (void *)qhandle; 121 return 0; 122 } 123 124 /** 125 * lpfc_nvme_delete_queue - 126 * @pnvme_lport: Transport localport that LS is to be issued from 127 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors. 128 * @handle: An opaque driver handle from lpfc_nvme_create_queue 129 * 130 * Driver registers this routine to free 131 * any internal data structures to bind the @qidx to its internal 132 * IO queues. 133 * 134 * Return value : 135 * 0 - Success 136 * TODO: What are the failure codes. 137 **/ 138 static void 139 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport, 140 unsigned int qidx, 141 void *handle) 142 { 143 struct lpfc_nvme_lport *lport; 144 struct lpfc_vport *vport; 145 146 if (!pnvme_lport->private) 147 return; 148 149 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 150 vport = lport->vport; 151 152 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 153 "6001 ENTER. lpfc_pnvme x%px, qidx x%x qhandle x%px\n", 154 lport, qidx, handle); 155 kfree(handle); 156 } 157 158 static void 159 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport) 160 { 161 struct lpfc_nvme_lport *lport = localport->private; 162 163 lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME, 164 "6173 localport x%px delete complete\n", 165 lport); 166 167 /* release any threads waiting for the unreg to complete */ 168 if (lport->vport->localport) 169 complete(lport->lport_unreg_cmp); 170 } 171 172 /* lpfc_nvme_remoteport_delete 173 * 174 * @remoteport: Pointer to an nvme transport remoteport instance. 175 * 176 * This is a template downcall. NVME transport calls this function 177 * when it has completed the unregistration of a previously 178 * registered remoteport. 179 * 180 * Return value : 181 * None 182 */ 183 static void 184 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport) 185 { 186 struct lpfc_nvme_rport *rport = remoteport->private; 187 struct lpfc_vport *vport; 188 struct lpfc_nodelist *ndlp; 189 u32 fc4_xpt_flags; 190 191 ndlp = rport->ndlp; 192 if (!ndlp) { 193 pr_err("**** %s: NULL ndlp on rport x%px remoteport x%px\n", 194 __func__, rport, remoteport); 195 goto rport_err; 196 } 197 198 vport = ndlp->vport; 199 if (!vport) { 200 pr_err("**** %s: Null vport on ndlp x%px, ste x%x rport x%px\n", 201 __func__, ndlp, ndlp->nlp_state, rport); 202 goto rport_err; 203 } 204 205 fc4_xpt_flags = NVME_XPT_REGD | SCSI_XPT_REGD; 206 207 /* Remove this rport from the lport's list - memory is owned by the 208 * transport. Remove the ndlp reference for the NVME transport before 209 * calling state machine to remove the node. 210 */ 211 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 212 "6146 remoteport delete of remoteport x%px\n", 213 remoteport); 214 spin_lock_irq(&ndlp->lock); 215 216 /* The register rebind might have occurred before the delete 217 * downcall. Guard against this race. 218 */ 219 if (ndlp->fc4_xpt_flags & NVME_XPT_UNREG_WAIT) 220 ndlp->fc4_xpt_flags &= ~(NVME_XPT_UNREG_WAIT | NVME_XPT_REGD); 221 222 spin_unlock_irq(&ndlp->lock); 223 224 /* On a devloss timeout event, one more put is executed provided the 225 * NVME and SCSI rport unregister requests are complete. If the vport 226 * is unloading, this extra put is executed by lpfc_drop_node. 227 */ 228 if (!(ndlp->fc4_xpt_flags & fc4_xpt_flags)) 229 lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM); 230 231 rport_err: 232 return; 233 } 234 235 /** 236 * lpfc_nvme_handle_lsreq - Process an unsolicited NVME LS request 237 * @phba: pointer to lpfc hba data structure. 238 * @axchg: pointer to exchange context for the NVME LS request 239 * 240 * This routine is used for processing an asychronously received NVME LS 241 * request. Any remaining validation is done and the LS is then forwarded 242 * to the nvme-fc transport via nvme_fc_rcv_ls_req(). 243 * 244 * The calling sequence should be: nvme_fc_rcv_ls_req() -> (processing) 245 * -> lpfc_nvme_xmt_ls_rsp/cmp -> req->done. 246 * __lpfc_nvme_xmt_ls_rsp_cmp should free the allocated axchg. 247 * 248 * Returns 0 if LS was handled and delivered to the transport 249 * Returns 1 if LS failed to be handled and should be dropped 250 */ 251 int 252 lpfc_nvme_handle_lsreq(struct lpfc_hba *phba, 253 struct lpfc_async_xchg_ctx *axchg) 254 { 255 #if (IS_ENABLED(CONFIG_NVME_FC)) 256 struct lpfc_vport *vport; 257 struct lpfc_nvme_rport *lpfc_rport; 258 struct nvme_fc_remote_port *remoteport; 259 struct lpfc_nvme_lport *lport; 260 uint32_t *payload = axchg->payload; 261 int rc; 262 263 vport = axchg->ndlp->vport; 264 lpfc_rport = axchg->ndlp->nrport; 265 if (!lpfc_rport) 266 return -EINVAL; 267 268 remoteport = lpfc_rport->remoteport; 269 if (!vport->localport) 270 return -EINVAL; 271 272 lport = vport->localport->private; 273 if (!lport) 274 return -EINVAL; 275 276 rc = nvme_fc_rcv_ls_req(remoteport, &axchg->ls_rsp, axchg->payload, 277 axchg->size); 278 279 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_DISC, 280 "6205 NVME Unsol rcv: sz %d rc %d: %08x %08x %08x " 281 "%08x %08x %08x\n", 282 axchg->size, rc, 283 *payload, *(payload+1), *(payload+2), 284 *(payload+3), *(payload+4), *(payload+5)); 285 286 if (!rc) 287 return 0; 288 #endif 289 return 1; 290 } 291 292 /** 293 * __lpfc_nvme_ls_req_cmp - Generic completion handler for a NVME 294 * LS request. 295 * @phba: Pointer to HBA context object 296 * @vport: The local port that issued the LS 297 * @cmdwqe: Pointer to driver command WQE object. 298 * @wcqe: Pointer to driver response CQE object. 299 * 300 * This function is the generic completion handler for NVME LS requests. 301 * The function updates any states and statistics, calls the transport 302 * ls_req done() routine, then tears down the command and buffers used 303 * for the LS request. 304 **/ 305 void 306 __lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_vport *vport, 307 struct lpfc_iocbq *cmdwqe, 308 struct lpfc_wcqe_complete *wcqe) 309 { 310 struct nvmefc_ls_req *pnvme_lsreq; 311 struct lpfc_dmabuf *buf_ptr; 312 struct lpfc_nodelist *ndlp; 313 uint32_t status; 314 315 pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2; 316 ndlp = (struct lpfc_nodelist *)cmdwqe->context1; 317 status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK; 318 319 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 320 "6047 NVMEx LS REQ x%px cmpl DID %x Xri: %x " 321 "status %x reason x%x cmd:x%px lsreg:x%px bmp:x%px " 322 "ndlp:x%px\n", 323 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 324 cmdwqe->sli4_xritag, status, 325 (wcqe->parameter & 0xffff), 326 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp); 327 328 lpfc_nvmeio_data(phba, "NVMEx LS CMPL: xri x%x stat x%x parm x%x\n", 329 cmdwqe->sli4_xritag, status, wcqe->parameter); 330 331 if (cmdwqe->context3) { 332 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3; 333 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys); 334 kfree(buf_ptr); 335 cmdwqe->context3 = NULL; 336 } 337 if (pnvme_lsreq->done) 338 pnvme_lsreq->done(pnvme_lsreq, status); 339 else 340 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 341 "6046 NVMEx cmpl without done call back? " 342 "Data x%px DID %x Xri: %x status %x\n", 343 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 344 cmdwqe->sli4_xritag, status); 345 if (ndlp) { 346 lpfc_nlp_put(ndlp); 347 cmdwqe->context1 = NULL; 348 } 349 lpfc_sli_release_iocbq(phba, cmdwqe); 350 } 351 352 static void 353 lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe, 354 struct lpfc_wcqe_complete *wcqe) 355 { 356 struct lpfc_vport *vport = cmdwqe->vport; 357 struct lpfc_nvme_lport *lport; 358 uint32_t status; 359 360 status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK; 361 362 if (vport->localport) { 363 lport = (struct lpfc_nvme_lport *)vport->localport->private; 364 if (lport) { 365 atomic_inc(&lport->fc4NvmeLsCmpls); 366 if (status) { 367 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 368 atomic_inc(&lport->cmpl_ls_xb); 369 atomic_inc(&lport->cmpl_ls_err); 370 } 371 } 372 } 373 374 __lpfc_nvme_ls_req_cmp(phba, vport, cmdwqe, wcqe); 375 } 376 377 static int 378 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp, 379 struct lpfc_dmabuf *inp, 380 struct nvmefc_ls_req *pnvme_lsreq, 381 void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *, 382 struct lpfc_wcqe_complete *), 383 struct lpfc_nodelist *ndlp, uint32_t num_entry, 384 uint32_t tmo, uint8_t retry) 385 { 386 struct lpfc_hba *phba = vport->phba; 387 union lpfc_wqe128 *wqe; 388 struct lpfc_iocbq *genwqe; 389 struct ulp_bde64 *bpl; 390 struct ulp_bde64 bde; 391 int i, rc, xmit_len, first_len; 392 393 /* Allocate buffer for command WQE */ 394 genwqe = lpfc_sli_get_iocbq(phba); 395 if (genwqe == NULL) 396 return 1; 397 398 wqe = &genwqe->wqe; 399 /* Initialize only 64 bytes */ 400 memset(wqe, 0, sizeof(union lpfc_wqe)); 401 402 genwqe->context3 = (uint8_t *)bmp; 403 genwqe->iocb_flag |= LPFC_IO_NVME_LS; 404 405 /* Save for completion so we can release these resources */ 406 genwqe->context1 = lpfc_nlp_get(ndlp); 407 if (!genwqe->context1) { 408 dev_warn(&phba->pcidev->dev, 409 "Warning: Failed node ref, not sending LS_REQ\n"); 410 lpfc_sli_release_iocbq(phba, genwqe); 411 return 1; 412 } 413 414 genwqe->context2 = (uint8_t *)pnvme_lsreq; 415 /* Fill in payload, bp points to frame payload */ 416 417 if (!tmo) 418 /* FC spec states we need 3 * ratov for CT requests */ 419 tmo = (3 * phba->fc_ratov); 420 421 /* For this command calculate the xmit length of the request bde. */ 422 xmit_len = 0; 423 first_len = 0; 424 bpl = (struct ulp_bde64 *)bmp->virt; 425 for (i = 0; i < num_entry; i++) { 426 bde.tus.w = bpl[i].tus.w; 427 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64) 428 break; 429 xmit_len += bde.tus.f.bdeSize; 430 if (i == 0) 431 first_len = xmit_len; 432 } 433 434 genwqe->rsvd2 = num_entry; 435 genwqe->hba_wqidx = 0; 436 437 /* Words 0 - 2 */ 438 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64; 439 wqe->generic.bde.tus.f.bdeSize = first_len; 440 wqe->generic.bde.addrLow = bpl[0].addrLow; 441 wqe->generic.bde.addrHigh = bpl[0].addrHigh; 442 443 /* Word 3 */ 444 wqe->gen_req.request_payload_len = first_len; 445 446 /* Word 4 */ 447 448 /* Word 5 */ 449 bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0); 450 bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1); 451 bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1); 452 bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ); 453 bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME); 454 455 /* Word 6 */ 456 bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com, 457 phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]); 458 bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag); 459 460 /* Word 7 */ 461 bf_set(wqe_tmo, &wqe->gen_req.wqe_com, tmo); 462 bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3); 463 bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE); 464 bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI); 465 466 /* Word 8 */ 467 wqe->gen_req.wqe_com.abort_tag = genwqe->iotag; 468 469 /* Word 9 */ 470 bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag); 471 472 /* Word 10 */ 473 bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1); 474 bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ); 475 bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1); 476 bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE); 477 bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0); 478 479 /* Word 11 */ 480 bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 481 bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND); 482 483 484 /* Issue GEN REQ WQE for NPORT <did> */ 485 genwqe->wqe_cmpl = cmpl; 486 genwqe->iocb_cmpl = NULL; 487 genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT; 488 genwqe->vport = vport; 489 genwqe->retry = retry; 490 491 lpfc_nvmeio_data(phba, "NVME LS XMIT: xri x%x iotag x%x to x%06x\n", 492 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID); 493 494 rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe); 495 if (rc) { 496 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 497 "6045 Issue GEN REQ WQE to NPORT x%x " 498 "Data: x%x x%x rc x%x\n", 499 ndlp->nlp_DID, genwqe->iotag, 500 vport->port_state, rc); 501 lpfc_nlp_put(ndlp); 502 lpfc_sli_release_iocbq(phba, genwqe); 503 return 1; 504 } 505 506 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_ELS, 507 "6050 Issue GEN REQ WQE to NPORT x%x " 508 "Data: oxid: x%x state: x%x wq:x%px lsreq:x%px " 509 "bmp:x%px xmit:%d 1st:%d\n", 510 ndlp->nlp_DID, genwqe->sli4_xritag, 511 vport->port_state, 512 genwqe, pnvme_lsreq, bmp, xmit_len, first_len); 513 return 0; 514 } 515 516 517 /** 518 * __lpfc_nvme_ls_req - Generic service routine to issue an NVME LS request 519 * @vport: The local port issuing the LS 520 * @ndlp: The remote port to send the LS to 521 * @pnvme_lsreq: Pointer to LS request structure from the transport 522 * @gen_req_cmp: Completion call-back 523 * 524 * Routine validates the ndlp, builds buffers and sends a GEN_REQUEST 525 * WQE to perform the LS operation. 526 * 527 * Return value : 528 * 0 - Success 529 * non-zero: various error codes, in form of -Exxx 530 **/ 531 int 532 __lpfc_nvme_ls_req(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp, 533 struct nvmefc_ls_req *pnvme_lsreq, 534 void (*gen_req_cmp)(struct lpfc_hba *phba, 535 struct lpfc_iocbq *cmdwqe, 536 struct lpfc_wcqe_complete *wcqe)) 537 { 538 struct lpfc_dmabuf *bmp; 539 struct ulp_bde64 *bpl; 540 int ret; 541 uint16_t ntype, nstate; 542 543 if (!ndlp) { 544 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 545 "6051 NVMEx LS REQ: Bad NDLP x%px, Failing " 546 "LS Req\n", 547 ndlp); 548 return -ENODEV; 549 } 550 551 ntype = ndlp->nlp_type; 552 nstate = ndlp->nlp_state; 553 if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) || 554 (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) { 555 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 556 "6088 NVMEx LS REQ: Fail DID x%06x not " 557 "ready for IO. Type x%x, State x%x\n", 558 ndlp->nlp_DID, ntype, nstate); 559 return -ENODEV; 560 } 561 562 if (!vport->phba->sli4_hba.nvmels_wq) 563 return -ENOMEM; 564 565 /* 566 * there are two dma buf in the request, actually there is one and 567 * the second one is just the start address + cmd size. 568 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped 569 * in a lpfc_dmabuf struct. When freeing we just free the wrapper 570 * because the nvem layer owns the data bufs. 571 * We do not have to break these packets open, we don't care what is 572 * in them. And we do not have to look at the resonse data, we only 573 * care that we got a response. All of the caring is going to happen 574 * in the nvme-fc layer. 575 */ 576 577 bmp = kmalloc(sizeof(*bmp), GFP_KERNEL); 578 if (!bmp) { 579 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 580 "6044 NVMEx LS REQ: Could not alloc LS buf " 581 "for DID %x\n", 582 ndlp->nlp_DID); 583 return -ENOMEM; 584 } 585 586 bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys)); 587 if (!bmp->virt) { 588 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 589 "6042 NVMEx LS REQ: Could not alloc mbuf " 590 "for DID %x\n", 591 ndlp->nlp_DID); 592 kfree(bmp); 593 return -ENOMEM; 594 } 595 596 INIT_LIST_HEAD(&bmp->list); 597 598 bpl = (struct ulp_bde64 *)bmp->virt; 599 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma)); 600 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma)); 601 bpl->tus.f.bdeFlags = 0; 602 bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen; 603 bpl->tus.w = le32_to_cpu(bpl->tus.w); 604 bpl++; 605 606 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma)); 607 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma)); 608 bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I; 609 bpl->tus.f.bdeSize = pnvme_lsreq->rsplen; 610 bpl->tus.w = le32_to_cpu(bpl->tus.w); 611 612 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 613 "6149 NVMEx LS REQ: Issue to DID 0x%06x lsreq x%px, " 614 "rqstlen:%d rsplen:%d %pad %pad\n", 615 ndlp->nlp_DID, pnvme_lsreq, pnvme_lsreq->rqstlen, 616 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 617 &pnvme_lsreq->rspdma); 618 619 ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr, 620 pnvme_lsreq, gen_req_cmp, ndlp, 2, 621 pnvme_lsreq->timeout, 0); 622 if (ret != WQE_SUCCESS) { 623 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 624 "6052 NVMEx REQ: EXIT. issue ls wqe failed " 625 "lsreq x%px Status %x DID %x\n", 626 pnvme_lsreq, ret, ndlp->nlp_DID); 627 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys); 628 kfree(bmp); 629 return -EIO; 630 } 631 632 return 0; 633 } 634 635 /** 636 * lpfc_nvme_ls_req - Issue an NVME Link Service request 637 * @pnvme_lport: Transport localport that LS is to be issued from. 638 * @pnvme_rport: Transport remoteport that LS is to be sent to. 639 * @pnvme_lsreq: the transport nvme_ls_req structure for the LS 640 * 641 * Driver registers this routine to handle any link service request 642 * from the nvme_fc transport to a remote nvme-aware port. 643 * 644 * Return value : 645 * 0 - Success 646 * non-zero: various error codes, in form of -Exxx 647 **/ 648 static int 649 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport, 650 struct nvme_fc_remote_port *pnvme_rport, 651 struct nvmefc_ls_req *pnvme_lsreq) 652 { 653 struct lpfc_nvme_lport *lport; 654 struct lpfc_nvme_rport *rport; 655 struct lpfc_vport *vport; 656 int ret; 657 658 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 659 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 660 if (unlikely(!lport) || unlikely(!rport)) 661 return -EINVAL; 662 663 vport = lport->vport; 664 if (vport->load_flag & FC_UNLOADING) 665 return -ENODEV; 666 667 atomic_inc(&lport->fc4NvmeLsRequests); 668 669 ret = __lpfc_nvme_ls_req(vport, rport->ndlp, pnvme_lsreq, 670 lpfc_nvme_ls_req_cmp); 671 if (ret) 672 atomic_inc(&lport->xmt_ls_err); 673 674 return ret; 675 } 676 677 /** 678 * __lpfc_nvme_ls_abort - Generic service routine to abort a prior 679 * NVME LS request 680 * @vport: The local port that issued the LS 681 * @ndlp: The remote port the LS was sent to 682 * @pnvme_lsreq: Pointer to LS request structure from the transport 683 * 684 * The driver validates the ndlp, looks for the LS, and aborts the 685 * LS if found. 686 * 687 * Returns: 688 * 0 : if LS found and aborted 689 * non-zero: various error conditions in form -Exxx 690 **/ 691 int 692 __lpfc_nvme_ls_abort(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp, 693 struct nvmefc_ls_req *pnvme_lsreq) 694 { 695 struct lpfc_hba *phba = vport->phba; 696 struct lpfc_sli_ring *pring; 697 struct lpfc_iocbq *wqe, *next_wqe; 698 bool foundit = false; 699 700 if (!ndlp) { 701 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 702 "6049 NVMEx LS REQ Abort: Bad NDLP x%px DID " 703 "x%06x, Failing LS Req\n", 704 ndlp, ndlp ? ndlp->nlp_DID : 0); 705 return -EINVAL; 706 } 707 708 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS, 709 "6040 NVMEx LS REQ Abort: Issue LS_ABORT for lsreq " 710 "x%px rqstlen:%d rsplen:%d %pad %pad\n", 711 pnvme_lsreq, pnvme_lsreq->rqstlen, 712 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 713 &pnvme_lsreq->rspdma); 714 715 /* 716 * Lock the ELS ring txcmplq and look for the wqe that matches 717 * this ELS. If found, issue an abort on the wqe. 718 */ 719 pring = phba->sli4_hba.nvmels_wq->pring; 720 spin_lock_irq(&phba->hbalock); 721 spin_lock(&pring->ring_lock); 722 list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) { 723 if (wqe->context2 == pnvme_lsreq) { 724 wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 725 foundit = true; 726 break; 727 } 728 } 729 spin_unlock(&pring->ring_lock); 730 731 if (foundit) 732 lpfc_sli_issue_abort_iotag(phba, pring, wqe, NULL); 733 spin_unlock_irq(&phba->hbalock); 734 735 if (foundit) 736 return 0; 737 738 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS, 739 "6213 NVMEx LS REQ Abort: Unable to locate req x%px\n", 740 pnvme_lsreq); 741 return -EINVAL; 742 } 743 744 static int 745 lpfc_nvme_xmt_ls_rsp(struct nvme_fc_local_port *localport, 746 struct nvme_fc_remote_port *remoteport, 747 struct nvmefc_ls_rsp *ls_rsp) 748 { 749 struct lpfc_async_xchg_ctx *axchg = 750 container_of(ls_rsp, struct lpfc_async_xchg_ctx, ls_rsp); 751 struct lpfc_nvme_lport *lport; 752 int rc; 753 754 if (axchg->phba->pport->load_flag & FC_UNLOADING) 755 return -ENODEV; 756 757 lport = (struct lpfc_nvme_lport *)localport->private; 758 759 rc = __lpfc_nvme_xmt_ls_rsp(axchg, ls_rsp, __lpfc_nvme_xmt_ls_rsp_cmp); 760 761 if (rc) { 762 /* 763 * unless the failure is due to having already sent 764 * the response, an abort will be generated for the 765 * exchange if the rsp can't be sent. 766 */ 767 if (rc != -EALREADY) 768 atomic_inc(&lport->xmt_ls_abort); 769 return rc; 770 } 771 772 return 0; 773 } 774 775 /** 776 * lpfc_nvme_ls_abort - Abort a prior NVME LS request 777 * @pnvme_lport: Transport localport that LS is to be issued from. 778 * @pnvme_rport: Transport remoteport that LS is to be sent to. 779 * @pnvme_lsreq: the transport nvme_ls_req structure for the LS 780 * 781 * Driver registers this routine to abort a NVME LS request that is 782 * in progress (from the transports perspective). 783 **/ 784 static void 785 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport, 786 struct nvme_fc_remote_port *pnvme_rport, 787 struct nvmefc_ls_req *pnvme_lsreq) 788 { 789 struct lpfc_nvme_lport *lport; 790 struct lpfc_vport *vport; 791 struct lpfc_nodelist *ndlp; 792 int ret; 793 794 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 795 if (unlikely(!lport)) 796 return; 797 vport = lport->vport; 798 799 if (vport->load_flag & FC_UNLOADING) 800 return; 801 802 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id); 803 804 ret = __lpfc_nvme_ls_abort(vport, ndlp, pnvme_lsreq); 805 if (!ret) 806 atomic_inc(&lport->xmt_ls_abort); 807 } 808 809 /* Fix up the existing sgls for NVME IO. */ 810 static inline void 811 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport, 812 struct lpfc_io_buf *lpfc_ncmd, 813 struct nvmefc_fcp_req *nCmd) 814 { 815 struct lpfc_hba *phba = vport->phba; 816 struct sli4_sge *sgl; 817 union lpfc_wqe128 *wqe; 818 uint32_t *wptr, *dptr; 819 820 /* 821 * Get a local pointer to the built-in wqe and correct 822 * the cmd size to match NVME's 96 bytes and fix 823 * the dma address. 824 */ 825 826 wqe = &lpfc_ncmd->cur_iocbq.wqe; 827 828 /* 829 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to 830 * match NVME. NVME sends 96 bytes. Also, use the 831 * nvme commands command and response dma addresses 832 * rather than the virtual memory to ease the restore 833 * operation. 834 */ 835 sgl = lpfc_ncmd->dma_sgl; 836 sgl->sge_len = cpu_to_le32(nCmd->cmdlen); 837 if (phba->cfg_nvme_embed_cmd) { 838 sgl->addr_hi = 0; 839 sgl->addr_lo = 0; 840 841 /* Word 0-2 - NVME CMND IU (embedded payload) */ 842 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED; 843 wqe->generic.bde.tus.f.bdeSize = 56; 844 wqe->generic.bde.addrHigh = 0; 845 wqe->generic.bde.addrLow = 64; /* Word 16 */ 846 847 /* Word 10 - dbde is 0, wqes is 1 in template */ 848 849 /* 850 * Embed the payload in the last half of the WQE 851 * WQE words 16-30 get the NVME CMD IU payload 852 * 853 * WQE words 16-19 get payload Words 1-4 854 * WQE words 20-21 get payload Words 6-7 855 * WQE words 22-29 get payload Words 16-23 856 */ 857 wptr = &wqe->words[16]; /* WQE ptr */ 858 dptr = (uint32_t *)nCmd->cmdaddr; /* payload ptr */ 859 dptr++; /* Skip Word 0 in payload */ 860 861 *wptr++ = *dptr++; /* Word 1 */ 862 *wptr++ = *dptr++; /* Word 2 */ 863 *wptr++ = *dptr++; /* Word 3 */ 864 *wptr++ = *dptr++; /* Word 4 */ 865 dptr++; /* Skip Word 5 in payload */ 866 *wptr++ = *dptr++; /* Word 6 */ 867 *wptr++ = *dptr++; /* Word 7 */ 868 dptr += 8; /* Skip Words 8-15 in payload */ 869 *wptr++ = *dptr++; /* Word 16 */ 870 *wptr++ = *dptr++; /* Word 17 */ 871 *wptr++ = *dptr++; /* Word 18 */ 872 *wptr++ = *dptr++; /* Word 19 */ 873 *wptr++ = *dptr++; /* Word 20 */ 874 *wptr++ = *dptr++; /* Word 21 */ 875 *wptr++ = *dptr++; /* Word 22 */ 876 *wptr = *dptr; /* Word 23 */ 877 } else { 878 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma)); 879 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma)); 880 881 /* Word 0-2 - NVME CMND IU Inline BDE */ 882 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64; 883 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen; 884 wqe->generic.bde.addrHigh = sgl->addr_hi; 885 wqe->generic.bde.addrLow = sgl->addr_lo; 886 887 /* Word 10 */ 888 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1); 889 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0); 890 } 891 892 sgl++; 893 894 /* Setup the physical region for the FCP RSP */ 895 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma)); 896 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma)); 897 sgl->word2 = le32_to_cpu(sgl->word2); 898 if (nCmd->sg_cnt) 899 bf_set(lpfc_sli4_sge_last, sgl, 0); 900 else 901 bf_set(lpfc_sli4_sge_last, sgl, 1); 902 sgl->word2 = cpu_to_le32(sgl->word2); 903 sgl->sge_len = cpu_to_le32(nCmd->rsplen); 904 } 905 906 907 /* 908 * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO 909 * 910 * Driver registers this routine as it io request handler. This 911 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 912 * data structure to the rport indicated in @lpfc_nvme_rport. 913 * 914 * Return value : 915 * 0 - Success 916 * TODO: What are the failure codes. 917 **/ 918 static void 919 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn, 920 struct lpfc_wcqe_complete *wcqe) 921 { 922 struct lpfc_io_buf *lpfc_ncmd = 923 (struct lpfc_io_buf *)pwqeIn->context1; 924 struct lpfc_vport *vport = pwqeIn->vport; 925 struct nvmefc_fcp_req *nCmd; 926 struct nvme_fc_ersp_iu *ep; 927 struct nvme_fc_cmd_iu *cp; 928 struct lpfc_nodelist *ndlp; 929 struct lpfc_nvme_fcpreq_priv *freqpriv; 930 struct lpfc_nvme_lport *lport; 931 uint32_t code, status, idx; 932 uint16_t cid, sqhd, data; 933 uint32_t *ptr; 934 uint32_t lat; 935 bool call_done = false; 936 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 937 int cpu; 938 #endif 939 940 /* Sanity check on return of outstanding command */ 941 if (!lpfc_ncmd) { 942 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 943 "6071 Null lpfc_ncmd pointer. No " 944 "release, skip completion\n"); 945 return; 946 } 947 948 /* Guard against abort handler being called at same time */ 949 spin_lock(&lpfc_ncmd->buf_lock); 950 951 if (!lpfc_ncmd->nvmeCmd) { 952 spin_unlock(&lpfc_ncmd->buf_lock); 953 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 954 "6066 Missing cmpl ptrs: lpfc_ncmd x%px, " 955 "nvmeCmd x%px\n", 956 lpfc_ncmd, lpfc_ncmd->nvmeCmd); 957 958 /* Release the lpfc_ncmd regardless of the missing elements. */ 959 lpfc_release_nvme_buf(phba, lpfc_ncmd); 960 return; 961 } 962 nCmd = lpfc_ncmd->nvmeCmd; 963 status = bf_get(lpfc_wcqe_c_status, wcqe); 964 965 idx = lpfc_ncmd->cur_iocbq.hba_wqidx; 966 phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++; 967 968 if (unlikely(status && vport->localport)) { 969 lport = (struct lpfc_nvme_lport *)vport->localport->private; 970 if (lport) { 971 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 972 atomic_inc(&lport->cmpl_fcp_xb); 973 atomic_inc(&lport->cmpl_fcp_err); 974 } 975 } 976 977 lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n", 978 lpfc_ncmd->cur_iocbq.sli4_xritag, 979 status, wcqe->parameter); 980 /* 981 * Catch race where our node has transitioned, but the 982 * transport is still transitioning. 983 */ 984 ndlp = lpfc_ncmd->ndlp; 985 if (!ndlp) { 986 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 987 "6062 Ignoring NVME cmpl. No ndlp\n"); 988 goto out_err; 989 } 990 991 code = bf_get(lpfc_wcqe_c_code, wcqe); 992 if (code == CQE_CODE_NVME_ERSP) { 993 /* For this type of CQE, we need to rebuild the rsp */ 994 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr; 995 996 /* 997 * Get Command Id from cmd to plug into response. This 998 * code is not needed in the next NVME Transport drop. 999 */ 1000 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr; 1001 cid = cp->sqe.common.command_id; 1002 1003 /* 1004 * RSN is in CQE word 2 1005 * SQHD is in CQE Word 3 bits 15:0 1006 * Cmd Specific info is in CQE Word 1 1007 * and in CQE Word 0 bits 15:0 1008 */ 1009 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe); 1010 1011 /* Now lets build the NVME ERSP IU */ 1012 ep->iu_len = cpu_to_be16(8); 1013 ep->rsn = wcqe->parameter; 1014 ep->xfrd_len = cpu_to_be32(nCmd->payload_length); 1015 ep->rsvd12 = 0; 1016 ptr = (uint32_t *)&ep->cqe.result.u64; 1017 *ptr++ = wcqe->total_data_placed; 1018 data = bf_get(lpfc_wcqe_c_ersp0, wcqe); 1019 *ptr = (uint32_t)data; 1020 ep->cqe.sq_head = sqhd; 1021 ep->cqe.sq_id = nCmd->sqid; 1022 ep->cqe.command_id = cid; 1023 ep->cqe.status = 0; 1024 1025 lpfc_ncmd->status = IOSTAT_SUCCESS; 1026 lpfc_ncmd->result = 0; 1027 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN; 1028 nCmd->transferred_length = nCmd->payload_length; 1029 } else { 1030 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK); 1031 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK); 1032 1033 /* For NVME, the only failure path that results in an 1034 * IO error is when the adapter rejects it. All other 1035 * conditions are a success case and resolved by the 1036 * transport. 1037 * IOSTAT_FCP_RSP_ERROR means: 1038 * 1. Length of data received doesn't match total 1039 * transfer length in WQE 1040 * 2. If the RSP payload does NOT match these cases: 1041 * a. RSP length 12/24 bytes and all zeros 1042 * b. NVME ERSP 1043 */ 1044 switch (lpfc_ncmd->status) { 1045 case IOSTAT_SUCCESS: 1046 nCmd->transferred_length = wcqe->total_data_placed; 1047 nCmd->rcv_rsplen = 0; 1048 nCmd->status = 0; 1049 break; 1050 case IOSTAT_FCP_RSP_ERROR: 1051 nCmd->transferred_length = wcqe->total_data_placed; 1052 nCmd->rcv_rsplen = wcqe->parameter; 1053 nCmd->status = 0; 1054 1055 /* Check if this is really an ERSP */ 1056 if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) { 1057 lpfc_ncmd->status = IOSTAT_SUCCESS; 1058 lpfc_ncmd->result = 0; 1059 1060 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 1061 "6084 NVME Completion ERSP: " 1062 "xri %x placed x%x\n", 1063 lpfc_ncmd->cur_iocbq.sli4_xritag, 1064 wcqe->total_data_placed); 1065 break; 1066 } 1067 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1068 "6081 NVME Completion Protocol Error: " 1069 "xri %x status x%x result x%x " 1070 "placed x%x\n", 1071 lpfc_ncmd->cur_iocbq.sli4_xritag, 1072 lpfc_ncmd->status, lpfc_ncmd->result, 1073 wcqe->total_data_placed); 1074 break; 1075 case IOSTAT_LOCAL_REJECT: 1076 /* Let fall through to set command final state. */ 1077 if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED) 1078 lpfc_printf_vlog(vport, KERN_INFO, 1079 LOG_NVME_IOERR, 1080 "6032 Delay Aborted cmd x%px " 1081 "nvme cmd x%px, xri x%x, " 1082 "xb %d\n", 1083 lpfc_ncmd, nCmd, 1084 lpfc_ncmd->cur_iocbq.sli4_xritag, 1085 bf_get(lpfc_wcqe_c_xb, wcqe)); 1086 fallthrough; 1087 default: 1088 out_err: 1089 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1090 "6072 NVME Completion Error: xri %x " 1091 "status x%x result x%x [x%x] " 1092 "placed x%x\n", 1093 lpfc_ncmd->cur_iocbq.sli4_xritag, 1094 lpfc_ncmd->status, lpfc_ncmd->result, 1095 wcqe->parameter, 1096 wcqe->total_data_placed); 1097 nCmd->transferred_length = 0; 1098 nCmd->rcv_rsplen = 0; 1099 nCmd->status = NVME_SC_INTERNAL; 1100 } 1101 } 1102 1103 /* pick up SLI4 exhange busy condition */ 1104 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 1105 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 1106 else 1107 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 1108 1109 /* Update stats and complete the IO. There is 1110 * no need for dma unprep because the nvme_transport 1111 * owns the dma address. 1112 */ 1113 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1114 if (lpfc_ncmd->ts_cmd_start) { 1115 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp; 1116 lpfc_ncmd->ts_data_io = ktime_get_ns(); 1117 phba->ktime_last_cmd = lpfc_ncmd->ts_data_io; 1118 lpfc_io_ktime(phba, lpfc_ncmd); 1119 } 1120 if (unlikely(phba->hdwqstat_on & LPFC_CHECK_NVME_IO)) { 1121 cpu = raw_smp_processor_id(); 1122 this_cpu_inc(phba->sli4_hba.c_stat->cmpl_io); 1123 if (lpfc_ncmd->cpu != cpu) 1124 lpfc_printf_vlog(vport, 1125 KERN_INFO, LOG_NVME_IOERR, 1126 "6701 CPU Check cmpl: " 1127 "cpu %d expect %d\n", 1128 cpu, lpfc_ncmd->cpu); 1129 } 1130 #endif 1131 1132 /* NVME targets need completion held off until the abort exchange 1133 * completes unless the NVME Rport is getting unregistered. 1134 */ 1135 1136 if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) { 1137 freqpriv = nCmd->private; 1138 freqpriv->nvme_buf = NULL; 1139 lpfc_ncmd->nvmeCmd = NULL; 1140 call_done = true; 1141 } 1142 spin_unlock(&lpfc_ncmd->buf_lock); 1143 1144 /* Check if IO qualified for CMF */ 1145 if (phba->cmf_active_mode != LPFC_CFG_OFF && 1146 nCmd->io_dir == NVMEFC_FCP_READ && 1147 nCmd->payload_length) { 1148 /* Used when calculating average latency */ 1149 lat = ktime_get_ns() - lpfc_ncmd->rx_cmd_start; 1150 lpfc_update_cmf_cmpl(phba, lat, nCmd->payload_length, NULL); 1151 } 1152 1153 if (call_done) 1154 nCmd->done(nCmd); 1155 1156 /* Call release with XB=1 to queue the IO into the abort list. */ 1157 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1158 } 1159 1160 1161 /** 1162 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO 1163 * @vport: pointer to a host virtual N_Port data structure 1164 * @lpfc_ncmd: Pointer to lpfc scsi command 1165 * @pnode: pointer to a node-list data structure 1166 * @cstat: pointer to the control status structure 1167 * 1168 * Driver registers this routine as it io request handler. This 1169 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1170 * data structure to the rport indicated in @lpfc_nvme_rport. 1171 * 1172 * Return value : 1173 * 0 - Success 1174 * TODO: What are the failure codes. 1175 **/ 1176 static int 1177 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport, 1178 struct lpfc_io_buf *lpfc_ncmd, 1179 struct lpfc_nodelist *pnode, 1180 struct lpfc_fc4_ctrl_stat *cstat) 1181 { 1182 struct lpfc_hba *phba = vport->phba; 1183 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1184 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq); 1185 union lpfc_wqe128 *wqe = &pwqeq->wqe; 1186 uint32_t req_len; 1187 1188 /* 1189 * There are three possibilities here - use scatter-gather segment, use 1190 * the single mapping, or neither. 1191 */ 1192 if (nCmd->sg_cnt) { 1193 if (nCmd->io_dir == NVMEFC_FCP_WRITE) { 1194 /* From the iwrite template, initialize words 7 - 11 */ 1195 memcpy(&wqe->words[7], 1196 &lpfc_iwrite_cmd_template.words[7], 1197 sizeof(uint32_t) * 5); 1198 1199 /* Word 4 */ 1200 wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length; 1201 1202 /* Word 5 */ 1203 if ((phba->cfg_nvme_enable_fb) && 1204 (pnode->nlp_flag & NLP_FIRSTBURST)) { 1205 req_len = lpfc_ncmd->nvmeCmd->payload_length; 1206 if (req_len < pnode->nvme_fb_size) 1207 wqe->fcp_iwrite.initial_xfer_len = 1208 req_len; 1209 else 1210 wqe->fcp_iwrite.initial_xfer_len = 1211 pnode->nvme_fb_size; 1212 } else { 1213 wqe->fcp_iwrite.initial_xfer_len = 0; 1214 } 1215 cstat->output_requests++; 1216 } else { 1217 /* From the iread template, initialize words 7 - 11 */ 1218 memcpy(&wqe->words[7], 1219 &lpfc_iread_cmd_template.words[7], 1220 sizeof(uint32_t) * 5); 1221 1222 /* Word 4 */ 1223 wqe->fcp_iread.total_xfer_len = nCmd->payload_length; 1224 1225 /* Word 5 */ 1226 wqe->fcp_iread.rsrvd5 = 0; 1227 1228 /* For a CMF Managed port, iod must be zero'ed */ 1229 if (phba->cmf_active_mode == LPFC_CFG_MANAGED) 1230 bf_set(wqe_iod, &wqe->fcp_iread.wqe_com, 1231 LPFC_WQE_IOD_NONE); 1232 cstat->input_requests++; 1233 } 1234 } else { 1235 /* From the icmnd template, initialize words 4 - 11 */ 1236 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4], 1237 sizeof(uint32_t) * 8); 1238 cstat->control_requests++; 1239 } 1240 1241 if (pnode->nlp_nvme_info & NLP_NVME_NSLER) 1242 bf_set(wqe_erp, &wqe->generic.wqe_com, 1); 1243 /* 1244 * Finish initializing those WQE fields that are independent 1245 * of the nvme_cmnd request_buffer 1246 */ 1247 1248 /* Word 3 */ 1249 bf_set(payload_offset_len, &wqe->fcp_icmd, 1250 (nCmd->rsplen + nCmd->cmdlen)); 1251 1252 /* Word 6 */ 1253 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com, 1254 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]); 1255 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag); 1256 1257 /* Word 8 */ 1258 wqe->generic.wqe_com.abort_tag = pwqeq->iotag; 1259 1260 /* Word 9 */ 1261 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag); 1262 1263 /* Word 10 */ 1264 bf_set(wqe_xchg, &wqe->fcp_iwrite.wqe_com, LPFC_NVME_XCHG); 1265 1266 /* Words 13 14 15 are for PBDE support */ 1267 1268 pwqeq->vport = vport; 1269 return 0; 1270 } 1271 1272 1273 /** 1274 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO 1275 * @vport: pointer to a host virtual N_Port data structure 1276 * @lpfc_ncmd: Pointer to lpfc scsi command 1277 * 1278 * Driver registers this routine as it io request handler. This 1279 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1280 * data structure to the rport indicated in @lpfc_nvme_rport. 1281 * 1282 * Return value : 1283 * 0 - Success 1284 * TODO: What are the failure codes. 1285 **/ 1286 static int 1287 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport, 1288 struct lpfc_io_buf *lpfc_ncmd) 1289 { 1290 struct lpfc_hba *phba = vport->phba; 1291 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1292 union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe; 1293 struct sli4_sge *sgl = lpfc_ncmd->dma_sgl; 1294 struct sli4_hybrid_sgl *sgl_xtra = NULL; 1295 struct scatterlist *data_sg; 1296 struct sli4_sge *first_data_sgl; 1297 struct ulp_bde64 *bde; 1298 dma_addr_t physaddr = 0; 1299 uint32_t num_bde = 0; 1300 uint32_t dma_len = 0; 1301 uint32_t dma_offset = 0; 1302 int nseg, i, j; 1303 bool lsp_just_set = false; 1304 1305 /* Fix up the command and response DMA stuff. */ 1306 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd); 1307 1308 /* 1309 * There are three possibilities here - use scatter-gather segment, use 1310 * the single mapping, or neither. 1311 */ 1312 if (nCmd->sg_cnt) { 1313 /* 1314 * Jump over the cmd and rsp SGEs. The fix routine 1315 * has already adjusted for this. 1316 */ 1317 sgl += 2; 1318 1319 first_data_sgl = sgl; 1320 lpfc_ncmd->seg_cnt = nCmd->sg_cnt; 1321 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) { 1322 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT, 1323 "6058 Too many sg segments from " 1324 "NVME Transport. Max %d, " 1325 "nvmeIO sg_cnt %d\n", 1326 phba->cfg_nvme_seg_cnt + 1, 1327 lpfc_ncmd->seg_cnt); 1328 lpfc_ncmd->seg_cnt = 0; 1329 return 1; 1330 } 1331 1332 /* 1333 * The driver established a maximum scatter-gather segment count 1334 * during probe that limits the number of sg elements in any 1335 * single nvme command. Just run through the seg_cnt and format 1336 * the sge's. 1337 */ 1338 nseg = nCmd->sg_cnt; 1339 data_sg = nCmd->first_sgl; 1340 1341 /* for tracking the segment boundaries */ 1342 j = 2; 1343 for (i = 0; i < nseg; i++) { 1344 if (data_sg == NULL) { 1345 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT, 1346 "6059 dptr err %d, nseg %d\n", 1347 i, nseg); 1348 lpfc_ncmd->seg_cnt = 0; 1349 return 1; 1350 } 1351 1352 sgl->word2 = 0; 1353 if ((num_bde + 1) == nseg) { 1354 bf_set(lpfc_sli4_sge_last, sgl, 1); 1355 bf_set(lpfc_sli4_sge_type, sgl, 1356 LPFC_SGE_TYPE_DATA); 1357 } else { 1358 bf_set(lpfc_sli4_sge_last, sgl, 0); 1359 1360 /* expand the segment */ 1361 if (!lsp_just_set && 1362 !((j + 1) % phba->border_sge_num) && 1363 ((nseg - 1) != i)) { 1364 /* set LSP type */ 1365 bf_set(lpfc_sli4_sge_type, sgl, 1366 LPFC_SGE_TYPE_LSP); 1367 1368 sgl_xtra = lpfc_get_sgl_per_hdwq( 1369 phba, lpfc_ncmd); 1370 1371 if (unlikely(!sgl_xtra)) { 1372 lpfc_ncmd->seg_cnt = 0; 1373 return 1; 1374 } 1375 sgl->addr_lo = cpu_to_le32(putPaddrLow( 1376 sgl_xtra->dma_phys_sgl)); 1377 sgl->addr_hi = cpu_to_le32(putPaddrHigh( 1378 sgl_xtra->dma_phys_sgl)); 1379 1380 } else { 1381 bf_set(lpfc_sli4_sge_type, sgl, 1382 LPFC_SGE_TYPE_DATA); 1383 } 1384 } 1385 1386 if (!(bf_get(lpfc_sli4_sge_type, sgl) & 1387 LPFC_SGE_TYPE_LSP)) { 1388 if ((nseg - 1) == i) 1389 bf_set(lpfc_sli4_sge_last, sgl, 1); 1390 1391 physaddr = data_sg->dma_address; 1392 dma_len = data_sg->length; 1393 sgl->addr_lo = cpu_to_le32( 1394 putPaddrLow(physaddr)); 1395 sgl->addr_hi = cpu_to_le32( 1396 putPaddrHigh(physaddr)); 1397 1398 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset); 1399 sgl->word2 = cpu_to_le32(sgl->word2); 1400 sgl->sge_len = cpu_to_le32(dma_len); 1401 1402 dma_offset += dma_len; 1403 data_sg = sg_next(data_sg); 1404 1405 sgl++; 1406 1407 lsp_just_set = false; 1408 } else { 1409 sgl->word2 = cpu_to_le32(sgl->word2); 1410 1411 sgl->sge_len = cpu_to_le32( 1412 phba->cfg_sg_dma_buf_size); 1413 1414 sgl = (struct sli4_sge *)sgl_xtra->dma_sgl; 1415 i = i - 1; 1416 1417 lsp_just_set = true; 1418 } 1419 1420 j++; 1421 } 1422 if (phba->cfg_enable_pbde) { 1423 /* Use PBDE support for first SGL only, offset == 0 */ 1424 /* Words 13-15 */ 1425 bde = (struct ulp_bde64 *) 1426 &wqe->words[13]; 1427 bde->addrLow = first_data_sgl->addr_lo; 1428 bde->addrHigh = first_data_sgl->addr_hi; 1429 bde->tus.f.bdeSize = 1430 le32_to_cpu(first_data_sgl->sge_len); 1431 bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64; 1432 bde->tus.w = cpu_to_le32(bde->tus.w); 1433 1434 /* Word 11 */ 1435 bf_set(wqe_pbde, &wqe->generic.wqe_com, 1); 1436 } else { 1437 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3)); 1438 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0); 1439 } 1440 1441 } else { 1442 lpfc_ncmd->seg_cnt = 0; 1443 1444 /* For this clause to be valid, the payload_length 1445 * and sg_cnt must zero. 1446 */ 1447 if (nCmd->payload_length != 0) { 1448 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT, 1449 "6063 NVME DMA Prep Err: sg_cnt %d " 1450 "payload_length x%x\n", 1451 nCmd->sg_cnt, nCmd->payload_length); 1452 return 1; 1453 } 1454 } 1455 return 0; 1456 } 1457 1458 /** 1459 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO 1460 * @pnvme_lport: Pointer to the driver's local port data 1461 * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1462 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1463 * @pnvme_fcreq: IO request from nvme fc to driver. 1464 * 1465 * Driver registers this routine as it io request handler. This 1466 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1467 * data structure to the rport indicated in @lpfc_nvme_rport. 1468 * 1469 * Return value : 1470 * 0 - Success 1471 * TODO: What are the failure codes. 1472 **/ 1473 static int 1474 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport, 1475 struct nvme_fc_remote_port *pnvme_rport, 1476 void *hw_queue_handle, 1477 struct nvmefc_fcp_req *pnvme_fcreq) 1478 { 1479 int ret = 0; 1480 int expedite = 0; 1481 int idx, cpu; 1482 struct lpfc_nvme_lport *lport; 1483 struct lpfc_fc4_ctrl_stat *cstat; 1484 struct lpfc_vport *vport; 1485 struct lpfc_hba *phba; 1486 struct lpfc_nodelist *ndlp; 1487 struct lpfc_io_buf *lpfc_ncmd; 1488 struct lpfc_nvme_rport *rport; 1489 struct lpfc_nvme_qhandle *lpfc_queue_info; 1490 struct lpfc_nvme_fcpreq_priv *freqpriv; 1491 struct nvme_common_command *sqe; 1492 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1493 uint64_t start = 0; 1494 #endif 1495 1496 /* Validate pointers. LLDD fault handling with transport does 1497 * have timing races. 1498 */ 1499 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1500 if (unlikely(!lport)) { 1501 ret = -EINVAL; 1502 goto out_fail; 1503 } 1504 1505 vport = lport->vport; 1506 1507 if (unlikely(!hw_queue_handle)) { 1508 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1509 "6117 Fail IO, NULL hw_queue_handle\n"); 1510 atomic_inc(&lport->xmt_fcp_err); 1511 ret = -EBUSY; 1512 goto out_fail; 1513 } 1514 1515 phba = vport->phba; 1516 1517 if (unlikely(vport->load_flag & FC_UNLOADING)) { 1518 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1519 "6124 Fail IO, Driver unload\n"); 1520 atomic_inc(&lport->xmt_fcp_err); 1521 ret = -ENODEV; 1522 goto out_fail; 1523 } 1524 1525 freqpriv = pnvme_fcreq->private; 1526 if (unlikely(!freqpriv)) { 1527 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1528 "6158 Fail IO, NULL request data\n"); 1529 atomic_inc(&lport->xmt_fcp_err); 1530 ret = -EINVAL; 1531 goto out_fail; 1532 } 1533 1534 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1535 if (phba->ktime_on) 1536 start = ktime_get_ns(); 1537 #endif 1538 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1539 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle; 1540 1541 /* 1542 * Catch race where our node has transitioned, but the 1543 * transport is still transitioning. 1544 */ 1545 ndlp = rport->ndlp; 1546 if (!ndlp) { 1547 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1548 "6053 Busy IO, ndlp not ready: rport x%px " 1549 "ndlp x%px, DID x%06x\n", 1550 rport, ndlp, pnvme_rport->port_id); 1551 atomic_inc(&lport->xmt_fcp_err); 1552 ret = -EBUSY; 1553 goto out_fail; 1554 } 1555 1556 /* The remote node has to be a mapped target or it's an error. */ 1557 if ((ndlp->nlp_type & NLP_NVME_TARGET) && 1558 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) { 1559 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1560 "6036 Fail IO, DID x%06x not ready for " 1561 "IO. State x%x, Type x%x Flg x%x\n", 1562 pnvme_rport->port_id, 1563 ndlp->nlp_state, ndlp->nlp_type, 1564 ndlp->fc4_xpt_flags); 1565 atomic_inc(&lport->xmt_fcp_bad_ndlp); 1566 ret = -EBUSY; 1567 goto out_fail; 1568 1569 } 1570 1571 /* Currently only NVME Keep alive commands should be expedited 1572 * if the driver runs out of a resource. These should only be 1573 * issued on the admin queue, qidx 0 1574 */ 1575 if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) { 1576 sqe = &((struct nvme_fc_cmd_iu *) 1577 pnvme_fcreq->cmdaddr)->sqe.common; 1578 if (sqe->opcode == nvme_admin_keep_alive) 1579 expedite = 1; 1580 } 1581 1582 /* Check if IO qualifies for CMF */ 1583 if (phba->cmf_active_mode != LPFC_CFG_OFF && 1584 pnvme_fcreq->io_dir == NVMEFC_FCP_READ && 1585 pnvme_fcreq->payload_length) { 1586 ret = lpfc_update_cmf_cmd(phba, pnvme_fcreq->payload_length); 1587 if (ret) { 1588 ret = -EBUSY; 1589 goto out_fail; 1590 } 1591 /* Get start time for IO latency */ 1592 start = ktime_get_ns(); 1593 } 1594 1595 /* The node is shared with FCP IO, make sure the IO pending count does 1596 * not exceed the programmed depth. 1597 */ 1598 if (lpfc_ndlp_check_qdepth(phba, ndlp)) { 1599 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) && 1600 !expedite) { 1601 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1602 "6174 Fail IO, ndlp qdepth exceeded: " 1603 "idx %d DID %x pend %d qdepth %d\n", 1604 lpfc_queue_info->index, ndlp->nlp_DID, 1605 atomic_read(&ndlp->cmd_pending), 1606 ndlp->cmd_qdepth); 1607 atomic_inc(&lport->xmt_fcp_qdepth); 1608 ret = -EBUSY; 1609 goto out_fail1; 1610 } 1611 } 1612 1613 /* Lookup Hardware Queue index based on fcp_io_sched module parameter */ 1614 if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) { 1615 idx = lpfc_queue_info->index; 1616 } else { 1617 cpu = raw_smp_processor_id(); 1618 idx = phba->sli4_hba.cpu_map[cpu].hdwq; 1619 } 1620 1621 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite); 1622 if (lpfc_ncmd == NULL) { 1623 atomic_inc(&lport->xmt_fcp_noxri); 1624 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1625 "6065 Fail IO, driver buffer pool is empty: " 1626 "idx %d DID %x\n", 1627 lpfc_queue_info->index, ndlp->nlp_DID); 1628 ret = -EBUSY; 1629 goto out_fail1; 1630 } 1631 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1632 if (start) { 1633 lpfc_ncmd->ts_cmd_start = start; 1634 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd; 1635 } else { 1636 lpfc_ncmd->ts_cmd_start = 0; 1637 } 1638 #endif 1639 lpfc_ncmd->rx_cmd_start = start; 1640 1641 /* 1642 * Store the data needed by the driver to issue, abort, and complete 1643 * an IO. 1644 * Do not let the IO hang out forever. There is no midlayer issuing 1645 * an abort so inform the FW of the maximum IO pending time. 1646 */ 1647 freqpriv->nvme_buf = lpfc_ncmd; 1648 lpfc_ncmd->nvmeCmd = pnvme_fcreq; 1649 lpfc_ncmd->ndlp = ndlp; 1650 lpfc_ncmd->qidx = lpfc_queue_info->qidx; 1651 1652 /* 1653 * Issue the IO on the WQ indicated by index in the hw_queue_handle. 1654 * This identfier was create in our hardware queue create callback 1655 * routine. The driver now is dependent on the IO queue steering from 1656 * the transport. We are trusting the upper NVME layers know which 1657 * index to use and that they have affinitized a CPU to this hardware 1658 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ. 1659 */ 1660 lpfc_ncmd->cur_iocbq.hba_wqidx = idx; 1661 cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat; 1662 1663 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat); 1664 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd); 1665 if (ret) { 1666 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1667 "6175 Fail IO, Prep DMA: " 1668 "idx %d DID %x\n", 1669 lpfc_queue_info->index, ndlp->nlp_DID); 1670 atomic_inc(&lport->xmt_fcp_err); 1671 ret = -ENOMEM; 1672 goto out_free_nvme_buf; 1673 } 1674 1675 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n", 1676 lpfc_ncmd->cur_iocbq.sli4_xritag, 1677 lpfc_queue_info->index, ndlp->nlp_DID); 1678 1679 ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq); 1680 if (ret) { 1681 atomic_inc(&lport->xmt_fcp_wqerr); 1682 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1683 "6113 Fail IO, Could not issue WQE err %x " 1684 "sid: x%x did: x%x oxid: x%x\n", 1685 ret, vport->fc_myDID, ndlp->nlp_DID, 1686 lpfc_ncmd->cur_iocbq.sli4_xritag); 1687 goto out_free_nvme_buf; 1688 } 1689 1690 if (phba->cfg_xri_rebalancing) 1691 lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no); 1692 1693 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1694 if (lpfc_ncmd->ts_cmd_start) 1695 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns(); 1696 1697 if (phba->hdwqstat_on & LPFC_CHECK_NVME_IO) { 1698 cpu = raw_smp_processor_id(); 1699 this_cpu_inc(phba->sli4_hba.c_stat->xmt_io); 1700 lpfc_ncmd->cpu = cpu; 1701 if (idx != cpu) 1702 lpfc_printf_vlog(vport, 1703 KERN_INFO, LOG_NVME_IOERR, 1704 "6702 CPU Check cmd: " 1705 "cpu %d wq %d\n", 1706 lpfc_ncmd->cpu, 1707 lpfc_queue_info->index); 1708 } 1709 #endif 1710 return 0; 1711 1712 out_free_nvme_buf: 1713 if (lpfc_ncmd->nvmeCmd->sg_cnt) { 1714 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE) 1715 cstat->output_requests--; 1716 else 1717 cstat->input_requests--; 1718 } else 1719 cstat->control_requests--; 1720 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1721 out_fail1: 1722 lpfc_update_cmf_cmpl(phba, LPFC_CGN_NOT_SENT, 1723 pnvme_fcreq->payload_length, NULL); 1724 out_fail: 1725 return ret; 1726 } 1727 1728 /** 1729 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request. 1730 * @phba: Pointer to HBA context object 1731 * @cmdiocb: Pointer to command iocb object. 1732 * @abts_cmpl: Pointer to wcqe complete object. 1733 * 1734 * This is the callback function for any NVME FCP IO that was aborted. 1735 * 1736 * Return value: 1737 * None 1738 **/ 1739 void 1740 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb, 1741 struct lpfc_wcqe_complete *abts_cmpl) 1742 { 1743 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 1744 "6145 ABORT_XRI_CN completing on rpi x%x " 1745 "original iotag x%x, abort cmd iotag x%x " 1746 "req_tag x%x, status x%x, hwstatus x%x\n", 1747 cmdiocb->iocb.un.acxri.abortContextTag, 1748 cmdiocb->iocb.un.acxri.abortIoTag, 1749 cmdiocb->iotag, 1750 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl), 1751 bf_get(lpfc_wcqe_c_status, abts_cmpl), 1752 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl)); 1753 lpfc_sli_release_iocbq(phba, cmdiocb); 1754 } 1755 1756 /** 1757 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS 1758 * @pnvme_lport: Pointer to the driver's local port data 1759 * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1760 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1761 * @pnvme_fcreq: IO request from nvme fc to driver. 1762 * 1763 * Driver registers this routine as its nvme request io abort handler. This 1764 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq 1765 * data structure to the rport indicated in @lpfc_nvme_rport. This routine 1766 * is executed asynchronously - one the target is validated as "MAPPED" and 1767 * ready for IO, the driver issues the abort request and returns. 1768 * 1769 * Return value: 1770 * None 1771 **/ 1772 static void 1773 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport, 1774 struct nvme_fc_remote_port *pnvme_rport, 1775 void *hw_queue_handle, 1776 struct nvmefc_fcp_req *pnvme_fcreq) 1777 { 1778 struct lpfc_nvme_lport *lport; 1779 struct lpfc_vport *vport; 1780 struct lpfc_hba *phba; 1781 struct lpfc_io_buf *lpfc_nbuf; 1782 struct lpfc_iocbq *nvmereq_wqe; 1783 struct lpfc_nvme_fcpreq_priv *freqpriv; 1784 unsigned long flags; 1785 int ret_val; 1786 1787 /* Validate pointers. LLDD fault handling with transport does 1788 * have timing races. 1789 */ 1790 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1791 if (unlikely(!lport)) 1792 return; 1793 1794 vport = lport->vport; 1795 1796 if (unlikely(!hw_queue_handle)) { 1797 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1798 "6129 Fail Abort, HW Queue Handle NULL.\n"); 1799 return; 1800 } 1801 1802 phba = vport->phba; 1803 freqpriv = pnvme_fcreq->private; 1804 1805 if (unlikely(!freqpriv)) 1806 return; 1807 if (vport->load_flag & FC_UNLOADING) 1808 return; 1809 1810 /* Announce entry to new IO submit field. */ 1811 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1812 "6002 Abort Request to rport DID x%06x " 1813 "for nvme_fc_req x%px\n", 1814 pnvme_rport->port_id, 1815 pnvme_fcreq); 1816 1817 /* If the hba is getting reset, this flag is set. It is 1818 * cleared when the reset is complete and rings reestablished. 1819 */ 1820 spin_lock_irqsave(&phba->hbalock, flags); 1821 /* driver queued commands are in process of being flushed */ 1822 if (phba->hba_flag & HBA_IOQ_FLUSH) { 1823 spin_unlock_irqrestore(&phba->hbalock, flags); 1824 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1825 "6139 Driver in reset cleanup - flushing " 1826 "NVME Req now. hba_flag x%x\n", 1827 phba->hba_flag); 1828 return; 1829 } 1830 1831 lpfc_nbuf = freqpriv->nvme_buf; 1832 if (!lpfc_nbuf) { 1833 spin_unlock_irqrestore(&phba->hbalock, flags); 1834 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1835 "6140 NVME IO req has no matching lpfc nvme " 1836 "io buffer. Skipping abort req.\n"); 1837 return; 1838 } else if (!lpfc_nbuf->nvmeCmd) { 1839 spin_unlock_irqrestore(&phba->hbalock, flags); 1840 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1841 "6141 lpfc NVME IO req has no nvme_fcreq " 1842 "io buffer. Skipping abort req.\n"); 1843 return; 1844 } 1845 nvmereq_wqe = &lpfc_nbuf->cur_iocbq; 1846 1847 /* Guard against IO completion being called at same time */ 1848 spin_lock(&lpfc_nbuf->buf_lock); 1849 1850 /* 1851 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's 1852 * state must match the nvme_fcreq passed by the nvme 1853 * transport. If they don't match, it is likely the driver 1854 * has already completed the NVME IO and the nvme transport 1855 * has not seen it yet. 1856 */ 1857 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) { 1858 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1859 "6143 NVME req mismatch: " 1860 "lpfc_nbuf x%px nvmeCmd x%px, " 1861 "pnvme_fcreq x%px. Skipping Abort xri x%x\n", 1862 lpfc_nbuf, lpfc_nbuf->nvmeCmd, 1863 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1864 goto out_unlock; 1865 } 1866 1867 /* Don't abort IOs no longer on the pending queue. */ 1868 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) { 1869 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1870 "6142 NVME IO req x%px not queued - skipping " 1871 "abort req xri x%x\n", 1872 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1873 goto out_unlock; 1874 } 1875 1876 atomic_inc(&lport->xmt_fcp_abort); 1877 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n", 1878 nvmereq_wqe->sli4_xritag, 1879 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id); 1880 1881 /* Outstanding abort is in progress */ 1882 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) { 1883 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1884 "6144 Outstanding NVME I/O Abort Request " 1885 "still pending on nvme_fcreq x%px, " 1886 "lpfc_ncmd x%px xri x%x\n", 1887 pnvme_fcreq, lpfc_nbuf, 1888 nvmereq_wqe->sli4_xritag); 1889 goto out_unlock; 1890 } 1891 1892 ret_val = lpfc_sli4_issue_abort_iotag(phba, nvmereq_wqe, 1893 lpfc_nvme_abort_fcreq_cmpl); 1894 1895 spin_unlock(&lpfc_nbuf->buf_lock); 1896 spin_unlock_irqrestore(&phba->hbalock, flags); 1897 1898 /* Make sure HBA is alive */ 1899 lpfc_issue_hb_tmo(phba); 1900 1901 if (ret_val != WQE_SUCCESS) { 1902 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 1903 "6137 Failed abts issue_wqe with status x%x " 1904 "for nvme_fcreq x%px.\n", 1905 ret_val, pnvme_fcreq); 1906 return; 1907 } 1908 1909 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1910 "6138 Transport Abort NVME Request Issued for " 1911 "ox_id x%x\n", 1912 nvmereq_wqe->sli4_xritag); 1913 return; 1914 1915 out_unlock: 1916 spin_unlock(&lpfc_nbuf->buf_lock); 1917 spin_unlock_irqrestore(&phba->hbalock, flags); 1918 return; 1919 } 1920 1921 /* Declare and initialization an instance of the FC NVME template. */ 1922 static struct nvme_fc_port_template lpfc_nvme_template = { 1923 /* initiator-based functions */ 1924 .localport_delete = lpfc_nvme_localport_delete, 1925 .remoteport_delete = lpfc_nvme_remoteport_delete, 1926 .create_queue = lpfc_nvme_create_queue, 1927 .delete_queue = lpfc_nvme_delete_queue, 1928 .ls_req = lpfc_nvme_ls_req, 1929 .fcp_io = lpfc_nvme_fcp_io_submit, 1930 .ls_abort = lpfc_nvme_ls_abort, 1931 .fcp_abort = lpfc_nvme_fcp_abort, 1932 .xmt_ls_rsp = lpfc_nvme_xmt_ls_rsp, 1933 1934 .max_hw_queues = 1, 1935 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1936 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1937 .dma_boundary = 0xFFFFFFFF, 1938 1939 /* Sizes of additional private data for data structures. 1940 * No use for the last two sizes at this time. 1941 */ 1942 .local_priv_sz = sizeof(struct lpfc_nvme_lport), 1943 .remote_priv_sz = sizeof(struct lpfc_nvme_rport), 1944 .lsrqst_priv_sz = 0, 1945 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv), 1946 }; 1947 1948 /* 1949 * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA 1950 * 1951 * This routine removes a nvme buffer from head of @hdwq io_buf_list 1952 * and returns to caller. 1953 * 1954 * Return codes: 1955 * NULL - Error 1956 * Pointer to lpfc_nvme_buf - Success 1957 **/ 1958 static struct lpfc_io_buf * 1959 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 1960 int idx, int expedite) 1961 { 1962 struct lpfc_io_buf *lpfc_ncmd; 1963 struct lpfc_sli4_hdw_queue *qp; 1964 struct sli4_sge *sgl; 1965 struct lpfc_iocbq *pwqeq; 1966 union lpfc_wqe128 *wqe; 1967 1968 lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite); 1969 1970 if (lpfc_ncmd) { 1971 pwqeq = &(lpfc_ncmd->cur_iocbq); 1972 wqe = &pwqeq->wqe; 1973 1974 /* Setup key fields in buffer that may have been changed 1975 * if other protocols used this buffer. 1976 */ 1977 pwqeq->iocb_flag = LPFC_IO_NVME; 1978 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl; 1979 lpfc_ncmd->start_time = jiffies; 1980 lpfc_ncmd->flags = 0; 1981 1982 /* Rsp SGE will be filled in when we rcv an IO 1983 * from the NVME Layer to be sent. 1984 * The cmd is going to be embedded so we need a SKIP SGE. 1985 */ 1986 sgl = lpfc_ncmd->dma_sgl; 1987 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP); 1988 bf_set(lpfc_sli4_sge_last, sgl, 0); 1989 sgl->word2 = cpu_to_le32(sgl->word2); 1990 /* Fill in word 3 / sgl_len during cmd submission */ 1991 1992 /* Initialize 64 bytes only */ 1993 memset(wqe, 0, sizeof(union lpfc_wqe)); 1994 1995 if (lpfc_ndlp_check_qdepth(phba, ndlp)) { 1996 atomic_inc(&ndlp->cmd_pending); 1997 lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH; 1998 } 1999 2000 } else { 2001 qp = &phba->sli4_hba.hdwq[idx]; 2002 qp->empty_io_bufs++; 2003 } 2004 2005 return lpfc_ncmd; 2006 } 2007 2008 /** 2009 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list. 2010 * @phba: The Hba for which this call is being executed. 2011 * @lpfc_ncmd: The nvme buffer which is being released. 2012 * 2013 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba 2014 * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer 2015 * and cannot be reused for at least RA_TOV amount of time if it was 2016 * aborted. 2017 **/ 2018 static void 2019 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd) 2020 { 2021 struct lpfc_sli4_hdw_queue *qp; 2022 unsigned long iflag = 0; 2023 2024 if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp) 2025 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending); 2026 2027 lpfc_ncmd->ndlp = NULL; 2028 lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH; 2029 2030 qp = lpfc_ncmd->hdwq; 2031 if (unlikely(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) { 2032 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2033 "6310 XB release deferred for " 2034 "ox_id x%x on reqtag x%x\n", 2035 lpfc_ncmd->cur_iocbq.sli4_xritag, 2036 lpfc_ncmd->cur_iocbq.iotag); 2037 2038 spin_lock_irqsave(&qp->abts_io_buf_list_lock, iflag); 2039 list_add_tail(&lpfc_ncmd->list, 2040 &qp->lpfc_abts_io_buf_list); 2041 qp->abts_nvme_io_bufs++; 2042 spin_unlock_irqrestore(&qp->abts_io_buf_list_lock, iflag); 2043 } else 2044 lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp); 2045 } 2046 2047 /** 2048 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance. 2049 * @vport: the lpfc_vport instance requesting a localport. 2050 * 2051 * This routine is invoked to create an nvme localport instance to bind 2052 * to the nvme_fc_transport. It is called once during driver load 2053 * like lpfc_create_shost after all other services are initialized. 2054 * It requires a vport, vpi, and wwns at call time. Other localport 2055 * parameters are modified as the driver's FCID and the Fabric WWN 2056 * are established. 2057 * 2058 * Return codes 2059 * 0 - successful 2060 * -ENOMEM - no heap memory available 2061 * other values - from nvme registration upcall 2062 **/ 2063 int 2064 lpfc_nvme_create_localport(struct lpfc_vport *vport) 2065 { 2066 int ret = 0; 2067 struct lpfc_hba *phba = vport->phba; 2068 struct nvme_fc_port_info nfcp_info; 2069 struct nvme_fc_local_port *localport; 2070 struct lpfc_nvme_lport *lport; 2071 2072 /* Initialize this localport instance. The vport wwn usage ensures 2073 * that NPIV is accounted for. 2074 */ 2075 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info)); 2076 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR; 2077 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn); 2078 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn); 2079 2080 /* We need to tell the transport layer + 1 because it takes page 2081 * alignment into account. When space for the SGL is allocated we 2082 * allocate + 3, one for cmd, one for rsp and one for this alignment 2083 */ 2084 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1; 2085 2086 /* Advertise how many hw queues we support based on cfg_hdw_queue, 2087 * which will not exceed cpu count. 2088 */ 2089 lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue; 2090 2091 if (!IS_ENABLED(CONFIG_NVME_FC)) 2092 return ret; 2093 2094 /* localport is allocated from the stack, but the registration 2095 * call allocates heap memory as well as the private area. 2096 */ 2097 2098 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template, 2099 &vport->phba->pcidev->dev, &localport); 2100 if (!ret) { 2101 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC, 2102 "6005 Successfully registered local " 2103 "NVME port num %d, localP x%px, private " 2104 "x%px, sg_seg %d\n", 2105 localport->port_num, localport, 2106 localport->private, 2107 lpfc_nvme_template.max_sgl_segments); 2108 2109 /* Private is our lport size declared in the template. */ 2110 lport = (struct lpfc_nvme_lport *)localport->private; 2111 vport->localport = localport; 2112 lport->vport = vport; 2113 vport->nvmei_support = 1; 2114 2115 atomic_set(&lport->xmt_fcp_noxri, 0); 2116 atomic_set(&lport->xmt_fcp_bad_ndlp, 0); 2117 atomic_set(&lport->xmt_fcp_qdepth, 0); 2118 atomic_set(&lport->xmt_fcp_err, 0); 2119 atomic_set(&lport->xmt_fcp_wqerr, 0); 2120 atomic_set(&lport->xmt_fcp_abort, 0); 2121 atomic_set(&lport->xmt_ls_abort, 0); 2122 atomic_set(&lport->xmt_ls_err, 0); 2123 atomic_set(&lport->cmpl_fcp_xb, 0); 2124 atomic_set(&lport->cmpl_fcp_err, 0); 2125 atomic_set(&lport->cmpl_ls_xb, 0); 2126 atomic_set(&lport->cmpl_ls_err, 0); 2127 2128 atomic_set(&lport->fc4NvmeLsRequests, 0); 2129 atomic_set(&lport->fc4NvmeLsCmpls, 0); 2130 } 2131 2132 return ret; 2133 } 2134 2135 #if (IS_ENABLED(CONFIG_NVME_FC)) 2136 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg. 2137 * 2138 * The driver has to wait for the host nvme transport to callback 2139 * indicating the localport has successfully unregistered all 2140 * resources. Since this is an uninterruptible wait, loop every ten 2141 * seconds and print a message indicating no progress. 2142 * 2143 * An uninterruptible wait is used because of the risk of transport-to- 2144 * driver state mismatch. 2145 */ 2146 static void 2147 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport, 2148 struct lpfc_nvme_lport *lport, 2149 struct completion *lport_unreg_cmp) 2150 { 2151 u32 wait_tmo; 2152 int ret, i, pending = 0; 2153 struct lpfc_sli_ring *pring; 2154 struct lpfc_hba *phba = vport->phba; 2155 struct lpfc_sli4_hdw_queue *qp; 2156 int abts_scsi, abts_nvme; 2157 2158 /* Host transport has to clean up and confirm requiring an indefinite 2159 * wait. Print a message if a 10 second wait expires and renew the 2160 * wait. This is unexpected. 2161 */ 2162 wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000); 2163 while (true) { 2164 ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo); 2165 if (unlikely(!ret)) { 2166 pending = 0; 2167 abts_scsi = 0; 2168 abts_nvme = 0; 2169 for (i = 0; i < phba->cfg_hdw_queue; i++) { 2170 qp = &phba->sli4_hba.hdwq[i]; 2171 pring = qp->io_wq->pring; 2172 if (!pring) 2173 continue; 2174 pending += pring->txcmplq_cnt; 2175 abts_scsi += qp->abts_scsi_io_bufs; 2176 abts_nvme += qp->abts_nvme_io_bufs; 2177 } 2178 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 2179 "6176 Lport x%px Localport x%px wait " 2180 "timed out. Pending %d [%d:%d]. " 2181 "Renewing.\n", 2182 lport, vport->localport, pending, 2183 abts_scsi, abts_nvme); 2184 continue; 2185 } 2186 break; 2187 } 2188 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 2189 "6177 Lport x%px Localport x%px Complete Success\n", 2190 lport, vport->localport); 2191 } 2192 #endif 2193 2194 /** 2195 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. 2196 * @vport: pointer to a host virtual N_Port data structure 2197 * 2198 * This routine is invoked to destroy all lports bound to the phba. 2199 * The lport memory was allocated by the nvme fc transport and is 2200 * released there. This routine ensures all rports bound to the 2201 * lport have been disconnected. 2202 * 2203 **/ 2204 void 2205 lpfc_nvme_destroy_localport(struct lpfc_vport *vport) 2206 { 2207 #if (IS_ENABLED(CONFIG_NVME_FC)) 2208 struct nvme_fc_local_port *localport; 2209 struct lpfc_nvme_lport *lport; 2210 int ret; 2211 DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp); 2212 2213 if (vport->nvmei_support == 0) 2214 return; 2215 2216 localport = vport->localport; 2217 lport = (struct lpfc_nvme_lport *)localport->private; 2218 2219 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2220 "6011 Destroying NVME localport x%px\n", 2221 localport); 2222 2223 /* lport's rport list is clear. Unregister 2224 * lport and release resources. 2225 */ 2226 lport->lport_unreg_cmp = &lport_unreg_cmp; 2227 ret = nvme_fc_unregister_localport(localport); 2228 2229 /* Wait for completion. This either blocks 2230 * indefinitely or succeeds 2231 */ 2232 lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp); 2233 vport->localport = NULL; 2234 2235 /* Regardless of the unregister upcall response, clear 2236 * nvmei_support. All rports are unregistered and the 2237 * driver will clean up. 2238 */ 2239 vport->nvmei_support = 0; 2240 if (ret == 0) { 2241 lpfc_printf_vlog(vport, 2242 KERN_INFO, LOG_NVME_DISC, 2243 "6009 Unregistered lport Success\n"); 2244 } else { 2245 lpfc_printf_vlog(vport, 2246 KERN_INFO, LOG_NVME_DISC, 2247 "6010 Unregistered lport " 2248 "Failed, status x%x\n", 2249 ret); 2250 } 2251 #endif 2252 } 2253 2254 void 2255 lpfc_nvme_update_localport(struct lpfc_vport *vport) 2256 { 2257 #if (IS_ENABLED(CONFIG_NVME_FC)) 2258 struct nvme_fc_local_port *localport; 2259 struct lpfc_nvme_lport *lport; 2260 2261 localport = vport->localport; 2262 if (!localport) { 2263 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2264 "6710 Update NVME fail. No localport\n"); 2265 return; 2266 } 2267 lport = (struct lpfc_nvme_lport *)localport->private; 2268 if (!lport) { 2269 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2270 "6171 Update NVME fail. localP x%px, No lport\n", 2271 localport); 2272 return; 2273 } 2274 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2275 "6012 Update NVME lport x%px did x%x\n", 2276 localport, vport->fc_myDID); 2277 2278 localport->port_id = vport->fc_myDID; 2279 if (localport->port_id == 0) 2280 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY; 2281 else 2282 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR; 2283 2284 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2285 "6030 bound lport x%px to DID x%06x\n", 2286 lport, localport->port_id); 2287 #endif 2288 } 2289 2290 int 2291 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2292 { 2293 #if (IS_ENABLED(CONFIG_NVME_FC)) 2294 int ret = 0; 2295 struct nvme_fc_local_port *localport; 2296 struct lpfc_nvme_lport *lport; 2297 struct lpfc_nvme_rport *rport; 2298 struct lpfc_nvme_rport *oldrport; 2299 struct nvme_fc_remote_port *remote_port; 2300 struct nvme_fc_port_info rpinfo; 2301 struct lpfc_nodelist *prev_ndlp = NULL; 2302 struct fc_rport *srport = ndlp->rport; 2303 2304 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC, 2305 "6006 Register NVME PORT. DID x%06x nlptype x%x\n", 2306 ndlp->nlp_DID, ndlp->nlp_type); 2307 2308 localport = vport->localport; 2309 if (!localport) 2310 return 0; 2311 2312 lport = (struct lpfc_nvme_lport *)localport->private; 2313 2314 /* NVME rports are not preserved across devloss. 2315 * Just register this instance. Note, rpinfo->dev_loss_tmo 2316 * is left 0 to indicate accept transport defaults. The 2317 * driver communicates port role capabilities consistent 2318 * with the PRLI response data. 2319 */ 2320 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info)); 2321 rpinfo.port_id = ndlp->nlp_DID; 2322 if (ndlp->nlp_type & NLP_NVME_TARGET) 2323 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET; 2324 if (ndlp->nlp_type & NLP_NVME_INITIATOR) 2325 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR; 2326 2327 if (ndlp->nlp_type & NLP_NVME_DISCOVERY) 2328 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY; 2329 2330 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn); 2331 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn); 2332 if (srport) 2333 rpinfo.dev_loss_tmo = srport->dev_loss_tmo; 2334 else 2335 rpinfo.dev_loss_tmo = vport->cfg_devloss_tmo; 2336 2337 spin_lock_irq(&ndlp->lock); 2338 oldrport = lpfc_ndlp_get_nrport(ndlp); 2339 if (oldrport) { 2340 prev_ndlp = oldrport->ndlp; 2341 spin_unlock_irq(&ndlp->lock); 2342 } else { 2343 spin_unlock_irq(&ndlp->lock); 2344 if (!lpfc_nlp_get(ndlp)) { 2345 dev_warn(&vport->phba->pcidev->dev, 2346 "Warning - No node ref - exit register\n"); 2347 return 0; 2348 } 2349 } 2350 2351 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port); 2352 if (!ret) { 2353 /* If the ndlp already has an nrport, this is just 2354 * a resume of the existing rport. Else this is a 2355 * new rport. 2356 */ 2357 /* Guard against an unregister/reregister 2358 * race that leaves the WAIT flag set. 2359 */ 2360 spin_lock_irq(&ndlp->lock); 2361 ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT; 2362 ndlp->fc4_xpt_flags |= NVME_XPT_REGD; 2363 spin_unlock_irq(&ndlp->lock); 2364 rport = remote_port->private; 2365 if (oldrport) { 2366 2367 /* Sever the ndlp<->rport association 2368 * before dropping the ndlp ref from 2369 * register. 2370 */ 2371 spin_lock_irq(&ndlp->lock); 2372 ndlp->nrport = NULL; 2373 ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT; 2374 spin_unlock_irq(&ndlp->lock); 2375 rport->ndlp = NULL; 2376 rport->remoteport = NULL; 2377 2378 /* Reference only removed if previous NDLP is no longer 2379 * active. It might be just a swap and removing the 2380 * reference would cause a premature cleanup. 2381 */ 2382 if (prev_ndlp && prev_ndlp != ndlp) { 2383 if (!prev_ndlp->nrport) 2384 lpfc_nlp_put(prev_ndlp); 2385 } 2386 } 2387 2388 /* Clean bind the rport to the ndlp. */ 2389 rport->remoteport = remote_port; 2390 rport->lport = lport; 2391 rport->ndlp = ndlp; 2392 spin_lock_irq(&ndlp->lock); 2393 ndlp->nrport = rport; 2394 spin_unlock_irq(&ndlp->lock); 2395 lpfc_printf_vlog(vport, KERN_INFO, 2396 LOG_NVME_DISC | LOG_NODE, 2397 "6022 Bind lport x%px to remoteport x%px " 2398 "rport x%px WWNN 0x%llx, " 2399 "Rport WWPN 0x%llx DID " 2400 "x%06x Role x%x, ndlp %p prev_ndlp x%px\n", 2401 lport, remote_port, rport, 2402 rpinfo.node_name, rpinfo.port_name, 2403 rpinfo.port_id, rpinfo.port_role, 2404 ndlp, prev_ndlp); 2405 } else { 2406 lpfc_printf_vlog(vport, KERN_ERR, 2407 LOG_TRACE_EVENT, 2408 "6031 RemotePort Registration failed " 2409 "err: %d, DID x%06x\n", 2410 ret, ndlp->nlp_DID); 2411 } 2412 2413 return ret; 2414 #else 2415 return 0; 2416 #endif 2417 } 2418 2419 /* 2420 * lpfc_nvme_rescan_port - Check to see if we should rescan this remoteport 2421 * 2422 * If the ndlp represents an NVME Target, that we are logged into, 2423 * ping the NVME FC Transport layer to initiate a device rescan 2424 * on this remote NPort. 2425 */ 2426 void 2427 lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2428 { 2429 #if (IS_ENABLED(CONFIG_NVME_FC)) 2430 struct lpfc_nvme_rport *nrport; 2431 struct nvme_fc_remote_port *remoteport = NULL; 2432 2433 spin_lock_irq(&ndlp->lock); 2434 nrport = lpfc_ndlp_get_nrport(ndlp); 2435 if (nrport) 2436 remoteport = nrport->remoteport; 2437 spin_unlock_irq(&ndlp->lock); 2438 2439 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2440 "6170 Rescan NPort DID x%06x type x%x " 2441 "state x%x nrport x%px remoteport x%px\n", 2442 ndlp->nlp_DID, ndlp->nlp_type, ndlp->nlp_state, 2443 nrport, remoteport); 2444 2445 if (!nrport || !remoteport) 2446 goto rescan_exit; 2447 2448 /* Only rescan if we are an NVME target in the MAPPED state */ 2449 if (remoteport->port_role & FC_PORT_ROLE_NVME_DISCOVERY && 2450 ndlp->nlp_state == NLP_STE_MAPPED_NODE) { 2451 nvme_fc_rescan_remoteport(remoteport); 2452 2453 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 2454 "6172 NVME rescanned DID x%06x " 2455 "port_state x%x\n", 2456 ndlp->nlp_DID, remoteport->port_state); 2457 } 2458 return; 2459 rescan_exit: 2460 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2461 "6169 Skip NVME Rport Rescan, NVME remoteport " 2462 "unregistered\n"); 2463 #endif 2464 } 2465 2466 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. 2467 * 2468 * There is no notion of Devloss or rport recovery from the current 2469 * nvme_transport perspective. Loss of an rport just means IO cannot 2470 * be sent and recovery is completely up to the initator. 2471 * For now, the driver just unbinds the DID and port_role so that 2472 * no further IO can be issued. Changes are planned for later. 2473 * 2474 * Notes - the ndlp reference count is not decremented here since 2475 * since there is no nvme_transport api for devloss. Node ref count 2476 * is only adjusted in driver unload. 2477 */ 2478 void 2479 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2480 { 2481 #if (IS_ENABLED(CONFIG_NVME_FC)) 2482 int ret; 2483 struct nvme_fc_local_port *localport; 2484 struct lpfc_nvme_lport *lport; 2485 struct lpfc_nvme_rport *rport; 2486 struct nvme_fc_remote_port *remoteport = NULL; 2487 2488 localport = vport->localport; 2489 2490 /* This is fundamental error. The localport is always 2491 * available until driver unload. Just exit. 2492 */ 2493 if (!localport) 2494 return; 2495 2496 lport = (struct lpfc_nvme_lport *)localport->private; 2497 if (!lport) 2498 goto input_err; 2499 2500 spin_lock_irq(&ndlp->lock); 2501 rport = lpfc_ndlp_get_nrport(ndlp); 2502 if (rport) 2503 remoteport = rport->remoteport; 2504 spin_unlock_irq(&ndlp->lock); 2505 if (!remoteport) 2506 goto input_err; 2507 2508 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2509 "6033 Unreg nvme remoteport x%px, portname x%llx, " 2510 "port_id x%06x, portstate x%x port type x%x " 2511 "refcnt %d\n", 2512 remoteport, remoteport->port_name, 2513 remoteport->port_id, remoteport->port_state, 2514 ndlp->nlp_type, kref_read(&ndlp->kref)); 2515 2516 /* Sanity check ndlp type. Only call for NVME ports. Don't 2517 * clear any rport state until the transport calls back. 2518 */ 2519 2520 if (ndlp->nlp_type & NLP_NVME_TARGET) { 2521 /* No concern about the role change on the nvme remoteport. 2522 * The transport will update it. 2523 */ 2524 spin_lock_irq(&vport->phba->hbalock); 2525 ndlp->fc4_xpt_flags |= NVME_XPT_UNREG_WAIT; 2526 spin_unlock_irq(&vport->phba->hbalock); 2527 2528 /* Don't let the host nvme transport keep sending keep-alives 2529 * on this remoteport. Vport is unloading, no recovery. The 2530 * return values is ignored. The upcall is a courtesy to the 2531 * transport. 2532 */ 2533 if (vport->load_flag & FC_UNLOADING) 2534 (void)nvme_fc_set_remoteport_devloss(remoteport, 0); 2535 2536 ret = nvme_fc_unregister_remoteport(remoteport); 2537 2538 /* The driver no longer knows if the nrport memory is valid. 2539 * because the controller teardown process has begun and 2540 * is asynchronous. Break the binding in the ndlp. Also 2541 * remove the register ndlp reference to setup node release. 2542 */ 2543 ndlp->nrport = NULL; 2544 lpfc_nlp_put(ndlp); 2545 if (ret != 0) { 2546 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 2547 "6167 NVME unregister failed %d " 2548 "port_state x%x\n", 2549 ret, remoteport->port_state); 2550 } 2551 } 2552 return; 2553 2554 input_err: 2555 #endif 2556 lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT, 2557 "6168 State error: lport x%px, rport x%px FCID x%06x\n", 2558 vport->localport, ndlp->rport, ndlp->nlp_DID); 2559 } 2560 2561 /** 2562 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort 2563 * @phba: pointer to lpfc hba data structure. 2564 * @axri: pointer to the fcp xri abort wcqe structure. 2565 * @lpfc_ncmd: The nvme job structure for the request being aborted. 2566 * 2567 * This routine is invoked by the worker thread to process a SLI4 fast-path 2568 * NVME aborted xri. Aborted NVME IO commands are completed to the transport 2569 * here. 2570 **/ 2571 void 2572 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba, 2573 struct sli4_wcqe_xri_aborted *axri, 2574 struct lpfc_io_buf *lpfc_ncmd) 2575 { 2576 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri); 2577 struct nvmefc_fcp_req *nvme_cmd = NULL; 2578 struct lpfc_nodelist *ndlp = lpfc_ncmd->ndlp; 2579 2580 2581 if (ndlp) 2582 lpfc_sli4_abts_err_handler(phba, ndlp, axri); 2583 2584 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2585 "6311 nvme_cmd %p xri x%x tag x%x abort complete and " 2586 "xri released\n", 2587 lpfc_ncmd->nvmeCmd, xri, 2588 lpfc_ncmd->cur_iocbq.iotag); 2589 2590 /* Aborted NVME commands are required to not complete 2591 * before the abort exchange command fully completes. 2592 * Once completed, it is available via the put list. 2593 */ 2594 if (lpfc_ncmd->nvmeCmd) { 2595 nvme_cmd = lpfc_ncmd->nvmeCmd; 2596 nvme_cmd->done(nvme_cmd); 2597 lpfc_ncmd->nvmeCmd = NULL; 2598 } 2599 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2600 } 2601 2602 /** 2603 * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete 2604 * @phba: Pointer to HBA context object. 2605 * 2606 * This function flushes all wqes in the nvme rings and frees all resources 2607 * in the txcmplq. This function does not issue abort wqes for the IO 2608 * commands in txcmplq, they will just be returned with 2609 * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI 2610 * slot has been permanently disabled. 2611 **/ 2612 void 2613 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba) 2614 { 2615 struct lpfc_sli_ring *pring; 2616 u32 i, wait_cnt = 0; 2617 2618 if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq) 2619 return; 2620 2621 /* Cycle through all IO rings and make sure all outstanding 2622 * WQEs have been removed from the txcmplqs. 2623 */ 2624 for (i = 0; i < phba->cfg_hdw_queue; i++) { 2625 if (!phba->sli4_hba.hdwq[i].io_wq) 2626 continue; 2627 pring = phba->sli4_hba.hdwq[i].io_wq->pring; 2628 2629 if (!pring) 2630 continue; 2631 2632 /* Retrieve everything on the txcmplq */ 2633 while (!list_empty(&pring->txcmplq)) { 2634 msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1); 2635 wait_cnt++; 2636 2637 /* The sleep is 10mS. Every ten seconds, 2638 * dump a message. Something is wrong. 2639 */ 2640 if ((wait_cnt % 1000) == 0) { 2641 lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT, 2642 "6178 NVME IO not empty, " 2643 "cnt %d\n", wait_cnt); 2644 } 2645 } 2646 } 2647 2648 /* Make sure HBA is alive */ 2649 lpfc_issue_hb_tmo(phba); 2650 2651 } 2652 2653 void 2654 lpfc_nvme_cancel_iocb(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn, 2655 uint32_t stat, uint32_t param) 2656 { 2657 #if (IS_ENABLED(CONFIG_NVME_FC)) 2658 struct lpfc_io_buf *lpfc_ncmd; 2659 struct nvmefc_fcp_req *nCmd; 2660 struct lpfc_wcqe_complete wcqe; 2661 struct lpfc_wcqe_complete *wcqep = &wcqe; 2662 2663 lpfc_ncmd = (struct lpfc_io_buf *)pwqeIn->context1; 2664 if (!lpfc_ncmd) { 2665 lpfc_sli_release_iocbq(phba, pwqeIn); 2666 return; 2667 } 2668 /* For abort iocb just return, IO iocb will do a done call */ 2669 if (bf_get(wqe_cmnd, &pwqeIn->wqe.gen_req.wqe_com) == 2670 CMD_ABORT_XRI_CX) { 2671 lpfc_sli_release_iocbq(phba, pwqeIn); 2672 return; 2673 } 2674 2675 spin_lock(&lpfc_ncmd->buf_lock); 2676 nCmd = lpfc_ncmd->nvmeCmd; 2677 if (!nCmd) { 2678 spin_unlock(&lpfc_ncmd->buf_lock); 2679 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2680 return; 2681 } 2682 spin_unlock(&lpfc_ncmd->buf_lock); 2683 2684 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_IOERR, 2685 "6194 NVME Cancel xri %x\n", 2686 lpfc_ncmd->cur_iocbq.sli4_xritag); 2687 2688 wcqep->word0 = 0; 2689 bf_set(lpfc_wcqe_c_status, wcqep, stat); 2690 wcqep->parameter = param; 2691 wcqep->word3 = 0; /* xb is 0 */ 2692 2693 /* Call release with XB=1 to queue the IO into the abort list. */ 2694 if (phba->sli.sli_flag & LPFC_SLI_ACTIVE) 2695 bf_set(lpfc_wcqe_c_xb, wcqep, 1); 2696 2697 (pwqeIn->wqe_cmpl)(phba, pwqeIn, wcqep); 2698 #endif 2699 } 2700