xref: /openbmc/linux/drivers/scsi/lpfc/lpfc_nvme.c (revision aac28965)
1 /*******************************************************************
2  * This file is part of the Emulex Linux Device Driver for         *
3  * Fibre Channel Host Bus Adapters.                                *
4  * Copyright (C) 2017-2022 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 
97 	if (!vport || vport->load_flag & FC_UNLOADING ||
98 	    vport->phba->hba_flag & HBA_IOQ_FLUSH)
99 		return -ENODEV;
100 
101 	qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
102 	if (qhandle == NULL)
103 		return -ENOMEM;
104 
105 	qhandle->cpu_id = raw_smp_processor_id();
106 	qhandle->qidx = qidx;
107 	/*
108 	 * NVME qidx == 0 is the admin queue, so both admin queue
109 	 * and first IO queue will use MSI-X vector and associated
110 	 * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
111 	 */
112 	if (qidx) {
113 		str = "IO ";  /* IO queue */
114 		qhandle->index = ((qidx - 1) %
115 			lpfc_nvme_template.max_hw_queues);
116 	} else {
117 		str = "ADM";  /* Admin queue */
118 		qhandle->index = qidx;
119 	}
120 
121 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
122 			 "6073 Binding %s HdwQueue %d  (cpu %d) to "
123 			 "hdw_queue %d qhandle x%px\n", str,
124 			 qidx, qhandle->cpu_id, qhandle->index, qhandle);
125 	*handle = (void *)qhandle;
126 	return 0;
127 }
128 
129 /**
130  * lpfc_nvme_delete_queue -
131  * @pnvme_lport: Transport localport that LS is to be issued from
132  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
133  * @handle: An opaque driver handle from lpfc_nvme_create_queue
134  *
135  * Driver registers this routine to free
136  * any internal data structures to bind the @qidx to its internal
137  * IO queues.
138  *
139  * Return value :
140  *   0 - Success
141  *   TODO:  What are the failure codes.
142  **/
143 static void
144 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
145 		       unsigned int qidx,
146 		       void *handle)
147 {
148 	struct lpfc_nvme_lport *lport;
149 	struct lpfc_vport *vport;
150 
151 	if (!pnvme_lport->private)
152 		return;
153 
154 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
155 	vport = lport->vport;
156 
157 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
158 			"6001 ENTER.  lpfc_pnvme x%px, qidx x%x qhandle x%px\n",
159 			lport, qidx, handle);
160 	kfree(handle);
161 }
162 
163 static void
164 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
165 {
166 	struct lpfc_nvme_lport *lport = localport->private;
167 
168 	lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
169 			 "6173 localport x%px delete complete\n",
170 			 lport);
171 
172 	/* release any threads waiting for the unreg to complete */
173 	if (lport->vport->localport)
174 		complete(lport->lport_unreg_cmp);
175 }
176 
177 /* lpfc_nvme_remoteport_delete
178  *
179  * @remoteport: Pointer to an nvme transport remoteport instance.
180  *
181  * This is a template downcall.  NVME transport calls this function
182  * when it has completed the unregistration of a previously
183  * registered remoteport.
184  *
185  * Return value :
186  * None
187  */
188 static void
189 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
190 {
191 	struct lpfc_nvme_rport *rport = remoteport->private;
192 	struct lpfc_vport *vport;
193 	struct lpfc_nodelist *ndlp;
194 	u32 fc4_xpt_flags;
195 
196 	ndlp = rport->ndlp;
197 	if (!ndlp) {
198 		pr_err("**** %s: NULL ndlp on rport x%px remoteport x%px\n",
199 		       __func__, rport, remoteport);
200 		goto rport_err;
201 	}
202 
203 	vport = ndlp->vport;
204 	if (!vport) {
205 		pr_err("**** %s: Null vport on ndlp x%px, ste x%x rport x%px\n",
206 		       __func__, ndlp, ndlp->nlp_state, rport);
207 		goto rport_err;
208 	}
209 
210 	fc4_xpt_flags = NVME_XPT_REGD | SCSI_XPT_REGD;
211 
212 	/* Remove this rport from the lport's list - memory is owned by the
213 	 * transport. Remove the ndlp reference for the NVME transport before
214 	 * calling state machine to remove the node.
215 	 */
216 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
217 			 "6146 remoteport delete of remoteport x%px, ndlp x%px "
218 			 "DID x%x xflags x%x\n",
219 			 remoteport, ndlp, ndlp->nlp_DID, ndlp->fc4_xpt_flags);
220 	spin_lock_irq(&ndlp->lock);
221 
222 	/* The register rebind might have occurred before the delete
223 	 * downcall.  Guard against this race.
224 	 */
225 	if (ndlp->fc4_xpt_flags & NVME_XPT_UNREG_WAIT)
226 		ndlp->fc4_xpt_flags &= ~(NVME_XPT_UNREG_WAIT | NVME_XPT_REGD);
227 
228 	spin_unlock_irq(&ndlp->lock);
229 
230 	/* On a devloss timeout event, one more put is executed provided the
231 	 * NVME and SCSI rport unregister requests are complete.  If the vport
232 	 * is unloading, this extra put is executed by lpfc_drop_node.
233 	 */
234 	if (!(ndlp->fc4_xpt_flags & fc4_xpt_flags))
235 		lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM);
236 
237  rport_err:
238 	return;
239 }
240 
241 /**
242  * lpfc_nvme_handle_lsreq - Process an unsolicited NVME LS request
243  * @phba: pointer to lpfc hba data structure.
244  * @axchg: pointer to exchange context for the NVME LS request
245  *
246  * This routine is used for processing an asychronously received NVME LS
247  * request. Any remaining validation is done and the LS is then forwarded
248  * to the nvme-fc transport via nvme_fc_rcv_ls_req().
249  *
250  * The calling sequence should be: nvme_fc_rcv_ls_req() -> (processing)
251  * -> lpfc_nvme_xmt_ls_rsp/cmp -> req->done.
252  * __lpfc_nvme_xmt_ls_rsp_cmp should free the allocated axchg.
253  *
254  * Returns 0 if LS was handled and delivered to the transport
255  * Returns 1 if LS failed to be handled and should be dropped
256  */
257 int
258 lpfc_nvme_handle_lsreq(struct lpfc_hba *phba,
259 			struct lpfc_async_xchg_ctx *axchg)
260 {
261 #if (IS_ENABLED(CONFIG_NVME_FC))
262 	struct lpfc_vport *vport;
263 	struct lpfc_nvme_rport *lpfc_rport;
264 	struct nvme_fc_remote_port *remoteport;
265 	struct lpfc_nvme_lport *lport;
266 	uint32_t *payload = axchg->payload;
267 	int rc;
268 
269 	vport = axchg->ndlp->vport;
270 	lpfc_rport = axchg->ndlp->nrport;
271 	if (!lpfc_rport)
272 		return -EINVAL;
273 
274 	remoteport = lpfc_rport->remoteport;
275 	if (!vport->localport ||
276 	    vport->phba->hba_flag & HBA_IOQ_FLUSH)
277 		return -EINVAL;
278 
279 	lport = vport->localport->private;
280 	if (!lport)
281 		return -EINVAL;
282 
283 	rc = nvme_fc_rcv_ls_req(remoteport, &axchg->ls_rsp, axchg->payload,
284 				axchg->size);
285 
286 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_DISC,
287 			"6205 NVME Unsol rcv: sz %d rc %d: %08x %08x %08x "
288 			"%08x %08x %08x\n",
289 			axchg->size, rc,
290 			*payload, *(payload+1), *(payload+2),
291 			*(payload+3), *(payload+4), *(payload+5));
292 
293 	if (!rc)
294 		return 0;
295 #endif
296 	return 1;
297 }
298 
299 /**
300  * __lpfc_nvme_ls_req_cmp - Generic completion handler for a NVME
301  *        LS request.
302  * @phba: Pointer to HBA context object
303  * @vport: The local port that issued the LS
304  * @cmdwqe: Pointer to driver command WQE object.
305  * @wcqe: Pointer to driver response CQE object.
306  *
307  * This function is the generic completion handler for NVME LS requests.
308  * The function updates any states and statistics, calls the transport
309  * ls_req done() routine, then tears down the command and buffers used
310  * for the LS request.
311  **/
312 void
313 __lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba,  struct lpfc_vport *vport,
314 			struct lpfc_iocbq *cmdwqe,
315 			struct lpfc_wcqe_complete *wcqe)
316 {
317 	struct nvmefc_ls_req *pnvme_lsreq;
318 	struct lpfc_dmabuf *buf_ptr;
319 	struct lpfc_nodelist *ndlp;
320 	uint32_t status;
321 
322 	pnvme_lsreq = cmdwqe->context_un.nvme_lsreq;
323 	ndlp = cmdwqe->ndlp;
324 	buf_ptr = cmdwqe->bpl_dmabuf;
325 
326 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
327 
328 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
329 			 "6047 NVMEx LS REQ x%px cmpl DID %x Xri: %x "
330 			 "status %x reason x%x cmd:x%px lsreg:x%px bmp:x%px "
331 			 "ndlp:x%px\n",
332 			 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
333 			 cmdwqe->sli4_xritag, status,
334 			 (wcqe->parameter & 0xffff),
335 			 cmdwqe, pnvme_lsreq, cmdwqe->bpl_dmabuf,
336 			 ndlp);
337 
338 	lpfc_nvmeio_data(phba, "NVMEx LS CMPL: xri x%x stat x%x parm x%x\n",
339 			 cmdwqe->sli4_xritag, status, wcqe->parameter);
340 
341 	if (buf_ptr) {
342 		lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
343 		kfree(buf_ptr);
344 		cmdwqe->bpl_dmabuf = NULL;
345 	}
346 	if (pnvme_lsreq->done)
347 		pnvme_lsreq->done(pnvme_lsreq, status);
348 	else
349 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
350 				 "6046 NVMEx cmpl without done call back? "
351 				 "Data x%px DID %x Xri: %x status %x\n",
352 				pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
353 				cmdwqe->sli4_xritag, status);
354 	if (ndlp) {
355 		lpfc_nlp_put(ndlp);
356 		cmdwqe->ndlp = NULL;
357 	}
358 	lpfc_sli_release_iocbq(phba, cmdwqe);
359 }
360 
361 static void
362 lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
363 		     struct lpfc_iocbq *rspwqe)
364 {
365 	struct lpfc_vport *vport = cmdwqe->vport;
366 	struct lpfc_nvme_lport *lport;
367 	uint32_t status;
368 	struct lpfc_wcqe_complete *wcqe = &rspwqe->wcqe_cmpl;
369 
370 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
371 
372 	if (vport->localport) {
373 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
374 		if (lport) {
375 			atomic_inc(&lport->fc4NvmeLsCmpls);
376 			if (status) {
377 				if (bf_get(lpfc_wcqe_c_xb, wcqe))
378 					atomic_inc(&lport->cmpl_ls_xb);
379 				atomic_inc(&lport->cmpl_ls_err);
380 			}
381 		}
382 	}
383 
384 	__lpfc_nvme_ls_req_cmp(phba, vport, cmdwqe, wcqe);
385 }
386 
387 static int
388 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
389 		  struct lpfc_dmabuf *inp,
390 		  struct nvmefc_ls_req *pnvme_lsreq,
391 		  void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
392 			       struct lpfc_iocbq *),
393 		  struct lpfc_nodelist *ndlp, uint32_t num_entry,
394 		  uint32_t tmo, uint8_t retry)
395 {
396 	struct lpfc_hba *phba = vport->phba;
397 	union lpfc_wqe128 *wqe;
398 	struct lpfc_iocbq *genwqe;
399 	struct ulp_bde64 *bpl;
400 	struct ulp_bde64 bde;
401 	int i, rc, xmit_len, first_len;
402 
403 	/* Allocate buffer for  command WQE */
404 	genwqe = lpfc_sli_get_iocbq(phba);
405 	if (genwqe == NULL)
406 		return 1;
407 
408 	wqe = &genwqe->wqe;
409 	/* Initialize only 64 bytes */
410 	memset(wqe, 0, sizeof(union lpfc_wqe));
411 
412 	genwqe->bpl_dmabuf = bmp;
413 	genwqe->cmd_flag |= LPFC_IO_NVME_LS;
414 
415 	/* Save for completion so we can release these resources */
416 	genwqe->ndlp = lpfc_nlp_get(ndlp);
417 	if (!genwqe->ndlp) {
418 		dev_warn(&phba->pcidev->dev,
419 			 "Warning: Failed node ref, not sending LS_REQ\n");
420 		lpfc_sli_release_iocbq(phba, genwqe);
421 		return 1;
422 	}
423 
424 	genwqe->context_un.nvme_lsreq = pnvme_lsreq;
425 	/* Fill in payload, bp points to frame payload */
426 
427 	if (!tmo)
428 		/* FC spec states we need 3 * ratov for CT requests */
429 		tmo = (3 * phba->fc_ratov);
430 
431 	/* For this command calculate the xmit length of the request bde. */
432 	xmit_len = 0;
433 	first_len = 0;
434 	bpl = (struct ulp_bde64 *)bmp->virt;
435 	for (i = 0; i < num_entry; i++) {
436 		bde.tus.w = bpl[i].tus.w;
437 		if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
438 			break;
439 		xmit_len += bde.tus.f.bdeSize;
440 		if (i == 0)
441 			first_len = xmit_len;
442 	}
443 
444 	genwqe->num_bdes = num_entry;
445 	genwqe->hba_wqidx = 0;
446 
447 	/* Words 0 - 2 */
448 	wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
449 	wqe->generic.bde.tus.f.bdeSize = first_len;
450 	wqe->generic.bde.addrLow = bpl[0].addrLow;
451 	wqe->generic.bde.addrHigh = bpl[0].addrHigh;
452 
453 	/* Word 3 */
454 	wqe->gen_req.request_payload_len = first_len;
455 
456 	/* Word 4 */
457 
458 	/* Word 5 */
459 	bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
460 	bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
461 	bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
462 	bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
463 	bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
464 
465 	/* Word 6 */
466 	bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
467 	       phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
468 	bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
469 
470 	/* Word 7 */
471 	bf_set(wqe_tmo, &wqe->gen_req.wqe_com, tmo);
472 	bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
473 	bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
474 	bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
475 
476 	/* Word 8 */
477 	wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
478 
479 	/* Word 9 */
480 	bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
481 
482 	/* Word 10 */
483 	bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
484 	bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
485 	bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
486 	bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
487 	bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
488 
489 	/* Word 11 */
490 	bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
491 	bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
492 
493 
494 	/* Issue GEN REQ WQE for NPORT <did> */
495 	genwqe->cmd_cmpl = cmpl;
496 	genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
497 	genwqe->vport = vport;
498 	genwqe->retry = retry;
499 
500 	lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
501 			 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
502 
503 	rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe);
504 	if (rc) {
505 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
506 				 "6045 Issue GEN REQ WQE to NPORT x%x "
507 				 "Data: x%x x%x  rc x%x\n",
508 				 ndlp->nlp_DID, genwqe->iotag,
509 				 vport->port_state, rc);
510 		lpfc_nlp_put(ndlp);
511 		lpfc_sli_release_iocbq(phba, genwqe);
512 		return 1;
513 	}
514 
515 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_ELS,
516 			 "6050 Issue GEN REQ WQE to NPORT x%x "
517 			 "Data: oxid: x%x state: x%x wq:x%px lsreq:x%px "
518 			 "bmp:x%px xmit:%d 1st:%d\n",
519 			 ndlp->nlp_DID, genwqe->sli4_xritag,
520 			 vport->port_state,
521 			 genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
522 	return 0;
523 }
524 
525 
526 /**
527  * __lpfc_nvme_ls_req - Generic service routine to issue an NVME LS request
528  * @vport: The local port issuing the LS
529  * @ndlp: The remote port to send the LS to
530  * @pnvme_lsreq: Pointer to LS request structure from the transport
531  * @gen_req_cmp: Completion call-back
532  *
533  * Routine validates the ndlp, builds buffers and sends a GEN_REQUEST
534  * WQE to perform the LS operation.
535  *
536  * Return value :
537  *   0 - Success
538  *   non-zero: various error codes, in form of -Exxx
539  **/
540 int
541 __lpfc_nvme_ls_req(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
542 		      struct nvmefc_ls_req *pnvme_lsreq,
543 		      void (*gen_req_cmp)(struct lpfc_hba *phba,
544 				struct lpfc_iocbq *cmdwqe,
545 				struct lpfc_iocbq *rspwqe))
546 {
547 	struct lpfc_dmabuf *bmp;
548 	struct ulp_bde64 *bpl;
549 	int ret;
550 	uint16_t ntype, nstate;
551 
552 	if (!ndlp) {
553 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
554 				 "6051 NVMEx LS REQ: Bad NDLP x%px, Failing "
555 				 "LS Req\n",
556 				 ndlp);
557 		return -ENODEV;
558 	}
559 
560 	ntype = ndlp->nlp_type;
561 	nstate = ndlp->nlp_state;
562 	if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
563 	    (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
564 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
565 				 "6088 NVMEx LS REQ: Fail DID x%06x not "
566 				 "ready for IO. Type x%x, State x%x\n",
567 				 ndlp->nlp_DID, ntype, nstate);
568 		return -ENODEV;
569 	}
570 	if (vport->phba->hba_flag & HBA_IOQ_FLUSH)
571 		return -ENODEV;
572 
573 	if (!vport->phba->sli4_hba.nvmels_wq)
574 		return -ENOMEM;
575 
576 	/*
577 	 * there are two dma buf in the request, actually there is one and
578 	 * the second one is just the start address + cmd size.
579 	 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
580 	 * in a lpfc_dmabuf struct. When freeing we just free the wrapper
581 	 * because the nvem layer owns the data bufs.
582 	 * We do not have to break these packets open, we don't care what is
583 	 * in them. And we do not have to look at the resonse data, we only
584 	 * care that we got a response. All of the caring is going to happen
585 	 * in the nvme-fc layer.
586 	 */
587 
588 	bmp = kmalloc(sizeof(*bmp), GFP_KERNEL);
589 	if (!bmp) {
590 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
591 				 "6044 NVMEx LS REQ: Could not alloc LS buf "
592 				 "for DID %x\n",
593 				 ndlp->nlp_DID);
594 		return -ENOMEM;
595 	}
596 
597 	bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
598 	if (!bmp->virt) {
599 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
600 				 "6042 NVMEx LS REQ: Could not alloc mbuf "
601 				 "for DID %x\n",
602 				 ndlp->nlp_DID);
603 		kfree(bmp);
604 		return -ENOMEM;
605 	}
606 
607 	INIT_LIST_HEAD(&bmp->list);
608 
609 	bpl = (struct ulp_bde64 *)bmp->virt;
610 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
611 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
612 	bpl->tus.f.bdeFlags = 0;
613 	bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
614 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
615 	bpl++;
616 
617 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
618 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
619 	bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
620 	bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
621 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
622 
623 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
624 			"6149 NVMEx LS REQ: Issue to DID 0x%06x lsreq x%px, "
625 			"rqstlen:%d rsplen:%d %pad %pad\n",
626 			ndlp->nlp_DID, pnvme_lsreq, pnvme_lsreq->rqstlen,
627 			pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
628 			&pnvme_lsreq->rspdma);
629 
630 	ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
631 				pnvme_lsreq, gen_req_cmp, ndlp, 2,
632 				pnvme_lsreq->timeout, 0);
633 	if (ret != WQE_SUCCESS) {
634 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
635 				 "6052 NVMEx REQ: EXIT. issue ls wqe failed "
636 				 "lsreq x%px Status %x DID %x\n",
637 				 pnvme_lsreq, ret, ndlp->nlp_DID);
638 		lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
639 		kfree(bmp);
640 		return -EIO;
641 	}
642 
643 	return 0;
644 }
645 
646 /**
647  * lpfc_nvme_ls_req - Issue an NVME Link Service request
648  * @pnvme_lport: Transport localport that LS is to be issued from.
649  * @pnvme_rport: Transport remoteport that LS is to be sent to.
650  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
651  *
652  * Driver registers this routine to handle any link service request
653  * from the nvme_fc transport to a remote nvme-aware port.
654  *
655  * Return value :
656  *   0 - Success
657  *   non-zero: various error codes, in form of -Exxx
658  **/
659 static int
660 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
661 		 struct nvme_fc_remote_port *pnvme_rport,
662 		 struct nvmefc_ls_req *pnvme_lsreq)
663 {
664 	struct lpfc_nvme_lport *lport;
665 	struct lpfc_nvme_rport *rport;
666 	struct lpfc_vport *vport;
667 	int ret;
668 
669 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
670 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
671 	if (unlikely(!lport) || unlikely(!rport))
672 		return -EINVAL;
673 
674 	vport = lport->vport;
675 	if (vport->load_flag & FC_UNLOADING ||
676 	    vport->phba->hba_flag & HBA_IOQ_FLUSH)
677 		return -ENODEV;
678 
679 	atomic_inc(&lport->fc4NvmeLsRequests);
680 
681 	ret = __lpfc_nvme_ls_req(vport, rport->ndlp, pnvme_lsreq,
682 				 lpfc_nvme_ls_req_cmp);
683 	if (ret)
684 		atomic_inc(&lport->xmt_ls_err);
685 
686 	return ret;
687 }
688 
689 /**
690  * __lpfc_nvme_ls_abort - Generic service routine to abort a prior
691  *         NVME LS request
692  * @vport: The local port that issued the LS
693  * @ndlp: The remote port the LS was sent to
694  * @pnvme_lsreq: Pointer to LS request structure from the transport
695  *
696  * The driver validates the ndlp, looks for the LS, and aborts the
697  * LS if found.
698  *
699  * Returns:
700  * 0 : if LS found and aborted
701  * non-zero: various error conditions in form -Exxx
702  **/
703 int
704 __lpfc_nvme_ls_abort(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
705 			struct nvmefc_ls_req *pnvme_lsreq)
706 {
707 	struct lpfc_hba *phba = vport->phba;
708 	struct lpfc_sli_ring *pring;
709 	struct lpfc_iocbq *wqe, *next_wqe;
710 	bool foundit = false;
711 
712 	if (!ndlp) {
713 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
714 				"6049 NVMEx LS REQ Abort: Bad NDLP x%px DID "
715 				"x%06x, Failing LS Req\n",
716 				ndlp, ndlp ? ndlp->nlp_DID : 0);
717 		return -EINVAL;
718 	}
719 
720 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
721 			 "6040 NVMEx LS REQ Abort: Issue LS_ABORT for lsreq "
722 			 "x%px rqstlen:%d rsplen:%d %pad %pad\n",
723 			 pnvme_lsreq, pnvme_lsreq->rqstlen,
724 			 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
725 			 &pnvme_lsreq->rspdma);
726 
727 	/*
728 	 * Lock the ELS ring txcmplq and look for the wqe that matches
729 	 * this ELS. If found, issue an abort on the wqe.
730 	 */
731 	pring = phba->sli4_hba.nvmels_wq->pring;
732 	spin_lock_irq(&phba->hbalock);
733 	spin_lock(&pring->ring_lock);
734 	list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
735 		if (wqe->context_un.nvme_lsreq == pnvme_lsreq) {
736 			wqe->cmd_flag |= LPFC_DRIVER_ABORTED;
737 			foundit = true;
738 			break;
739 		}
740 	}
741 	spin_unlock(&pring->ring_lock);
742 
743 	if (foundit)
744 		lpfc_sli_issue_abort_iotag(phba, pring, wqe, NULL);
745 	spin_unlock_irq(&phba->hbalock);
746 
747 	if (foundit)
748 		return 0;
749 
750 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
751 			 "6213 NVMEx LS REQ Abort: Unable to locate req x%px\n",
752 			 pnvme_lsreq);
753 	return -EINVAL;
754 }
755 
756 static int
757 lpfc_nvme_xmt_ls_rsp(struct nvme_fc_local_port *localport,
758 		     struct nvme_fc_remote_port *remoteport,
759 		     struct nvmefc_ls_rsp *ls_rsp)
760 {
761 	struct lpfc_async_xchg_ctx *axchg =
762 		container_of(ls_rsp, struct lpfc_async_xchg_ctx, ls_rsp);
763 	struct lpfc_nvme_lport *lport;
764 	int rc;
765 
766 	if (axchg->phba->pport->load_flag & FC_UNLOADING)
767 		return -ENODEV;
768 
769 	lport = (struct lpfc_nvme_lport *)localport->private;
770 
771 	rc = __lpfc_nvme_xmt_ls_rsp(axchg, ls_rsp, __lpfc_nvme_xmt_ls_rsp_cmp);
772 
773 	if (rc) {
774 		/*
775 		 * unless the failure is due to having already sent
776 		 * the response, an abort will be generated for the
777 		 * exchange if the rsp can't be sent.
778 		 */
779 		if (rc != -EALREADY)
780 			atomic_inc(&lport->xmt_ls_abort);
781 		return rc;
782 	}
783 
784 	return 0;
785 }
786 
787 /**
788  * lpfc_nvme_ls_abort - Abort a prior NVME LS request
789  * @pnvme_lport: Transport localport that LS is to be issued from.
790  * @pnvme_rport: Transport remoteport that LS is to be sent to.
791  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
792  *
793  * Driver registers this routine to abort a NVME LS request that is
794  * in progress (from the transports perspective).
795  **/
796 static void
797 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
798 		   struct nvme_fc_remote_port *pnvme_rport,
799 		   struct nvmefc_ls_req *pnvme_lsreq)
800 {
801 	struct lpfc_nvme_lport *lport;
802 	struct lpfc_vport *vport;
803 	struct lpfc_nodelist *ndlp;
804 	int ret;
805 
806 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
807 	if (unlikely(!lport))
808 		return;
809 	vport = lport->vport;
810 
811 	if (vport->load_flag & FC_UNLOADING)
812 		return;
813 
814 	ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
815 
816 	ret = __lpfc_nvme_ls_abort(vport, ndlp, pnvme_lsreq);
817 	if (!ret)
818 		atomic_inc(&lport->xmt_ls_abort);
819 }
820 
821 /* Fix up the existing sgls for NVME IO. */
822 static inline void
823 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
824 		       struct lpfc_io_buf *lpfc_ncmd,
825 		       struct nvmefc_fcp_req *nCmd)
826 {
827 	struct lpfc_hba  *phba = vport->phba;
828 	struct sli4_sge *sgl;
829 	union lpfc_wqe128 *wqe;
830 	uint32_t *wptr, *dptr;
831 
832 	/*
833 	 * Get a local pointer to the built-in wqe and correct
834 	 * the cmd size to match NVME's 96 bytes and fix
835 	 * the dma address.
836 	 */
837 
838 	wqe = &lpfc_ncmd->cur_iocbq.wqe;
839 
840 	/*
841 	 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
842 	 * match NVME.  NVME sends 96 bytes. Also, use the
843 	 * nvme commands command and response dma addresses
844 	 * rather than the virtual memory to ease the restore
845 	 * operation.
846 	 */
847 	sgl = lpfc_ncmd->dma_sgl;
848 	sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
849 	if (phba->cfg_nvme_embed_cmd) {
850 		sgl->addr_hi = 0;
851 		sgl->addr_lo = 0;
852 
853 		/* Word 0-2 - NVME CMND IU (embedded payload) */
854 		wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
855 		wqe->generic.bde.tus.f.bdeSize = 56;
856 		wqe->generic.bde.addrHigh = 0;
857 		wqe->generic.bde.addrLow =  64;  /* Word 16 */
858 
859 		/* Word 10  - dbde is 0, wqes is 1 in template */
860 
861 		/*
862 		 * Embed the payload in the last half of the WQE
863 		 * WQE words 16-30 get the NVME CMD IU payload
864 		 *
865 		 * WQE words 16-19 get payload Words 1-4
866 		 * WQE words 20-21 get payload Words 6-7
867 		 * WQE words 22-29 get payload Words 16-23
868 		 */
869 		wptr = &wqe->words[16];  /* WQE ptr */
870 		dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
871 		dptr++;			/* Skip Word 0 in payload */
872 
873 		*wptr++ = *dptr++;	/* Word 1 */
874 		*wptr++ = *dptr++;	/* Word 2 */
875 		*wptr++ = *dptr++;	/* Word 3 */
876 		*wptr++ = *dptr++;	/* Word 4 */
877 		dptr++;			/* Skip Word 5 in payload */
878 		*wptr++ = *dptr++;	/* Word 6 */
879 		*wptr++ = *dptr++;	/* Word 7 */
880 		dptr += 8;		/* Skip Words 8-15 in payload */
881 		*wptr++ = *dptr++;	/* Word 16 */
882 		*wptr++ = *dptr++;	/* Word 17 */
883 		*wptr++ = *dptr++;	/* Word 18 */
884 		*wptr++ = *dptr++;	/* Word 19 */
885 		*wptr++ = *dptr++;	/* Word 20 */
886 		*wptr++ = *dptr++;	/* Word 21 */
887 		*wptr++ = *dptr++;	/* Word 22 */
888 		*wptr   = *dptr;	/* Word 23 */
889 	} else {
890 		sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
891 		sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
892 
893 		/* Word 0-2 - NVME CMND IU Inline BDE */
894 		wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
895 		wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
896 		wqe->generic.bde.addrHigh = sgl->addr_hi;
897 		wqe->generic.bde.addrLow =  sgl->addr_lo;
898 
899 		/* Word 10 */
900 		bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
901 		bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
902 	}
903 
904 	sgl++;
905 
906 	/* Setup the physical region for the FCP RSP */
907 	sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
908 	sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
909 	sgl->word2 = le32_to_cpu(sgl->word2);
910 	if (nCmd->sg_cnt)
911 		bf_set(lpfc_sli4_sge_last, sgl, 0);
912 	else
913 		bf_set(lpfc_sli4_sge_last, sgl, 1);
914 	sgl->word2 = cpu_to_le32(sgl->word2);
915 	sgl->sge_len = cpu_to_le32(nCmd->rsplen);
916 }
917 
918 
919 /*
920  * lpfc_nvme_io_cmd_cmpl - Complete an NVME-over-FCP IO
921  *
922  * Driver registers this routine as it io request handler.  This
923  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
924  * data structure to the rport indicated in @lpfc_nvme_rport.
925  *
926  * Return value :
927  *   0 - Success
928  *   TODO: What are the failure codes.
929  **/
930 static void
931 lpfc_nvme_io_cmd_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
932 		      struct lpfc_iocbq *pwqeOut)
933 {
934 	struct lpfc_io_buf *lpfc_ncmd = pwqeIn->io_buf;
935 	struct lpfc_wcqe_complete *wcqe = &pwqeOut->wcqe_cmpl;
936 	struct lpfc_vport *vport = pwqeIn->vport;
937 	struct nvmefc_fcp_req *nCmd;
938 	struct nvme_fc_ersp_iu *ep;
939 	struct nvme_fc_cmd_iu *cp;
940 	struct lpfc_nodelist *ndlp;
941 	struct lpfc_nvme_fcpreq_priv *freqpriv;
942 	struct lpfc_nvme_lport *lport;
943 	uint32_t code, status, idx;
944 	uint16_t cid, sqhd, data;
945 	uint32_t *ptr;
946 	uint32_t lat;
947 	bool call_done = false;
948 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
949 	int cpu;
950 #endif
951 	int offline = 0;
952 
953 	/* Sanity check on return of outstanding command */
954 	if (!lpfc_ncmd) {
955 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
956 				 "6071 Null lpfc_ncmd pointer. No "
957 				 "release, skip completion\n");
958 		return;
959 	}
960 
961 	/* Guard against abort handler being called at same time */
962 	spin_lock(&lpfc_ncmd->buf_lock);
963 
964 	if (!lpfc_ncmd->nvmeCmd) {
965 		spin_unlock(&lpfc_ncmd->buf_lock);
966 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
967 				 "6066 Missing cmpl ptrs: lpfc_ncmd x%px, "
968 				 "nvmeCmd x%px\n",
969 				 lpfc_ncmd, lpfc_ncmd->nvmeCmd);
970 
971 		/* Release the lpfc_ncmd regardless of the missing elements. */
972 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
973 		return;
974 	}
975 	nCmd = lpfc_ncmd->nvmeCmd;
976 	status = bf_get(lpfc_wcqe_c_status, wcqe);
977 
978 	idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
979 	phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++;
980 
981 	if (unlikely(status && vport->localport)) {
982 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
983 		if (lport) {
984 			if (bf_get(lpfc_wcqe_c_xb, wcqe))
985 				atomic_inc(&lport->cmpl_fcp_xb);
986 			atomic_inc(&lport->cmpl_fcp_err);
987 		}
988 	}
989 
990 	lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
991 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
992 			 status, wcqe->parameter);
993 	/*
994 	 * Catch race where our node has transitioned, but the
995 	 * transport is still transitioning.
996 	 */
997 	ndlp = lpfc_ncmd->ndlp;
998 	if (!ndlp) {
999 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1000 				 "6062 Ignoring NVME cmpl.  No ndlp\n");
1001 		goto out_err;
1002 	}
1003 
1004 	code = bf_get(lpfc_wcqe_c_code, wcqe);
1005 	if (code == CQE_CODE_NVME_ERSP) {
1006 		/* For this type of CQE, we need to rebuild the rsp */
1007 		ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1008 
1009 		/*
1010 		 * Get Command Id from cmd to plug into response. This
1011 		 * code is not needed in the next NVME Transport drop.
1012 		 */
1013 		cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1014 		cid = cp->sqe.common.command_id;
1015 
1016 		/*
1017 		 * RSN is in CQE word 2
1018 		 * SQHD is in CQE Word 3 bits 15:0
1019 		 * Cmd Specific info is in CQE Word 1
1020 		 * and in CQE Word 0 bits 15:0
1021 		 */
1022 		sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1023 
1024 		/* Now lets build the NVME ERSP IU */
1025 		ep->iu_len = cpu_to_be16(8);
1026 		ep->rsn = wcqe->parameter;
1027 		ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1028 		ep->rsvd12 = 0;
1029 		ptr = (uint32_t *)&ep->cqe.result.u64;
1030 		*ptr++ = wcqe->total_data_placed;
1031 		data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1032 		*ptr = (uint32_t)data;
1033 		ep->cqe.sq_head = sqhd;
1034 		ep->cqe.sq_id =  nCmd->sqid;
1035 		ep->cqe.command_id = cid;
1036 		ep->cqe.status = 0;
1037 
1038 		lpfc_ncmd->status = IOSTAT_SUCCESS;
1039 		lpfc_ncmd->result = 0;
1040 		nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1041 		nCmd->transferred_length = nCmd->payload_length;
1042 	} else {
1043 		lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1044 		lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1045 
1046 		/* For NVME, the only failure path that results in an
1047 		 * IO error is when the adapter rejects it.  All other
1048 		 * conditions are a success case and resolved by the
1049 		 * transport.
1050 		 * IOSTAT_FCP_RSP_ERROR means:
1051 		 * 1. Length of data received doesn't match total
1052 		 *    transfer length in WQE
1053 		 * 2. If the RSP payload does NOT match these cases:
1054 		 *    a. RSP length 12/24 bytes and all zeros
1055 		 *    b. NVME ERSP
1056 		 */
1057 		switch (lpfc_ncmd->status) {
1058 		case IOSTAT_SUCCESS:
1059 			nCmd->transferred_length = wcqe->total_data_placed;
1060 			nCmd->rcv_rsplen = 0;
1061 			nCmd->status = 0;
1062 			break;
1063 		case IOSTAT_FCP_RSP_ERROR:
1064 			nCmd->transferred_length = wcqe->total_data_placed;
1065 			nCmd->rcv_rsplen = wcqe->parameter;
1066 			nCmd->status = 0;
1067 
1068 			/* Get the NVME cmd details for this unique error. */
1069 			cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1070 			ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1071 
1072 			/* Check if this is really an ERSP */
1073 			if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) {
1074 				lpfc_ncmd->status = IOSTAT_SUCCESS;
1075 				lpfc_ncmd->result = 0;
1076 
1077 				lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
1078 					"6084 NVME FCP_ERR ERSP: "
1079 					"xri %x placed x%x opcode x%x cmd_id "
1080 					"x%x cqe_status x%x\n",
1081 					lpfc_ncmd->cur_iocbq.sli4_xritag,
1082 					wcqe->total_data_placed,
1083 					cp->sqe.common.opcode,
1084 					cp->sqe.common.command_id,
1085 					ep->cqe.status);
1086 				break;
1087 			}
1088 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1089 					 "6081 NVME Completion Protocol Error: "
1090 					 "xri %x status x%x result x%x "
1091 					 "placed x%x opcode x%x cmd_id x%x, "
1092 					 "cqe_status x%x\n",
1093 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1094 					 lpfc_ncmd->status, lpfc_ncmd->result,
1095 					 wcqe->total_data_placed,
1096 					 cp->sqe.common.opcode,
1097 					 cp->sqe.common.command_id,
1098 					 ep->cqe.status);
1099 			break;
1100 		case IOSTAT_LOCAL_REJECT:
1101 			/* Let fall through to set command final state. */
1102 			if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1103 				lpfc_printf_vlog(vport, KERN_INFO,
1104 					 LOG_NVME_IOERR,
1105 					 "6032 Delay Aborted cmd x%px "
1106 					 "nvme cmd x%px, xri x%x, "
1107 					 "xb %d\n",
1108 					 lpfc_ncmd, nCmd,
1109 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1110 					 bf_get(lpfc_wcqe_c_xb, wcqe));
1111 			fallthrough;
1112 		default:
1113 out_err:
1114 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1115 					 "6072 NVME Completion Error: xri %x "
1116 					 "status x%x result x%x [x%x] "
1117 					 "placed x%x\n",
1118 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1119 					 lpfc_ncmd->status, lpfc_ncmd->result,
1120 					 wcqe->parameter,
1121 					 wcqe->total_data_placed);
1122 			nCmd->transferred_length = 0;
1123 			nCmd->rcv_rsplen = 0;
1124 			nCmd->status = NVME_SC_INTERNAL;
1125 			offline = pci_channel_offline(vport->phba->pcidev);
1126 		}
1127 	}
1128 
1129 	/* pick up SLI4 exhange busy condition */
1130 	if (bf_get(lpfc_wcqe_c_xb, wcqe) && !offline)
1131 		lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1132 	else
1133 		lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1134 
1135 	/* Update stats and complete the IO.  There is
1136 	 * no need for dma unprep because the nvme_transport
1137 	 * owns the dma address.
1138 	 */
1139 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1140 	if (lpfc_ncmd->ts_cmd_start) {
1141 		lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1142 		lpfc_ncmd->ts_data_io = ktime_get_ns();
1143 		phba->ktime_last_cmd = lpfc_ncmd->ts_data_io;
1144 		lpfc_io_ktime(phba, lpfc_ncmd);
1145 	}
1146 	if (unlikely(phba->hdwqstat_on & LPFC_CHECK_NVME_IO)) {
1147 		cpu = raw_smp_processor_id();
1148 		this_cpu_inc(phba->sli4_hba.c_stat->cmpl_io);
1149 		if (lpfc_ncmd->cpu != cpu)
1150 			lpfc_printf_vlog(vport,
1151 					 KERN_INFO, LOG_NVME_IOERR,
1152 					 "6701 CPU Check cmpl: "
1153 					 "cpu %d expect %d\n",
1154 					 cpu, lpfc_ncmd->cpu);
1155 	}
1156 #endif
1157 
1158 	/* NVME targets need completion held off until the abort exchange
1159 	 * completes unless the NVME Rport is getting unregistered.
1160 	 */
1161 
1162 	if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1163 		freqpriv = nCmd->private;
1164 		freqpriv->nvme_buf = NULL;
1165 		lpfc_ncmd->nvmeCmd = NULL;
1166 		call_done = true;
1167 	}
1168 	spin_unlock(&lpfc_ncmd->buf_lock);
1169 
1170 	/* Check if IO qualified for CMF */
1171 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1172 	    nCmd->io_dir == NVMEFC_FCP_READ &&
1173 	    nCmd->payload_length) {
1174 		/* Used when calculating average latency */
1175 		lat = ktime_get_ns() - lpfc_ncmd->rx_cmd_start;
1176 		lpfc_update_cmf_cmpl(phba, lat, nCmd->payload_length, NULL);
1177 	}
1178 
1179 	if (call_done)
1180 		nCmd->done(nCmd);
1181 
1182 	/* Call release with XB=1 to queue the IO into the abort list. */
1183 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1184 }
1185 
1186 
1187 /**
1188  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1189  * @vport: pointer to a host virtual N_Port data structure
1190  * @lpfc_ncmd: Pointer to lpfc scsi command
1191  * @pnode: pointer to a node-list data structure
1192  * @cstat: pointer to the control status structure
1193  *
1194  * Driver registers this routine as it io request handler.  This
1195  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1196  * data structure to the rport indicated in @lpfc_nvme_rport.
1197  *
1198  * Return value :
1199  *   0 - Success
1200  *   TODO: What are the failure codes.
1201  **/
1202 static int
1203 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1204 		      struct lpfc_io_buf *lpfc_ncmd,
1205 		      struct lpfc_nodelist *pnode,
1206 		      struct lpfc_fc4_ctrl_stat *cstat)
1207 {
1208 	struct lpfc_hba *phba = vport->phba;
1209 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1210 	struct nvme_common_command *sqe;
1211 	struct lpfc_iocbq *pwqeq = &lpfc_ncmd->cur_iocbq;
1212 	union lpfc_wqe128 *wqe = &pwqeq->wqe;
1213 	uint32_t req_len;
1214 
1215 	/*
1216 	 * There are three possibilities here - use scatter-gather segment, use
1217 	 * the single mapping, or neither.
1218 	 */
1219 	if (nCmd->sg_cnt) {
1220 		if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1221 			/* From the iwrite template, initialize words 7 - 11 */
1222 			memcpy(&wqe->words[7],
1223 			       &lpfc_iwrite_cmd_template.words[7],
1224 			       sizeof(uint32_t) * 5);
1225 
1226 			/* Word 4 */
1227 			wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1228 
1229 			/* Word 5 */
1230 			if ((phba->cfg_nvme_enable_fb) &&
1231 			    (pnode->nlp_flag & NLP_FIRSTBURST)) {
1232 				req_len = lpfc_ncmd->nvmeCmd->payload_length;
1233 				if (req_len < pnode->nvme_fb_size)
1234 					wqe->fcp_iwrite.initial_xfer_len =
1235 						req_len;
1236 				else
1237 					wqe->fcp_iwrite.initial_xfer_len =
1238 						pnode->nvme_fb_size;
1239 			} else {
1240 				wqe->fcp_iwrite.initial_xfer_len = 0;
1241 			}
1242 			cstat->output_requests++;
1243 		} else {
1244 			/* From the iread template, initialize words 7 - 11 */
1245 			memcpy(&wqe->words[7],
1246 			       &lpfc_iread_cmd_template.words[7],
1247 			       sizeof(uint32_t) * 5);
1248 
1249 			/* Word 4 */
1250 			wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1251 
1252 			/* Word 5 */
1253 			wqe->fcp_iread.rsrvd5 = 0;
1254 
1255 			/* For a CMF Managed port, iod must be zero'ed */
1256 			if (phba->cmf_active_mode == LPFC_CFG_MANAGED)
1257 				bf_set(wqe_iod, &wqe->fcp_iread.wqe_com,
1258 				       LPFC_WQE_IOD_NONE);
1259 			cstat->input_requests++;
1260 		}
1261 	} else {
1262 		/* From the icmnd template, initialize words 4 - 11 */
1263 		memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1264 		       sizeof(uint32_t) * 8);
1265 		cstat->control_requests++;
1266 	}
1267 
1268 	if (pnode->nlp_nvme_info & NLP_NVME_NSLER) {
1269 		bf_set(wqe_erp, &wqe->generic.wqe_com, 1);
1270 		sqe = &((struct nvme_fc_cmd_iu *)
1271 			nCmd->cmdaddr)->sqe.common;
1272 		if (sqe->opcode == nvme_admin_async_event)
1273 			bf_set(wqe_ffrq, &wqe->generic.wqe_com, 1);
1274 	}
1275 
1276 	/*
1277 	 * Finish initializing those WQE fields that are independent
1278 	 * of the nvme_cmnd request_buffer
1279 	 */
1280 
1281 	/* Word 3 */
1282 	bf_set(payload_offset_len, &wqe->fcp_icmd,
1283 	       (nCmd->rsplen + nCmd->cmdlen));
1284 
1285 	/* Word 6 */
1286 	bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1287 	       phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1288 	bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1289 
1290 	/* Word 8 */
1291 	wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1292 
1293 	/* Word 9 */
1294 	bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1295 
1296 	/* Word 10 */
1297 	bf_set(wqe_xchg, &wqe->fcp_iwrite.wqe_com, LPFC_NVME_XCHG);
1298 
1299 	/* Words 13 14 15 are for PBDE support */
1300 
1301 	/* add the VMID tags as per switch response */
1302 	if (unlikely(lpfc_ncmd->cur_iocbq.cmd_flag & LPFC_IO_VMID)) {
1303 		if (phba->pport->vmid_priority_tagging) {
1304 			bf_set(wqe_ccpe, &wqe->fcp_iwrite.wqe_com, 1);
1305 			bf_set(wqe_ccp, &wqe->fcp_iwrite.wqe_com,
1306 			       lpfc_ncmd->cur_iocbq.vmid_tag.cs_ctl_vmid);
1307 		} else {
1308 			bf_set(wqe_appid, &wqe->fcp_iwrite.wqe_com, 1);
1309 			bf_set(wqe_wqes, &wqe->fcp_iwrite.wqe_com, 1);
1310 			wqe->words[31] = lpfc_ncmd->cur_iocbq.vmid_tag.app_id;
1311 		}
1312 	}
1313 
1314 	pwqeq->vport = vport;
1315 	return 0;
1316 }
1317 
1318 
1319 /**
1320  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1321  * @vport: pointer to a host virtual N_Port data structure
1322  * @lpfc_ncmd: Pointer to lpfc scsi command
1323  *
1324  * Driver registers this routine as it io request handler.  This
1325  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1326  * data structure to the rport indicated in @lpfc_nvme_rport.
1327  *
1328  * Return value :
1329  *   0 - Success
1330  *   TODO: What are the failure codes.
1331  **/
1332 static int
1333 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1334 		      struct lpfc_io_buf *lpfc_ncmd)
1335 {
1336 	struct lpfc_hba *phba = vport->phba;
1337 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1338 	union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1339 	struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1340 	struct sli4_hybrid_sgl *sgl_xtra = NULL;
1341 	struct scatterlist *data_sg;
1342 	struct sli4_sge *first_data_sgl;
1343 	struct ulp_bde64 *bde;
1344 	dma_addr_t physaddr = 0;
1345 	uint32_t dma_len = 0;
1346 	uint32_t dma_offset = 0;
1347 	int nseg, i, j;
1348 	bool lsp_just_set = false;
1349 
1350 	/* Fix up the command and response DMA stuff. */
1351 	lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1352 
1353 	/*
1354 	 * There are three possibilities here - use scatter-gather segment, use
1355 	 * the single mapping, or neither.
1356 	 */
1357 	if (nCmd->sg_cnt) {
1358 		/*
1359 		 * Jump over the cmd and rsp SGEs.  The fix routine
1360 		 * has already adjusted for this.
1361 		 */
1362 		sgl += 2;
1363 
1364 		first_data_sgl = sgl;
1365 		lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1366 		if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1367 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1368 					"6058 Too many sg segments from "
1369 					"NVME Transport.  Max %d, "
1370 					"nvmeIO sg_cnt %d\n",
1371 					phba->cfg_nvme_seg_cnt + 1,
1372 					lpfc_ncmd->seg_cnt);
1373 			lpfc_ncmd->seg_cnt = 0;
1374 			return 1;
1375 		}
1376 
1377 		/*
1378 		 * The driver established a maximum scatter-gather segment count
1379 		 * during probe that limits the number of sg elements in any
1380 		 * single nvme command.  Just run through the seg_cnt and format
1381 		 * the sge's.
1382 		 */
1383 		nseg = nCmd->sg_cnt;
1384 		data_sg = nCmd->first_sgl;
1385 
1386 		/* for tracking the segment boundaries */
1387 		j = 2;
1388 		for (i = 0; i < nseg; i++) {
1389 			if (data_sg == NULL) {
1390 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1391 						"6059 dptr err %d, nseg %d\n",
1392 						i, nseg);
1393 				lpfc_ncmd->seg_cnt = 0;
1394 				return 1;
1395 			}
1396 
1397 			sgl->word2 = 0;
1398 			if (nseg == 1) {
1399 				bf_set(lpfc_sli4_sge_last, sgl, 1);
1400 				bf_set(lpfc_sli4_sge_type, sgl,
1401 				       LPFC_SGE_TYPE_DATA);
1402 			} else {
1403 				bf_set(lpfc_sli4_sge_last, sgl, 0);
1404 
1405 				/* expand the segment */
1406 				if (!lsp_just_set &&
1407 				    !((j + 1) % phba->border_sge_num) &&
1408 				    ((nseg - 1) != i)) {
1409 					/* set LSP type */
1410 					bf_set(lpfc_sli4_sge_type, sgl,
1411 					       LPFC_SGE_TYPE_LSP);
1412 
1413 					sgl_xtra = lpfc_get_sgl_per_hdwq(
1414 							phba, lpfc_ncmd);
1415 
1416 					if (unlikely(!sgl_xtra)) {
1417 						lpfc_ncmd->seg_cnt = 0;
1418 						return 1;
1419 					}
1420 					sgl->addr_lo = cpu_to_le32(putPaddrLow(
1421 						       sgl_xtra->dma_phys_sgl));
1422 					sgl->addr_hi = cpu_to_le32(putPaddrHigh(
1423 						       sgl_xtra->dma_phys_sgl));
1424 
1425 				} else {
1426 					bf_set(lpfc_sli4_sge_type, sgl,
1427 					       LPFC_SGE_TYPE_DATA);
1428 				}
1429 			}
1430 
1431 			if (!(bf_get(lpfc_sli4_sge_type, sgl) &
1432 				     LPFC_SGE_TYPE_LSP)) {
1433 				if ((nseg - 1) == i)
1434 					bf_set(lpfc_sli4_sge_last, sgl, 1);
1435 
1436 				physaddr = sg_dma_address(data_sg);
1437 				dma_len = sg_dma_len(data_sg);
1438 				sgl->addr_lo = cpu_to_le32(
1439 							 putPaddrLow(physaddr));
1440 				sgl->addr_hi = cpu_to_le32(
1441 							putPaddrHigh(physaddr));
1442 
1443 				bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1444 				sgl->word2 = cpu_to_le32(sgl->word2);
1445 				sgl->sge_len = cpu_to_le32(dma_len);
1446 
1447 				dma_offset += dma_len;
1448 				data_sg = sg_next(data_sg);
1449 
1450 				sgl++;
1451 
1452 				lsp_just_set = false;
1453 			} else {
1454 				sgl->word2 = cpu_to_le32(sgl->word2);
1455 
1456 				sgl->sge_len = cpu_to_le32(
1457 						     phba->cfg_sg_dma_buf_size);
1458 
1459 				sgl = (struct sli4_sge *)sgl_xtra->dma_sgl;
1460 				i = i - 1;
1461 
1462 				lsp_just_set = true;
1463 			}
1464 
1465 			j++;
1466 		}
1467 
1468 		/* PBDE support for first data SGE only */
1469 		if (nseg == 1 && phba->cfg_enable_pbde) {
1470 			/* Words 13-15 */
1471 			bde = (struct ulp_bde64 *)
1472 				&wqe->words[13];
1473 			bde->addrLow = first_data_sgl->addr_lo;
1474 			bde->addrHigh = first_data_sgl->addr_hi;
1475 			bde->tus.f.bdeSize =
1476 				le32_to_cpu(first_data_sgl->sge_len);
1477 			bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1478 			bde->tus.w = cpu_to_le32(bde->tus.w);
1479 
1480 			/* Word 11 - set PBDE bit */
1481 			bf_set(wqe_pbde, &wqe->generic.wqe_com, 1);
1482 		} else {
1483 			memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1484 			/* Word 11 - PBDE bit disabled by default template */
1485 		}
1486 
1487 	} else {
1488 		lpfc_ncmd->seg_cnt = 0;
1489 
1490 		/* For this clause to be valid, the payload_length
1491 		 * and sg_cnt must zero.
1492 		 */
1493 		if (nCmd->payload_length != 0) {
1494 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1495 					"6063 NVME DMA Prep Err: sg_cnt %d "
1496 					"payload_length x%x\n",
1497 					nCmd->sg_cnt, nCmd->payload_length);
1498 			return 1;
1499 		}
1500 	}
1501 	return 0;
1502 }
1503 
1504 /**
1505  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1506  * @pnvme_lport: Pointer to the driver's local port data
1507  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1508  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1509  * @pnvme_fcreq: IO request from nvme fc to driver.
1510  *
1511  * Driver registers this routine as it io request handler.  This
1512  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1513  * data structure to the rport indicated in @lpfc_nvme_rport.
1514  *
1515  * Return value :
1516  *   0 - Success
1517  *   TODO: What are the failure codes.
1518  **/
1519 static int
1520 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1521 			struct nvme_fc_remote_port *pnvme_rport,
1522 			void *hw_queue_handle,
1523 			struct nvmefc_fcp_req *pnvme_fcreq)
1524 {
1525 	int ret = 0;
1526 	int expedite = 0;
1527 	int idx, cpu;
1528 	struct lpfc_nvme_lport *lport;
1529 	struct lpfc_fc4_ctrl_stat *cstat;
1530 	struct lpfc_vport *vport;
1531 	struct lpfc_hba *phba;
1532 	struct lpfc_nodelist *ndlp;
1533 	struct lpfc_io_buf *lpfc_ncmd;
1534 	struct lpfc_nvme_rport *rport;
1535 	struct lpfc_nvme_qhandle *lpfc_queue_info;
1536 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1537 	struct nvme_common_command *sqe;
1538 	uint64_t start = 0;
1539 #if (IS_ENABLED(CONFIG_NVME_FC))
1540 	u8 *uuid = NULL;
1541 	int err;
1542 	enum dma_data_direction iodir;
1543 #endif
1544 
1545 	/* Validate pointers. LLDD fault handling with transport does
1546 	 * have timing races.
1547 	 */
1548 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1549 	if (unlikely(!lport)) {
1550 		ret = -EINVAL;
1551 		goto out_fail;
1552 	}
1553 
1554 	vport = lport->vport;
1555 
1556 	if (unlikely(!hw_queue_handle)) {
1557 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1558 				 "6117 Fail IO, NULL hw_queue_handle\n");
1559 		atomic_inc(&lport->xmt_fcp_err);
1560 		ret = -EBUSY;
1561 		goto out_fail;
1562 	}
1563 
1564 	phba = vport->phba;
1565 
1566 	if ((unlikely(vport->load_flag & FC_UNLOADING)) ||
1567 	    phba->hba_flag & HBA_IOQ_FLUSH) {
1568 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1569 				 "6124 Fail IO, Driver unload\n");
1570 		atomic_inc(&lport->xmt_fcp_err);
1571 		ret = -ENODEV;
1572 		goto out_fail;
1573 	}
1574 
1575 	freqpriv = pnvme_fcreq->private;
1576 	if (unlikely(!freqpriv)) {
1577 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1578 				 "6158 Fail IO, NULL request data\n");
1579 		atomic_inc(&lport->xmt_fcp_err);
1580 		ret = -EINVAL;
1581 		goto out_fail;
1582 	}
1583 
1584 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1585 	if (phba->ktime_on)
1586 		start = ktime_get_ns();
1587 #endif
1588 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1589 	lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1590 
1591 	/*
1592 	 * Catch race where our node has transitioned, but the
1593 	 * transport is still transitioning.
1594 	 */
1595 	ndlp = rport->ndlp;
1596 	if (!ndlp) {
1597 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1598 				 "6053 Busy IO, ndlp not ready: rport x%px "
1599 				  "ndlp x%px, DID x%06x\n",
1600 				 rport, ndlp, pnvme_rport->port_id);
1601 		atomic_inc(&lport->xmt_fcp_err);
1602 		ret = -EBUSY;
1603 		goto out_fail;
1604 	}
1605 
1606 	/* The remote node has to be a mapped target or it's an error. */
1607 	if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1608 	    (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1609 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1610 				 "6036 Fail IO, DID x%06x not ready for "
1611 				 "IO. State x%x, Type x%x Flg x%x\n",
1612 				 pnvme_rport->port_id,
1613 				 ndlp->nlp_state, ndlp->nlp_type,
1614 				 ndlp->fc4_xpt_flags);
1615 		atomic_inc(&lport->xmt_fcp_bad_ndlp);
1616 		ret = -EBUSY;
1617 		goto out_fail;
1618 
1619 	}
1620 
1621 	/* Currently only NVME Keep alive commands should be expedited
1622 	 * if the driver runs out of a resource. These should only be
1623 	 * issued on the admin queue, qidx 0
1624 	 */
1625 	if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1626 		sqe = &((struct nvme_fc_cmd_iu *)
1627 			pnvme_fcreq->cmdaddr)->sqe.common;
1628 		if (sqe->opcode == nvme_admin_keep_alive)
1629 			expedite = 1;
1630 	}
1631 
1632 	/* Check if IO qualifies for CMF */
1633 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1634 	    pnvme_fcreq->io_dir == NVMEFC_FCP_READ &&
1635 	    pnvme_fcreq->payload_length) {
1636 		ret = lpfc_update_cmf_cmd(phba, pnvme_fcreq->payload_length);
1637 		if (ret) {
1638 			ret = -EBUSY;
1639 			goto out_fail;
1640 		}
1641 		/* Get start time for IO latency */
1642 		start = ktime_get_ns();
1643 	}
1644 
1645 	/* The node is shared with FCP IO, make sure the IO pending count does
1646 	 * not exceed the programmed depth.
1647 	 */
1648 	if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1649 		if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1650 		    !expedite) {
1651 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1652 					 "6174 Fail IO, ndlp qdepth exceeded: "
1653 					 "idx %d DID %x pend %d qdepth %d\n",
1654 					 lpfc_queue_info->index, ndlp->nlp_DID,
1655 					 atomic_read(&ndlp->cmd_pending),
1656 					 ndlp->cmd_qdepth);
1657 			atomic_inc(&lport->xmt_fcp_qdepth);
1658 			ret = -EBUSY;
1659 			goto out_fail1;
1660 		}
1661 	}
1662 
1663 	/* Lookup Hardware Queue index based on fcp_io_sched module parameter */
1664 	if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) {
1665 		idx = lpfc_queue_info->index;
1666 	} else {
1667 		cpu = raw_smp_processor_id();
1668 		idx = phba->sli4_hba.cpu_map[cpu].hdwq;
1669 	}
1670 
1671 	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite);
1672 	if (lpfc_ncmd == NULL) {
1673 		atomic_inc(&lport->xmt_fcp_noxri);
1674 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1675 				 "6065 Fail IO, driver buffer pool is empty: "
1676 				 "idx %d DID %x\n",
1677 				 lpfc_queue_info->index, ndlp->nlp_DID);
1678 		ret = -EBUSY;
1679 		goto out_fail1;
1680 	}
1681 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1682 	if (start) {
1683 		lpfc_ncmd->ts_cmd_start = start;
1684 		lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1685 	} else {
1686 		lpfc_ncmd->ts_cmd_start = 0;
1687 	}
1688 #endif
1689 	lpfc_ncmd->rx_cmd_start = start;
1690 
1691 	/*
1692 	 * Store the data needed by the driver to issue, abort, and complete
1693 	 * an IO.
1694 	 * Do not let the IO hang out forever.  There is no midlayer issuing
1695 	 * an abort so inform the FW of the maximum IO pending time.
1696 	 */
1697 	freqpriv->nvme_buf = lpfc_ncmd;
1698 	lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1699 	lpfc_ncmd->ndlp = ndlp;
1700 	lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1701 
1702 #if (IS_ENABLED(CONFIG_NVME_FC))
1703 	/* check the necessary and sufficient condition to support VMID */
1704 	if (lpfc_is_vmid_enabled(phba) &&
1705 	    (ndlp->vmid_support ||
1706 	     phba->pport->vmid_priority_tagging ==
1707 	     LPFC_VMID_PRIO_TAG_ALL_TARGETS)) {
1708 		/* is the I/O generated by a VM, get the associated virtual */
1709 		/* entity id */
1710 		uuid = nvme_fc_io_getuuid(pnvme_fcreq);
1711 
1712 		if (uuid) {
1713 			if (pnvme_fcreq->io_dir == NVMEFC_FCP_WRITE)
1714 				iodir = DMA_TO_DEVICE;
1715 			else if (pnvme_fcreq->io_dir == NVMEFC_FCP_READ)
1716 				iodir = DMA_FROM_DEVICE;
1717 			else
1718 				iodir = DMA_NONE;
1719 
1720 			err = lpfc_vmid_get_appid(vport, uuid, iodir,
1721 					(union lpfc_vmid_io_tag *)
1722 						&lpfc_ncmd->cur_iocbq.vmid_tag);
1723 			if (!err)
1724 				lpfc_ncmd->cur_iocbq.cmd_flag |= LPFC_IO_VMID;
1725 		}
1726 	}
1727 #endif
1728 
1729 	/*
1730 	 * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1731 	 * This identfier was create in our hardware queue create callback
1732 	 * routine. The driver now is dependent on the IO queue steering from
1733 	 * the transport.  We are trusting the upper NVME layers know which
1734 	 * index to use and that they have affinitized a CPU to this hardware
1735 	 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1736 	 */
1737 	lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1738 	cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat;
1739 
1740 	lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1741 	ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1742 	if (ret) {
1743 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1744 				 "6175 Fail IO, Prep DMA: "
1745 				 "idx %d DID %x\n",
1746 				 lpfc_queue_info->index, ndlp->nlp_DID);
1747 		atomic_inc(&lport->xmt_fcp_err);
1748 		ret = -ENOMEM;
1749 		goto out_free_nvme_buf;
1750 	}
1751 
1752 	lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1753 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
1754 			 lpfc_queue_info->index, ndlp->nlp_DID);
1755 
1756 	ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq);
1757 	if (ret) {
1758 		atomic_inc(&lport->xmt_fcp_wqerr);
1759 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1760 				 "6113 Fail IO, Could not issue WQE err %x "
1761 				 "sid: x%x did: x%x oxid: x%x\n",
1762 				 ret, vport->fc_myDID, ndlp->nlp_DID,
1763 				 lpfc_ncmd->cur_iocbq.sli4_xritag);
1764 		goto out_free_nvme_buf;
1765 	}
1766 
1767 	if (phba->cfg_xri_rebalancing)
1768 		lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no);
1769 
1770 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1771 	if (lpfc_ncmd->ts_cmd_start)
1772 		lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1773 
1774 	if (phba->hdwqstat_on & LPFC_CHECK_NVME_IO) {
1775 		cpu = raw_smp_processor_id();
1776 		this_cpu_inc(phba->sli4_hba.c_stat->xmt_io);
1777 		lpfc_ncmd->cpu = cpu;
1778 		if (idx != cpu)
1779 			lpfc_printf_vlog(vport,
1780 					 KERN_INFO, LOG_NVME_IOERR,
1781 					"6702 CPU Check cmd: "
1782 					"cpu %d wq %d\n",
1783 					lpfc_ncmd->cpu,
1784 					lpfc_queue_info->index);
1785 	}
1786 #endif
1787 	return 0;
1788 
1789  out_free_nvme_buf:
1790 	if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1791 		if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1792 			cstat->output_requests--;
1793 		else
1794 			cstat->input_requests--;
1795 	} else
1796 		cstat->control_requests--;
1797 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1798  out_fail1:
1799 	lpfc_update_cmf_cmpl(phba, LPFC_CGN_NOT_SENT,
1800 			     pnvme_fcreq->payload_length, NULL);
1801  out_fail:
1802 	return ret;
1803 }
1804 
1805 /**
1806  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1807  * @phba: Pointer to HBA context object
1808  * @cmdiocb: Pointer to command iocb object.
1809  * @rspiocb: Pointer to response iocb object.
1810  *
1811  * This is the callback function for any NVME FCP IO that was aborted.
1812  *
1813  * Return value:
1814  *   None
1815  **/
1816 void
1817 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1818 			   struct lpfc_iocbq *rspiocb)
1819 {
1820 	struct lpfc_wcqe_complete *abts_cmpl = &rspiocb->wcqe_cmpl;
1821 
1822 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1823 			"6145 ABORT_XRI_CN completing on rpi x%x "
1824 			"original iotag x%x, abort cmd iotag x%x "
1825 			"req_tag x%x, status x%x, hwstatus x%x\n",
1826 			bf_get(wqe_ctxt_tag, &cmdiocb->wqe.generic.wqe_com),
1827 			get_job_abtsiotag(phba, cmdiocb), cmdiocb->iotag,
1828 			bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1829 			bf_get(lpfc_wcqe_c_status, abts_cmpl),
1830 			bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1831 	lpfc_sli_release_iocbq(phba, cmdiocb);
1832 }
1833 
1834 /**
1835  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1836  * @pnvme_lport: Pointer to the driver's local port data
1837  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1838  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1839  * @pnvme_fcreq: IO request from nvme fc to driver.
1840  *
1841  * Driver registers this routine as its nvme request io abort handler.  This
1842  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1843  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1844  * is executed asynchronously - one the target is validated as "MAPPED" and
1845  * ready for IO, the driver issues the abort request and returns.
1846  *
1847  * Return value:
1848  *   None
1849  **/
1850 static void
1851 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1852 		    struct nvme_fc_remote_port *pnvme_rport,
1853 		    void *hw_queue_handle,
1854 		    struct nvmefc_fcp_req *pnvme_fcreq)
1855 {
1856 	struct lpfc_nvme_lport *lport;
1857 	struct lpfc_vport *vport;
1858 	struct lpfc_hba *phba;
1859 	struct lpfc_io_buf *lpfc_nbuf;
1860 	struct lpfc_iocbq *nvmereq_wqe;
1861 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1862 	unsigned long flags;
1863 	int ret_val;
1864 	struct nvme_fc_cmd_iu *cp;
1865 
1866 	/* Validate pointers. LLDD fault handling with transport does
1867 	 * have timing races.
1868 	 */
1869 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1870 	if (unlikely(!lport))
1871 		return;
1872 
1873 	vport = lport->vport;
1874 
1875 	if (unlikely(!hw_queue_handle)) {
1876 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1877 				 "6129 Fail Abort, HW Queue Handle NULL.\n");
1878 		return;
1879 	}
1880 
1881 	phba = vport->phba;
1882 	freqpriv = pnvme_fcreq->private;
1883 
1884 	if (unlikely(!freqpriv))
1885 		return;
1886 	if (vport->load_flag & FC_UNLOADING)
1887 		return;
1888 
1889 	/* Announce entry to new IO submit field. */
1890 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1891 			 "6002 Abort Request to rport DID x%06x "
1892 			 "for nvme_fc_req x%px\n",
1893 			 pnvme_rport->port_id,
1894 			 pnvme_fcreq);
1895 
1896 	/* If the hba is getting reset, this flag is set.  It is
1897 	 * cleared when the reset is complete and rings reestablished.
1898 	 */
1899 	spin_lock_irqsave(&phba->hbalock, flags);
1900 	/* driver queued commands are in process of being flushed */
1901 	if (phba->hba_flag & HBA_IOQ_FLUSH) {
1902 		spin_unlock_irqrestore(&phba->hbalock, flags);
1903 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1904 				 "6139 Driver in reset cleanup - flushing "
1905 				 "NVME Req now.  hba_flag x%x\n",
1906 				 phba->hba_flag);
1907 		return;
1908 	}
1909 
1910 	lpfc_nbuf = freqpriv->nvme_buf;
1911 	if (!lpfc_nbuf) {
1912 		spin_unlock_irqrestore(&phba->hbalock, flags);
1913 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1914 				 "6140 NVME IO req has no matching lpfc nvme "
1915 				 "io buffer.  Skipping abort req.\n");
1916 		return;
1917 	} else if (!lpfc_nbuf->nvmeCmd) {
1918 		spin_unlock_irqrestore(&phba->hbalock, flags);
1919 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1920 				 "6141 lpfc NVME IO req has no nvme_fcreq "
1921 				 "io buffer.  Skipping abort req.\n");
1922 		return;
1923 	}
1924 	nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1925 
1926 	/* Guard against IO completion being called at same time */
1927 	spin_lock(&lpfc_nbuf->buf_lock);
1928 
1929 	/*
1930 	 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1931 	 * state must match the nvme_fcreq passed by the nvme
1932 	 * transport.  If they don't match, it is likely the driver
1933 	 * has already completed the NVME IO and the nvme transport
1934 	 * has not seen it yet.
1935 	 */
1936 	if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1937 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1938 				 "6143 NVME req mismatch: "
1939 				 "lpfc_nbuf x%px nvmeCmd x%px, "
1940 				 "pnvme_fcreq x%px.  Skipping Abort xri x%x\n",
1941 				 lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1942 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1943 		goto out_unlock;
1944 	}
1945 
1946 	/* Don't abort IOs no longer on the pending queue. */
1947 	if (!(nvmereq_wqe->cmd_flag & LPFC_IO_ON_TXCMPLQ)) {
1948 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1949 				 "6142 NVME IO req x%px not queued - skipping "
1950 				 "abort req xri x%x\n",
1951 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1952 		goto out_unlock;
1953 	}
1954 
1955 	atomic_inc(&lport->xmt_fcp_abort);
1956 	lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1957 			 nvmereq_wqe->sli4_xritag,
1958 			 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1959 
1960 	/* Outstanding abort is in progress */
1961 	if (nvmereq_wqe->cmd_flag & LPFC_DRIVER_ABORTED) {
1962 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1963 				 "6144 Outstanding NVME I/O Abort Request "
1964 				 "still pending on nvme_fcreq x%px, "
1965 				 "lpfc_ncmd x%px xri x%x\n",
1966 				 pnvme_fcreq, lpfc_nbuf,
1967 				 nvmereq_wqe->sli4_xritag);
1968 		goto out_unlock;
1969 	}
1970 
1971 	ret_val = lpfc_sli4_issue_abort_iotag(phba, nvmereq_wqe,
1972 					      lpfc_nvme_abort_fcreq_cmpl);
1973 
1974 	spin_unlock(&lpfc_nbuf->buf_lock);
1975 	spin_unlock_irqrestore(&phba->hbalock, flags);
1976 
1977 	/* Make sure HBA is alive */
1978 	lpfc_issue_hb_tmo(phba);
1979 
1980 	if (ret_val != WQE_SUCCESS) {
1981 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1982 				 "6137 Failed abts issue_wqe with status x%x "
1983 				 "for nvme_fcreq x%px.\n",
1984 				 ret_val, pnvme_fcreq);
1985 		return;
1986 	}
1987 
1988 	/*
1989 	 * Get Command Id from cmd to plug into response. This
1990 	 * code is not needed in the next NVME Transport drop.
1991 	 */
1992 	cp = (struct nvme_fc_cmd_iu *)lpfc_nbuf->nvmeCmd->cmdaddr;
1993 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1994 			 "6138 Transport Abort NVME Request Issued for "
1995 			 "ox_id x%x nvme opcode x%x nvme cmd_id x%x\n",
1996 			 nvmereq_wqe->sli4_xritag, cp->sqe.common.opcode,
1997 			 cp->sqe.common.command_id);
1998 	return;
1999 
2000 out_unlock:
2001 	spin_unlock(&lpfc_nbuf->buf_lock);
2002 	spin_unlock_irqrestore(&phba->hbalock, flags);
2003 	return;
2004 }
2005 
2006 /* Declare and initialization an instance of the FC NVME template. */
2007 static struct nvme_fc_port_template lpfc_nvme_template = {
2008 	/* initiator-based functions */
2009 	.localport_delete  = lpfc_nvme_localport_delete,
2010 	.remoteport_delete = lpfc_nvme_remoteport_delete,
2011 	.create_queue = lpfc_nvme_create_queue,
2012 	.delete_queue = lpfc_nvme_delete_queue,
2013 	.ls_req       = lpfc_nvme_ls_req,
2014 	.fcp_io       = lpfc_nvme_fcp_io_submit,
2015 	.ls_abort     = lpfc_nvme_ls_abort,
2016 	.fcp_abort    = lpfc_nvme_fcp_abort,
2017 	.xmt_ls_rsp   = lpfc_nvme_xmt_ls_rsp,
2018 
2019 	.max_hw_queues = 1,
2020 	.max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
2021 	.max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
2022 	.dma_boundary = 0xFFFFFFFF,
2023 
2024 	/* Sizes of additional private data for data structures.
2025 	 * No use for the last two sizes at this time.
2026 	 */
2027 	.local_priv_sz = sizeof(struct lpfc_nvme_lport),
2028 	.remote_priv_sz = sizeof(struct lpfc_nvme_rport),
2029 	.lsrqst_priv_sz = 0,
2030 	.fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
2031 };
2032 
2033 /*
2034  * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
2035  *
2036  * This routine removes a nvme buffer from head of @hdwq io_buf_list
2037  * and returns to caller.
2038  *
2039  * Return codes:
2040  *   NULL - Error
2041  *   Pointer to lpfc_nvme_buf - Success
2042  **/
2043 static struct lpfc_io_buf *
2044 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
2045 		  int idx, int expedite)
2046 {
2047 	struct lpfc_io_buf *lpfc_ncmd;
2048 	struct lpfc_sli4_hdw_queue *qp;
2049 	struct sli4_sge *sgl;
2050 	struct lpfc_iocbq *pwqeq;
2051 	union lpfc_wqe128 *wqe;
2052 
2053 	lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite);
2054 
2055 	if (lpfc_ncmd) {
2056 		pwqeq = &(lpfc_ncmd->cur_iocbq);
2057 		wqe = &pwqeq->wqe;
2058 
2059 		/* Setup key fields in buffer that may have been changed
2060 		 * if other protocols used this buffer.
2061 		 */
2062 		pwqeq->cmd_flag = LPFC_IO_NVME;
2063 		pwqeq->cmd_cmpl = lpfc_nvme_io_cmd_cmpl;
2064 		lpfc_ncmd->start_time = jiffies;
2065 		lpfc_ncmd->flags = 0;
2066 
2067 		/* Rsp SGE will be filled in when we rcv an IO
2068 		 * from the NVME Layer to be sent.
2069 		 * The cmd is going to be embedded so we need a SKIP SGE.
2070 		 */
2071 		sgl = lpfc_ncmd->dma_sgl;
2072 		bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
2073 		bf_set(lpfc_sli4_sge_last, sgl, 0);
2074 		sgl->word2 = cpu_to_le32(sgl->word2);
2075 		/* Fill in word 3 / sgl_len during cmd submission */
2076 
2077 		/* Initialize 64 bytes only */
2078 		memset(wqe, 0, sizeof(union lpfc_wqe));
2079 
2080 		if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
2081 			atomic_inc(&ndlp->cmd_pending);
2082 			lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH;
2083 		}
2084 
2085 	} else {
2086 		qp = &phba->sli4_hba.hdwq[idx];
2087 		qp->empty_io_bufs++;
2088 	}
2089 
2090 	return  lpfc_ncmd;
2091 }
2092 
2093 /**
2094  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2095  * @phba: The Hba for which this call is being executed.
2096  * @lpfc_ncmd: The nvme buffer which is being released.
2097  *
2098  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2099  * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2100  * and cannot be reused for at least RA_TOV amount of time if it was
2101  * aborted.
2102  **/
2103 static void
2104 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd)
2105 {
2106 	struct lpfc_sli4_hdw_queue *qp;
2107 	unsigned long iflag = 0;
2108 
2109 	if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2110 		atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2111 
2112 	lpfc_ncmd->ndlp = NULL;
2113 	lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH;
2114 
2115 	qp = lpfc_ncmd->hdwq;
2116 	if (unlikely(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
2117 		lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2118 				"6310 XB release deferred for "
2119 				"ox_id x%x on reqtag x%x\n",
2120 				lpfc_ncmd->cur_iocbq.sli4_xritag,
2121 				lpfc_ncmd->cur_iocbq.iotag);
2122 
2123 		spin_lock_irqsave(&qp->abts_io_buf_list_lock, iflag);
2124 		list_add_tail(&lpfc_ncmd->list,
2125 			&qp->lpfc_abts_io_buf_list);
2126 		qp->abts_nvme_io_bufs++;
2127 		spin_unlock_irqrestore(&qp->abts_io_buf_list_lock, iflag);
2128 	} else
2129 		lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp);
2130 }
2131 
2132 /**
2133  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2134  * @vport: the lpfc_vport instance requesting a localport.
2135  *
2136  * This routine is invoked to create an nvme localport instance to bind
2137  * to the nvme_fc_transport.  It is called once during driver load
2138  * like lpfc_create_shost after all other services are initialized.
2139  * It requires a vport, vpi, and wwns at call time.  Other localport
2140  * parameters are modified as the driver's FCID and the Fabric WWN
2141  * are established.
2142  *
2143  * Return codes
2144  *      0 - successful
2145  *      -ENOMEM - no heap memory available
2146  *      other values - from nvme registration upcall
2147  **/
2148 int
2149 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2150 {
2151 	int ret = 0;
2152 	struct lpfc_hba  *phba = vport->phba;
2153 	struct nvme_fc_port_info nfcp_info;
2154 	struct nvme_fc_local_port *localport;
2155 	struct lpfc_nvme_lport *lport;
2156 
2157 	/* Initialize this localport instance.  The vport wwn usage ensures
2158 	 * that NPIV is accounted for.
2159 	 */
2160 	memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2161 	nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2162 	nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2163 	nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2164 
2165 	/* We need to tell the transport layer + 1 because it takes page
2166 	 * alignment into account. When space for the SGL is allocated we
2167 	 * allocate + 3, one for cmd, one for rsp and one for this alignment
2168 	 */
2169 	lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2170 
2171 	/* Advertise how many hw queues we support based on cfg_hdw_queue,
2172 	 * which will not exceed cpu count.
2173 	 */
2174 	lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2175 
2176 	if (!IS_ENABLED(CONFIG_NVME_FC))
2177 		return ret;
2178 
2179 	/* localport is allocated from the stack, but the registration
2180 	 * call allocates heap memory as well as the private area.
2181 	 */
2182 
2183 	ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2184 					 &vport->phba->pcidev->dev, &localport);
2185 	if (!ret) {
2186 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2187 				 "6005 Successfully registered local "
2188 				 "NVME port num %d, localP x%px, private "
2189 				 "x%px, sg_seg %d\n",
2190 				 localport->port_num, localport,
2191 				 localport->private,
2192 				 lpfc_nvme_template.max_sgl_segments);
2193 
2194 		/* Private is our lport size declared in the template. */
2195 		lport = (struct lpfc_nvme_lport *)localport->private;
2196 		vport->localport = localport;
2197 		lport->vport = vport;
2198 		vport->nvmei_support = 1;
2199 
2200 		atomic_set(&lport->xmt_fcp_noxri, 0);
2201 		atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2202 		atomic_set(&lport->xmt_fcp_qdepth, 0);
2203 		atomic_set(&lport->xmt_fcp_err, 0);
2204 		atomic_set(&lport->xmt_fcp_wqerr, 0);
2205 		atomic_set(&lport->xmt_fcp_abort, 0);
2206 		atomic_set(&lport->xmt_ls_abort, 0);
2207 		atomic_set(&lport->xmt_ls_err, 0);
2208 		atomic_set(&lport->cmpl_fcp_xb, 0);
2209 		atomic_set(&lport->cmpl_fcp_err, 0);
2210 		atomic_set(&lport->cmpl_ls_xb, 0);
2211 		atomic_set(&lport->cmpl_ls_err, 0);
2212 
2213 		atomic_set(&lport->fc4NvmeLsRequests, 0);
2214 		atomic_set(&lport->fc4NvmeLsCmpls, 0);
2215 	}
2216 
2217 	return ret;
2218 }
2219 
2220 #if (IS_ENABLED(CONFIG_NVME_FC))
2221 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2222  *
2223  * The driver has to wait for the host nvme transport to callback
2224  * indicating the localport has successfully unregistered all
2225  * resources.  Since this is an uninterruptible wait, loop every ten
2226  * seconds and print a message indicating no progress.
2227  *
2228  * An uninterruptible wait is used because of the risk of transport-to-
2229  * driver state mismatch.
2230  */
2231 static void
2232 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2233 			   struct lpfc_nvme_lport *lport,
2234 			   struct completion *lport_unreg_cmp)
2235 {
2236 	u32 wait_tmo;
2237 	int ret, i, pending = 0;
2238 	struct lpfc_sli_ring  *pring;
2239 	struct lpfc_hba  *phba = vport->phba;
2240 	struct lpfc_sli4_hdw_queue *qp;
2241 	int abts_scsi, abts_nvme;
2242 
2243 	/* Host transport has to clean up and confirm requiring an indefinite
2244 	 * wait. Print a message if a 10 second wait expires and renew the
2245 	 * wait. This is unexpected.
2246 	 */
2247 	wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2248 	while (true) {
2249 		ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2250 		if (unlikely(!ret)) {
2251 			pending = 0;
2252 			abts_scsi = 0;
2253 			abts_nvme = 0;
2254 			for (i = 0; i < phba->cfg_hdw_queue; i++) {
2255 				qp = &phba->sli4_hba.hdwq[i];
2256 				if (!vport->localport || !qp || !qp->io_wq)
2257 					return;
2258 
2259 				pring = qp->io_wq->pring;
2260 				if (!pring)
2261 					continue;
2262 				pending += pring->txcmplq_cnt;
2263 				abts_scsi += qp->abts_scsi_io_bufs;
2264 				abts_nvme += qp->abts_nvme_io_bufs;
2265 			}
2266 			if (!vport->localport ||
2267 			    test_bit(HBA_PCI_ERR, &vport->phba->bit_flags) ||
2268 			    vport->load_flag & FC_UNLOADING)
2269 				return;
2270 
2271 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2272 					 "6176 Lport x%px Localport x%px wait "
2273 					 "timed out. Pending %d [%d:%d]. "
2274 					 "Renewing.\n",
2275 					 lport, vport->localport, pending,
2276 					 abts_scsi, abts_nvme);
2277 			continue;
2278 		}
2279 		break;
2280 	}
2281 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2282 			 "6177 Lport x%px Localport x%px Complete Success\n",
2283 			 lport, vport->localport);
2284 }
2285 #endif
2286 
2287 /**
2288  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2289  * @vport: pointer to a host virtual N_Port data structure
2290  *
2291  * This routine is invoked to destroy all lports bound to the phba.
2292  * The lport memory was allocated by the nvme fc transport and is
2293  * released there.  This routine ensures all rports bound to the
2294  * lport have been disconnected.
2295  *
2296  **/
2297 void
2298 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2299 {
2300 #if (IS_ENABLED(CONFIG_NVME_FC))
2301 	struct nvme_fc_local_port *localport;
2302 	struct lpfc_nvme_lport *lport;
2303 	int ret;
2304 	DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2305 
2306 	if (vport->nvmei_support == 0)
2307 		return;
2308 
2309 	localport = vport->localport;
2310 	if (!localport)
2311 		return;
2312 	lport = (struct lpfc_nvme_lport *)localport->private;
2313 
2314 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2315 			 "6011 Destroying NVME localport x%px\n",
2316 			 localport);
2317 
2318 	/* lport's rport list is clear.  Unregister
2319 	 * lport and release resources.
2320 	 */
2321 	lport->lport_unreg_cmp = &lport_unreg_cmp;
2322 	ret = nvme_fc_unregister_localport(localport);
2323 
2324 	/* Wait for completion.  This either blocks
2325 	 * indefinitely or succeeds
2326 	 */
2327 	lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2328 	vport->localport = NULL;
2329 
2330 	/* Regardless of the unregister upcall response, clear
2331 	 * nvmei_support.  All rports are unregistered and the
2332 	 * driver will clean up.
2333 	 */
2334 	vport->nvmei_support = 0;
2335 	if (ret == 0) {
2336 		lpfc_printf_vlog(vport,
2337 				 KERN_INFO, LOG_NVME_DISC,
2338 				 "6009 Unregistered lport Success\n");
2339 	} else {
2340 		lpfc_printf_vlog(vport,
2341 				 KERN_INFO, LOG_NVME_DISC,
2342 				 "6010 Unregistered lport "
2343 				 "Failed, status x%x\n",
2344 				 ret);
2345 	}
2346 #endif
2347 }
2348 
2349 void
2350 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2351 {
2352 #if (IS_ENABLED(CONFIG_NVME_FC))
2353 	struct nvme_fc_local_port *localport;
2354 	struct lpfc_nvme_lport *lport;
2355 
2356 	localport = vport->localport;
2357 	if (!localport) {
2358 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2359 				 "6710 Update NVME fail. No localport\n");
2360 		return;
2361 	}
2362 	lport = (struct lpfc_nvme_lport *)localport->private;
2363 	if (!lport) {
2364 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2365 				 "6171 Update NVME fail. localP x%px, No lport\n",
2366 				 localport);
2367 		return;
2368 	}
2369 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2370 			 "6012 Update NVME lport x%px did x%x\n",
2371 			 localport, vport->fc_myDID);
2372 
2373 	localport->port_id = vport->fc_myDID;
2374 	if (localport->port_id == 0)
2375 		localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2376 	else
2377 		localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2378 
2379 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2380 			 "6030 bound lport x%px to DID x%06x\n",
2381 			 lport, localport->port_id);
2382 #endif
2383 }
2384 
2385 int
2386 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2387 {
2388 #if (IS_ENABLED(CONFIG_NVME_FC))
2389 	int ret = 0;
2390 	struct nvme_fc_local_port *localport;
2391 	struct lpfc_nvme_lport *lport;
2392 	struct lpfc_nvme_rport *rport;
2393 	struct lpfc_nvme_rport *oldrport;
2394 	struct nvme_fc_remote_port *remote_port;
2395 	struct nvme_fc_port_info rpinfo;
2396 	struct lpfc_nodelist *prev_ndlp = NULL;
2397 	struct fc_rport *srport = ndlp->rport;
2398 
2399 	lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2400 			 "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2401 			 ndlp->nlp_DID, ndlp->nlp_type);
2402 
2403 	localport = vport->localport;
2404 	if (!localport)
2405 		return 0;
2406 
2407 	lport = (struct lpfc_nvme_lport *)localport->private;
2408 
2409 	/* NVME rports are not preserved across devloss.
2410 	 * Just register this instance.  Note, rpinfo->dev_loss_tmo
2411 	 * is left 0 to indicate accept transport defaults.  The
2412 	 * driver communicates port role capabilities consistent
2413 	 * with the PRLI response data.
2414 	 */
2415 	memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2416 	rpinfo.port_id = ndlp->nlp_DID;
2417 	if (ndlp->nlp_type & NLP_NVME_TARGET)
2418 		rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2419 	if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2420 		rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2421 
2422 	if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2423 		rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2424 
2425 	rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2426 	rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2427 	if (srport)
2428 		rpinfo.dev_loss_tmo = srport->dev_loss_tmo;
2429 	else
2430 		rpinfo.dev_loss_tmo = vport->cfg_devloss_tmo;
2431 
2432 	spin_lock_irq(&ndlp->lock);
2433 
2434 	/* If an oldrport exists, so does the ndlp reference.  If not
2435 	 * a new reference is needed because either the node has never
2436 	 * been registered or it's been unregistered and getting deleted.
2437 	 */
2438 	oldrport = lpfc_ndlp_get_nrport(ndlp);
2439 	if (oldrport) {
2440 		prev_ndlp = oldrport->ndlp;
2441 		spin_unlock_irq(&ndlp->lock);
2442 	} else {
2443 		spin_unlock_irq(&ndlp->lock);
2444 		if (!lpfc_nlp_get(ndlp)) {
2445 			dev_warn(&vport->phba->pcidev->dev,
2446 				 "Warning - No node ref - exit register\n");
2447 			return 0;
2448 		}
2449 	}
2450 
2451 	ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2452 	if (!ret) {
2453 		/* If the ndlp already has an nrport, this is just
2454 		 * a resume of the existing rport.  Else this is a
2455 		 * new rport.
2456 		 */
2457 		/* Guard against an unregister/reregister
2458 		 * race that leaves the WAIT flag set.
2459 		 */
2460 		spin_lock_irq(&ndlp->lock);
2461 		ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2462 		ndlp->fc4_xpt_flags |= NVME_XPT_REGD;
2463 		spin_unlock_irq(&ndlp->lock);
2464 		rport = remote_port->private;
2465 		if (oldrport) {
2466 
2467 			/* Sever the ndlp<->rport association
2468 			 * before dropping the ndlp ref from
2469 			 * register.
2470 			 */
2471 			spin_lock_irq(&ndlp->lock);
2472 			ndlp->nrport = NULL;
2473 			ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2474 			spin_unlock_irq(&ndlp->lock);
2475 			rport->ndlp = NULL;
2476 			rport->remoteport = NULL;
2477 
2478 			/* Reference only removed if previous NDLP is no longer
2479 			 * active. It might be just a swap and removing the
2480 			 * reference would cause a premature cleanup.
2481 			 */
2482 			if (prev_ndlp && prev_ndlp != ndlp) {
2483 				if (!prev_ndlp->nrport)
2484 					lpfc_nlp_put(prev_ndlp);
2485 			}
2486 		}
2487 
2488 		/* Clean bind the rport to the ndlp. */
2489 		rport->remoteport = remote_port;
2490 		rport->lport = lport;
2491 		rport->ndlp = ndlp;
2492 		spin_lock_irq(&ndlp->lock);
2493 		ndlp->nrport = rport;
2494 		spin_unlock_irq(&ndlp->lock);
2495 		lpfc_printf_vlog(vport, KERN_INFO,
2496 				 LOG_NVME_DISC | LOG_NODE,
2497 				 "6022 Bind lport x%px to remoteport x%px "
2498 				 "rport x%px WWNN 0x%llx, "
2499 				 "Rport WWPN 0x%llx DID "
2500 				 "x%06x Role x%x, ndlp %p prev_ndlp x%px\n",
2501 				 lport, remote_port, rport,
2502 				 rpinfo.node_name, rpinfo.port_name,
2503 				 rpinfo.port_id, rpinfo.port_role,
2504 				 ndlp, prev_ndlp);
2505 	} else {
2506 		lpfc_printf_vlog(vport, KERN_ERR,
2507 				 LOG_TRACE_EVENT,
2508 				 "6031 RemotePort Registration failed "
2509 				 "err: %d, DID x%06x\n",
2510 				 ret, ndlp->nlp_DID);
2511 	}
2512 
2513 	return ret;
2514 #else
2515 	return 0;
2516 #endif
2517 }
2518 
2519 /*
2520  * lpfc_nvme_rescan_port - Check to see if we should rescan this remoteport
2521  *
2522  * If the ndlp represents an NVME Target, that we are logged into,
2523  * ping the NVME FC Transport layer to initiate a device rescan
2524  * on this remote NPort.
2525  */
2526 void
2527 lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2528 {
2529 #if (IS_ENABLED(CONFIG_NVME_FC))
2530 	struct lpfc_nvme_rport *nrport;
2531 	struct nvme_fc_remote_port *remoteport = NULL;
2532 
2533 	spin_lock_irq(&ndlp->lock);
2534 	nrport = lpfc_ndlp_get_nrport(ndlp);
2535 	if (nrport)
2536 		remoteport = nrport->remoteport;
2537 	spin_unlock_irq(&ndlp->lock);
2538 
2539 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2540 			 "6170 Rescan NPort DID x%06x type x%x "
2541 			 "state x%x nrport x%px remoteport x%px\n",
2542 			 ndlp->nlp_DID, ndlp->nlp_type, ndlp->nlp_state,
2543 			 nrport, remoteport);
2544 
2545 	if (!nrport || !remoteport)
2546 		goto rescan_exit;
2547 
2548 	/* Rescan an NVME target in MAPPED state with DISCOVERY role set */
2549 	if (remoteport->port_role & FC_PORT_ROLE_NVME_DISCOVERY &&
2550 	    ndlp->nlp_state == NLP_STE_MAPPED_NODE) {
2551 		nvme_fc_rescan_remoteport(remoteport);
2552 
2553 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2554 				 "6172 NVME rescanned DID x%06x "
2555 				 "port_state x%x\n",
2556 				 ndlp->nlp_DID, remoteport->port_state);
2557 	}
2558 	return;
2559  rescan_exit:
2560 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2561 			 "6169 Skip NVME Rport Rescan, NVME remoteport "
2562 			 "unregistered\n");
2563 #endif
2564 }
2565 
2566 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2567  *
2568  * There is no notion of Devloss or rport recovery from the current
2569  * nvme_transport perspective.  Loss of an rport just means IO cannot
2570  * be sent and recovery is completely up to the initator.
2571  * For now, the driver just unbinds the DID and port_role so that
2572  * no further IO can be issued.  Changes are planned for later.
2573  *
2574  * Notes - the ndlp reference count is not decremented here since
2575  * since there is no nvme_transport api for devloss.  Node ref count
2576  * is only adjusted in driver unload.
2577  */
2578 void
2579 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2580 {
2581 #if (IS_ENABLED(CONFIG_NVME_FC))
2582 	int ret;
2583 	struct nvme_fc_local_port *localport;
2584 	struct lpfc_nvme_lport *lport;
2585 	struct lpfc_nvme_rport *rport;
2586 	struct nvme_fc_remote_port *remoteport = NULL;
2587 
2588 	localport = vport->localport;
2589 
2590 	/* This is fundamental error.  The localport is always
2591 	 * available until driver unload.  Just exit.
2592 	 */
2593 	if (!localport)
2594 		return;
2595 
2596 	lport = (struct lpfc_nvme_lport *)localport->private;
2597 	if (!lport)
2598 		goto input_err;
2599 
2600 	spin_lock_irq(&ndlp->lock);
2601 	rport = lpfc_ndlp_get_nrport(ndlp);
2602 	if (rport)
2603 		remoteport = rport->remoteport;
2604 	spin_unlock_irq(&ndlp->lock);
2605 	if (!remoteport)
2606 		goto input_err;
2607 
2608 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2609 			 "6033 Unreg nvme remoteport x%px, portname x%llx, "
2610 			 "port_id x%06x, portstate x%x port type x%x "
2611 			 "refcnt %d\n",
2612 			 remoteport, remoteport->port_name,
2613 			 remoteport->port_id, remoteport->port_state,
2614 			 ndlp->nlp_type, kref_read(&ndlp->kref));
2615 
2616 	/* Sanity check ndlp type.  Only call for NVME ports. Don't
2617 	 * clear any rport state until the transport calls back.
2618 	 */
2619 
2620 	if (ndlp->nlp_type & NLP_NVME_TARGET) {
2621 		/* No concern about the role change on the nvme remoteport.
2622 		 * The transport will update it.
2623 		 */
2624 		spin_lock_irq(&vport->phba->hbalock);
2625 		ndlp->fc4_xpt_flags |= NVME_XPT_UNREG_WAIT;
2626 		spin_unlock_irq(&vport->phba->hbalock);
2627 
2628 		/* Don't let the host nvme transport keep sending keep-alives
2629 		 * on this remoteport. Vport is unloading, no recovery. The
2630 		 * return values is ignored.  The upcall is a courtesy to the
2631 		 * transport.
2632 		 */
2633 		if (vport->load_flag & FC_UNLOADING)
2634 			(void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2635 
2636 		ret = nvme_fc_unregister_remoteport(remoteport);
2637 
2638 		/* The driver no longer knows if the nrport memory is valid.
2639 		 * because the controller teardown process has begun and
2640 		 * is asynchronous.  Break the binding in the ndlp. Also
2641 		 * remove the register ndlp reference to setup node release.
2642 		 */
2643 		ndlp->nrport = NULL;
2644 		lpfc_nlp_put(ndlp);
2645 		if (ret != 0) {
2646 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2647 					 "6167 NVME unregister failed %d "
2648 					 "port_state x%x\n",
2649 					 ret, remoteport->port_state);
2650 		}
2651 	}
2652 	return;
2653 
2654  input_err:
2655 #endif
2656 	lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2657 			 "6168 State error: lport x%px, rport x%px FCID x%06x\n",
2658 			 vport->localport, ndlp->rport, ndlp->nlp_DID);
2659 }
2660 
2661 /**
2662  * lpfc_sli4_nvme_pci_offline_aborted - Fast-path process of NVME xri abort
2663  * @phba: pointer to lpfc hba data structure.
2664  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2665  *
2666  * This routine is invoked by the worker thread to process a SLI4 fast-path
2667  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2668  * here.
2669  **/
2670 void
2671 lpfc_sli4_nvme_pci_offline_aborted(struct lpfc_hba *phba,
2672 				   struct lpfc_io_buf *lpfc_ncmd)
2673 {
2674 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2675 
2676 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2677 			"6533 %s nvme_cmd %p tag x%x abort complete and "
2678 			"xri released\n", __func__,
2679 			lpfc_ncmd->nvmeCmd,
2680 			lpfc_ncmd->cur_iocbq.iotag);
2681 
2682 	/* Aborted NVME commands are required to not complete
2683 	 * before the abort exchange command fully completes.
2684 	 * Once completed, it is available via the put list.
2685 	 */
2686 	if (lpfc_ncmd->nvmeCmd) {
2687 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2688 		nvme_cmd->transferred_length = 0;
2689 		nvme_cmd->rcv_rsplen = 0;
2690 		nvme_cmd->status = NVME_SC_INTERNAL;
2691 		nvme_cmd->done(nvme_cmd);
2692 		lpfc_ncmd->nvmeCmd = NULL;
2693 	}
2694 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2695 }
2696 
2697 /**
2698  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2699  * @phba: pointer to lpfc hba data structure.
2700  * @axri: pointer to the fcp xri abort wcqe structure.
2701  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2702  *
2703  * This routine is invoked by the worker thread to process a SLI4 fast-path
2704  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2705  * here.
2706  **/
2707 void
2708 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2709 			   struct sli4_wcqe_xri_aborted *axri,
2710 			   struct lpfc_io_buf *lpfc_ncmd)
2711 {
2712 	uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2713 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2714 	struct lpfc_nodelist *ndlp = lpfc_ncmd->ndlp;
2715 
2716 
2717 	if (ndlp)
2718 		lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2719 
2720 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2721 			"6311 nvme_cmd %p xri x%x tag x%x abort complete and "
2722 			"xri released\n",
2723 			lpfc_ncmd->nvmeCmd, xri,
2724 			lpfc_ncmd->cur_iocbq.iotag);
2725 
2726 	/* Aborted NVME commands are required to not complete
2727 	 * before the abort exchange command fully completes.
2728 	 * Once completed, it is available via the put list.
2729 	 */
2730 	if (lpfc_ncmd->nvmeCmd) {
2731 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2732 		nvme_cmd->done(nvme_cmd);
2733 		lpfc_ncmd->nvmeCmd = NULL;
2734 	}
2735 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2736 }
2737 
2738 /**
2739  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2740  * @phba: Pointer to HBA context object.
2741  *
2742  * This function flushes all wqes in the nvme rings and frees all resources
2743  * in the txcmplq. This function does not issue abort wqes for the IO
2744  * commands in txcmplq, they will just be returned with
2745  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2746  * slot has been permanently disabled.
2747  **/
2748 void
2749 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2750 {
2751 	struct lpfc_sli_ring  *pring;
2752 	u32 i, wait_cnt = 0;
2753 
2754 	if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2755 		return;
2756 
2757 	/* Cycle through all IO rings and make sure all outstanding
2758 	 * WQEs have been removed from the txcmplqs.
2759 	 */
2760 	for (i = 0; i < phba->cfg_hdw_queue; i++) {
2761 		if (!phba->sli4_hba.hdwq[i].io_wq)
2762 			continue;
2763 		pring = phba->sli4_hba.hdwq[i].io_wq->pring;
2764 
2765 		if (!pring)
2766 			continue;
2767 
2768 		/* Retrieve everything on the txcmplq */
2769 		while (!list_empty(&pring->txcmplq)) {
2770 			msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2771 			wait_cnt++;
2772 
2773 			/* The sleep is 10mS.  Every ten seconds,
2774 			 * dump a message.  Something is wrong.
2775 			 */
2776 			if ((wait_cnt % 1000) == 0) {
2777 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
2778 						"6178 NVME IO not empty, "
2779 						"cnt %d\n", wait_cnt);
2780 			}
2781 		}
2782 	}
2783 
2784 	/* Make sure HBA is alive */
2785 	lpfc_issue_hb_tmo(phba);
2786 
2787 }
2788 
2789 void
2790 lpfc_nvme_cancel_iocb(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
2791 		      uint32_t stat, uint32_t param)
2792 {
2793 #if (IS_ENABLED(CONFIG_NVME_FC))
2794 	struct lpfc_io_buf *lpfc_ncmd;
2795 	struct nvmefc_fcp_req *nCmd;
2796 	struct lpfc_wcqe_complete wcqe;
2797 	struct lpfc_wcqe_complete *wcqep = &wcqe;
2798 
2799 	lpfc_ncmd = pwqeIn->io_buf;
2800 	if (!lpfc_ncmd) {
2801 		lpfc_sli_release_iocbq(phba, pwqeIn);
2802 		return;
2803 	}
2804 	/* For abort iocb just return, IO iocb will do a done call */
2805 	if (bf_get(wqe_cmnd, &pwqeIn->wqe.gen_req.wqe_com) ==
2806 	    CMD_ABORT_XRI_CX) {
2807 		lpfc_sli_release_iocbq(phba, pwqeIn);
2808 		return;
2809 	}
2810 
2811 	spin_lock(&lpfc_ncmd->buf_lock);
2812 	nCmd = lpfc_ncmd->nvmeCmd;
2813 	if (!nCmd) {
2814 		spin_unlock(&lpfc_ncmd->buf_lock);
2815 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
2816 		return;
2817 	}
2818 	spin_unlock(&lpfc_ncmd->buf_lock);
2819 
2820 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_IOERR,
2821 			"6194 NVME Cancel xri %x\n",
2822 			lpfc_ncmd->cur_iocbq.sli4_xritag);
2823 
2824 	wcqep->word0 = 0;
2825 	bf_set(lpfc_wcqe_c_status, wcqep, stat);
2826 	wcqep->parameter = param;
2827 	wcqep->word3 = 0; /* xb is 0 */
2828 
2829 	/* Call release with XB=1 to queue the IO into the abort list. */
2830 	if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
2831 		bf_set(lpfc_wcqe_c_xb, wcqep, 1);
2832 
2833 	memcpy(&pwqeIn->wcqe_cmpl, wcqep, sizeof(*wcqep));
2834 	(pwqeIn->cmd_cmpl)(phba, pwqeIn, pwqeIn);
2835 #endif
2836 }
2837