xref: /openbmc/linux/drivers/scsi/lpfc/lpfc_nvme.c (revision cef69974)
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 = (struct nvmefc_ls_req *)cmdwqe->context2;
323 	ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
324 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
325 
326 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
327 			 "6047 NVMEx LS REQ x%px cmpl DID %x Xri: %x "
328 			 "status %x reason x%x cmd:x%px lsreg:x%px bmp:x%px "
329 			 "ndlp:x%px\n",
330 			 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
331 			 cmdwqe->sli4_xritag, status,
332 			 (wcqe->parameter & 0xffff),
333 			 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
334 
335 	lpfc_nvmeio_data(phba, "NVMEx LS CMPL: xri x%x stat x%x parm x%x\n",
336 			 cmdwqe->sli4_xritag, status, wcqe->parameter);
337 
338 	if (cmdwqe->context3) {
339 		buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
340 		lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
341 		kfree(buf_ptr);
342 		cmdwqe->context3 = NULL;
343 	}
344 	if (pnvme_lsreq->done)
345 		pnvme_lsreq->done(pnvme_lsreq, status);
346 	else
347 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
348 				 "6046 NVMEx cmpl without done call back? "
349 				 "Data x%px DID %x Xri: %x status %x\n",
350 				pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
351 				cmdwqe->sli4_xritag, status);
352 	if (ndlp) {
353 		lpfc_nlp_put(ndlp);
354 		cmdwqe->context1 = NULL;
355 	}
356 	lpfc_sli_release_iocbq(phba, cmdwqe);
357 }
358 
359 static void
360 lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
361 		     struct lpfc_iocbq *rspwqe)
362 {
363 	struct lpfc_vport *vport = cmdwqe->vport;
364 	struct lpfc_nvme_lport *lport;
365 	uint32_t status;
366 	struct lpfc_wcqe_complete *wcqe = &rspwqe->wcqe_cmpl;
367 
368 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
369 
370 	if (vport->localport) {
371 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
372 		if (lport) {
373 			atomic_inc(&lport->fc4NvmeLsCmpls);
374 			if (status) {
375 				if (bf_get(lpfc_wcqe_c_xb, wcqe))
376 					atomic_inc(&lport->cmpl_ls_xb);
377 				atomic_inc(&lport->cmpl_ls_err);
378 			}
379 		}
380 	}
381 
382 	__lpfc_nvme_ls_req_cmp(phba, vport, cmdwqe, wcqe);
383 }
384 
385 static int
386 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
387 		  struct lpfc_dmabuf *inp,
388 		  struct nvmefc_ls_req *pnvme_lsreq,
389 		  void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
390 			       struct lpfc_iocbq *),
391 		  struct lpfc_nodelist *ndlp, uint32_t num_entry,
392 		  uint32_t tmo, uint8_t retry)
393 {
394 	struct lpfc_hba *phba = vport->phba;
395 	union lpfc_wqe128 *wqe;
396 	struct lpfc_iocbq *genwqe;
397 	struct ulp_bde64 *bpl;
398 	struct ulp_bde64 bde;
399 	int i, rc, xmit_len, first_len;
400 
401 	/* Allocate buffer for  command WQE */
402 	genwqe = lpfc_sli_get_iocbq(phba);
403 	if (genwqe == NULL)
404 		return 1;
405 
406 	wqe = &genwqe->wqe;
407 	/* Initialize only 64 bytes */
408 	memset(wqe, 0, sizeof(union lpfc_wqe));
409 
410 	genwqe->context3 = (uint8_t *)bmp;
411 	genwqe->cmd_flag |= LPFC_IO_NVME_LS;
412 
413 	/* Save for completion so we can release these resources */
414 	genwqe->context1 = lpfc_nlp_get(ndlp);
415 	if (!genwqe->context1) {
416 		dev_warn(&phba->pcidev->dev,
417 			 "Warning: Failed node ref, not sending LS_REQ\n");
418 		lpfc_sli_release_iocbq(phba, genwqe);
419 		return 1;
420 	}
421 
422 	genwqe->context2 = (uint8_t *)pnvme_lsreq;
423 	/* Fill in payload, bp points to frame payload */
424 
425 	if (!tmo)
426 		/* FC spec states we need 3 * ratov for CT requests */
427 		tmo = (3 * phba->fc_ratov);
428 
429 	/* For this command calculate the xmit length of the request bde. */
430 	xmit_len = 0;
431 	first_len = 0;
432 	bpl = (struct ulp_bde64 *)bmp->virt;
433 	for (i = 0; i < num_entry; i++) {
434 		bde.tus.w = bpl[i].tus.w;
435 		if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
436 			break;
437 		xmit_len += bde.tus.f.bdeSize;
438 		if (i == 0)
439 			first_len = xmit_len;
440 	}
441 
442 	genwqe->num_bdes = num_entry;
443 	genwqe->hba_wqidx = 0;
444 
445 	/* Words 0 - 2 */
446 	wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
447 	wqe->generic.bde.tus.f.bdeSize = first_len;
448 	wqe->generic.bde.addrLow = bpl[0].addrLow;
449 	wqe->generic.bde.addrHigh = bpl[0].addrHigh;
450 
451 	/* Word 3 */
452 	wqe->gen_req.request_payload_len = first_len;
453 
454 	/* Word 4 */
455 
456 	/* Word 5 */
457 	bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
458 	bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
459 	bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
460 	bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
461 	bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
462 
463 	/* Word 6 */
464 	bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
465 	       phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
466 	bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
467 
468 	/* Word 7 */
469 	bf_set(wqe_tmo, &wqe->gen_req.wqe_com, tmo);
470 	bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
471 	bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
472 	bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
473 
474 	/* Word 8 */
475 	wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
476 
477 	/* Word 9 */
478 	bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
479 
480 	/* Word 10 */
481 	bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
482 	bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
483 	bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
484 	bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
485 	bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
486 
487 	/* Word 11 */
488 	bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
489 	bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
490 
491 
492 	/* Issue GEN REQ WQE for NPORT <did> */
493 	genwqe->cmd_cmpl = cmpl;
494 	genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
495 	genwqe->vport = vport;
496 	genwqe->retry = retry;
497 
498 	lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
499 			 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
500 
501 	rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe);
502 	if (rc) {
503 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
504 				 "6045 Issue GEN REQ WQE to NPORT x%x "
505 				 "Data: x%x x%x  rc x%x\n",
506 				 ndlp->nlp_DID, genwqe->iotag,
507 				 vport->port_state, rc);
508 		lpfc_nlp_put(ndlp);
509 		lpfc_sli_release_iocbq(phba, genwqe);
510 		return 1;
511 	}
512 
513 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_ELS,
514 			 "6050 Issue GEN REQ WQE to NPORT x%x "
515 			 "Data: oxid: x%x state: x%x wq:x%px lsreq:x%px "
516 			 "bmp:x%px xmit:%d 1st:%d\n",
517 			 ndlp->nlp_DID, genwqe->sli4_xritag,
518 			 vport->port_state,
519 			 genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
520 	return 0;
521 }
522 
523 
524 /**
525  * __lpfc_nvme_ls_req - Generic service routine to issue an NVME LS request
526  * @vport: The local port issuing the LS
527  * @ndlp: The remote port to send the LS to
528  * @pnvme_lsreq: Pointer to LS request structure from the transport
529  * @gen_req_cmp: Completion call-back
530  *
531  * Routine validates the ndlp, builds buffers and sends a GEN_REQUEST
532  * WQE to perform the LS operation.
533  *
534  * Return value :
535  *   0 - Success
536  *   non-zero: various error codes, in form of -Exxx
537  **/
538 int
539 __lpfc_nvme_ls_req(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
540 		      struct nvmefc_ls_req *pnvme_lsreq,
541 		      void (*gen_req_cmp)(struct lpfc_hba *phba,
542 				struct lpfc_iocbq *cmdwqe,
543 				struct lpfc_iocbq *rspwqe))
544 {
545 	struct lpfc_dmabuf *bmp;
546 	struct ulp_bde64 *bpl;
547 	int ret;
548 	uint16_t ntype, nstate;
549 
550 	if (!ndlp) {
551 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
552 				 "6051 NVMEx LS REQ: Bad NDLP x%px, Failing "
553 				 "LS Req\n",
554 				 ndlp);
555 		return -ENODEV;
556 	}
557 
558 	ntype = ndlp->nlp_type;
559 	nstate = ndlp->nlp_state;
560 	if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
561 	    (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
562 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
563 				 "6088 NVMEx LS REQ: Fail DID x%06x not "
564 				 "ready for IO. Type x%x, State x%x\n",
565 				 ndlp->nlp_DID, ntype, nstate);
566 		return -ENODEV;
567 	}
568 	if (vport->phba->hba_flag & HBA_IOQ_FLUSH)
569 		return -ENODEV;
570 
571 	if (!vport->phba->sli4_hba.nvmels_wq)
572 		return -ENOMEM;
573 
574 	/*
575 	 * there are two dma buf in the request, actually there is one and
576 	 * the second one is just the start address + cmd size.
577 	 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
578 	 * in a lpfc_dmabuf struct. When freeing we just free the wrapper
579 	 * because the nvem layer owns the data bufs.
580 	 * We do not have to break these packets open, we don't care what is
581 	 * in them. And we do not have to look at the resonse data, we only
582 	 * care that we got a response. All of the caring is going to happen
583 	 * in the nvme-fc layer.
584 	 */
585 
586 	bmp = kmalloc(sizeof(*bmp), GFP_KERNEL);
587 	if (!bmp) {
588 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
589 				 "6044 NVMEx LS REQ: Could not alloc LS buf "
590 				 "for DID %x\n",
591 				 ndlp->nlp_DID);
592 		return -ENOMEM;
593 	}
594 
595 	bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
596 	if (!bmp->virt) {
597 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
598 				 "6042 NVMEx LS REQ: Could not alloc mbuf "
599 				 "for DID %x\n",
600 				 ndlp->nlp_DID);
601 		kfree(bmp);
602 		return -ENOMEM;
603 	}
604 
605 	INIT_LIST_HEAD(&bmp->list);
606 
607 	bpl = (struct ulp_bde64 *)bmp->virt;
608 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
609 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
610 	bpl->tus.f.bdeFlags = 0;
611 	bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
612 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
613 	bpl++;
614 
615 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
616 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
617 	bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
618 	bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
619 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
620 
621 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
622 			"6149 NVMEx LS REQ: Issue to DID 0x%06x lsreq x%px, "
623 			"rqstlen:%d rsplen:%d %pad %pad\n",
624 			ndlp->nlp_DID, pnvme_lsreq, pnvme_lsreq->rqstlen,
625 			pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
626 			&pnvme_lsreq->rspdma);
627 
628 	ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
629 				pnvme_lsreq, gen_req_cmp, ndlp, 2,
630 				pnvme_lsreq->timeout, 0);
631 	if (ret != WQE_SUCCESS) {
632 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
633 				 "6052 NVMEx REQ: EXIT. issue ls wqe failed "
634 				 "lsreq x%px Status %x DID %x\n",
635 				 pnvme_lsreq, ret, ndlp->nlp_DID);
636 		lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
637 		kfree(bmp);
638 		return -EIO;
639 	}
640 
641 	return 0;
642 }
643 
644 /**
645  * lpfc_nvme_ls_req - Issue an NVME Link Service request
646  * @pnvme_lport: Transport localport that LS is to be issued from.
647  * @pnvme_rport: Transport remoteport that LS is to be sent to.
648  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
649  *
650  * Driver registers this routine to handle any link service request
651  * from the nvme_fc transport to a remote nvme-aware port.
652  *
653  * Return value :
654  *   0 - Success
655  *   non-zero: various error codes, in form of -Exxx
656  **/
657 static int
658 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
659 		 struct nvme_fc_remote_port *pnvme_rport,
660 		 struct nvmefc_ls_req *pnvme_lsreq)
661 {
662 	struct lpfc_nvme_lport *lport;
663 	struct lpfc_nvme_rport *rport;
664 	struct lpfc_vport *vport;
665 	int ret;
666 
667 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
668 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
669 	if (unlikely(!lport) || unlikely(!rport))
670 		return -EINVAL;
671 
672 	vport = lport->vport;
673 	if (vport->load_flag & FC_UNLOADING ||
674 	    vport->phba->hba_flag & HBA_IOQ_FLUSH)
675 		return -ENODEV;
676 
677 	atomic_inc(&lport->fc4NvmeLsRequests);
678 
679 	ret = __lpfc_nvme_ls_req(vport, rport->ndlp, pnvme_lsreq,
680 				 lpfc_nvme_ls_req_cmp);
681 	if (ret)
682 		atomic_inc(&lport->xmt_ls_err);
683 
684 	return ret;
685 }
686 
687 /**
688  * __lpfc_nvme_ls_abort - Generic service routine to abort a prior
689  *         NVME LS request
690  * @vport: The local port that issued the LS
691  * @ndlp: The remote port the LS was sent to
692  * @pnvme_lsreq: Pointer to LS request structure from the transport
693  *
694  * The driver validates the ndlp, looks for the LS, and aborts the
695  * LS if found.
696  *
697  * Returns:
698  * 0 : if LS found and aborted
699  * non-zero: various error conditions in form -Exxx
700  **/
701 int
702 __lpfc_nvme_ls_abort(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
703 			struct nvmefc_ls_req *pnvme_lsreq)
704 {
705 	struct lpfc_hba *phba = vport->phba;
706 	struct lpfc_sli_ring *pring;
707 	struct lpfc_iocbq *wqe, *next_wqe;
708 	bool foundit = false;
709 
710 	if (!ndlp) {
711 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
712 				"6049 NVMEx LS REQ Abort: Bad NDLP x%px DID "
713 				"x%06x, Failing LS Req\n",
714 				ndlp, ndlp ? ndlp->nlp_DID : 0);
715 		return -EINVAL;
716 	}
717 
718 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
719 			 "6040 NVMEx LS REQ Abort: Issue LS_ABORT for lsreq "
720 			 "x%px rqstlen:%d rsplen:%d %pad %pad\n",
721 			 pnvme_lsreq, pnvme_lsreq->rqstlen,
722 			 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
723 			 &pnvme_lsreq->rspdma);
724 
725 	/*
726 	 * Lock the ELS ring txcmplq and look for the wqe that matches
727 	 * this ELS. If found, issue an abort on the wqe.
728 	 */
729 	pring = phba->sli4_hba.nvmels_wq->pring;
730 	spin_lock_irq(&phba->hbalock);
731 	spin_lock(&pring->ring_lock);
732 	list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
733 		if (wqe->context2 == pnvme_lsreq) {
734 			wqe->cmd_flag |= LPFC_DRIVER_ABORTED;
735 			foundit = true;
736 			break;
737 		}
738 	}
739 	spin_unlock(&pring->ring_lock);
740 
741 	if (foundit)
742 		lpfc_sli_issue_abort_iotag(phba, pring, wqe, NULL);
743 	spin_unlock_irq(&phba->hbalock);
744 
745 	if (foundit)
746 		return 0;
747 
748 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
749 			 "6213 NVMEx LS REQ Abort: Unable to locate req x%px\n",
750 			 pnvme_lsreq);
751 	return -EINVAL;
752 }
753 
754 static int
755 lpfc_nvme_xmt_ls_rsp(struct nvme_fc_local_port *localport,
756 		     struct nvme_fc_remote_port *remoteport,
757 		     struct nvmefc_ls_rsp *ls_rsp)
758 {
759 	struct lpfc_async_xchg_ctx *axchg =
760 		container_of(ls_rsp, struct lpfc_async_xchg_ctx, ls_rsp);
761 	struct lpfc_nvme_lport *lport;
762 	int rc;
763 
764 	if (axchg->phba->pport->load_flag & FC_UNLOADING)
765 		return -ENODEV;
766 
767 	lport = (struct lpfc_nvme_lport *)localport->private;
768 
769 	rc = __lpfc_nvme_xmt_ls_rsp(axchg, ls_rsp, __lpfc_nvme_xmt_ls_rsp_cmp);
770 
771 	if (rc) {
772 		/*
773 		 * unless the failure is due to having already sent
774 		 * the response, an abort will be generated for the
775 		 * exchange if the rsp can't be sent.
776 		 */
777 		if (rc != -EALREADY)
778 			atomic_inc(&lport->xmt_ls_abort);
779 		return rc;
780 	}
781 
782 	return 0;
783 }
784 
785 /**
786  * lpfc_nvme_ls_abort - Abort a prior NVME LS request
787  * @pnvme_lport: Transport localport that LS is to be issued from.
788  * @pnvme_rport: Transport remoteport that LS is to be sent to.
789  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
790  *
791  * Driver registers this routine to abort a NVME LS request that is
792  * in progress (from the transports perspective).
793  **/
794 static void
795 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
796 		   struct nvme_fc_remote_port *pnvme_rport,
797 		   struct nvmefc_ls_req *pnvme_lsreq)
798 {
799 	struct lpfc_nvme_lport *lport;
800 	struct lpfc_vport *vport;
801 	struct lpfc_nodelist *ndlp;
802 	int ret;
803 
804 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
805 	if (unlikely(!lport))
806 		return;
807 	vport = lport->vport;
808 
809 	if (vport->load_flag & FC_UNLOADING)
810 		return;
811 
812 	ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
813 
814 	ret = __lpfc_nvme_ls_abort(vport, ndlp, pnvme_lsreq);
815 	if (!ret)
816 		atomic_inc(&lport->xmt_ls_abort);
817 }
818 
819 /* Fix up the existing sgls for NVME IO. */
820 static inline void
821 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
822 		       struct lpfc_io_buf *lpfc_ncmd,
823 		       struct nvmefc_fcp_req *nCmd)
824 {
825 	struct lpfc_hba  *phba = vport->phba;
826 	struct sli4_sge *sgl;
827 	union lpfc_wqe128 *wqe;
828 	uint32_t *wptr, *dptr;
829 
830 	/*
831 	 * Get a local pointer to the built-in wqe and correct
832 	 * the cmd size to match NVME's 96 bytes and fix
833 	 * the dma address.
834 	 */
835 
836 	wqe = &lpfc_ncmd->cur_iocbq.wqe;
837 
838 	/*
839 	 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
840 	 * match NVME.  NVME sends 96 bytes. Also, use the
841 	 * nvme commands command and response dma addresses
842 	 * rather than the virtual memory to ease the restore
843 	 * operation.
844 	 */
845 	sgl = lpfc_ncmd->dma_sgl;
846 	sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
847 	if (phba->cfg_nvme_embed_cmd) {
848 		sgl->addr_hi = 0;
849 		sgl->addr_lo = 0;
850 
851 		/* Word 0-2 - NVME CMND IU (embedded payload) */
852 		wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
853 		wqe->generic.bde.tus.f.bdeSize = 56;
854 		wqe->generic.bde.addrHigh = 0;
855 		wqe->generic.bde.addrLow =  64;  /* Word 16 */
856 
857 		/* Word 10  - dbde is 0, wqes is 1 in template */
858 
859 		/*
860 		 * Embed the payload in the last half of the WQE
861 		 * WQE words 16-30 get the NVME CMD IU payload
862 		 *
863 		 * WQE words 16-19 get payload Words 1-4
864 		 * WQE words 20-21 get payload Words 6-7
865 		 * WQE words 22-29 get payload Words 16-23
866 		 */
867 		wptr = &wqe->words[16];  /* WQE ptr */
868 		dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
869 		dptr++;			/* Skip Word 0 in payload */
870 
871 		*wptr++ = *dptr++;	/* Word 1 */
872 		*wptr++ = *dptr++;	/* Word 2 */
873 		*wptr++ = *dptr++;	/* Word 3 */
874 		*wptr++ = *dptr++;	/* Word 4 */
875 		dptr++;			/* Skip Word 5 in payload */
876 		*wptr++ = *dptr++;	/* Word 6 */
877 		*wptr++ = *dptr++;	/* Word 7 */
878 		dptr += 8;		/* Skip Words 8-15 in payload */
879 		*wptr++ = *dptr++;	/* Word 16 */
880 		*wptr++ = *dptr++;	/* Word 17 */
881 		*wptr++ = *dptr++;	/* Word 18 */
882 		*wptr++ = *dptr++;	/* Word 19 */
883 		*wptr++ = *dptr++;	/* Word 20 */
884 		*wptr++ = *dptr++;	/* Word 21 */
885 		*wptr++ = *dptr++;	/* Word 22 */
886 		*wptr   = *dptr;	/* Word 23 */
887 	} else {
888 		sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
889 		sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
890 
891 		/* Word 0-2 - NVME CMND IU Inline BDE */
892 		wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
893 		wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
894 		wqe->generic.bde.addrHigh = sgl->addr_hi;
895 		wqe->generic.bde.addrLow =  sgl->addr_lo;
896 
897 		/* Word 10 */
898 		bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
899 		bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
900 	}
901 
902 	sgl++;
903 
904 	/* Setup the physical region for the FCP RSP */
905 	sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
906 	sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
907 	sgl->word2 = le32_to_cpu(sgl->word2);
908 	if (nCmd->sg_cnt)
909 		bf_set(lpfc_sli4_sge_last, sgl, 0);
910 	else
911 		bf_set(lpfc_sli4_sge_last, sgl, 1);
912 	sgl->word2 = cpu_to_le32(sgl->word2);
913 	sgl->sge_len = cpu_to_le32(nCmd->rsplen);
914 }
915 
916 
917 /*
918  * lpfc_nvme_io_cmd_cmpl - Complete an NVME-over-FCP IO
919  *
920  * Driver registers this routine as it io request handler.  This
921  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
922  * data structure to the rport indicated in @lpfc_nvme_rport.
923  *
924  * Return value :
925  *   0 - Success
926  *   TODO: What are the failure codes.
927  **/
928 static void
929 lpfc_nvme_io_cmd_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
930 		      struct lpfc_iocbq *pwqeOut)
931 {
932 	struct lpfc_io_buf *lpfc_ncmd =
933 		(struct lpfc_io_buf *)pwqeIn->context1;
934 	struct lpfc_wcqe_complete *wcqe = &pwqeOut->wcqe_cmpl;
935 	struct lpfc_vport *vport = pwqeIn->vport;
936 	struct nvmefc_fcp_req *nCmd;
937 	struct nvme_fc_ersp_iu *ep;
938 	struct nvme_fc_cmd_iu *cp;
939 	struct lpfc_nodelist *ndlp;
940 	struct lpfc_nvme_fcpreq_priv *freqpriv;
941 	struct lpfc_nvme_lport *lport;
942 	uint32_t code, status, idx;
943 	uint16_t cid, sqhd, data;
944 	uint32_t *ptr;
945 	uint32_t lat;
946 	bool call_done = false;
947 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
948 	int cpu;
949 #endif
950 	int offline = 0;
951 
952 	/* Sanity check on return of outstanding command */
953 	if (!lpfc_ncmd) {
954 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
955 				 "6071 Null lpfc_ncmd pointer. No "
956 				 "release, skip completion\n");
957 		return;
958 	}
959 
960 	/* Guard against abort handler being called at same time */
961 	spin_lock(&lpfc_ncmd->buf_lock);
962 
963 	if (!lpfc_ncmd->nvmeCmd) {
964 		spin_unlock(&lpfc_ncmd->buf_lock);
965 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
966 				 "6066 Missing cmpl ptrs: lpfc_ncmd x%px, "
967 				 "nvmeCmd x%px\n",
968 				 lpfc_ncmd, lpfc_ncmd->nvmeCmd);
969 
970 		/* Release the lpfc_ncmd regardless of the missing elements. */
971 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
972 		return;
973 	}
974 	nCmd = lpfc_ncmd->nvmeCmd;
975 	status = bf_get(lpfc_wcqe_c_status, wcqe);
976 
977 	idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
978 	phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++;
979 
980 	if (unlikely(status && vport->localport)) {
981 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
982 		if (lport) {
983 			if (bf_get(lpfc_wcqe_c_xb, wcqe))
984 				atomic_inc(&lport->cmpl_fcp_xb);
985 			atomic_inc(&lport->cmpl_fcp_err);
986 		}
987 	}
988 
989 	lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
990 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
991 			 status, wcqe->parameter);
992 	/*
993 	 * Catch race where our node has transitioned, but the
994 	 * transport is still transitioning.
995 	 */
996 	ndlp = lpfc_ncmd->ndlp;
997 	if (!ndlp) {
998 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
999 				 "6062 Ignoring NVME cmpl.  No ndlp\n");
1000 		goto out_err;
1001 	}
1002 
1003 	code = bf_get(lpfc_wcqe_c_code, wcqe);
1004 	if (code == CQE_CODE_NVME_ERSP) {
1005 		/* For this type of CQE, we need to rebuild the rsp */
1006 		ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1007 
1008 		/*
1009 		 * Get Command Id from cmd to plug into response. This
1010 		 * code is not needed in the next NVME Transport drop.
1011 		 */
1012 		cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1013 		cid = cp->sqe.common.command_id;
1014 
1015 		/*
1016 		 * RSN is in CQE word 2
1017 		 * SQHD is in CQE Word 3 bits 15:0
1018 		 * Cmd Specific info is in CQE Word 1
1019 		 * and in CQE Word 0 bits 15:0
1020 		 */
1021 		sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1022 
1023 		/* Now lets build the NVME ERSP IU */
1024 		ep->iu_len = cpu_to_be16(8);
1025 		ep->rsn = wcqe->parameter;
1026 		ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1027 		ep->rsvd12 = 0;
1028 		ptr = (uint32_t *)&ep->cqe.result.u64;
1029 		*ptr++ = wcqe->total_data_placed;
1030 		data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1031 		*ptr = (uint32_t)data;
1032 		ep->cqe.sq_head = sqhd;
1033 		ep->cqe.sq_id =  nCmd->sqid;
1034 		ep->cqe.command_id = cid;
1035 		ep->cqe.status = 0;
1036 
1037 		lpfc_ncmd->status = IOSTAT_SUCCESS;
1038 		lpfc_ncmd->result = 0;
1039 		nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1040 		nCmd->transferred_length = nCmd->payload_length;
1041 	} else {
1042 		lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1043 		lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1044 
1045 		/* For NVME, the only failure path that results in an
1046 		 * IO error is when the adapter rejects it.  All other
1047 		 * conditions are a success case and resolved by the
1048 		 * transport.
1049 		 * IOSTAT_FCP_RSP_ERROR means:
1050 		 * 1. Length of data received doesn't match total
1051 		 *    transfer length in WQE
1052 		 * 2. If the RSP payload does NOT match these cases:
1053 		 *    a. RSP length 12/24 bytes and all zeros
1054 		 *    b. NVME ERSP
1055 		 */
1056 		switch (lpfc_ncmd->status) {
1057 		case IOSTAT_SUCCESS:
1058 			nCmd->transferred_length = wcqe->total_data_placed;
1059 			nCmd->rcv_rsplen = 0;
1060 			nCmd->status = 0;
1061 			break;
1062 		case IOSTAT_FCP_RSP_ERROR:
1063 			nCmd->transferred_length = wcqe->total_data_placed;
1064 			nCmd->rcv_rsplen = wcqe->parameter;
1065 			nCmd->status = 0;
1066 
1067 			/* Check if this is really an ERSP */
1068 			if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) {
1069 				lpfc_ncmd->status = IOSTAT_SUCCESS;
1070 				lpfc_ncmd->result = 0;
1071 
1072 				lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
1073 					 "6084 NVME Completion ERSP: "
1074 					 "xri %x placed x%x\n",
1075 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1076 					 wcqe->total_data_placed);
1077 				break;
1078 			}
1079 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1080 					 "6081 NVME Completion Protocol Error: "
1081 					 "xri %x status x%x result x%x "
1082 					 "placed x%x\n",
1083 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1084 					 lpfc_ncmd->status, lpfc_ncmd->result,
1085 					 wcqe->total_data_placed);
1086 			break;
1087 		case IOSTAT_LOCAL_REJECT:
1088 			/* Let fall through to set command final state. */
1089 			if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1090 				lpfc_printf_vlog(vport, KERN_INFO,
1091 					 LOG_NVME_IOERR,
1092 					 "6032 Delay Aborted cmd x%px "
1093 					 "nvme cmd x%px, xri x%x, "
1094 					 "xb %d\n",
1095 					 lpfc_ncmd, nCmd,
1096 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1097 					 bf_get(lpfc_wcqe_c_xb, wcqe));
1098 			fallthrough;
1099 		default:
1100 out_err:
1101 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1102 					 "6072 NVME Completion Error: xri %x "
1103 					 "status x%x result x%x [x%x] "
1104 					 "placed x%x\n",
1105 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1106 					 lpfc_ncmd->status, lpfc_ncmd->result,
1107 					 wcqe->parameter,
1108 					 wcqe->total_data_placed);
1109 			nCmd->transferred_length = 0;
1110 			nCmd->rcv_rsplen = 0;
1111 			nCmd->status = NVME_SC_INTERNAL;
1112 			offline = pci_channel_offline(vport->phba->pcidev);
1113 		}
1114 	}
1115 
1116 	/* pick up SLI4 exhange busy condition */
1117 	if (bf_get(lpfc_wcqe_c_xb, wcqe) && !offline)
1118 		lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1119 	else
1120 		lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1121 
1122 	/* Update stats and complete the IO.  There is
1123 	 * no need for dma unprep because the nvme_transport
1124 	 * owns the dma address.
1125 	 */
1126 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1127 	if (lpfc_ncmd->ts_cmd_start) {
1128 		lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1129 		lpfc_ncmd->ts_data_io = ktime_get_ns();
1130 		phba->ktime_last_cmd = lpfc_ncmd->ts_data_io;
1131 		lpfc_io_ktime(phba, lpfc_ncmd);
1132 	}
1133 	if (unlikely(phba->hdwqstat_on & LPFC_CHECK_NVME_IO)) {
1134 		cpu = raw_smp_processor_id();
1135 		this_cpu_inc(phba->sli4_hba.c_stat->cmpl_io);
1136 		if (lpfc_ncmd->cpu != cpu)
1137 			lpfc_printf_vlog(vport,
1138 					 KERN_INFO, LOG_NVME_IOERR,
1139 					 "6701 CPU Check cmpl: "
1140 					 "cpu %d expect %d\n",
1141 					 cpu, lpfc_ncmd->cpu);
1142 	}
1143 #endif
1144 
1145 	/* NVME targets need completion held off until the abort exchange
1146 	 * completes unless the NVME Rport is getting unregistered.
1147 	 */
1148 
1149 	if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1150 		freqpriv = nCmd->private;
1151 		freqpriv->nvme_buf = NULL;
1152 		lpfc_ncmd->nvmeCmd = NULL;
1153 		call_done = true;
1154 	}
1155 	spin_unlock(&lpfc_ncmd->buf_lock);
1156 
1157 	/* Check if IO qualified for CMF */
1158 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1159 	    nCmd->io_dir == NVMEFC_FCP_READ &&
1160 	    nCmd->payload_length) {
1161 		/* Used when calculating average latency */
1162 		lat = ktime_get_ns() - lpfc_ncmd->rx_cmd_start;
1163 		lpfc_update_cmf_cmpl(phba, lat, nCmd->payload_length, NULL);
1164 	}
1165 
1166 	if (call_done)
1167 		nCmd->done(nCmd);
1168 
1169 	/* Call release with XB=1 to queue the IO into the abort list. */
1170 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1171 }
1172 
1173 
1174 /**
1175  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1176  * @vport: pointer to a host virtual N_Port data structure
1177  * @lpfc_ncmd: Pointer to lpfc scsi command
1178  * @pnode: pointer to a node-list data structure
1179  * @cstat: pointer to the control status structure
1180  *
1181  * Driver registers this routine as it io request handler.  This
1182  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1183  * data structure to the rport indicated in @lpfc_nvme_rport.
1184  *
1185  * Return value :
1186  *   0 - Success
1187  *   TODO: What are the failure codes.
1188  **/
1189 static int
1190 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1191 		      struct lpfc_io_buf *lpfc_ncmd,
1192 		      struct lpfc_nodelist *pnode,
1193 		      struct lpfc_fc4_ctrl_stat *cstat)
1194 {
1195 	struct lpfc_hba *phba = vport->phba;
1196 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1197 	struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1198 	union lpfc_wqe128 *wqe = &pwqeq->wqe;
1199 	uint32_t req_len;
1200 
1201 	/*
1202 	 * There are three possibilities here - use scatter-gather segment, use
1203 	 * the single mapping, or neither.
1204 	 */
1205 	if (nCmd->sg_cnt) {
1206 		if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1207 			/* From the iwrite template, initialize words 7 - 11 */
1208 			memcpy(&wqe->words[7],
1209 			       &lpfc_iwrite_cmd_template.words[7],
1210 			       sizeof(uint32_t) * 5);
1211 
1212 			/* Word 4 */
1213 			wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1214 
1215 			/* Word 5 */
1216 			if ((phba->cfg_nvme_enable_fb) &&
1217 			    (pnode->nlp_flag & NLP_FIRSTBURST)) {
1218 				req_len = lpfc_ncmd->nvmeCmd->payload_length;
1219 				if (req_len < pnode->nvme_fb_size)
1220 					wqe->fcp_iwrite.initial_xfer_len =
1221 						req_len;
1222 				else
1223 					wqe->fcp_iwrite.initial_xfer_len =
1224 						pnode->nvme_fb_size;
1225 			} else {
1226 				wqe->fcp_iwrite.initial_xfer_len = 0;
1227 			}
1228 			cstat->output_requests++;
1229 		} else {
1230 			/* From the iread template, initialize words 7 - 11 */
1231 			memcpy(&wqe->words[7],
1232 			       &lpfc_iread_cmd_template.words[7],
1233 			       sizeof(uint32_t) * 5);
1234 
1235 			/* Word 4 */
1236 			wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1237 
1238 			/* Word 5 */
1239 			wqe->fcp_iread.rsrvd5 = 0;
1240 
1241 			/* For a CMF Managed port, iod must be zero'ed */
1242 			if (phba->cmf_active_mode == LPFC_CFG_MANAGED)
1243 				bf_set(wqe_iod, &wqe->fcp_iread.wqe_com,
1244 				       LPFC_WQE_IOD_NONE);
1245 			cstat->input_requests++;
1246 		}
1247 	} else {
1248 		/* From the icmnd template, initialize words 4 - 11 */
1249 		memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1250 		       sizeof(uint32_t) * 8);
1251 		cstat->control_requests++;
1252 	}
1253 
1254 	if (pnode->nlp_nvme_info & NLP_NVME_NSLER)
1255 		bf_set(wqe_erp, &wqe->generic.wqe_com, 1);
1256 	/*
1257 	 * Finish initializing those WQE fields that are independent
1258 	 * of the nvme_cmnd request_buffer
1259 	 */
1260 
1261 	/* Word 3 */
1262 	bf_set(payload_offset_len, &wqe->fcp_icmd,
1263 	       (nCmd->rsplen + nCmd->cmdlen));
1264 
1265 	/* Word 6 */
1266 	bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1267 	       phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1268 	bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1269 
1270 	/* Word 8 */
1271 	wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1272 
1273 	/* Word 9 */
1274 	bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1275 
1276 	/* Word 10 */
1277 	bf_set(wqe_xchg, &wqe->fcp_iwrite.wqe_com, LPFC_NVME_XCHG);
1278 
1279 	/* Words 13 14 15 are for PBDE support */
1280 
1281 	pwqeq->vport = vport;
1282 	return 0;
1283 }
1284 
1285 
1286 /**
1287  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1288  * @vport: pointer to a host virtual N_Port data structure
1289  * @lpfc_ncmd: Pointer to lpfc scsi command
1290  *
1291  * Driver registers this routine as it io request handler.  This
1292  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1293  * data structure to the rport indicated in @lpfc_nvme_rport.
1294  *
1295  * Return value :
1296  *   0 - Success
1297  *   TODO: What are the failure codes.
1298  **/
1299 static int
1300 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1301 		      struct lpfc_io_buf *lpfc_ncmd)
1302 {
1303 	struct lpfc_hba *phba = vport->phba;
1304 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1305 	union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1306 	struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1307 	struct sli4_hybrid_sgl *sgl_xtra = NULL;
1308 	struct scatterlist *data_sg;
1309 	struct sli4_sge *first_data_sgl;
1310 	struct ulp_bde64 *bde;
1311 	dma_addr_t physaddr = 0;
1312 	uint32_t dma_len = 0;
1313 	uint32_t dma_offset = 0;
1314 	int nseg, i, j;
1315 	bool lsp_just_set = false;
1316 
1317 	/* Fix up the command and response DMA stuff. */
1318 	lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1319 
1320 	/*
1321 	 * There are three possibilities here - use scatter-gather segment, use
1322 	 * the single mapping, or neither.
1323 	 */
1324 	if (nCmd->sg_cnt) {
1325 		/*
1326 		 * Jump over the cmd and rsp SGEs.  The fix routine
1327 		 * has already adjusted for this.
1328 		 */
1329 		sgl += 2;
1330 
1331 		first_data_sgl = sgl;
1332 		lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1333 		if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1334 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1335 					"6058 Too many sg segments from "
1336 					"NVME Transport.  Max %d, "
1337 					"nvmeIO sg_cnt %d\n",
1338 					phba->cfg_nvme_seg_cnt + 1,
1339 					lpfc_ncmd->seg_cnt);
1340 			lpfc_ncmd->seg_cnt = 0;
1341 			return 1;
1342 		}
1343 
1344 		/*
1345 		 * The driver established a maximum scatter-gather segment count
1346 		 * during probe that limits the number of sg elements in any
1347 		 * single nvme command.  Just run through the seg_cnt and format
1348 		 * the sge's.
1349 		 */
1350 		nseg = nCmd->sg_cnt;
1351 		data_sg = nCmd->first_sgl;
1352 
1353 		/* for tracking the segment boundaries */
1354 		j = 2;
1355 		for (i = 0; i < nseg; i++) {
1356 			if (data_sg == NULL) {
1357 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1358 						"6059 dptr err %d, nseg %d\n",
1359 						i, nseg);
1360 				lpfc_ncmd->seg_cnt = 0;
1361 				return 1;
1362 			}
1363 
1364 			sgl->word2 = 0;
1365 			if (nseg == 1) {
1366 				bf_set(lpfc_sli4_sge_last, sgl, 1);
1367 				bf_set(lpfc_sli4_sge_type, sgl,
1368 				       LPFC_SGE_TYPE_DATA);
1369 			} else {
1370 				bf_set(lpfc_sli4_sge_last, sgl, 0);
1371 
1372 				/* expand the segment */
1373 				if (!lsp_just_set &&
1374 				    !((j + 1) % phba->border_sge_num) &&
1375 				    ((nseg - 1) != i)) {
1376 					/* set LSP type */
1377 					bf_set(lpfc_sli4_sge_type, sgl,
1378 					       LPFC_SGE_TYPE_LSP);
1379 
1380 					sgl_xtra = lpfc_get_sgl_per_hdwq(
1381 							phba, lpfc_ncmd);
1382 
1383 					if (unlikely(!sgl_xtra)) {
1384 						lpfc_ncmd->seg_cnt = 0;
1385 						return 1;
1386 					}
1387 					sgl->addr_lo = cpu_to_le32(putPaddrLow(
1388 						       sgl_xtra->dma_phys_sgl));
1389 					sgl->addr_hi = cpu_to_le32(putPaddrHigh(
1390 						       sgl_xtra->dma_phys_sgl));
1391 
1392 				} else {
1393 					bf_set(lpfc_sli4_sge_type, sgl,
1394 					       LPFC_SGE_TYPE_DATA);
1395 				}
1396 			}
1397 
1398 			if (!(bf_get(lpfc_sli4_sge_type, sgl) &
1399 				     LPFC_SGE_TYPE_LSP)) {
1400 				if ((nseg - 1) == i)
1401 					bf_set(lpfc_sli4_sge_last, sgl, 1);
1402 
1403 				physaddr = data_sg->dma_address;
1404 				dma_len = data_sg->length;
1405 				sgl->addr_lo = cpu_to_le32(
1406 							 putPaddrLow(physaddr));
1407 				sgl->addr_hi = cpu_to_le32(
1408 							putPaddrHigh(physaddr));
1409 
1410 				bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1411 				sgl->word2 = cpu_to_le32(sgl->word2);
1412 				sgl->sge_len = cpu_to_le32(dma_len);
1413 
1414 				dma_offset += dma_len;
1415 				data_sg = sg_next(data_sg);
1416 
1417 				sgl++;
1418 
1419 				lsp_just_set = false;
1420 			} else {
1421 				sgl->word2 = cpu_to_le32(sgl->word2);
1422 
1423 				sgl->sge_len = cpu_to_le32(
1424 						     phba->cfg_sg_dma_buf_size);
1425 
1426 				sgl = (struct sli4_sge *)sgl_xtra->dma_sgl;
1427 				i = i - 1;
1428 
1429 				lsp_just_set = true;
1430 			}
1431 
1432 			j++;
1433 		}
1434 
1435 		/* PBDE support for first data SGE only */
1436 		if (nseg == 1 && phba->cfg_enable_pbde) {
1437 			/* Words 13-15 */
1438 			bde = (struct ulp_bde64 *)
1439 				&wqe->words[13];
1440 			bde->addrLow = first_data_sgl->addr_lo;
1441 			bde->addrHigh = first_data_sgl->addr_hi;
1442 			bde->tus.f.bdeSize =
1443 				le32_to_cpu(first_data_sgl->sge_len);
1444 			bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1445 			bde->tus.w = cpu_to_le32(bde->tus.w);
1446 
1447 			/* Word 11 - set PBDE bit */
1448 			bf_set(wqe_pbde, &wqe->generic.wqe_com, 1);
1449 		} else {
1450 			memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1451 			/* Word 11 - PBDE bit disabled by default template */
1452 		}
1453 
1454 	} else {
1455 		lpfc_ncmd->seg_cnt = 0;
1456 
1457 		/* For this clause to be valid, the payload_length
1458 		 * and sg_cnt must zero.
1459 		 */
1460 		if (nCmd->payload_length != 0) {
1461 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1462 					"6063 NVME DMA Prep Err: sg_cnt %d "
1463 					"payload_length x%x\n",
1464 					nCmd->sg_cnt, nCmd->payload_length);
1465 			return 1;
1466 		}
1467 	}
1468 	return 0;
1469 }
1470 
1471 /**
1472  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1473  * @pnvme_lport: Pointer to the driver's local port data
1474  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1475  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1476  * @pnvme_fcreq: IO request from nvme fc to driver.
1477  *
1478  * Driver registers this routine as it io request handler.  This
1479  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1480  * data structure to the rport indicated in @lpfc_nvme_rport.
1481  *
1482  * Return value :
1483  *   0 - Success
1484  *   TODO: What are the failure codes.
1485  **/
1486 static int
1487 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1488 			struct nvme_fc_remote_port *pnvme_rport,
1489 			void *hw_queue_handle,
1490 			struct nvmefc_fcp_req *pnvme_fcreq)
1491 {
1492 	int ret = 0;
1493 	int expedite = 0;
1494 	int idx, cpu;
1495 	struct lpfc_nvme_lport *lport;
1496 	struct lpfc_fc4_ctrl_stat *cstat;
1497 	struct lpfc_vport *vport;
1498 	struct lpfc_hba *phba;
1499 	struct lpfc_nodelist *ndlp;
1500 	struct lpfc_io_buf *lpfc_ncmd;
1501 	struct lpfc_nvme_rport *rport;
1502 	struct lpfc_nvme_qhandle *lpfc_queue_info;
1503 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1504 	struct nvme_common_command *sqe;
1505 	uint64_t start = 0;
1506 
1507 	/* Validate pointers. LLDD fault handling with transport does
1508 	 * have timing races.
1509 	 */
1510 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1511 	if (unlikely(!lport)) {
1512 		ret = -EINVAL;
1513 		goto out_fail;
1514 	}
1515 
1516 	vport = lport->vport;
1517 
1518 	if (unlikely(!hw_queue_handle)) {
1519 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1520 				 "6117 Fail IO, NULL hw_queue_handle\n");
1521 		atomic_inc(&lport->xmt_fcp_err);
1522 		ret = -EBUSY;
1523 		goto out_fail;
1524 	}
1525 
1526 	phba = vport->phba;
1527 
1528 	if ((unlikely(vport->load_flag & FC_UNLOADING)) ||
1529 	    phba->hba_flag & HBA_IOQ_FLUSH) {
1530 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1531 				 "6124 Fail IO, Driver unload\n");
1532 		atomic_inc(&lport->xmt_fcp_err);
1533 		ret = -ENODEV;
1534 		goto out_fail;
1535 	}
1536 
1537 	freqpriv = pnvme_fcreq->private;
1538 	if (unlikely(!freqpriv)) {
1539 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1540 				 "6158 Fail IO, NULL request data\n");
1541 		atomic_inc(&lport->xmt_fcp_err);
1542 		ret = -EINVAL;
1543 		goto out_fail;
1544 	}
1545 
1546 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1547 	if (phba->ktime_on)
1548 		start = ktime_get_ns();
1549 #endif
1550 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1551 	lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1552 
1553 	/*
1554 	 * Catch race where our node has transitioned, but the
1555 	 * transport is still transitioning.
1556 	 */
1557 	ndlp = rport->ndlp;
1558 	if (!ndlp) {
1559 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1560 				 "6053 Busy IO, ndlp not ready: rport x%px "
1561 				  "ndlp x%px, DID x%06x\n",
1562 				 rport, ndlp, pnvme_rport->port_id);
1563 		atomic_inc(&lport->xmt_fcp_err);
1564 		ret = -EBUSY;
1565 		goto out_fail;
1566 	}
1567 
1568 	/* The remote node has to be a mapped target or it's an error. */
1569 	if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1570 	    (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1571 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1572 				 "6036 Fail IO, DID x%06x not ready for "
1573 				 "IO. State x%x, Type x%x Flg x%x\n",
1574 				 pnvme_rport->port_id,
1575 				 ndlp->nlp_state, ndlp->nlp_type,
1576 				 ndlp->fc4_xpt_flags);
1577 		atomic_inc(&lport->xmt_fcp_bad_ndlp);
1578 		ret = -EBUSY;
1579 		goto out_fail;
1580 
1581 	}
1582 
1583 	/* Currently only NVME Keep alive commands should be expedited
1584 	 * if the driver runs out of a resource. These should only be
1585 	 * issued on the admin queue, qidx 0
1586 	 */
1587 	if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1588 		sqe = &((struct nvme_fc_cmd_iu *)
1589 			pnvme_fcreq->cmdaddr)->sqe.common;
1590 		if (sqe->opcode == nvme_admin_keep_alive)
1591 			expedite = 1;
1592 	}
1593 
1594 	/* Check if IO qualifies for CMF */
1595 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1596 	    pnvme_fcreq->io_dir == NVMEFC_FCP_READ &&
1597 	    pnvme_fcreq->payload_length) {
1598 		ret = lpfc_update_cmf_cmd(phba, pnvme_fcreq->payload_length);
1599 		if (ret) {
1600 			ret = -EBUSY;
1601 			goto out_fail;
1602 		}
1603 		/* Get start time for IO latency */
1604 		start = ktime_get_ns();
1605 	}
1606 
1607 	/* The node is shared with FCP IO, make sure the IO pending count does
1608 	 * not exceed the programmed depth.
1609 	 */
1610 	if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1611 		if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1612 		    !expedite) {
1613 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1614 					 "6174 Fail IO, ndlp qdepth exceeded: "
1615 					 "idx %d DID %x pend %d qdepth %d\n",
1616 					 lpfc_queue_info->index, ndlp->nlp_DID,
1617 					 atomic_read(&ndlp->cmd_pending),
1618 					 ndlp->cmd_qdepth);
1619 			atomic_inc(&lport->xmt_fcp_qdepth);
1620 			ret = -EBUSY;
1621 			goto out_fail1;
1622 		}
1623 	}
1624 
1625 	/* Lookup Hardware Queue index based on fcp_io_sched module parameter */
1626 	if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) {
1627 		idx = lpfc_queue_info->index;
1628 	} else {
1629 		cpu = raw_smp_processor_id();
1630 		idx = phba->sli4_hba.cpu_map[cpu].hdwq;
1631 	}
1632 
1633 	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite);
1634 	if (lpfc_ncmd == NULL) {
1635 		atomic_inc(&lport->xmt_fcp_noxri);
1636 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1637 				 "6065 Fail IO, driver buffer pool is empty: "
1638 				 "idx %d DID %x\n",
1639 				 lpfc_queue_info->index, ndlp->nlp_DID);
1640 		ret = -EBUSY;
1641 		goto out_fail1;
1642 	}
1643 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1644 	if (start) {
1645 		lpfc_ncmd->ts_cmd_start = start;
1646 		lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1647 	} else {
1648 		lpfc_ncmd->ts_cmd_start = 0;
1649 	}
1650 #endif
1651 	lpfc_ncmd->rx_cmd_start = start;
1652 
1653 	/*
1654 	 * Store the data needed by the driver to issue, abort, and complete
1655 	 * an IO.
1656 	 * Do not let the IO hang out forever.  There is no midlayer issuing
1657 	 * an abort so inform the FW of the maximum IO pending time.
1658 	 */
1659 	freqpriv->nvme_buf = lpfc_ncmd;
1660 	lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1661 	lpfc_ncmd->ndlp = ndlp;
1662 	lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1663 
1664 	/*
1665 	 * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1666 	 * This identfier was create in our hardware queue create callback
1667 	 * routine. The driver now is dependent on the IO queue steering from
1668 	 * the transport.  We are trusting the upper NVME layers know which
1669 	 * index to use and that they have affinitized a CPU to this hardware
1670 	 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1671 	 */
1672 	lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1673 	cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat;
1674 
1675 	lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1676 	ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1677 	if (ret) {
1678 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1679 				 "6175 Fail IO, Prep DMA: "
1680 				 "idx %d DID %x\n",
1681 				 lpfc_queue_info->index, ndlp->nlp_DID);
1682 		atomic_inc(&lport->xmt_fcp_err);
1683 		ret = -ENOMEM;
1684 		goto out_free_nvme_buf;
1685 	}
1686 
1687 	lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1688 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
1689 			 lpfc_queue_info->index, ndlp->nlp_DID);
1690 
1691 	ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq);
1692 	if (ret) {
1693 		atomic_inc(&lport->xmt_fcp_wqerr);
1694 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1695 				 "6113 Fail IO, Could not issue WQE err %x "
1696 				 "sid: x%x did: x%x oxid: x%x\n",
1697 				 ret, vport->fc_myDID, ndlp->nlp_DID,
1698 				 lpfc_ncmd->cur_iocbq.sli4_xritag);
1699 		goto out_free_nvme_buf;
1700 	}
1701 
1702 	if (phba->cfg_xri_rebalancing)
1703 		lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no);
1704 
1705 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1706 	if (lpfc_ncmd->ts_cmd_start)
1707 		lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1708 
1709 	if (phba->hdwqstat_on & LPFC_CHECK_NVME_IO) {
1710 		cpu = raw_smp_processor_id();
1711 		this_cpu_inc(phba->sli4_hba.c_stat->xmt_io);
1712 		lpfc_ncmd->cpu = cpu;
1713 		if (idx != cpu)
1714 			lpfc_printf_vlog(vport,
1715 					 KERN_INFO, LOG_NVME_IOERR,
1716 					"6702 CPU Check cmd: "
1717 					"cpu %d wq %d\n",
1718 					lpfc_ncmd->cpu,
1719 					lpfc_queue_info->index);
1720 	}
1721 #endif
1722 	return 0;
1723 
1724  out_free_nvme_buf:
1725 	if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1726 		if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1727 			cstat->output_requests--;
1728 		else
1729 			cstat->input_requests--;
1730 	} else
1731 		cstat->control_requests--;
1732 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1733  out_fail1:
1734 	lpfc_update_cmf_cmpl(phba, LPFC_CGN_NOT_SENT,
1735 			     pnvme_fcreq->payload_length, NULL);
1736  out_fail:
1737 	return ret;
1738 }
1739 
1740 /**
1741  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1742  * @phba: Pointer to HBA context object
1743  * @cmdiocb: Pointer to command iocb object.
1744  * @abts_cmpl: Pointer to wcqe complete object.
1745  *
1746  * This is the callback function for any NVME FCP IO that was aborted.
1747  *
1748  * Return value:
1749  *   None
1750  **/
1751 void
1752 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1753 			   struct lpfc_wcqe_complete *abts_cmpl)
1754 {
1755 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1756 			"6145 ABORT_XRI_CN completing on rpi x%x "
1757 			"original iotag x%x, abort cmd iotag x%x "
1758 			"req_tag x%x, status x%x, hwstatus x%x\n",
1759 			bf_get(wqe_ctxt_tag, &cmdiocb->wqe.generic.wqe_com),
1760 			get_job_abtsiotag(phba, cmdiocb), cmdiocb->iotag,
1761 			bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1762 			bf_get(lpfc_wcqe_c_status, abts_cmpl),
1763 			bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1764 	lpfc_sli_release_iocbq(phba, cmdiocb);
1765 }
1766 
1767 /**
1768  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1769  * @pnvme_lport: Pointer to the driver's local port data
1770  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1771  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1772  * @pnvme_fcreq: IO request from nvme fc to driver.
1773  *
1774  * Driver registers this routine as its nvme request io abort handler.  This
1775  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1776  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1777  * is executed asynchronously - one the target is validated as "MAPPED" and
1778  * ready for IO, the driver issues the abort request and returns.
1779  *
1780  * Return value:
1781  *   None
1782  **/
1783 static void
1784 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1785 		    struct nvme_fc_remote_port *pnvme_rport,
1786 		    void *hw_queue_handle,
1787 		    struct nvmefc_fcp_req *pnvme_fcreq)
1788 {
1789 	struct lpfc_nvme_lport *lport;
1790 	struct lpfc_vport *vport;
1791 	struct lpfc_hba *phba;
1792 	struct lpfc_io_buf *lpfc_nbuf;
1793 	struct lpfc_iocbq *nvmereq_wqe;
1794 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1795 	unsigned long flags;
1796 	int ret_val;
1797 
1798 	/* Validate pointers. LLDD fault handling with transport does
1799 	 * have timing races.
1800 	 */
1801 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1802 	if (unlikely(!lport))
1803 		return;
1804 
1805 	vport = lport->vport;
1806 
1807 	if (unlikely(!hw_queue_handle)) {
1808 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1809 				 "6129 Fail Abort, HW Queue Handle NULL.\n");
1810 		return;
1811 	}
1812 
1813 	phba = vport->phba;
1814 	freqpriv = pnvme_fcreq->private;
1815 
1816 	if (unlikely(!freqpriv))
1817 		return;
1818 	if (vport->load_flag & FC_UNLOADING)
1819 		return;
1820 
1821 	/* Announce entry to new IO submit field. */
1822 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1823 			 "6002 Abort Request to rport DID x%06x "
1824 			 "for nvme_fc_req x%px\n",
1825 			 pnvme_rport->port_id,
1826 			 pnvme_fcreq);
1827 
1828 	/* If the hba is getting reset, this flag is set.  It is
1829 	 * cleared when the reset is complete and rings reestablished.
1830 	 */
1831 	spin_lock_irqsave(&phba->hbalock, flags);
1832 	/* driver queued commands are in process of being flushed */
1833 	if (phba->hba_flag & HBA_IOQ_FLUSH) {
1834 		spin_unlock_irqrestore(&phba->hbalock, flags);
1835 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1836 				 "6139 Driver in reset cleanup - flushing "
1837 				 "NVME Req now.  hba_flag x%x\n",
1838 				 phba->hba_flag);
1839 		return;
1840 	}
1841 
1842 	lpfc_nbuf = freqpriv->nvme_buf;
1843 	if (!lpfc_nbuf) {
1844 		spin_unlock_irqrestore(&phba->hbalock, flags);
1845 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1846 				 "6140 NVME IO req has no matching lpfc nvme "
1847 				 "io buffer.  Skipping abort req.\n");
1848 		return;
1849 	} else if (!lpfc_nbuf->nvmeCmd) {
1850 		spin_unlock_irqrestore(&phba->hbalock, flags);
1851 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1852 				 "6141 lpfc NVME IO req has no nvme_fcreq "
1853 				 "io buffer.  Skipping abort req.\n");
1854 		return;
1855 	}
1856 	nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1857 
1858 	/* Guard against IO completion being called at same time */
1859 	spin_lock(&lpfc_nbuf->buf_lock);
1860 
1861 	/*
1862 	 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1863 	 * state must match the nvme_fcreq passed by the nvme
1864 	 * transport.  If they don't match, it is likely the driver
1865 	 * has already completed the NVME IO and the nvme transport
1866 	 * has not seen it yet.
1867 	 */
1868 	if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1869 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1870 				 "6143 NVME req mismatch: "
1871 				 "lpfc_nbuf x%px nvmeCmd x%px, "
1872 				 "pnvme_fcreq x%px.  Skipping Abort xri x%x\n",
1873 				 lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1874 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1875 		goto out_unlock;
1876 	}
1877 
1878 	/* Don't abort IOs no longer on the pending queue. */
1879 	if (!(nvmereq_wqe->cmd_flag & LPFC_IO_ON_TXCMPLQ)) {
1880 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1881 				 "6142 NVME IO req x%px not queued - skipping "
1882 				 "abort req xri x%x\n",
1883 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1884 		goto out_unlock;
1885 	}
1886 
1887 	atomic_inc(&lport->xmt_fcp_abort);
1888 	lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1889 			 nvmereq_wqe->sli4_xritag,
1890 			 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1891 
1892 	/* Outstanding abort is in progress */
1893 	if (nvmereq_wqe->cmd_flag & LPFC_DRIVER_ABORTED) {
1894 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1895 				 "6144 Outstanding NVME I/O Abort Request "
1896 				 "still pending on nvme_fcreq x%px, "
1897 				 "lpfc_ncmd x%px xri x%x\n",
1898 				 pnvme_fcreq, lpfc_nbuf,
1899 				 nvmereq_wqe->sli4_xritag);
1900 		goto out_unlock;
1901 	}
1902 
1903 	ret_val = lpfc_sli4_issue_abort_iotag(phba, nvmereq_wqe,
1904 					      lpfc_nvme_abort_fcreq_cmpl);
1905 
1906 	spin_unlock(&lpfc_nbuf->buf_lock);
1907 	spin_unlock_irqrestore(&phba->hbalock, flags);
1908 
1909 	/* Make sure HBA is alive */
1910 	lpfc_issue_hb_tmo(phba);
1911 
1912 	if (ret_val != WQE_SUCCESS) {
1913 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1914 				 "6137 Failed abts issue_wqe with status x%x "
1915 				 "for nvme_fcreq x%px.\n",
1916 				 ret_val, pnvme_fcreq);
1917 		return;
1918 	}
1919 
1920 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1921 			 "6138 Transport Abort NVME Request Issued for "
1922 			 "ox_id x%x\n",
1923 			 nvmereq_wqe->sli4_xritag);
1924 	return;
1925 
1926 out_unlock:
1927 	spin_unlock(&lpfc_nbuf->buf_lock);
1928 	spin_unlock_irqrestore(&phba->hbalock, flags);
1929 	return;
1930 }
1931 
1932 /* Declare and initialization an instance of the FC NVME template. */
1933 static struct nvme_fc_port_template lpfc_nvme_template = {
1934 	/* initiator-based functions */
1935 	.localport_delete  = lpfc_nvme_localport_delete,
1936 	.remoteport_delete = lpfc_nvme_remoteport_delete,
1937 	.create_queue = lpfc_nvme_create_queue,
1938 	.delete_queue = lpfc_nvme_delete_queue,
1939 	.ls_req       = lpfc_nvme_ls_req,
1940 	.fcp_io       = lpfc_nvme_fcp_io_submit,
1941 	.ls_abort     = lpfc_nvme_ls_abort,
1942 	.fcp_abort    = lpfc_nvme_fcp_abort,
1943 	.xmt_ls_rsp   = lpfc_nvme_xmt_ls_rsp,
1944 
1945 	.max_hw_queues = 1,
1946 	.max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1947 	.max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1948 	.dma_boundary = 0xFFFFFFFF,
1949 
1950 	/* Sizes of additional private data for data structures.
1951 	 * No use for the last two sizes at this time.
1952 	 */
1953 	.local_priv_sz = sizeof(struct lpfc_nvme_lport),
1954 	.remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1955 	.lsrqst_priv_sz = 0,
1956 	.fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1957 };
1958 
1959 /*
1960  * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
1961  *
1962  * This routine removes a nvme buffer from head of @hdwq io_buf_list
1963  * and returns to caller.
1964  *
1965  * Return codes:
1966  *   NULL - Error
1967  *   Pointer to lpfc_nvme_buf - Success
1968  **/
1969 static struct lpfc_io_buf *
1970 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
1971 		  int idx, int expedite)
1972 {
1973 	struct lpfc_io_buf *lpfc_ncmd;
1974 	struct lpfc_sli4_hdw_queue *qp;
1975 	struct sli4_sge *sgl;
1976 	struct lpfc_iocbq *pwqeq;
1977 	union lpfc_wqe128 *wqe;
1978 
1979 	lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite);
1980 
1981 	if (lpfc_ncmd) {
1982 		pwqeq = &(lpfc_ncmd->cur_iocbq);
1983 		wqe = &pwqeq->wqe;
1984 
1985 		/* Setup key fields in buffer that may have been changed
1986 		 * if other protocols used this buffer.
1987 		 */
1988 		pwqeq->cmd_flag = LPFC_IO_NVME;
1989 		pwqeq->cmd_cmpl = lpfc_nvme_io_cmd_cmpl;
1990 		lpfc_ncmd->start_time = jiffies;
1991 		lpfc_ncmd->flags = 0;
1992 
1993 		/* Rsp SGE will be filled in when we rcv an IO
1994 		 * from the NVME Layer to be sent.
1995 		 * The cmd is going to be embedded so we need a SKIP SGE.
1996 		 */
1997 		sgl = lpfc_ncmd->dma_sgl;
1998 		bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1999 		bf_set(lpfc_sli4_sge_last, sgl, 0);
2000 		sgl->word2 = cpu_to_le32(sgl->word2);
2001 		/* Fill in word 3 / sgl_len during cmd submission */
2002 
2003 		/* Initialize 64 bytes only */
2004 		memset(wqe, 0, sizeof(union lpfc_wqe));
2005 
2006 		if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
2007 			atomic_inc(&ndlp->cmd_pending);
2008 			lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH;
2009 		}
2010 
2011 	} else {
2012 		qp = &phba->sli4_hba.hdwq[idx];
2013 		qp->empty_io_bufs++;
2014 	}
2015 
2016 	return  lpfc_ncmd;
2017 }
2018 
2019 /**
2020  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2021  * @phba: The Hba for which this call is being executed.
2022  * @lpfc_ncmd: The nvme buffer which is being released.
2023  *
2024  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2025  * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2026  * and cannot be reused for at least RA_TOV amount of time if it was
2027  * aborted.
2028  **/
2029 static void
2030 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd)
2031 {
2032 	struct lpfc_sli4_hdw_queue *qp;
2033 	unsigned long iflag = 0;
2034 
2035 	if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2036 		atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2037 
2038 	lpfc_ncmd->ndlp = NULL;
2039 	lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH;
2040 
2041 	qp = lpfc_ncmd->hdwq;
2042 	if (unlikely(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
2043 		lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2044 				"6310 XB release deferred for "
2045 				"ox_id x%x on reqtag x%x\n",
2046 				lpfc_ncmd->cur_iocbq.sli4_xritag,
2047 				lpfc_ncmd->cur_iocbq.iotag);
2048 
2049 		spin_lock_irqsave(&qp->abts_io_buf_list_lock, iflag);
2050 		list_add_tail(&lpfc_ncmd->list,
2051 			&qp->lpfc_abts_io_buf_list);
2052 		qp->abts_nvme_io_bufs++;
2053 		spin_unlock_irqrestore(&qp->abts_io_buf_list_lock, iflag);
2054 	} else
2055 		lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp);
2056 }
2057 
2058 /**
2059  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2060  * @vport: the lpfc_vport instance requesting a localport.
2061  *
2062  * This routine is invoked to create an nvme localport instance to bind
2063  * to the nvme_fc_transport.  It is called once during driver load
2064  * like lpfc_create_shost after all other services are initialized.
2065  * It requires a vport, vpi, and wwns at call time.  Other localport
2066  * parameters are modified as the driver's FCID and the Fabric WWN
2067  * are established.
2068  *
2069  * Return codes
2070  *      0 - successful
2071  *      -ENOMEM - no heap memory available
2072  *      other values - from nvme registration upcall
2073  **/
2074 int
2075 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2076 {
2077 	int ret = 0;
2078 	struct lpfc_hba  *phba = vport->phba;
2079 	struct nvme_fc_port_info nfcp_info;
2080 	struct nvme_fc_local_port *localport;
2081 	struct lpfc_nvme_lport *lport;
2082 
2083 	/* Initialize this localport instance.  The vport wwn usage ensures
2084 	 * that NPIV is accounted for.
2085 	 */
2086 	memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2087 	nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2088 	nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2089 	nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2090 
2091 	/* We need to tell the transport layer + 1 because it takes page
2092 	 * alignment into account. When space for the SGL is allocated we
2093 	 * allocate + 3, one for cmd, one for rsp and one for this alignment
2094 	 */
2095 	lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2096 
2097 	/* Advertise how many hw queues we support based on cfg_hdw_queue,
2098 	 * which will not exceed cpu count.
2099 	 */
2100 	lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2101 
2102 	if (!IS_ENABLED(CONFIG_NVME_FC))
2103 		return ret;
2104 
2105 	/* localport is allocated from the stack, but the registration
2106 	 * call allocates heap memory as well as the private area.
2107 	 */
2108 
2109 	ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2110 					 &vport->phba->pcidev->dev, &localport);
2111 	if (!ret) {
2112 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2113 				 "6005 Successfully registered local "
2114 				 "NVME port num %d, localP x%px, private "
2115 				 "x%px, sg_seg %d\n",
2116 				 localport->port_num, localport,
2117 				 localport->private,
2118 				 lpfc_nvme_template.max_sgl_segments);
2119 
2120 		/* Private is our lport size declared in the template. */
2121 		lport = (struct lpfc_nvme_lport *)localport->private;
2122 		vport->localport = localport;
2123 		lport->vport = vport;
2124 		vport->nvmei_support = 1;
2125 
2126 		atomic_set(&lport->xmt_fcp_noxri, 0);
2127 		atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2128 		atomic_set(&lport->xmt_fcp_qdepth, 0);
2129 		atomic_set(&lport->xmt_fcp_err, 0);
2130 		atomic_set(&lport->xmt_fcp_wqerr, 0);
2131 		atomic_set(&lport->xmt_fcp_abort, 0);
2132 		atomic_set(&lport->xmt_ls_abort, 0);
2133 		atomic_set(&lport->xmt_ls_err, 0);
2134 		atomic_set(&lport->cmpl_fcp_xb, 0);
2135 		atomic_set(&lport->cmpl_fcp_err, 0);
2136 		atomic_set(&lport->cmpl_ls_xb, 0);
2137 		atomic_set(&lport->cmpl_ls_err, 0);
2138 
2139 		atomic_set(&lport->fc4NvmeLsRequests, 0);
2140 		atomic_set(&lport->fc4NvmeLsCmpls, 0);
2141 	}
2142 
2143 	return ret;
2144 }
2145 
2146 #if (IS_ENABLED(CONFIG_NVME_FC))
2147 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2148  *
2149  * The driver has to wait for the host nvme transport to callback
2150  * indicating the localport has successfully unregistered all
2151  * resources.  Since this is an uninterruptible wait, loop every ten
2152  * seconds and print a message indicating no progress.
2153  *
2154  * An uninterruptible wait is used because of the risk of transport-to-
2155  * driver state mismatch.
2156  */
2157 static void
2158 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2159 			   struct lpfc_nvme_lport *lport,
2160 			   struct completion *lport_unreg_cmp)
2161 {
2162 	u32 wait_tmo;
2163 	int ret, i, pending = 0;
2164 	struct lpfc_sli_ring  *pring;
2165 	struct lpfc_hba  *phba = vport->phba;
2166 	struct lpfc_sli4_hdw_queue *qp;
2167 	int abts_scsi, abts_nvme;
2168 
2169 	/* Host transport has to clean up and confirm requiring an indefinite
2170 	 * wait. Print a message if a 10 second wait expires and renew the
2171 	 * wait. This is unexpected.
2172 	 */
2173 	wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2174 	while (true) {
2175 		ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2176 		if (unlikely(!ret)) {
2177 			pending = 0;
2178 			abts_scsi = 0;
2179 			abts_nvme = 0;
2180 			for (i = 0; i < phba->cfg_hdw_queue; i++) {
2181 				qp = &phba->sli4_hba.hdwq[i];
2182 				if (!vport->localport || !qp || !qp->io_wq)
2183 					return;
2184 
2185 				pring = qp->io_wq->pring;
2186 				if (!pring)
2187 					continue;
2188 				pending += pring->txcmplq_cnt;
2189 				abts_scsi += qp->abts_scsi_io_bufs;
2190 				abts_nvme += qp->abts_nvme_io_bufs;
2191 			}
2192 			if (!vport->localport ||
2193 			    test_bit(HBA_PCI_ERR, &vport->phba->bit_flags) ||
2194 			    vport->load_flag & FC_UNLOADING)
2195 				return;
2196 
2197 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2198 					 "6176 Lport x%px Localport x%px wait "
2199 					 "timed out. Pending %d [%d:%d]. "
2200 					 "Renewing.\n",
2201 					 lport, vport->localport, pending,
2202 					 abts_scsi, abts_nvme);
2203 			continue;
2204 		}
2205 		break;
2206 	}
2207 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2208 			 "6177 Lport x%px Localport x%px Complete Success\n",
2209 			 lport, vport->localport);
2210 }
2211 #endif
2212 
2213 /**
2214  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2215  * @vport: pointer to a host virtual N_Port data structure
2216  *
2217  * This routine is invoked to destroy all lports bound to the phba.
2218  * The lport memory was allocated by the nvme fc transport and is
2219  * released there.  This routine ensures all rports bound to the
2220  * lport have been disconnected.
2221  *
2222  **/
2223 void
2224 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2225 {
2226 #if (IS_ENABLED(CONFIG_NVME_FC))
2227 	struct nvme_fc_local_port *localport;
2228 	struct lpfc_nvme_lport *lport;
2229 	int ret;
2230 	DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2231 
2232 	if (vport->nvmei_support == 0)
2233 		return;
2234 
2235 	localport = vport->localport;
2236 	if (!localport)
2237 		return;
2238 	lport = (struct lpfc_nvme_lport *)localport->private;
2239 
2240 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2241 			 "6011 Destroying NVME localport x%px\n",
2242 			 localport);
2243 
2244 	/* lport's rport list is clear.  Unregister
2245 	 * lport and release resources.
2246 	 */
2247 	lport->lport_unreg_cmp = &lport_unreg_cmp;
2248 	ret = nvme_fc_unregister_localport(localport);
2249 
2250 	/* Wait for completion.  This either blocks
2251 	 * indefinitely or succeeds
2252 	 */
2253 	lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2254 	vport->localport = NULL;
2255 
2256 	/* Regardless of the unregister upcall response, clear
2257 	 * nvmei_support.  All rports are unregistered and the
2258 	 * driver will clean up.
2259 	 */
2260 	vport->nvmei_support = 0;
2261 	if (ret == 0) {
2262 		lpfc_printf_vlog(vport,
2263 				 KERN_INFO, LOG_NVME_DISC,
2264 				 "6009 Unregistered lport Success\n");
2265 	} else {
2266 		lpfc_printf_vlog(vport,
2267 				 KERN_INFO, LOG_NVME_DISC,
2268 				 "6010 Unregistered lport "
2269 				 "Failed, status x%x\n",
2270 				 ret);
2271 	}
2272 #endif
2273 }
2274 
2275 void
2276 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2277 {
2278 #if (IS_ENABLED(CONFIG_NVME_FC))
2279 	struct nvme_fc_local_port *localport;
2280 	struct lpfc_nvme_lport *lport;
2281 
2282 	localport = vport->localport;
2283 	if (!localport) {
2284 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2285 				 "6710 Update NVME fail. No localport\n");
2286 		return;
2287 	}
2288 	lport = (struct lpfc_nvme_lport *)localport->private;
2289 	if (!lport) {
2290 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2291 				 "6171 Update NVME fail. localP x%px, No lport\n",
2292 				 localport);
2293 		return;
2294 	}
2295 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2296 			 "6012 Update NVME lport x%px did x%x\n",
2297 			 localport, vport->fc_myDID);
2298 
2299 	localport->port_id = vport->fc_myDID;
2300 	if (localport->port_id == 0)
2301 		localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2302 	else
2303 		localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2304 
2305 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2306 			 "6030 bound lport x%px to DID x%06x\n",
2307 			 lport, localport->port_id);
2308 #endif
2309 }
2310 
2311 int
2312 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2313 {
2314 #if (IS_ENABLED(CONFIG_NVME_FC))
2315 	int ret = 0;
2316 	struct nvme_fc_local_port *localport;
2317 	struct lpfc_nvme_lport *lport;
2318 	struct lpfc_nvme_rport *rport;
2319 	struct lpfc_nvme_rport *oldrport;
2320 	struct nvme_fc_remote_port *remote_port;
2321 	struct nvme_fc_port_info rpinfo;
2322 	struct lpfc_nodelist *prev_ndlp = NULL;
2323 	struct fc_rport *srport = ndlp->rport;
2324 
2325 	lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2326 			 "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2327 			 ndlp->nlp_DID, ndlp->nlp_type);
2328 
2329 	localport = vport->localport;
2330 	if (!localport)
2331 		return 0;
2332 
2333 	lport = (struct lpfc_nvme_lport *)localport->private;
2334 
2335 	/* NVME rports are not preserved across devloss.
2336 	 * Just register this instance.  Note, rpinfo->dev_loss_tmo
2337 	 * is left 0 to indicate accept transport defaults.  The
2338 	 * driver communicates port role capabilities consistent
2339 	 * with the PRLI response data.
2340 	 */
2341 	memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2342 	rpinfo.port_id = ndlp->nlp_DID;
2343 	if (ndlp->nlp_type & NLP_NVME_TARGET)
2344 		rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2345 	if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2346 		rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2347 
2348 	if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2349 		rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2350 
2351 	rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2352 	rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2353 	if (srport)
2354 		rpinfo.dev_loss_tmo = srport->dev_loss_tmo;
2355 	else
2356 		rpinfo.dev_loss_tmo = vport->cfg_devloss_tmo;
2357 
2358 	spin_lock_irq(&ndlp->lock);
2359 	oldrport = lpfc_ndlp_get_nrport(ndlp);
2360 	if (oldrport) {
2361 		prev_ndlp = oldrport->ndlp;
2362 		spin_unlock_irq(&ndlp->lock);
2363 	} else {
2364 		spin_unlock_irq(&ndlp->lock);
2365 		if (!lpfc_nlp_get(ndlp)) {
2366 			dev_warn(&vport->phba->pcidev->dev,
2367 				 "Warning - No node ref - exit register\n");
2368 			return 0;
2369 		}
2370 	}
2371 
2372 	ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2373 	if (!ret) {
2374 		/* If the ndlp already has an nrport, this is just
2375 		 * a resume of the existing rport.  Else this is a
2376 		 * new rport.
2377 		 */
2378 		/* Guard against an unregister/reregister
2379 		 * race that leaves the WAIT flag set.
2380 		 */
2381 		spin_lock_irq(&ndlp->lock);
2382 		ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2383 		ndlp->fc4_xpt_flags |= NVME_XPT_REGD;
2384 		spin_unlock_irq(&ndlp->lock);
2385 		rport = remote_port->private;
2386 		if (oldrport) {
2387 
2388 			/* Sever the ndlp<->rport association
2389 			 * before dropping the ndlp ref from
2390 			 * register.
2391 			 */
2392 			spin_lock_irq(&ndlp->lock);
2393 			ndlp->nrport = NULL;
2394 			ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2395 			spin_unlock_irq(&ndlp->lock);
2396 			rport->ndlp = NULL;
2397 			rport->remoteport = NULL;
2398 
2399 			/* Reference only removed if previous NDLP is no longer
2400 			 * active. It might be just a swap and removing the
2401 			 * reference would cause a premature cleanup.
2402 			 */
2403 			if (prev_ndlp && prev_ndlp != ndlp) {
2404 				if (!prev_ndlp->nrport)
2405 					lpfc_nlp_put(prev_ndlp);
2406 			}
2407 		}
2408 
2409 		/* Clean bind the rport to the ndlp. */
2410 		rport->remoteport = remote_port;
2411 		rport->lport = lport;
2412 		rport->ndlp = ndlp;
2413 		spin_lock_irq(&ndlp->lock);
2414 		ndlp->nrport = rport;
2415 		spin_unlock_irq(&ndlp->lock);
2416 		lpfc_printf_vlog(vport, KERN_INFO,
2417 				 LOG_NVME_DISC | LOG_NODE,
2418 				 "6022 Bind lport x%px to remoteport x%px "
2419 				 "rport x%px WWNN 0x%llx, "
2420 				 "Rport WWPN 0x%llx DID "
2421 				 "x%06x Role x%x, ndlp %p prev_ndlp x%px\n",
2422 				 lport, remote_port, rport,
2423 				 rpinfo.node_name, rpinfo.port_name,
2424 				 rpinfo.port_id, rpinfo.port_role,
2425 				 ndlp, prev_ndlp);
2426 	} else {
2427 		lpfc_printf_vlog(vport, KERN_ERR,
2428 				 LOG_TRACE_EVENT,
2429 				 "6031 RemotePort Registration failed "
2430 				 "err: %d, DID x%06x\n",
2431 				 ret, ndlp->nlp_DID);
2432 	}
2433 
2434 	return ret;
2435 #else
2436 	return 0;
2437 #endif
2438 }
2439 
2440 /*
2441  * lpfc_nvme_rescan_port - Check to see if we should rescan this remoteport
2442  *
2443  * If the ndlp represents an NVME Target, that we are logged into,
2444  * ping the NVME FC Transport layer to initiate a device rescan
2445  * on this remote NPort.
2446  */
2447 void
2448 lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2449 {
2450 #if (IS_ENABLED(CONFIG_NVME_FC))
2451 	struct lpfc_nvme_rport *nrport;
2452 	struct nvme_fc_remote_port *remoteport = NULL;
2453 
2454 	spin_lock_irq(&ndlp->lock);
2455 	nrport = lpfc_ndlp_get_nrport(ndlp);
2456 	if (nrport)
2457 		remoteport = nrport->remoteport;
2458 	spin_unlock_irq(&ndlp->lock);
2459 
2460 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2461 			 "6170 Rescan NPort DID x%06x type x%x "
2462 			 "state x%x nrport x%px remoteport x%px\n",
2463 			 ndlp->nlp_DID, ndlp->nlp_type, ndlp->nlp_state,
2464 			 nrport, remoteport);
2465 
2466 	if (!nrport || !remoteport)
2467 		goto rescan_exit;
2468 
2469 	/* Only rescan if we are an NVME target in the MAPPED state */
2470 	if (remoteport->port_role & FC_PORT_ROLE_NVME_DISCOVERY &&
2471 	    ndlp->nlp_state == NLP_STE_MAPPED_NODE) {
2472 		nvme_fc_rescan_remoteport(remoteport);
2473 
2474 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2475 				 "6172 NVME rescanned DID x%06x "
2476 				 "port_state x%x\n",
2477 				 ndlp->nlp_DID, remoteport->port_state);
2478 	}
2479 	return;
2480  rescan_exit:
2481 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2482 			 "6169 Skip NVME Rport Rescan, NVME remoteport "
2483 			 "unregistered\n");
2484 #endif
2485 }
2486 
2487 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2488  *
2489  * There is no notion of Devloss or rport recovery from the current
2490  * nvme_transport perspective.  Loss of an rport just means IO cannot
2491  * be sent and recovery is completely up to the initator.
2492  * For now, the driver just unbinds the DID and port_role so that
2493  * no further IO can be issued.  Changes are planned for later.
2494  *
2495  * Notes - the ndlp reference count is not decremented here since
2496  * since there is no nvme_transport api for devloss.  Node ref count
2497  * is only adjusted in driver unload.
2498  */
2499 void
2500 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2501 {
2502 #if (IS_ENABLED(CONFIG_NVME_FC))
2503 	int ret;
2504 	struct nvme_fc_local_port *localport;
2505 	struct lpfc_nvme_lport *lport;
2506 	struct lpfc_nvme_rport *rport;
2507 	struct nvme_fc_remote_port *remoteport = NULL;
2508 
2509 	localport = vport->localport;
2510 
2511 	/* This is fundamental error.  The localport is always
2512 	 * available until driver unload.  Just exit.
2513 	 */
2514 	if (!localport)
2515 		return;
2516 
2517 	lport = (struct lpfc_nvme_lport *)localport->private;
2518 	if (!lport)
2519 		goto input_err;
2520 
2521 	spin_lock_irq(&ndlp->lock);
2522 	rport = lpfc_ndlp_get_nrport(ndlp);
2523 	if (rport)
2524 		remoteport = rport->remoteport;
2525 	spin_unlock_irq(&ndlp->lock);
2526 	if (!remoteport)
2527 		goto input_err;
2528 
2529 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2530 			 "6033 Unreg nvme remoteport x%px, portname x%llx, "
2531 			 "port_id x%06x, portstate x%x port type x%x "
2532 			 "refcnt %d\n",
2533 			 remoteport, remoteport->port_name,
2534 			 remoteport->port_id, remoteport->port_state,
2535 			 ndlp->nlp_type, kref_read(&ndlp->kref));
2536 
2537 	/* Sanity check ndlp type.  Only call for NVME ports. Don't
2538 	 * clear any rport state until the transport calls back.
2539 	 */
2540 
2541 	if (ndlp->nlp_type & NLP_NVME_TARGET) {
2542 		/* No concern about the role change on the nvme remoteport.
2543 		 * The transport will update it.
2544 		 */
2545 		spin_lock_irq(&vport->phba->hbalock);
2546 		ndlp->fc4_xpt_flags |= NVME_XPT_UNREG_WAIT;
2547 		spin_unlock_irq(&vport->phba->hbalock);
2548 
2549 		/* Don't let the host nvme transport keep sending keep-alives
2550 		 * on this remoteport. Vport is unloading, no recovery. The
2551 		 * return values is ignored.  The upcall is a courtesy to the
2552 		 * transport.
2553 		 */
2554 		if (vport->load_flag & FC_UNLOADING)
2555 			(void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2556 
2557 		ret = nvme_fc_unregister_remoteport(remoteport);
2558 
2559 		/* The driver no longer knows if the nrport memory is valid.
2560 		 * because the controller teardown process has begun and
2561 		 * is asynchronous.  Break the binding in the ndlp. Also
2562 		 * remove the register ndlp reference to setup node release.
2563 		 */
2564 		ndlp->nrport = NULL;
2565 		lpfc_nlp_put(ndlp);
2566 		if (ret != 0) {
2567 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2568 					 "6167 NVME unregister failed %d "
2569 					 "port_state x%x\n",
2570 					 ret, remoteport->port_state);
2571 		}
2572 	}
2573 	return;
2574 
2575  input_err:
2576 #endif
2577 	lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2578 			 "6168 State error: lport x%px, rport x%px FCID x%06x\n",
2579 			 vport->localport, ndlp->rport, ndlp->nlp_DID);
2580 }
2581 
2582 /**
2583  * lpfc_sli4_nvme_pci_offline_aborted - Fast-path process of NVME xri abort
2584  * @phba: pointer to lpfc hba data structure.
2585  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2586  *
2587  * This routine is invoked by the worker thread to process a SLI4 fast-path
2588  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2589  * here.
2590  **/
2591 void
2592 lpfc_sli4_nvme_pci_offline_aborted(struct lpfc_hba *phba,
2593 				   struct lpfc_io_buf *lpfc_ncmd)
2594 {
2595 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2596 
2597 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2598 			"6533 %s nvme_cmd %p tag x%x abort complete and "
2599 			"xri released\n", __func__,
2600 			lpfc_ncmd->nvmeCmd,
2601 			lpfc_ncmd->cur_iocbq.iotag);
2602 
2603 	/* Aborted NVME commands are required to not complete
2604 	 * before the abort exchange command fully completes.
2605 	 * Once completed, it is available via the put list.
2606 	 */
2607 	if (lpfc_ncmd->nvmeCmd) {
2608 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2609 		nvme_cmd->transferred_length = 0;
2610 		nvme_cmd->rcv_rsplen = 0;
2611 		nvme_cmd->status = NVME_SC_INTERNAL;
2612 		nvme_cmd->done(nvme_cmd);
2613 		lpfc_ncmd->nvmeCmd = NULL;
2614 	}
2615 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2616 }
2617 
2618 /**
2619  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2620  * @phba: pointer to lpfc hba data structure.
2621  * @axri: pointer to the fcp xri abort wcqe structure.
2622  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2623  *
2624  * This routine is invoked by the worker thread to process a SLI4 fast-path
2625  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2626  * here.
2627  **/
2628 void
2629 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2630 			   struct sli4_wcqe_xri_aborted *axri,
2631 			   struct lpfc_io_buf *lpfc_ncmd)
2632 {
2633 	uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2634 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2635 	struct lpfc_nodelist *ndlp = lpfc_ncmd->ndlp;
2636 
2637 
2638 	if (ndlp)
2639 		lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2640 
2641 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2642 			"6311 nvme_cmd %p xri x%x tag x%x abort complete and "
2643 			"xri released\n",
2644 			lpfc_ncmd->nvmeCmd, xri,
2645 			lpfc_ncmd->cur_iocbq.iotag);
2646 
2647 	/* Aborted NVME commands are required to not complete
2648 	 * before the abort exchange command fully completes.
2649 	 * Once completed, it is available via the put list.
2650 	 */
2651 	if (lpfc_ncmd->nvmeCmd) {
2652 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2653 		nvme_cmd->done(nvme_cmd);
2654 		lpfc_ncmd->nvmeCmd = NULL;
2655 	}
2656 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2657 }
2658 
2659 /**
2660  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2661  * @phba: Pointer to HBA context object.
2662  *
2663  * This function flushes all wqes in the nvme rings and frees all resources
2664  * in the txcmplq. This function does not issue abort wqes for the IO
2665  * commands in txcmplq, they will just be returned with
2666  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2667  * slot has been permanently disabled.
2668  **/
2669 void
2670 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2671 {
2672 	struct lpfc_sli_ring  *pring;
2673 	u32 i, wait_cnt = 0;
2674 
2675 	if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2676 		return;
2677 
2678 	/* Cycle through all IO rings and make sure all outstanding
2679 	 * WQEs have been removed from the txcmplqs.
2680 	 */
2681 	for (i = 0; i < phba->cfg_hdw_queue; i++) {
2682 		if (!phba->sli4_hba.hdwq[i].io_wq)
2683 			continue;
2684 		pring = phba->sli4_hba.hdwq[i].io_wq->pring;
2685 
2686 		if (!pring)
2687 			continue;
2688 
2689 		/* Retrieve everything on the txcmplq */
2690 		while (!list_empty(&pring->txcmplq)) {
2691 			msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2692 			wait_cnt++;
2693 
2694 			/* The sleep is 10mS.  Every ten seconds,
2695 			 * dump a message.  Something is wrong.
2696 			 */
2697 			if ((wait_cnt % 1000) == 0) {
2698 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
2699 						"6178 NVME IO not empty, "
2700 						"cnt %d\n", wait_cnt);
2701 			}
2702 		}
2703 	}
2704 
2705 	/* Make sure HBA is alive */
2706 	lpfc_issue_hb_tmo(phba);
2707 
2708 }
2709 
2710 void
2711 lpfc_nvme_cancel_iocb(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
2712 		      uint32_t stat, uint32_t param)
2713 {
2714 #if (IS_ENABLED(CONFIG_NVME_FC))
2715 	struct lpfc_io_buf *lpfc_ncmd;
2716 	struct nvmefc_fcp_req *nCmd;
2717 	struct lpfc_wcqe_complete wcqe;
2718 	struct lpfc_wcqe_complete *wcqep = &wcqe;
2719 
2720 	lpfc_ncmd = (struct lpfc_io_buf *)pwqeIn->context1;
2721 	if (!lpfc_ncmd) {
2722 		lpfc_sli_release_iocbq(phba, pwqeIn);
2723 		return;
2724 	}
2725 	/* For abort iocb just return, IO iocb will do a done call */
2726 	if (bf_get(wqe_cmnd, &pwqeIn->wqe.gen_req.wqe_com) ==
2727 	    CMD_ABORT_XRI_CX) {
2728 		lpfc_sli_release_iocbq(phba, pwqeIn);
2729 		return;
2730 	}
2731 
2732 	spin_lock(&lpfc_ncmd->buf_lock);
2733 	nCmd = lpfc_ncmd->nvmeCmd;
2734 	if (!nCmd) {
2735 		spin_unlock(&lpfc_ncmd->buf_lock);
2736 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
2737 		return;
2738 	}
2739 	spin_unlock(&lpfc_ncmd->buf_lock);
2740 
2741 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_IOERR,
2742 			"6194 NVME Cancel xri %x\n",
2743 			lpfc_ncmd->cur_iocbq.sli4_xritag);
2744 
2745 	wcqep->word0 = 0;
2746 	bf_set(lpfc_wcqe_c_status, wcqep, stat);
2747 	wcqep->parameter = param;
2748 	wcqep->word3 = 0; /* xb is 0 */
2749 
2750 	/* Call release with XB=1 to queue the IO into the abort list. */
2751 	if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
2752 		bf_set(lpfc_wcqe_c_xb, wcqep, 1);
2753 
2754 	memcpy(&pwqeIn->wcqe_cmpl, wcqep, sizeof(*wcqep));
2755 	(pwqeIn->cmd_cmpl)(phba, pwqeIn, pwqeIn);
2756 #endif
2757 }
2758